Internet Protocol (IP), operators and researchers are seeking solutions to meet this demand. One of the most promising approaches is Heterogeneous Cloud Radio Access Networks (HCRAN), which has the potential to address the issues of the current generation while offering improvements such as centralized processing and greater energy efficiency. However, the deployment of these new architectures faces two critical challenges: high energy consumption and the underutilization of PRRHs (Remote Radio Heads), caused by the tidal effect, and the processing capacity limitation of BBUs (BaseBand Units), known as Hard Capacity, which can negatively impact network performance when exceeded. This thesis proposes a multilayer approach that enables the reconfiguration of underutilized PRRHs and an optimized mapping between PRRHs and BBUs, aiming for high availability, energy savings, and reduced high-speed processing. The results obtained, compared to other approaches in the literature, demonstrate that the proposed model can optimize the resources of PRRHs and BBUs without compromising the user's minimum QoS.

The network slicing at the radio access network (RAN) domain, called RAN slicing, requires elasticity, efficient resource sharing, and customization in order to deal with scarce and limited frequency spectrum resources while fulfilling the slice intents defined in a service-level agreement (SLA). In this scenario, radio resource scheduling (RRS) is an essential function to provide the resource management needed to prevent SLA violations, hence providing sufficient radio resources for RAN slices to accomplish their goals. The wide variety of scenarios supported in 5G and beyond 5G (B5G) networks makes the RRS problem in RAN slicing scenarios a significant challenge. This thesis proposal provides an intent-based RRS for RAN slicing using reinforcement learning (RL) with the slice requirements defined in a SLA. The proposed method aims to prevent SLA violations by making the management of resource block groups (RBGs) available between slices and users equipment (UEs) using inter-slice and intra-slice schedulers, respectively. It also proposes a slice prioritization structure to assure SLA requirements of more important slices when the available radio resources are not sufficient to guarantee all slice’s requirements fulfillment. It presents partial results obtained using an intent-based RRS using RL for a fixed number of slices and UEs to prevent SLA violations, aiming to demonstrate the importance of intent-based RRS design for scenarios with RAN slicing. It also presents a discussion about the related work limitations and improvements in the implemented method, proposing a new approach to deal with slices and UEs insertions in the network while avoiding the SLA  iolations and improving the network performance.

O trabalho tem por objetivo analisar comparativamente as metodologias de definição dos conjuntos para estabelecimento de Metas de Qualidade do setor elétrico relativas aos indicadores de DEC e de FEC, avaliando se os índices e reflexos financeiros estão aderentes às realidades locais e propor o aprimoramento da metodologia vigente para definição dos limites de continuidade dos conjuntos elétricos das distribuidoras de energia elétrica no Brasil, através da análises de mais atributos que no momento não são adotados pela agencia reguladora, porém podem capturar particularidades principalmente das distribuidoras da região Norte e Nordeste, que não são capturadas na metodologia vigente.

Conforme previsto por especialistas, houve um crescimento exponencial do número de dispositivos IoT usados pela população em geral. Entre estes dispositivos, um dos grupos mais notáveis é o de dispositivos Long Range (LoRa). Entre as características da tecnologia LoRa destaca-se o baixo consumo de energia, ao mesmo tempo em que o dispositivo proporciona uma grande área de cobertura. Essa tecnologia, quando aplicada na região amazônica – uma área densamente arborizada, úmida e com populações afastadas dos centros urbanos – pode oferecer uma nova gama de serviços aos nativos da região. Partindo do princípio que os modelos de propagação mais conhecidos não estão ajustados às especificidades da região amazônica, o que torna  os enlaces de rádio de difíceis predições, visamos apresentar soluções para este problema através deste trabalho que apresenta ajustes no modelo de perda de propagação Terra-Plana para atender comunicações LoRa sob os rios amazônicos, quando os dispositivos receptores estão em movimento.

O avanço das discussões sobre mudanças climáticas e impactos ambientais, atrelada às questões energéticas, evoluiu as metodologias de diagnóstico que vêm incorporando nas análises não só sistemas de consumo, mas também de geração de energia. Essas incorporações possibilitaram a classificação de Edificações autossuficientes de energia, que são caracterizadas por possuírem um sistema de geração de energia renovável que supra, por completo ou parcialmente, a sua demanda de consumo. No âmbito da mobilidade elétrica, a maior adesão de veículos elétricos traz desafios e oportunidades na área de consumo, gestão e eficientização de energia elétrica. O impacto que essa carga veicular robusta e crescente pode acarretar aos ser integrada nos edifícios, novos e existentes, ainda não é contemplada nas avaliações de desempenho. Consequentemente as metodologias de obtenção das certificações e etiquetagens não levam em consideração a carga desse sistema como uso final. Portanto, analisar o impacto da integração de veículos elétricos (VE) no segmento de edificações, visa subsidiar a formulação ou revisão de metodologias de diagnósticos energéticos, que passem a contemplar o sistema carregamento de VE integrado às edificações. Como objetivo o estudo visa a elaboração de uma metodologia de diagnóstico energético de edificações brasileiras, para avaliar a integração do carregamento de veículos elétricos VE como um sistema predial de uso final. A metodologia escolhida foi a da Etiquetagem PBE Edifica, visando complementar a metodologia de diagnóstico energético de edificações do Programa Brasileiro de Etiquetagem de Edificações (PBE Edifica), apresentada através da Instrução Normativa Inmetro para edificações comerciais, de serviços e públicas (INI-C), que avalia atualmente sistemas que possuem alto consumo de energia elétrica como: envoltória, iluminação, condicionamento de ar (calefação e/ou arrefecimento) e aquecimento de água. Através da incorporação do sistema de abastecimento de veículo elétrico, esse estudo prospecta uma contribuição positiva ao abranger mais um sistema que apresenta uma significativa carga de consumo elétrico predial. 

A gestão eficiente de resíduos sólidos é crucial para manter a cidade limpa e sustentável. Este trabalho apresenta uma metodologia que utiliza algoritmos bem estabelecidos para a estimativa de volume na gestão de resíduos sólidos urbanos a partir de imagens de única visualização. O sistema proposto baseia-se em conceitos e modelos de visão computacional de última geração, incluindo segmentação de instâncias, estimativa de profundidade e cálculo de volume baseado em nuvem de pontos. A metodologia demonstrou a capacidade de estimar com precisão o volume de objetos de resíduos sólidos, tanto individuais quanto múltiplos, em imagens. Avaliamos nossa abordagem utilizando dados do mundo real. Apesar dos desafios, como o reescalonamento manual de distâncias e conjuntos de dados limitados, nosso sistema possui um potencial considerável para refinamento e aprimoramento, visando cenários complexos como os urbanos reais. Os resultados numéricos mostraram que o sistema proposto é promissor mesmo em cenários complexos, com valores de erro percentual absoluto médio (MAPE) de 8,60% para resíduos únicos e 9.23% para resíduos múltiplos, resultando em uma média geral de 8.91%. O coeficiente de determinação foi de 95.11% para instâncias únicas e 87.64% para múltiplas. A metodologia proposta contribui significativamente para o avanço das tecnologias de gestão em smart cities

Conversion systems are critically important devices in electrical systems across various environments, especially in modern times, as multiple components with different voltage levels and sources are interconnected within a single system. Consequently, dynamic study methods of this network are examined using approaches that simplify the network based on speed modes and switching of other conversion systems, wherein fast systems are simplified to constant power loads (CPL). This method evaluates the network’s stability conditions. The study reveals that CPLs act as negative incremental resistances, and when analyzed through a linear model, it is observed that such loads reduce system damping, thereby decreasing stability margins and potentially rendering the system unstable. Additionally, uncertainties in the physical components of the circuit further affect the stability and performance of microgrids. Hence, designing regulators to mitigate oscillations caused by these effects becomes crucial to ensure the proper performance of these systems.In this work, a robust controller is designed to handle uncertainties and attenuate oscillations in the presence of constant power loads. This controller is implemented in a microgrid composed of two cascaded DC-DC buck converters, one of which is modeled as a CPL. The system model is utilized for both stability analysis and robust controller design in state space, where the compensator synthesis is structured in the form of a linear matrix inequality, solved using system optimization tools. The controller’s results are compared with another controller based on pole placement in both linear and nonlinear switched models, within the Matlab/Simulink simulation platform. Transient response and control signals are evaluated graphically and through performance indices under various operating conditions, including load disturbances and system parameter variations.

Federated Learning (FL) emerges as a promising solution to enable collaborative model training for autonomous vehicles while preserving privacy and addressing communication overhead issues. Efficient client selection for participation in the training process remains challenging, especially in scenarios with statistical heterogeneity of data distribution and client failure events. Client failure, an uncontrollable event during training, reduces accuracy, convergence, and speed. This master thesis introduces an entropy-based client selection mechanisms for FL over Vehicular Network environments with client failure and non-IID data distributions. The proposed method is compared to a random selection mechanism in both IID and non-IID scenarios, as well as scenarios with random client drops. The results demonstrate that entropy-based selection outperforms other methods regarding training loss, accuracy, and Area Under the Roc Curve, particularly in high client dropout and non-IID scenarios. These findings highlight the importance of considering entropy data for client selection to address the challenges posed by client failure and statistical heterogeneity in  FL over Vehicular Network.

UTILIZAÇÃO DE SATÉLITE GEOESTACIONÁRIO PARA AVALIAR A INCIDÊNCIA DE RAIOS EM LINHAS DE TRANSMISSÃO DO SISTEMA INTERLIGADO NACIONAL BRASILEIRO

INIBIDOR BIDIRECIONAL DE EVENTOS DE RUNAWAY NO COMUTADOR DE TAP DE REGULADORES DE TENSÃO EM REDES DE DISTRIBUIÇÃO RECONFIGURÁVEIS COM GERAÇÃO DISTRIBUÍDA

Cell-free massive multiple-input multiple-output (mMIMO) networks are a
promising solution for sixth-generation (6G) communication systems and beyond.
These networks use multiple distributed antennas to transmit and receive signals,
which improves spectral efficiency and ensures a fairer distribution of network resources
among users. However, these innovative networks introduce a new mobile
communication paradigm that extinguishes the concept of cells. This shift poses significant
deployment challenges, as conventional tools designed for cellular systems
are inadequate for planning and evaluating feasible cell-free mMIMO architectures
since they often fail to address the unique aspects of distributed joint transmission
characteristics of cell-free environments. In this sense, the literature has been
developing separated models specific to cell-free systems that deal with system coordination,
fronthaul signaling, and required computational complexities of processing
procedures, segmented fronthaul, transitioning from cellular network deployments,
and integration to O-RAN technologies. These advancements are instrumental in
moving cell-free mMIMO from theoretical exploration to practical application. Despite
this, further study is needed to integrate existing models and develop practical
evaluation tools to assess the viability of these innovative networks and their enablers.
This thesis aims to address this gap by proposing new tools designed to
evaluate the feasibility of cell-free mMIMO networks regarding reliability and costs.
The first tool focuses on evaluating the reliability of cell-free systems and is essential
for segmented cell-free systems, which have serial fronthaul connections between
Transmission and Reception Points (TRPs), as these configurations are prone to reliability
issues. This tool is used in this thesis to improve the understanding of failures
and the development of protection schemes for these connections. The second tool
assesses the total cost of ownership (TCO) of cell-free systems and enablers across
various user demands, considering fronthaul bandwidth limitations, TRP, and CPU
processing capacities. This thesis uses this tool to evaluate the cost implications
of two functional splits equivalent to distributed and centralized processing architectures
and determine under which conditions each configuration might be more
cost-effective. This thesis analysis and results suggest that segmented fronthaul
cell-free systems require protection strategies, even when the TRP serial chain is
integrated into a single piece of equipment. It is shown that a partial duplication
scheme where 40% of the chain is likely the most feasible protection scheme. Besides
that, the interconnection between serial chains is shown to be an excellent feasible
protection alternative when the system has at least seven chains or integrated systems.
Finally, the cost analysis reveals that a centralized processing approach with
partial minimum mean square error (P-MMSE) precoding is the most feasible way
to deploy cell-free systems in terms of processing options most of the time.

 User-centric (UC) cell-free (CF) massive multiple-input multiple-output (MIMO) systems are promising technologies for beyond 5G (B5G) networks. In these systems, the user equipment (UE) is associated with a subset of access points (APs) distributed into the coverage area, leading to improvements in macro-diversity and spectral efficiency (SE) compared to conventional cell-based systems. Despite the benefits, challenges such as scalable AP selection strategies, computational complexity (CC), and inter-central processing unit (CPU) coordination may still exist in these systems. In this regard, this thesis proposes a novel and general AP selection framework that affords scalability for UC systems, enabling more efficient use of the network resources, such as transmission power and reduced processing demands. The solution is based on a matched-decision among the most suitable connections for APs and UEs. Moreover, three strategies to fine-tune the AP clusters of UEs are proposed, aiming to reduce the number of APs connected to each UE without compromising the SE. Simulation results reveal that the matched-decision framework improves up to 163% the SE of the 95% likely UEs compared with baseline schemes. A heuristic approach that reduces the effects of inter-CPU coordination is also proposed. It decreases the number of inter-coordinated UEs (i.e., UEs connected to multiple CPUs) on each CPU to reduce signaling demands on backhaul links. Numerical results indicate that the proposed method mitigates inter-CPU coordination while yielding slight degradation in SE and improving energy efficiency (EE). Finally, this thesis investigates the performance of UC systems with limited processing capacity. Specifically, it is assumed that the CC of performing channel estimation and precoding signals does not increase with the number of APs. Thus, the UE can only be associated with a finite number of APs. Furthermore, a method is proposed for adjusting the AP clusters according to the network implementation, i.e., centralized or distributed. The results show that UC systems can keep the SE under minor degradation even if the CC up to 96%. Besides, the proposed method for adjusting the AP cluster leads to further reductions in CC.

A maioria dos desastres naturais resulta de eventos geodinâmicos, como deslizamentos de terra e colapso de estruturas geotécnicas. Essas falhas causam catástrofes que impactam diretamente o meio ambiente e causam perdas financeiras e humanas. A inspeção visual é o principal método para detectar falhas em superfícies de estruturas geotécnicas. Todavia, as visitas no local podem ser arriscadas devido à possibilidade de solo instável. Além disso, o design do terreno e as condições de instalação hostis e remotas tornam inviável o acesso a essas estruturas. Quando uma avaliação rápida e segura é necessária, a análise por visão computacional torna-se uma alternativa. No entanto, estudos em técnicas de visão computacional ainda precisam ser explorados neste campo devido às particularidades da engenharia geotécnica, como conjuntos de dados públicos limitados, redundantes e escassos. Neste contexto, esta tese apresenta uma abordagem CNN para a identificação de defeitos na superfície de estruturas geotécnicas com o objetivo de reduzir a dependência de inspeções no local conduzidas por humanos. Para tanto, foram coletadas imagens de indicadores de falha superficial em taludes às margens de uma rodovia brasileira, com o auxílio de VANT e dispositivos móveis. Em seguida, foram experimentadas arquiteturas CNNde baixa complexidade para construir um classificador binário capaz de detectar em imagens falhas aparentes a olho nu humano. A arquitetura composta por três camadas convolucionais, cada uma com 32 filtros, seguidas por duas camadas totalmente conectadas, cada uma composta por 128 neurônios apresentou acurácia de 94,26%. A avaliação de desempenho do modelo com o conjunto de teste obteve métricas de AUC de 0,99 e matriz de confusão que indicam desempenho robusto do classificador na detecção de danos, ao mesmo tempo que mantém uma baixa complexidade computacional, tornando-a adequada para aplicações práticas em campo. As contribuições da tese incluem a disponibilização de banco de imagens que retratam danos visíveis em taludes, a obtenção de um modelo de classificação adequado para dados escassos e processamento com recursos computacionais limitados, e a exploração de estratégias para inspeção remota e identificação de indicadores de falhas em superfícies de estruturas
geotécnicas.

O panorama da energia elétrica no Brasil é influenciado por uma variedade de fatores complexos e relações não lineares, o que torna a previsão desafiadora. Com o aumento da demanda por energia e a crescente preocupação ambiental, é crucial buscar soluções baseadas em práticas de energia limpa e renovável, visando tornar o mercado de energia mais sustentável. Essas práticas visam reduzir o desperdício e otimizar a eficiência dos processos envolvidos na operação das tecnologias de distribuição e geração de energia elétrica. Uma abordagem promissora para viabilizar a energia sustentável é a aplicação de técnicas de previsão para diversas variáveis do mercado energético. Esta pesquisa propõe uma análise empírica do uso de regressores para realizar previsões nas bases de dados do Preço de Liquidação das Diferenças (PLD) do mercado brasileiro e da velocidade do vento em aerogeradores do Nordeste do Brasil. Busca-se contribuir com informações significativas sobre as técnicas de aprendizagem de máquina, que podem ser empregadas como ferramentas eficazes para a previsão de séries temporais no setor elétrico. Os resultados obtidos podem incentivar a implantação dessas técnicas para extrair conhecimento sobre o comportamento do sistema de energia brasileiro. Isso é particularmente relevante, dado que o preço da energia frequentemente exibe sazonalidade, alta volatilidade e picos, e a geração de energia eólica é amplamente influenciada pelas condições climáticas. Para modelar a previsão dessas duas séries temporais, utilizamos o banco de dados sobre o PLD, focando especialmente no preço médio da energia do Sistema Nacional Brasileiro. As variáveis mais relevantes estão relacionadas às condições hidrológicas, carga elétrica e preço dos combustíveis das unidades térmicas. Para a coleta das variáveis relacionadas à energia eólica, foram considerados dois locais distintos na região nordeste do Brasil: Macau e Petrolina. Para o estudo de previsão, utilizamos uma Rede Neural Perceptron Multicamadas (MLP), uma Long Short Term Memory (LSTM), o Auto-Regressive Integrado de Média Móveis (ARIMA) e a Máquina de Suporte de Vetores (SVM) para determinar as linhas bases nos resultados da predição. Para aprimorar os resultados destes regressores, utilizamos duas abordagens distintas de previsão. Uma das abordagens consistiu na combinação das técnicas de Redes Neurais Artificiais Profundas, baseada na Meta-Heurística do Algoritmo Genético Canônico (AG), para ajustar os hiperparâmetros dos regressores MLP e LSTM. Já a segunda estratégia focou em comitês de máquinas, os quais incluíam MLP, Árvore de Decisão, Regressão Linear e SVM em um comitê, e MLP, LSTM, SVM e ARIMA em outro. Essas abordagens consideraram dois tipos de votação, voting average (VO) e voting weighted average (VOWA), para avaliar o impacto no desempenho do comitê de máquinas.

Nowadays, grid resilience as a feature has become non-negotiable, significantly when power interruptions can impact the economy and the society. Smart Grids (SGs) widespread popularity enables an immense amount of fine-grained electricity consumption data to be collected. However, risks can still exist in the Smart Grid (SG), since SG systems exchange valuable data, and distribution system loses a substantial amount of electrical energy. We divide this loss into two categories: technical and non-technical loss. A substantial amount of electrical energy is lost throughout the distribution system, and these losses are divided into two types: technical and non-technical. Non-technical losses (NTL) are any electrical energy consumed and not invoiced. They may occur due to illegal connections, fraudulent activities, issues with energy meters such as delay in the installation or reading errors, contaminated, defective, or non-adapted measuring equipment, very low valid consumption estimates, faulty connections, and disregarded customers. Non-technical losses are the primary cause of revenue loss in the SG. Annually, electrical utilities incur billions in losses due to non-technical reasons. This thesis presents two methods of detection of NTL, namely classification and characterization. For the classification, we create an ensemble predictor-based time series classifier for NTL detection. This predictor uses the user’s energy consumption as a data input for classification, from splitting the data to executing the classifier. Also, it assumes the temporal aspects of energy consumption data during pre-processing, training, testing, and validation stages. The classification method has the advantage of classifying heterogeneous features in data. The characterization method proposes a study based on Information Theory Quantifiers (ITQ) to mitigate this challenge. First, we use a sliding window to convert the user’s energy consumption time series into a Bandt-Pompe (BP) probability distribution function. Then, we extract the used ITQ. Finally, we then apply each metric to the Probability Density Function (PDF) and map the layers to characterize their behavior. The characterization method is advantageous to be used when we have big data.

Esta dissertação estuda a previsão do consumo energético (CE) e maximização da produtividade na pintura automotiva, utilizando uma abordagem que combina seleção de variáveis, modelos híbridos, hiperparâmetros destes modelos e otimização por meta-heurística em uma arquitetura com 3 steps. Os processos de pintura automotiva apresentam variáveis em forma de séries temporais que descrevem o histórico do CE. No step 1, escolhe-se o melhor modelo de aprendizado de máquina (RF, LSTM, XGBoost e GRU-LSTM) para prever séries temporais do CE em t+1. No step 2, avalia-se os modelos RF, XGBoost e RNA Densa para selecionar o melhor preditor de quantidade de veículos produzidos (ciclos). No step 3, seleciona-se a melhor meta-heurística entre GA, DE e PSO para otimizar o CE previsto pelo melhor modelo do step 1, usando como medida de fitness o melhor modelo do step 2. A arquitetura final reduziu o CE em até 16% e aumentou o ciclo em 127%, usando os modelos GRU-LSTM no step 1, RNA Densa no step 2 e DE no step 3. Os resultados ressaltam a oportunidade de uso da abordagem proposta para otimizar o CE e a produtividade na pintura automotiva.

Ransomware é um software malicioso que tem como objetivo criptografar os arquivos do usuário e exigir um resgate para desbloqueá-los. Trata-se de uma ameaça cibernética que pode causar significativos danos financeiros, além do comprometimento de privacidade e integridade dos dados. Embora os scanners de detecção baseados em assinaturas comumente combatam essa ameaça, eles falham na identificação de famílias (variantes) desconhecidas de ransomware. Um método para detectar novas ameaças sem a necessidade de executá-las é a análise estática, que inspeciona o código e a estrutura do software, juntamente com a classificação através de abordagens inteligentes. Para avaliar a Detecção de Novas Famílias de Ransomware (DNFR), é possível simular um cenário realista e desafiador pela categorização e isolamento de famílias de ransomware para treinamento e teste. Desta forma, esta tese tem como objetivo desenvolver um modelo eficiente de análise estática para a DNFR, que pode ser aplicado em sistemas Windows como uma camada adicional de segurança, verificando os arquivos executáveis no momento do recebimento ou antes de sua execução. A detecção precoce do ransomware é fundamental para reduzir a probabilidade de um ataque bem-sucedido. A abordagem proposta analisa abrangentemente os binários executáveis, extraindo e combinando diversas características estruturais, e os distingue entre ransomware ou software benigno empregando um modelo de votação suave que compreende três técnicas de aprendizado de máquina: Logistic Regression (LR), Random Forest (RF) e eXtreme Gradient Boosting (XGB). Os resultados para a DNFR demonstraram médias de 97,53% de acurácia, 96,36% de precisão, 97,52% de recall e 96,41% de F-measure. Além disso, a varredura e a predição de amostras individuais levaram uma média de 0,37 segundos. Essa performance indica sucesso na identificação rápida de variantes desconhecidas de ransomware e na adaptabilidade do modelo ao cenário em constante evolução, o que sugere sua aplicabilidade em sistemas de proteção antivírus, mesmo em dispositivos com recursos limitados. Portanto, o método oferece vantagens significativas e pode ajudar desenvolvedores de sistemas de detecção de ransomware na criação de soluções mais resilientes, confiáveis e com rápido tempo de resposta.

Projeto associado: Aplicação e Concepção de Meta-heurísticas, Técnicas de visualização de dados e Métodos de Inteligência Computacional para Problemas de Engenharia e Computação

Neste trabalho, é desenvolvido um modelo numérico tridimensional de fluidodinâmica turbulenta para uma unidade de 311 MVA de uma máquina geradora completa de usina hidrelétrica, utilizando o método dos elementos finitos. O objetivo é estudar e compreender os mecanismos de transporte de ozônio dentro da estrutura da máquina elétrica fechada. O ozônio é produzido por descargas parciais relacionadas a falhas nas barras estatóricas. Para analisar o transporte de ozônio a partir de fontes localizadas, é desenvolvido e apresentado pela primeira vez um modelo tridimensional de fluidodinâmica de um hidrogerador em operação. O modelo possui um alto nível de detalhes geométricos. Além disso, uma nova proposta para simplificar a modelagem de radiadores é implementada e validada. A estrutura modelada é baseada em uma máquina elétrica do hidrogerador de Campos Novos e consiste em 378 barras estatóricas do tipo bobina feitas de cobre revestido por mica e, mais externamente, por uma camada de revestimento semicondutor. Outras partes também são representadas, incluindo o núcleo estatórico e direcionadores de ar feitos de aço inoxidável, radiadores de cobre, o rotor com sua superfície de epóxi, e o piso e paredes externas de concreto. No modelo de dinâmica de fluidos, uma malha de elementos finitos foi projetada para representar as regiões de ar dentro do hidrogerador e as superfícies dos materiais que reagem com o ozônio (com suas respectivas taxas de reação), onde o fluxo de ar e o transporte de ozônio são modelados usando as equações de Navier-Stokes e a lei de conservação de massa. As fontes de descargas parciais são representadas por fontes de ozônio com formas prismáticas, posicionadas nas superfícies das barras estatóricas. As concentrações de ozônio foram calculadas dentro e ao redor da máquina geradora. O raio do rotor é de 3,8075 m e sua frequência de rotação é de 200 RPM. A velocidade do ar radial devido à ventilação interpolar também é considerada (2,2 m/s, conforme verificado experimentalmente no local). A velocidade radial nas proximidades dos radiadores é de 3 m/s. Concluiu-se que o perfil de transporte de ozônio é influenciado pela posição da fonte nas barras estatóricas, de modo que a localização da fonte é possível e depende da determinação das áreas de concentração máxima locais e globais de ozônio nos radiadores.

Artigos publicados:

[1] DE OLIVEIRA, RODRIGO M. S.; GIROTTO, G. G. ; ALCANTARA, L. D. S. ; LOPES, N. M. ; DMITRIEV, VICTOR, "Ozone Transport in 311 MVA Hydrogenerator: Computational Fluid Dynamics Modelling of Three-Dimensional Electric Machine". Energies, v. 16, p. 1-35, 2023.

(QUALIS A2).

https://www.mdpi.com/1996-1073/16/24/8072

[2] DE OLIVEIRA, RODRIGO M. S.; ZAMPOLO, RONALDO F. ; ALCANTARA, L. D. S. ; GIROTTO, G. G. ; LOPES, F. H. R. ; LOPES, N. M. ; BRASIL, F. S. ; NASCIMENTO, J. A. S. ; DMITRIEV, V., "Analysis of Ozone Production Reaction Rate and Partial Discharge Power in a Dielectric-Barrier Acrylic Chamber with 60 Hz High-Voltage Electrodes: CFD and Experimental Investigations", Energies, v. 16, p. 6947, 2023.

(QUALIS A2).

http://dx.doi.org/10.3390/en16196947

Resumo

Hidrogeradores são ativos cruciais tanto para empresas geradoras de energia quanto para a população que depende desse abastecimento. No entanto, essas máquinas estão suscetíveis a vários tipos de defeitos que podem resultar em interrupções inesperadas, se nada for feito a respeito. A análise de descargas parciais é uma abordagem já consolidada para avaliar a condição de equipamentos de alta tensão, sendo essencial a detecção automática de diferentes tipos de defeitos, uma vez que diferentes níveis de risco à operação variam de acordo com o tipo de descarga. Redes neurais profundas têm sido propostas visando à classificação de descargas parciais usando diagramas PRPD (phase-resolved partial discharge). Contudo, a obtenção de conjuntos de dados rotulados com grande número de exemplos é um problema que impacta diretamente no desempenho de classificadores treinados de maneira supervisionada. Neste contexto, esta dissertação propõe uma técnica semiautomática para a rotulagem de PRPDs, baseada em estratégias de redução de dimensionalidade e agrupamento de dados, bem como investiga o uso de GAN (generative adversarial network) na ampliação artificial do conjunto de treinamento. O conjunto de dados usado no trabalho é composto por PRPDs reais obtidos por meio de procedimentos de monitoramento online de descargas parciais em hidrogeradores. O desempenho dos sistemas propostos é contrastado com resultados recentes representativos do estado da arte na área. Os resultados demonstram que a aplicação da técnica proposta para rotulagem semiautomática reduz consideravelmente a carga de trabalho e o tempo associados à classificação manual. Além disso, o uso de PRPDs artificiais gerados pela GAN resultou notável melhoria no desempenho do classificador que alcançou 94,72% de acurácia média, em comparação com 89,44% obtido com a melhor técnica concorrente. Foram observados ganhos semelhantes também nas acurácias por classe.

Artigos publicados:

O desenvolvimento de biofilme na superfície dos sensores em contato prolongado com a água pode acarretar na coleta de dados inconsistentes e até mesmo no desgaste do equipamento. Por isso o monitoramento constante dos equipamentos se torna necessário, e para manter a integridade dos dados a manutenção preditiva é fundamental. Diante deste cenário a IoT (Internet of Things) e suas aplicações se apresentam como uma das alternativas viáveis para o envio de dados em tempo real, pois ela apresenta soluções de comunicação distribuídas, de menor custo e fácil acesso por seus usuários. Este estudo analisa os dados dos sensores de temperatura armazenados na nuvem através de cálculos estatísticos. A partir destas análises são definidos os limiares do sistema Fuzzy desenvolvido que indicará a necessidade de limpeza dos sensores de acordo com o nível de ruído gerado pela presença de biofilme, para que se mantenha a integridade dos dados coletados.

With the increase in the number of Unmanned Aerial Vehicles performing various automated activities in social life, it is necessary to create algorithms with efficiency and safety in order to avoid damage and losses in society. In this work, non-adaptive and adaptive digital controllers are implemented for the autopilot system of the four-engine unmanned aerial vehicle model AR Drone 2.0 from Parrot, aiming to analyze the impact of different techniques for identifying state systems on the control performance of a multiple inputs and multiple outputs (MIMO) process. The work uses parametric identification techniques such as Least Squares, Recursive Least Squares, Recursive Extended Least Squares, Instrumental Variables and adaptive joint estimation techniques for states and parameters such as the Extended Kalman Filter and the Instrumental Adaptive Kalman Filter, which is the product of the techniques previous ones. The controller project chosen to be used as the basis of the analysis was LQG, as it is a well-known algorithm in the literature for implementations of real-time controllers for MIMO systems, with the characteristics of optimal control and optimal state estimation. , thus, minimizing the designer's interference in tuning the controller gains and using the system parameters as a determinant for tuning. Therefore, the performance and robustness of the control design may vary depending on the parameters arising from the system identification technique used, which, when appropriate, favors the LQG design and, otherwise, compromises the reliability of the algorithm. To support the study carried out on the proposed topic, experiments were carried out using the indices "Integral of Squared Error", "Integral of Squared Control", "Total Variation of Control" and "Integral of Absolute Error", to evaluate performance , and sensitivity and complementary sensitivity curves to evaluate robustness, in addition to an index, called R, designed to evaluate the robustness of adaptive control algorithms for MIMO systems at each moment of the experiment, sample by sample, enabling discussion and comparison of control algorithms. Flight simulations and experiments tested the various parametric estimation techniques associated with LQG control, most of the techniques had acceptable performance and robustness, although suboptimal, while the LQG control project associated with the Instrumental Adaptive Kalman Filter obtained a better result in the segment reference and adaptive robustness due to its theoretical commitment to finding the real parameters of the system, therefore avoiding parametric polarizations.

Keywords: System identification, state estimation, four-engine, adaptive control, autopilot.

Published papers:

SILVEIRA, A. S.; NOGUEIRA, C.; SODRÉ, L.; da SILVA, D. "Predictive Minimum Variance Control: a Case Study on Wireless Altitude Hold Autopilot". 16º Simpósio Brasileiro de Automação Inteligente (SBAI 2023), 15 a 18 Outubro, Manaus-AM, 2023.

SILVEIRA, A. S.; SODRÉ, L.; SILVA, A. F.; CONDE, L.; BORGES, J. P.; YURI, S.; KLAUTAU, A. "Smith Predictor-based Adaptive Control of Network-Controlled UAVs". 15º Simpósio Brasileiro de Automação Inteligente (SBAI 2021), 2021, doi: 10.20906/sbai2021/216460.

AVALIAÇÃO TEÓRICA EXPERIMENTAL DE SISTEMAS FOTOVOLTAICOS DE BOMBEAMENTO DE ÁGUA

The Brazilian educational system faces significant challenges, as evidenced by low educational development assessment scores. Due to the traditional educational model employed in the country, there are difficulties in the effective transmission of complex content, leading to high rates of academic failure and subsequent school dropout. The lack of innovation, especially in basic education settings, contributes to a scenario of low mathematical proficiency among Brazilian students. In this context, this work arises as a result of an innovation developed to enhance the Geometa application, developed by the Inteceleri company, through the integration of Metaverse and Artificial Intelligence technologies to create an immersive and interactive educational  invironment. The intention is to fine-tune Artificial Intelligence for the real-time recognition of three-dimensional geometric shapes from real objects. The proposal aims to mitigate challenges faced in basic Mathematics education in Brazil by adopting innovative technological approaches aligned with Education 5.0, which can be replicated for similar technologies involving the Metaverse. Furthermore, it is also intended to create a dynamic and sustainable educational environment that not only facilitates the understanding of mathematical concepts but also promotes active student participation, encouraging their creativity and autonomy in the learning process. As contributions of this work, the following were accomplished: (i) the training of defined models for the classification of three-dimensional geometric shapes; (ii) the evaluation of models through performance metrics, inference time, and dimension; and (iii) the selection of the model with the best performance based on accuracy and inference efficiency. According to the obtained results, the ResNet model exhibited the 84% precision, presenting the highest performance obtained among the evaluated models, in addition to 9 inferences per second, which reflects on the lowest consumption of computational resources for the task.

O desmatamento é um dos principais problemas ambientais enfrentados no mundo, e na Amazônia a perda da cobertura vegetal tem graves consequências para o clima, biodiversidade e sociedade. Nesse contexto, a detecção e classificação de áreas desmatadas é fundamental para o monitoramento e controle do desmatamento. O objetivo deste trabalho foi identificar áreas degradadas e não degradadas na região nordeste paraense por meio de imagens do satélite Landsat 5, 7 e 8, utilizando a metodologia ReLU (Unidade Linear Retificada). O banco de dados abrangeu 210 imagens. Após os resultados foi possível gerar a matriz de confusão, para realizar a classificação digital para avaliação dos resultados das imagens de satélite. Os dados foram extraídos em KDD, o código fonte para o processamento dos dados foi a linguagem Python e rodado na plataforma Colaboratory. Por meio da matriz confusão estimou-se com 69,7% de acurácia os resultados. Os valores determinaram um índice eficaz de classificação das áreas degradadas e não degradadas. O modelo obtido  para o treinamento apresentou algumas interferências na imagens sendo possível identificar áreas desmatadas e não desmatadas. Para estudos futuros recomenda-se novas imagens de satélite da região, com destaque para os índices de resolução espectral correspondentes com os sensores determinados pelos satélites do estudo.  

Projeto: Soluções Inteligentes para Sensores Ópticos e Redes de Próxima Geração

This dissertation aims to investigate the use of machine learning models using multimodal data as input to optimize the Beam-selection process in millimeter-wave based networks. The use of data of different natures is interesting as we can adjust the model according to the quality/availability of these data. After executing the experiments and obtaining the results, it was observed that it is possible to obtain significant performance in different metrics even with simpler data such as image and coordinate.

In this dissertation, the terahertz potential of graphene was explored, especially its unique ability to control the radiation diagram and impedance of a graphene dipole antenna, through varying the chemical potential in six controllable segments. The study used the Method of Moments with graphene surface impedance values. The chemical potentials were divided into symmetric and asymmetric groups by adjusting the second resonance and the maximum gain angle of the radiation diagram. Compared to four-segment antennas, the gain of symmetric antennas increased with little variation in the second resonance. Furthermore, the gain of the six-segment antennas had a notable increase at the point of greatest deviation, keeping the angular displacement practically constant. This study paves the way for highly tunable and efficient graphene antennas, with promising applications in communications and radiation technology. Future work can explore varying chemical potentials, other antenna geometries and optimization techniques in the simulation.

Indicadores de sustentabilidade energética são ferramentas utilizadas para avaliar e medir o desempenho de sistemas energéticos, incluindo aspectos ambientais, sociais e econômicos. Eles podem ser utilizados para avaliar a sustentabilidade de diferentes fontes de energia, como fontes renováveis ou fontes fósseis, ou para avaliar o desempenho de sistemas energéticos em diferentes regiões ou países. A presente tese teve como objetivo desenvolver um quadro de indicadores de sustentabilidade energética e elaborar um índice final para a Amazônia utilizando lógica fuzzy. A metodologia utilizada incluiu a revisão da literatura com uma seleção de indicadores de sustentabilidade energética adaptados à região amazônica, coleta e análise dos dados utilizando lógica fuzzy e a proposição do Índice Amazônico de Sustentabilidade Energética. O sistema de inferência fuzzy utilizado permitiu operacionalizar os conceitos subjetivos da sustentabilidade através da criação do índice final, utilizado para avaliar o desempenho dos estados da Amazônia Legal e compará-los entre si. Como resultado, temos o Mato Grosso em primeiro lugar no ranqueamento, seguido pelos estados do Tocantins, Amapá, Rondônia, Roraima, Pará, Acre, Amazonas e na última posição o Maranhão. Os resultados encontrados mostraram que os indicadores selecionados e o índice final elaborado são úteis para avaliar a sustentabilidade energética na Amazônia e balizar os estados da Amazônia Legal, no sentido de auxiliar os gestores públicos para a tomada de decisões e elaboração de políticas públicas voltadas ao desenvolvimento regional sustentável da região amazônica.

Artigo Publicado:
MUNIZ, R. N. et al. Tools for Measuring Energy Sustainability: A Comparative Review. Energies, v. 13, n. 9, p. 2366, jan. 2020. Qualis A1.

Artigo Submetido:
MUNIZ, R. N. et al. Amazon Energy Sustainability Index. Journal of Intelligent Fuzzy Systems. Qualis A2.
.

The exploratory research aims to analyze existing regulations regarding nanotechnology and its impact on worker safety. The study will focus on examining the specific challenges associated with nanotechnology, including the lack of adequate toxicological data to assess risks, the complexity of characterizing nanomaterials, and the limitations of existing monitoring techniques. One of the main challenges faced is the lack of sufficient data on the health effects of workers exposed to nanomaterials. Nanotechnology is a relatively new field, and much remains to be learned about the potential adverse health effects these materials can cause. The lack of adequate toxicological data makes it difficult to accurately assess risks and implement effective protective measures. Furthermore, the characterization of nanomaterials is also a complex task. Its unique physical and chemical properties require specialized analysis methods and advanced techniques. The lack of standardization in this field makes it difficult to fully understand nanomaterials and their relationship to health effects. Another challenge is the limited monitoring techniques available to assess workers' exposure to nanomaterials. Traditional monitoring methods may not be adequate to detect and quantify the presence of nanomaterials in the work environment due to their small size and unique properties. This makes it difficult to accurately assess worker exposure and implement appropriate control measures. The study also sets out to examine the strategies and approaches adopted by different countries and international organizations to address these challenges and ensure the safety of workers exposed to nanotechnology. This may include developing specific regulations, promoting toxicological research and studies, creating good practice guidelines, and implementing engineering control and personal protection measures. In summary, the exploratory research aims to analyze existing regulations and challenges related to worker safety in the field of nanotechnology. Understanding and addressing these challenges is critical to ensuring worker protection and promoting the safe and responsible use of nanotechnology in industry.

The handover process was designed with the aim of ensuring service quality in cellular telephony networks. In scenarios with multiple Base Stations (BS), the handover decision-making process is a problem that has been studied in various researches, given that neglecting the use of efficient handover strategies can influence performance metrics and, consequently, lead to a decrease in the quality of services provided. To develop efficient handover strategies, the correct use of a set of metrics is essential, as well as the choice of decision-making methods that can optimize handover and result in a better connection for users. With this in mind, this thesis proposal seeks to identify and analyze a set of handover metrics that, when combined with decision-making methods, can optimize performance metrics for heterogeneous mobile networks. As results, two heuristics for handover were developed. The first one is called H2ATF (Heuristic for handover based on AHP-TOPSIS-FUZZY), which uses a specific set of metrics (Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ), Signal-to-Interference-plus-Noise Ratio (SINR), and User Speed) and employs the following methods: (a) Analytical Hierarchical Process (AHP) for defining criteria weights; (b) Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) for ranking the Base Stations; and (c) the use of an adaptive hysteresis, calculated through Fuzzy logic based on parameters that directly impact the handover process. The second heuristic, H2AG (Heuristic for handover based on Genetic Algorithm), utilizes genetic algorithms to minimize handover metrics. Through the development of these handover strategies, it was possible to determine the best timing and, collectively, the best BS to perform the handover. The performance evaluation results of the algorithms demonstrate that H2ATF achieved better results in terms of Handover Ping Pong rate (HPP), while H2AG showed lower average Handover Failure rate (HOF) in the evaluated scenarios.

IDENTIFICAÇÃO DE LARVAS DE MOSQUITO DO GÊNERO AEDES UTILIZANDO REDES NEURAIS CONVOLUCIONAIS

A teoria de controle é utilizada para estabilizar sistemas e obter respostas específicas para cada tipo de processo. Controladores clássicos, como o PID e o IMC utilizados nesta pesquisa, são difundidos globalmente nas indústrias, isto por possuírem topologias bem estudadas pela literatura e serem facilmente aplicados em microcontroladores; já os avançados, como GMV, GPC e LQR também utilizados neste trabalho, possuem certa resistência em aplicações comuns das indústrias de base, mas são muito utilizados em sistemas de energia, aerospaciais e robóticos, pois a complexidade e estrutura desses métodos gera robustez e alcança desempenhos satisfatórios para processos de difícil controle. Neste trabalho, esses métodos são estudados e avaliados com uma abordagem de sintonia que utiliza o aprendizado por reforço. São aplicadas duas formas de sintonia para os controladores, estas são o método do laço offline e o método de jogos diferenciais. O primeiro utiliza iterações offline, onde o agente do processo é a técnica de controle utilizada, que utiliza os índices de desempenho e robustez como ambiente (métrica de como o processo está evoluindo), sendo capaz de organizar uma política de ajuste para o controlador, que se baseia em recompensar o fator de ponderação até obter o critério de parada do processo (resposta desejada). O segundo método se  baseia em utilizar estratégias de reforço que recompensam o controlador conforme a resposta se modifica, assim o LQR aprende as políticas de controle ideais, adaptando-se às mudanças do ambiente, o que permite obter melhor desempenho por recalcular os tradicionais ganhos encontrados com a equação de Ricatti para sintonia do regulador; neste método, os jogos diferenciais são utilizados como uma estrutura para modelar e analisar sistemas dinâmicos com múltiplos agentes. Para validar o que é apresentado, o Motor Tacogerador e o Ar Drone são utilizados. O MTG é modelado utilizando a estimação dos mínimos quadrados com uma estrutura ARX-SISO para avaliação do primeiro método de sintonia. O Ar Drone é modelado utilizando o estimador de Kalman com uma estrutura ARX-MIMO para avaliação do segundo método de sintonia.

Palavras-chave: Teoria de Controle. Aprendizado por Reforço. Laço Offline. Jogos Diferenciais. Equação de Ricatti.

Artigos publicados/submetidos:

1. SILVA, D. A. M. ; SILVEIRA, A. S. ; NASCIMENTO, A. C. . State Space Predictive Minimum Variance Controller Applied to a Tacho Generator Motor. In: Seminar on Power Electronics and Control, 2022, Santa Maria. 14th Seminar on Power Electronics and Control, SEPOC 2022, 2022. [Publicado]

2. SILVA, D. A. M. ; ARAÚJO, R. B. ; SILVEIRA, A. S. . Análise de Desempenho e Robustez do Controlador Preditivo Generalizado Aplicado em Plantas Benchmarks. In: 15th IEEE/IAS International Conference on Industry Applications, INDUSCON 2023. November 22-24, São Bernardo do Campo, São Paulo, Brazil, 2023. [Em revisão] 

3.SILVA, D. A. M; ALMEIDA, K.; SILVEIRA, A. S.. Analysis of Reinforcement Learning for One Parameter Controllers Tuning. In: 30th Brazilian Symposium on Intelligent Automation, SBAI 2023. October 15-18, Manaus, Amazônas, Brazil, 2023. [Em revisão]

To serve the Internet of Things (IoT) networks, there are several intelligent systems that are still under development and when implemented in a practical way, they will require the use of robust structures for storing, managing and presenting the data collected in the form of information/knowledge. Such systems are based on the use of different technologies to ensure an orchestrated operation, that is, mainly considering the efficient use of resources under network monitoring, such as battery level and device memory. To this end, this work presents a web system for managing data captured in real time with support for decision making based on Artificial Intelligence and Statistics. For this, it makes use of General Regression Neural Network (GRNN), Linear Regression and Moving Average techniques, among others. The results indicate that the proposed system can be a real-time decision support tool.

-

The increase in digital multimedia information flow over wireless networks, driven by the IEEE 802.11 standards and the widespread use of devices operating in these networks, has led to the common use of ultra-resolution video streaming. This has driven the development of more efficient CODECs for data compression in wireless transmission. While many research efforts are focused on improving video codecs, the perceived video quality experienced by users in wireless networks is affected by the characteristics of the wireless channel. Many of these research studies aim to establish a relationship between Quality of Service (QoS) metrics and user Quality of Experience (QoE) in video streaming services, proposing enhancements to network techniques and protocols. However, few studies consider specific video information, such as frame loss, the Group of Picture (GOP) structure, and video characteristics themselves. In this thesis proposal, we analyze the behavior of video streaming using the H.264/AVC CODEC at 2160p (4K) resolution, with variations in the GOP structure and a fixed Quantization Parameter (QP) value. The selected videos are subjected to simulated packet loss to mimic a network with transmission errors, and QoE metrics are extracted to correlate with QoS metrics. We propose a mathematical model to predict video quality loss in error-prone networks, utilizing the Peak Signal-to-Noise Ratio (PSNR) metric as a QoE measure, taking into account the total number of lost frames and the variation in the GOP structure for different videos. The accuracy of the model is evaluated using the Root Mean Square Error (RMSE) metric.

The fourth industrial revolution has several pillars, the Internet of Things and Big Data
being one of the most prominent. These technologies make it possible to collect and
analyze large data sets in real time, allowing the development of models for the most
diverse situations, from consumer behavior to the prevention of factory failures. In this
context, the present work proposes an architecture for the implementation of an
automation system for industry 4.0, based on the collection of temperature, hour meter,
vibration and current data in electric motors of a primary aluminum industry. From the
measured variables, vibration data in the frequency domain, phase imbalance and
constant with Kalman filter can be obtained. A multicriteria decision algorithm was
adopted to assist in choosing the programming language. After the elaboration of this
systematic, a set of solutions was obtained that achieved the development of the
system. Therefore, an automation system was developed, called IOT CORE, which
performs from the collection of variables in real time, with low latency, high
performance, and makes it possible to transmit, store and visualize the data in several
supervisory systems.

Keywords: Supervisory, Microcontroller, Database, Communication Protocols, Cyber-
Physical Systems.

Ransomware é um tipo de malware que impede ou limita o acesso do usuário ao sistema e
arquivos até que um resgate seja pago. Combater essa ameaça é difícil devido à sua disseminação rápida e às constantes mudanças nas técnicas de criptografia utilizadas. Algoritmos de aprendizado de máquina, como Redes Neurais Artificiais, têm sido apontados como ferramentas promissoras na classificação de ransomware, porque elas podem aprender a identificar padrões e características complexas em grandes quantidades de dados. Isso permite que as redes neurais sejam treinadas com exemplos de amostras de software malicioso, incluindo ransomware, e depois sejam capazes de classificar novos exemplos com alta precisão. Além disso, as redes neurais também são capazes de aprender e se adaptar a mudanças no comportamento do malware, tornando-as ferramentas eficazes para a detecção de novos tipos de ransomware. Neste trabalho, é explorado três tipos de classificação de ransomware por RNA dentro de um pipeline composto com redução de dimensionalidade por Kernel PCA e balanceamento de classes com a abordagem de superamostragem aleatória. A MLP ( Multi-layer Perceptron) alcançou uma média de 98% de acurácia na classificação binária e 85% de acurácia na classificação de família com goodware, onde tais valores superam os resultados anteriores e demonstram assim a eficácia da inclusão do balanceamento de classes na melhoria do modelo de detecção de ransomware.

Artigo publicado para a defesa:
        1) PEREIRA, G.; SALES, JR., C.Ransomware classification by machine learning and dimensionality reduction. Journal of Engineering Research, v. 2, n. 25, 2022. ISSN 2764-1317. DOI: 10.22533/at.ed.3172252201116

Projeto vínculado: Aplicação e Concepção de Meta-heurísticas, Técnicas de visualização de dados e Métodos de Inteligência Computacional para Problemas de Engenharia e Computação

In smart city paradigms, it is essential to ensure the well-being of the population in terms of social infrastructure. Among the actions that can be taken to make a city more resilient are preventive measures, mechanisms for informing the population, and technologies that accelerate action in emergency situations. The establishment of a communication structure that survives disasters is of utmost importance so that authorities can act immediately in providing assistance and saving lives. In this sense, the use of UAVs as a communication alternative is an ideal solution for scenarios such as natural disasters or intentional attacks that may cause partial or complete disruption of telecommunications services. However, energy autonomy is a limitation that affects the mission's lifespan. With this in mind, our group has developed a new method based on reinforcement learning that aims to reduce energy consumption in UAV missions in disaster scenarios, circumventing the negative effects of wind variations and optimizing the air mesh time in locations affected by interruption of fiber optic-based telecommunications services. The thesis proposal aims to develop a new methodology that considers an emergency solution aimed at dimensioning backup elements in an optical metro network, considering the adoption of UAVs and wireless interfaces to provide temporary air links, and also provides efficient management of backup resources to extend the flying mesh network's life during the catastrophe. The method considers K-means to scale the position of resource stations - from which UAVs are launched. For UAV locomotion, the Q-learning method was used to investigate possible actions that UAVs could take due to randomly distributed urban obstacles in the scenario and due to wind speed, which is related to how UAVs are organized during the mission. For the tests, a metropolitan optical network topology from Stockholm, Sweden was used. The heuristic was combined using three reinforced learning methods (Simple Q-learning, ε-greedy and SARSA) and a naive solution. Numerical results of the experimented simulations proved that the reinforcement learning based solution was able to reduce the energy consumption by up to 15.93% compared to the naive solution, which can lead to a significant increase in the service life of voluntary UAV routes. Results of this work contributed to increase the efficiency of UAV routing, which in turn favors the increased use of UAVs in civil applications, as expected for smart cities.

In wireless network planning, the most important factor for estimating signal coverage area is the use of an accurately modeled propagation model. It should be adapted to the environment in which it will be implemented, capable of predicting signal nuances along the path to accurately predict signal degradation. Therefore, the objective of this work is to assist in the planning of future IoT networks, similar to the case study of the Smart Intelligent Multimodal Electric Mobility Management System (SIMA) project, which aims to create a Smart Campus to monitor parameters (location, battery level, temperature measurement, etc.) of electric vehicles, such as buses and boats, serving the academic community. Thus, the proposed work develops an empirical model based on measurements in LoRa networks operating at a frequency of 915MHz, specifically for Amazonian environments. The measurements were conducted at the campus of the Federal University of Pará (UFPA), in Belém, Pará, Brazil. To model the wireless signal path loss, an artificial neural network (ANN) technique, a computational intelligence approach, was employed. Four input parameters were used: distance between transmitter and receiver, Spreading Factor (SF), link type (uplink or downlink), and transmitter height. These parameters were evaluated to determine the accuracy of the proposed model, which was compared to Close-in (CI) and Floating Intercept (FI) models, as well as the UFPA Model adjusted through Genetic Algorithm (GA). The proposed ANN-based model stood out and was further employed in a multi-objective bio-inspired optimization algorithm called Evolutionary Particle Swarm Optimization (EPSO) to determine the positioning of gateways, aiming to maximize the LoRa network coverage area on the Smart Campus with the fewest possible gateways. The best result was obtained with four gateways and their respective positions, covering 100% of the established area. 

High Definition Video on Demand (VoD) and immersive multimedia services,
such as Virtual Reality (VR) and Augmented Reality (AR), continuously enhance multimedia visual quality and leads to an exponential usage growth of computational resources of processing, networking, and storage. Low-latency multimedia services rely on heterogeneous multi-tier edge computing datacenters to provide computational resources close to consumers, improving user-perceived Quality of Experience (QoE) and avoiding degraded network Quality of Service (QoS). Multimedia services are composed of specialized functions embedded into a monolithic service or decomposed into Service Function (SF) chains of ordered microservices inter-connected with high-speed network links. Monolithic and SF chains deployed at the network edges can reduce delay and improve the use of computational resources. In this context, multi-tier edge nodes can host services according to consumers’ and providers’ best interests. QoE, QoS, mobility, and multiple criteria are important in instantiating, re-instantiating, and terminating services in multi-tier edge
computing. Service orchestration controls and manages service requests and chooses edge nodes to execute services fulfilling requirements guarantees and efficiently using computational resources. The orchestrator can be configured to consider service aspects, such as QoE, QoS, fairness, computational edge node constraints, energy, and mobility. There are several issues of providing efficient service orchestration for multimedia in multi-tier edge computing environments, including (i) service orchestration in diverse changing network conditions intended to improve the user-perceived experience and QoS; (ii) mobile users assistance by energy-constrained mobile and static edge nodes in temporary and poor connectivity conditions; and (iii) mobility support in real-time for immersive multimedia SF chaining orchestration to mobile users in urban scenarios.
This thesis contributes to efficiently meeting future VoD, VR, and AR services needs in multi-tier edge computing. The first contribution lies in choosing an appropriate edge node to stream VoD, considering QoE, QoS, and cost to instantiate VoD services. The second lies in selecting and placing flying edge nodes to assist poorly-served users regarding QoS, considering distance, energy, and service priority. The third lies in mobility-aware Service Function Chaining (SFC) orchestration by mapping, instantiating, and re-instantiating SFs into multiple edge nodes considering QoS. The proposed solutions were widely compared to related works on different scenarios. The results show that the proposed schemes outperform the state-of-the-art.

The increase in the volume of literature related to biological and health sciences is a challenge for researchers and biocurators, even queries in specialized databases in biomedical literature, such as PubMed (scientific and medical abstracts/citations) and PubMed Central (full text journal articles) present a large volume of retrieved items that tends to make it difficult to locate relevant information about biological entities. Document selection, also known as like screening, is the first and one of the most important in the biocuration process steps, therefore, developing automated information extraction methods to support the construction of logical databases and discovering new knowledge from online journal collections is, not only a challenge, but a necessity, and the adoption of text mining techniques are viable alternatives, as they can streamline and optimize document screening, as well as assist in different steps of the standard biocuration workflow. In this scenario, this thesis proposal presents a methodology that adopts a text mining approach based on unsupervised learning, to classify the relevance of scientific articles in specific biomedical contexts and generate training sets accurate enough to maximize the efficiency of supervised classifiers. As a way of showing the effectiveness of the proposal, the context of antimicrobial resistance genes is used as a case study.

The sixth-generation (6G) technology of mobile networks will establish new standards to fulfill unreachable performance requirements by fifth-generation (5G) mobile networks. This is due to the high requirements for a more intelligent network, ultra-lower latency, extreme network communication speed, and supporting a massive number of connected applications. In this context, Mobile Edge Computing (MEC) and service migration strategies have shown promising results in improving the user experience, meeting requirements of beyond 5G networks (B5G), and optimizing infrastructure resources. In addition, Unmanned Aerial Vehicles (UAVs) are a promising solution to provide cloud and edge services in collaboration with MEC scenarios. However, most strategies have ignored resources available on users' devices and contextual information. Motivated by the use of massive-extended reality (XR) and virtual reality (VR) applications in B5G networks, this thesis proposal presents a video service migration strategy based on contextual information and evaluates the influence of user mobility on migration strategies, called Dynamic Video Service Migration (DVSM). Simulation results highlight $74\%$ higher performance of the DVSM compared to state-of-the-art algorithms and a performance equivalent to the optimal solution when the collection and analysis of context information (remaining device's energy, user's location, network conditions, etc) are carried out correctly.

This study proposes a method for improving the capability of a data-driven multi-agent system (MAS) to perform condition monitoring and fault detection in industrial processes. To mitigate the false fault-detection alarms, a co-operation strategy among software agents is proposed because it performs better than the individual agents. A few steps transform this method into a valuable procedure for improving diagnostic certainty. First, a failure mode and effects analysis are performed to select physical monitoring signals of the industrial process that allow agents to collaborate via shared signals. Next, several artificial neural networks (ANN) models are generated based on the normal behavior operation conditions of various industrial subsystems equipped with monitoring sensors. Thereafter, the agents use the ANN-based expected behavior models to prevent false alarms by continuously monitoring the measurement samples of physical signals that deviate from normal behavior. Finally, this method is applied to a wind turbine. The system and tests use actual data from a wind farm in Spain. The results show that the collaboration among agents facilitates the effective detection of faults and can significantly reduce false alarms, indicating a notable advancement in the industrial maintenance and monitoring strategy.

The rapid growth of data traffic due to the demands of new services and applications poses new challenges to the wireless network. Unmanned aerial vehicles (UAVs) can be a solution to support wireless networks during congestion, especially in scenarios where the region has high traffic peaks due to the temporal and spatial flow of users. In this paper, an intelligent machine-learning-based system is proposed to deploy UAV base stations (UAV-BS) to temporarily support the mobile network in regions suffering from the congestion effect caused by the high density of users. The system includes two main steps, the load prediction algorithm (LPA) and the UAV-BSs clustering and positioning algorithm (UCPA). In LPA, the load history generated by the mobile network is used to predict which base stations are congested. In UCPA, planning is performed to calculate the number of UAV BSs needed based on two strategies: naïve and optimized, in addition to calculating the optimal positioning for each device requested to support the overloaded base stations. For prediction, we used two models, generalized regression neural networks (GRNN) and Random Forest (RF), and the results showed that both models were able to make accurate predictions, and the random forest model was better with an accuracy of over 85%. The results showed that the intelligent system significantly reduced the overhead of the affected base stations, improved the quality of service (QoS), and reduced the probability of blocking users, as well as defined the preventive scheduling for the UAV BSs, which benefited the scheduling and energy efficiency.

In this work a library capable of automatically generating test cases for the Python programming language is developed, these tests are generated utilizing an genetic algorithm which uses an ad-hoc mutation operator based on social interaction. The algorithm is applied with success to the problem of automatically generating software tests, showing promising results when compared to the state of the art, revealing itself as a possible path of exploration when solving this category of problem. 

This work proposes a stochastic model predictive control using the Laguerre function with optimal Kalman filter state estimation. The controller design uses an ARMAX state-space model, incorporating moving average into the stochastic formulation by a state disturbance matrix in innovation form, which calculates a stochastic term introduced in the control law. The optimal Kalman filter gain design copes with the minimum variance case, where the Kalman filter weighting matrices are tuned based on the state disturbance matrix and the covariance of estimated states of an ARMAX model. Furthermore, it shows that the proposed strategy also can be applied in the classic MPC design.

Hildreth’s Quadratic Programming is the method used to solve the constrained optimization problem in a stochastic scenario. This method is used together with the Laguerre function, simplifying finding the optimal problem in constrained cases. Moreover, the Laguerre function improves the control horizon prediction, reducing the output variance, and preserving a better trade-off between the control effort and closed-loop performance. It is because of its orthogonal property, making it a universal approximator that results in a parsimonious representation of the control trajectory.

This work uses a highly oscillatory mechanical system, a robot joint, and a Hydropower MIMO system as numerical examples to demonstrate the proposed method’s efficiency. Furthermore, two experimental tests are introduced with two different plants that confirm these results: a circuit representing an under-damped coupled multivariable system with two inputs and two outputs and a Twin Rotor MIMO system, comparing the proposed strategy’s effects to the Laguerre deterministic MPC approach, classic MPC, and classic MPC using the ARMAX model with minimum variance Kalman filter estimation.

DEEP LEARNING SOFTWARE-BASED HOLDOVER FOR PTP IEEE 1588 SYNCHRONIZATION IN 5G NETWORKS

With population growth in urban areas, the more extensive city infrastructure faces several problems affecting the population's health and quality of life. Urban mobility changes became intense since the worldwide technological revolution brought many tools and methods to prevent harmful tendencies regarding urban transportation. In this context, Internet of Things (IoT) solutions perform a ubiquitous way of sensing the population mobility and the local mobility context as criminality, accidents, and air quality near the road infrastructure, complementing the city mobility. Likewise, Location-based Social Networks (LBSN) dispose of users' geolocated data, allowing the identification of mobility patterns, traffic flows, and alternative modal transport recommendations. In this matter, novel mobility solutions must attend city issues regarding public transportation, criminality, traffic influencing factors, and air quality compromising. Also, route selection methods must consider comfort features, making more pleasant urban trips. This master's dissertation proposes and evaluates two pollution-aware route selection approaches, a multi-modal hybrid routes method and a multi-criteria personalized route selection method, for urban citizens' mobility flow improvement. The hybrid multi-modal solution surpasses the single-modal, offering less expensive and less polluted trip options. The multi-criteria solution personalized profiles outperform the single-criterion choice in the same context considering all calculated route possibilities. 

With the growing scenario of the Internet of Things (IoT) and the context of low-power wide area networks (LPWAN), LoRa technology stands out as one of the most used to build communication networks for IoT applications. This communication technology uses open radio spectrum to transmit data between devices. However, LoRa technology presents a challenge in its physical layer or LoRaWAN protocol, as there is no official tool that deals with communication stability in mobility situations. To address this issue, the present research aims to develop real-time adaptive controls based on Fuzzy logic for LoRa channel optimization in mobility situations. The goal is to maintain communication between devices and, possibly, reduce response time and efficiently manage LoRa channel attributes. The research uses wireless signal quality metrics, such as signal-to-noise ratio (SNR) and received signal strength indicator (RSSI), along with spreading factors (7, 9, and 11) as inputs to the Fuzzy control. Based on these inputs, three Fuzzy controls were developed and experimentally tested in a target scenario at UFPA. The results obtained through field data collection demonstrate how LoRa technology can benefit from computational intelligence tools for resource optimization, increased reliability, and especially for mobility situations.

Society has been undergoing major transformations in the era of industry 4.0, and the growth of the internet of things has been accentuated, especially in cities, giving rise to smart cities and university campuses (smart campuses). In this work, a data transmission architecture was developed for the UFPA smart Campus, combining different protocols and communication technologies to meet the different requirements of IoT applications found on the Campus. The following technologies were used: LoRa, GSM/GPRS, MODBUS and the Dojot development middleware platform. The results were restricted, since the data were kept under observation in Dojot

This work presents an investigation directed to the analysis and design of Generalized Minimum Variance Control (GMV) with Parametric Robust Control applied to plants with interval parameters. The proposed control structure, in addition to using the GMV controller, utilizes stochastic augmentation, which adds stochastic properties to deterministic controllers, providing a higher degree of robustness in reference tracking and the ability to properly handle noise and uncertainties. The Project presented in this work uses the GMV based on the analog proportional-integral-derivative (PID) controller, which, through a “digital-analog mask”, transfers its tuning characteristics and robust performance to GMV, and in standard RST form. For this, the robust pole placement technique, integrated with the linear programming techniques solution, is used to calculate the controller parameters. Stability analysis and robust system performance analysis are performed considering factors such as load disturbance, transport delay and non-minimal phase dynamics. With this, the Parametric Robust Generalized Minimum Variance (PRGMV) controller is designed to reduce the variability in dynamics and increase the robustness of system operation for various operating conditions, considering the factors mentioned. Comparisons of this proposed technique with classical controllers such as PID, GMV and LQG are performed to evaluate the results obtained and validate the advantages of using the proposed technique. Results and comparisons are performed in a simulation environment using MATLAB software.

Desde seu isolamento em 2004, o grafeno tem se mostrado um material promissor, pois permite forte interação de portadores de carga com radiação terahertz (THz). Esta extensa interação com a luz se deve à excitação de polaritons de plasmon de superfície (SPPs) proporcionando assim diversas aplicações nas áreas de fotônica THz. Na presença de um campo magnético externo, a ressonância de cíclotron domina nos espectros THz. Neste trabalho analisamos vários modos de ressonância (dipolo, quadrupolo, hexapolo e os modos com simetria azimutal) que podem existir no disco de grafeno em um arranjo com células unitárias quadradas periódicas. Calculamos as ressonâncias com magnetização por campo magnético externo DC variando até 3T e as comparamos com o caso da mesma estrutura sem magnetização onde observou-se que a presença do campo magnético resulta no desdobramento de alguns modos ressonantes. As características ressonantes são obtidas numericamente usando o software COMSOL Multiphysics em dois estudos distintos e complementares: o regime de cálculo de autofrequência referente aos modos naturais de ressonância e o de excitação por onda plana (plane wave) com incidência normal. As condições de contorno Floquet nos quatro lados das células unitárias são usadas para simular a estrutura periódica no plano do grafeno. O lado da célula unitária (ou seja, o período da matriz) usado no cálculo numérico é A = 9µm, o raio do disco de grafeno é D = 3 µm. O problema da excitação de alguns modos por incidência de onda plana é discutido do ponto de vista da simetria circular que, para gerar os picos de ressonância neste regime, precisou ser quebrada através de um corte sutil no formato de fenda retangular de tamanho = nm por = 550nm partindo axialmente da borda. O disco é modelado como elemento bidimensional, e isso fornece uma boa aproximação da espessura atômica de uma única camada desse material. A condutividade é modelada pelo modelo semi-clássico de Drude. A discussão das propriedades do ressonador é cumprida em termos de correntes no grafeno e campos eletromagnéticos fora do grafeno. As aplicações potenciais dos resultados obtidos são rotadores Faraday e Kerr, filtros controláveis e absorvedores para circuitos fotônicos THz.

After the occurrence of the fifth generation auction (5G), carried out by the National Telecommunications Agency (ANATEL), operators started to implement the technology on Brazilian soil and it is expected that smart devices will become more connected, promoting the advancement and improvement of internet of things (IoT). However, Narrowband-IoT (NB-IoT) technology, used by 5G for IoT applications, is still not enough to meet all user requirements, with that in mind, LoRa technology emerges as an auxiliary to meet the requirements of users. In addition, the need to combat climate change generated by global warming grows, which causes an increase in the number of forested areas around the world. Given this scenario, this work aims to analyze the behavior of the LoRa signal in a suburban and densely wooded environment. For this, measurement campaigns are carried out at the Federal University of Pará (UFPA) and from the collected data a neural network model capable of reproducing this behavior is proposed. The standard model is compared to baseline models and proves to be superior in the downlink and uplink scenarios with a minimum RMSE error of 1.6623 dB for the first and 1.3891 dB for the second.

O Despacho Econômico de Carga (ELD) é um processo do sistema elétrico de potência que visa o planejamento e as operações das Usinas Termelétricas (UTE), para atender à demanda com menor custo operacional. Por outro lado, o ELD tornou-se um problema obsoleto por considerar apenas o custo de combustível na geração de energia, desconsiderando parâmetros importantes como custos ambientais e de segurança da rede elétrica. A energia produzida pelas usinas termelétricas é considerada muito cara e poluente, ou seja, possui impactos negativos para a esfera ambiental e o cenário econômico. Considerando a pressão global para redução de emissões de poluentes na atmosfera e sustentabilidade ambiental, a incorporação da geração de Energias Renováveis (ER) ou Energias Verdes na rede elétrica é indispensável. A energia solar está se tornando uma parte importante do portfólio de geração de energia em muitas regiões devido à grande redução de seus custos e aos incentivos políticos que favorecem a geração de fontes de energia limpa e renováveis. Nesse sentido, a proposta da tese é apresentar um comparativo entre diversas técnicas metaheurísticas na otimização do DEA soluções, incluindo a geração de energia solar fotovoltaico e o desligamentos dos motores menos eficientes, bem como ajudar o especialista na tomada de decisão para o funcionamento da planta geradora. Foram utilizados as metaheurísticas baseados em métodos estocásticos como: Ant Lion Optmizer (ALO), Dragonfly Algorithm (DA) e Differential Evolution (DE). Na busca por melhores soluções, foi implementado um novo algoritmo que verifica de forma inteligente a potência dos motores da termoelétrica. Com base na potência demandada para o horário, é utilizado até 70% da capacidade da geração da usina fotovoltaica e o restante é atendido pela usina termelétrica. Assim, encontra-se a potência ideal para o número mínimo de motores suficientes para atender a demanda, reduzindo o nível de combustível e poluentes na atmosfera. As simulações foram geradas no ambiente do MATLAB, utilizando um modelo híbrido composto por seis (06) Unidades Geradoras (UG) e treze (13) Usinas Solares Fotovoltaicas (USF). Por esta razão, os resultados foram comparados com os resultados das simulações de Khan, que utilizaram PSO. Neste sentido, esta tese apresenta resultados comparativos dos métodos utilizados, dentre as técnicas utilizadas, o ED foi o método que apresentou os melhores resultados, garantindo uma redução no combustível fóssil em 3,02%, correspondendo a US$ 8,557.08 e uma redução de poluentes na atmosfera de 1.42%.

Neste trabalho, é desenvolvida uma formulação semi-analítica para auxiliar no processo de projeto de antenas dipolos retangulares baseadas em grafeno com substratos de vidro. A formulação fornece diretamente o comprimento do dipolo necessário para obter ressonância em uma frequência desejada, uma vez que a largura da antena e o potencial químico das folhas de grafeno são conhecidos. Usando o método das Diferenças-Finitas no Domínio do Tempo (Finite-Difference Time-Domain: FDTD), foi realizado um grande número de simulações computacionais considerando várias combinações de dimensões de antena e valores de potencial químico, para obter os valores de referência. A formulação é baseada em: lei de escalonamento eletrostático, cálculo da constante de fase plasmônica, capacitância entre eletrodos metálicos e cálculo de suas auto-indutâncias. O método dos mínimos quadrados é aplicado para otimizar os coeficientes da formulação. Com os coeficientes otimizados, foram obtidos níveis de precisão muito satisfatórios. No intervalo de frequências entre 0,5 THz e 3,0 THz, o erro médio relativo absoluto é de 1,50%, com um erro relativo absoluto máximo de 6,77%.

Publicação:

GARCIA, M. E. C. ; DE OLIVEIRA, RODRIGO M. S. ; RODRIGUES, N. R. N. M. . Semi-analytical Equations for Designing Terahertz Graphene Dipole Antennas on Glass Substrate. Journal Of Microwaves, Optoelectronics and Electromagnetic Applications, v. 21, p. 11-34, 2022.

http://dx.doi.org/10.1590/2179-10742022v21i11335

The loss of an upper limb incapacitates an individual to do functions of grip and manipulate objects, causing disorders and limiting the capacity to do daily tasks. Therefore, in this work it is proposed a complete control system for multifunctional myoelectric hand prosthesis with the user. The proposed system presents one superior layer of decision-making, an inferior layer of prosthesis actuators control, and the integration logic of layers. In the superior layer, it is realized both the classification of 8 movements of the hand and the estimation of the grasping force. In the inferior layer, two control loops are implemented with the predictive controller MPCSS, one for the prosthesis fingers’ angular control position, and the other to control the grasping force of it. The integration logic runs intuitively to control the prosthesis, according to the user’s grasping force and movements patterns. For the development of the proposed system, this work investigated the methods of real-time classification with the neural networks MLP and ELM, and the Machine learning techniques of the KNN and SVM types. To estimate the user’s grasping force, a model in the state space with Kalman filter is proposed. In the control layers, it is utilized a multivariable predictive controller in state space formulation to control the angular position model and the grasping force model of the prosthesis. Furthermore, in the grasping force control loop, it is proposed the application of the Kalman filter to estimate and filter the system’s global force, by sensor fusion of sensors placed in the prosthesis fingers. The performance of the system was evaluated with data recorded from ten experimental healthy subjects. In the classification results, it was shown that the most viable technique and with the best performance was the SVM, that obtained a mean accuracy and  of 97%. The estimation force model, proposed in the state space with Kalman Filter, presented a Pearson’s correlation coefficient, R^2, of 0.92 ± 0,0318, and a NRMSE index of 0,277 ± 0,056, demonstrating good performance and a better option in comparison with other models previously seen in the literature. The proposed control loops presented robust performance indices and precision in the tracking of both angular position and force control. Furthermore, the Kalman filter presented in the force control loop enhanced the stability and the system performance to hold objects. The proposed system runs in real-time with a mean processing time less than 300 ms, does the classification of 8 patterns of movements of the hand, estimates the user's force, and executes the commands of gesture and force in the prosthesis similarly to the dexterity of the human hand.

Keywords: multifunctional prosthesis, real-time control, movement classification, Kalman filter, predictive control, linear regression, machine learning.

Vários países do mundo iniciaram a transição energética utilizando geração distribuída com fontes renováveis e sistemas de armazenamento com baterias, para proporcionar eficiência energética, flexibilidade, segurança operacional, sustentabilidade e redução de custos com energia elétrica. Monitoradas e gerenciadas por um sistema de gestão, as novas tecnologias têm trazido empoderamento ao cotidiano do consumidor, de modo que este possa participar ativamente das decisões de oferta e de demanda de energia em seu convívio individual ou em grupo. Sob a ótica do consumidor, este trabalho apresenta uma metodologia para desenvolvimento e gestão de uma comunidade de energia, operando com geração híbrida de energia elétrica, advinda da rede externa e da geração fotovoltaica, e sistemas de armazenamento. Dada suas características e seu perfil de cidade, um estudo de caso com dados reais do sistema de distribuição do campus da Universidade Federal do Pará (UFPA) é considerado e os sistemas fotovoltaicos e de armazenamento são modelados, com base em uma abordagem probabilística, de acordo com a irradiação solar da região e a demanda de energia da Universidade. A operação do sistema é gerenciada durante o período de doze meses, para simular as características, limitações e particularidades da comunidade de energia. Como conclusão do trabalho, é apresentado o retorno do investimento da comunidade de energia implementada na UFPA, evidenciando a possibilidade de escalabilidade de um novo modelo de negócio.

Este trabalho aborda uma técnica de otimização multiobjetivo híbrida bioinspirada, associada a uma rede neural de regressão geral como proposta para sintetizar a geometria e dimensões de uma superfície seletiva de frequência (FSS), para filtragem de ondas eletromagnéticas em aplicações 5G. Essa nova técnica híbrida associa o algoritmo bio-inspirado conhecido como Sailfish Optmizador (SFO), em conjunto de uma rede GRNN para a obtenção dos parâmetros de construção do filtro. Neste estudo o foco é destinado na aplicação do da técnica como ferramenta para o projeto e síntese da FSS, sendo ela com a forma de uma espira quadrada na célula unitária, impressa em uma placa de substrato de fibra de vidro (FR4). Os objetivos do processo de otimização consistem em ajustar a frequência ressonante do FSS para 3,5 GHz e a largura de banda de operação de 0,8 GHz. Uma Boa concordância entre os resultados simulados e medidos é relatada.

This work evaluates the yields of photovoltaic generators, using different configurations of control in load controllers used in isolated systems, being these controllers interconnected, forming a nanogrid of distribution in direct current. It contextualised the uses of microgrids and the presentation of the nanogrid used, which is installed in the laboratory of the Group of Studies and Development of Energy Alternatives (GEDAE) of the Federal University of Pará. Throughout the work, the main equipment responsible for the formation of a network in this type of system are indicated, being the load controllers, associated with energy storage systems and photovoltaic generators. In this sense, different configurations from those recommended by the manufacturers are adopted, but within acceptable limits, to increase the productivity of photovoltaic generators and thus increase the autonomy of energy storage systems. According to the results obtained, when only one generation and storage system was in operation, an increase in generation productivity was observed; when analyzing the nanogrid with all systems in operation, with different configurations, an increase was observed, however, it was difficult to identify, due to the exchange of energy present in the nanogrid, making it more advantageous to choose a single configuration for all systems.

O para-raios de Óxido de Zinco é um dos dispositivos de proteção de vital importância para os Sistemas Elétricos de Potência e os equipamentos que os compõe, uma vez que ele assegura a confiabilidade e a continuidade de operação desses sistemas a partir da sua capacidade de atenuar as sobretensões transitórias resultantes de descargas atmosféricas ou surtos de manobra. Esse equipamento é formado por blocos de resistores variáveis de ZnO, associados em série ou em paralelo, cuja curva de tensão-corrente apresenta alta não-linearidade. Tendo em vista a relevância da utilização do para-raios, faz-se necessária a investigação das condições de operação deste. Na literatura, são encontrados diversos procedimentos e técnicas destinadas a tal finalidade, entre as quais destaca-se o monitoramento por meio da medição e análise da corrente de fuga, dado que esta representa um dos fatores principais que contribuem para a degradação deste equipamento. Diante deste contexto, a presente dissertação de mestrado implementa um modelo bidimensional de um para-raios de Óxido de Zinco de 30 kV, utilizando o Método de Elementos Finitos, para obter-se as curvas da corrente de fuga, as distribuições do potencial elétrico, da densidade de corrente e a das linhas de campo elétrico, quando este dispositivo se encontra sujeito às condições de variações de tensão e poluições. Os resultados mostram um aumento das grandezas adquiridas(que aumento? Quais grandezas?) à medida que o nível de poluição se intensifica, bem como, com o aumento dos níveis de tensão.

Em muitos projetos dinâmicos e de diagnóstico de falhas de estruturas mecânicas, é necessário o conhecimento da força externa atuante. No entanto, a medição dessas forças muitas vezes é difícil ou impossível de ser realizada e, nesse caso, um problema inverso deve ser resolvido. Este artigo propõe um método de identificação de força usando a Metodologia de Superfície de Resposta com base em Design Composto Central em conjunto com o algoritmo de Regressão de Floresta Aleatória. Esse procedimento requer, inicialmente, o modelo modal em elementos finitos da estrutura forçada. Em seguida, análises harmônicas são realizadas, variando os parâmetros de forças e, então, aplica-se a Metodologia de Superfície de Resposta para a geração de um conjunto de dados contendo os valores de amplitude, frequência e localização das forças, além dos valores de aceleração de vibração em vários pontos da estrutura. Este conjunto de dados foi utilizado para treinar e testar um modelo de Regressão de Floresta Aleatória utilizado para predizer qualquer localização, amplitude e frequência da força a ser identificada, tendo como informações somente a aquisição da vibração em determinados pontos da estrutura. Os resultados numéricos mostraram excelente precisão na identificação da força aplicada à estrutura.

The Internet of Things (IoT) allows the ubiquitous monitoring of environments through sensors arranged in a certain area of interest. Such data collection generates unprecedented content of information that is presented to different algorithms that serve to assist in decision-making associated with economy, health, well-being, among others. To ensure the success of this communication chain, defined from the collection of data to the extraction of valuable decisions, it is necessary to implement an end-to-end communication. For this, the IoT makes use of the Long-Range communication technology (LoRa), which in turn guarantees wireless and cost-free communication between the sensors installed in the endnodes arranged in the area of interest and the data traffic aggregation points installed in the area to be monitored, ie the gateway. Although the solution is practical, it generates the challenges of minimizing the costs associated with the implementation of the smallest number of gateways in the area to be covered, as well as the task of planning the IoT network taking into account the optimal positioning of the gateways. Given this context and to respond to the challenges imposed by the planning of IoT networks, this work aims to propose an optimizing software for planning IoT networks based on Genetic Algorithm, Evolutionary Particle Swarm Optimization (EPSO) and Multiple Trials algorithm, in order to to minimize the number of gateways and determine the geolocation for their installation, thus aiming to guarantee the coverage of all endnodes and their respective sensors arranged in the field.

In this work we developed a protocol for the analysis of a cognitive training
(TC), in order to raise the performance in bulldozer operators of Vale S.A. This research took part
of the POAD (High performance operator’s program) Innovation Project of the Vale
Technological Institute (ITV). The protocols of the TC were based in Neurofeedback (NFB), in
order to develop the ability to self-regulate cerebral frequencies, based on
electroencephalogram (EEG) analysis. In this research, the Holo-Hilbert Spectral Analysis (HHSA)
for the study of amplitude modulation (AM) band frequency (FM) of the rhythms that compose
the cerebral frequencies. The HHSA was based on empirical mode decomposition (EMD) in two
layers. First the EEG signal has been decomposed in a series of intrinsic mode function in
modulated frequency (IMFs) and then every IMF modulated in frequency have been

Sistemas de geração solar fotovoltaica desempenham um papel essencial para a melhoria da eficiência energética de uma edificação. Nesse sentido, o presente estudo busca apresentar uma análise de balanço energético em diferentes cenários cronológicos para o prédio do Centro de Excelência em Eficiência Energética da Amazônia - CEAMAZON, entidade pública de Pesquisa, Desenvolvimento e Inovação, vinculada à UFPA, localizado no Parque de Ciência e Tecnologia Guamá. Além disso, é realizada uma breve análise em relação ao novo INI-C relacionada à metodologia de avaliação nZEB, com recente implantação de sistema de geração distribuída, destacando o conjunto de características da edificação que possam classificar sua eficiência energética e demonstrando parâmetros essenciais para certificação.

This work aims to present the implementation of a framework capable of synthesizing
data for machine learning models, demonstrate results obtained with semantic segmentation networks
trained with these synthetic data, and describe preliminary results that exemplify how this methodology
allows segmentation to be used for classification. In this work, the amount of weeds that infest certain
areas in planting zones were classified. The methodology takes advantage from the deep convolutional
neural networks, performing pixel-level calculations in the images to determine infestation percentages
and estimating when the network does not correctly define the amount of weeds in its predictions.

In the present work, the modeling of equivalent electromagnetic structures is presented, used in the analysis and description of the response of plasmonic sensors operating in the optical and terahertz (Far Infrared) ranges. Simulations were carried out to evaluate the electromagnetic response of four sensors, which operate according to the Surface Plasmon Resonance (SPR), and Surface Plasmon Coupled Emission (SPCE) phenomenon, these are: SPCE THz sensor based on DLG (Double-Layer Graphene); Kretschmann THz sensor based on MLG (Multi-Layer Graphene); Gold-based SPCE Optical Sensor; and SLG/Gold based Kretschmann Optical Sensor. The electromagnetic description of the sensors was performed by combining analytical models derived from the Multimodal method, where it is possible to calculate the intensity of the EH (Electric and Magnetic) fields, due to the incidence of a plane wave in a multilayer structure, and from the Out-of-Phase Periodic Green’s Function method, which uses the spectral representation of the EH fields, by the complex Fourier series, over the Floquet modes. In terahertz sensors, graphene is modeled as a surface impedance described by the Kubo model. In optical sensors, graphene and gold are characterized as materials of finite thickness and complex refractive index (RI), both experimental values made available by Laboratório de Síntese e Caracterização Laser de Nanomateriais, at Puc-Rio. The results show the EH field profiles in the structure of plasmonic sensors. In the analysis of the PGF of the SPCE Thz sensor, we verified the appearance of symmetrical and asymmetrical plasmonic poles in the spectral representation of their EH fields. For the Kretschmann THz sensor, we verified a sensitivity S=2.42°RIU −1 for the case with three graphene layers, in addition to a FWHM (Full Width at Half Maximum) of 5.75° for the case with five graphene layers. For the optical SPCE sensor, it was verified that the SPCE excitation occurs for the monochromatic laser source with an incidence ≥ 30◦. Finally, for the SPR sensor based on SLG/Gold, there was an increase of 2% in sensitivity when compared to the case without graphene.

In this work, it is proposed and studied all-dielectric metamaterials composed by square, rectangular, rhomboidal and hexagonal unit cells based on trimers. These metamaterials are all-dielectric periodic planar arrays of dielectric disks on a dielectric substrate, which are able to sustain two types of toroidal dipole orders (called toroidal - TO and antitoroidal - ATO), that are differentiated by the distribution of toroidal dipole moments in the trimers (symmetric and asymmetric, respectively). The structures, as well as the resonance modes, were analyzed theoretically by using group theory techniques, symmetry-adapted linear combination method (SALC), magnetic dipole moments approximation and circuit theory, for the numerical simulations was used the commercial Comsol Multiphysics software. In particular, a new theoretical approach is proposed to understand the excitations of symmetry-protected toroidal modes, through the so-called magnetic groups. The theoretical and numerical results were validated through microwave experiments in the 8 - 15 GHz range, with good agreement of the results. Due to the unique configuration of the fields of these modes strongly confined in the proposed metasurfaces, they can be considered as an efficient light-matter interaction platform for next generation technologies, with regard to enhanced absorption, non-linear switching, (bio-)sensing and in other applications of photonics.

DIMENSIONAMENTO, SIMULAÇÃO E ANÁLISE ECONÔMICA DE UM SISTEMA FOTOVOLTAICO CONECTADO À REDE COM ARMAZENAMENTO DE ENERGIA

The Driven-Right-Leg (DRL) system is widely applied to mitigate the effects of common mode voltage in biopotential amplifiers. It works as a closed-loop controller, whose objective is to reject disturbances caused by the capacitive coupling of the patient with the power line. In this work, the DRL system is evaluated from a robust parametric control point of view, with the intention of doing a more complete evaluation than the one found in the literature, measuring gain, phase and module extremal margins. The range of interval parametric variations, found in the literature, were used to describe the parametric uncertainties that disturb the studied system. Furthermore, a Lead-Lag controller was designed based on the model under parametric variation obtained, showing how both the analysis and synthesis of DRL controllers can be done with the presented theory. The results obtained were discussed in comparison with the DRL systems found in the specialized literature.

This thesis proposes an excitation scheme for multiple plasmonic resonances along the optical spectrum for use in the monitoring refractive index and to enabling the determination of the dispersion properties of the analyzed medium. The sensing configuration consists of a D-type photonic Crystal fiber whose flat face is partially covered by distinct metallic layers. Using simulations based on Finite Element Method, it is possible to demonstrate how to customize plasmon resonances at different wavelengths, allowing to measure with low cross-talk more than one parameter in a simple and direct way, such as the mean refractive index and the first or second order optical dispersion. The central aspect of this sensing configuration is to balance miniaturization with low coupling between the different localized plasmon modes in adjacent metallic nanostructures. Furthermore, the determination of the optical dispersion of a given medium in a large spectral range provides information about the concentration of the medium constituents, which is of crucial importance for the monitoring of media in real-time, such as fluids.

Fenômenos que desequilibrem o sistema elétrico de potência ocorrem constantemente, o que gera a necessidade de estudá-los com o objetivo de permitir o desenvolvimento de redes e equipamentos elétricos que se comportem de forma mais confiável e robusta diante de tais distúrbios. Estes estudos são, em geral, realizados em laboratórios especializados de Alta Tensão e Alta Potência. No decorrer dos anos, o mercado passou a demandar que estes estudos também fornecessem resultados que cumprissem padrões de qualidade certificados por órgãos oficiais. Nesse sentido, esta dissertação objetiva propor a adequação necessária do Laboratório de Ensaios em Alta e Extra Alta Tensão da Universidade Federal do Pará aos requisitos para a competência de laboratórios de ensaio e calibração definidos pela norma ABNT NBR ISO/IEC 17025, com foco no desenvolvimento e implantação de um Sistema de Gestão da Qualidade; criação de procedimentos operacionais e diretrizes de segurança; estabelecimento de um programa de calibração para os equipamentos do laboratório; e identificação das motivações, vantagens e impactos relacionados a implementação dos requisitos da referida norma, bem como as dificuldades e soluções encontradas. Para isso, primeiramente foi realizado um levantamento literário que buscou conhecer o panorama nacional de instituições de ensino superior com laboratórios acreditados sob a referida norma. Em seguida, iniciou-se o processo de implementação dos requisitos da norma, o qual se baseou na metodologia adaptada de Grochau (2011). Por fim, a experiência vivenciada permitiu observar que a adequação aos requisitos da norma em laboratórios vinculados a instituições públicas, apesar de complexa em um primeiro momento, é plenamente capaz de proporcionar diversos benefícios e vantagens, desde que sejam respeitadas as peculiaridades deste tipo de instituição como: burocracia; limitação de recursos; capacitação do pessoal envolvido e; conciliação das atividades de pesquisa e ensino com a prestação de serviço.

Artigos Publicados:

BRASIL, A, S; BRASIL, F, S; MANITO, A, R, A; NUNES, M. V. A. Processo de Acreditação de um Laboratório de Ensaios em Alta Tensão conforme a ABNT NBR ISO/IEC 17025: Implantação de um Sistema de Gestão da Qualidade e Calibração de Equipamentos. In: VIII Simpósio Brasileiro de Sistemas Elétricos 2020 (SBSE 2020). Santo André – SP.

BRASIL, A, S; BRASIL, F, S; MANITO, A, R, A; NUNES, M. V. A. Gestão de Segurança em Alta Tensão – Adequação à NR 10: Segurança em Instalações e Serviços em Eletricidade do Laboratório de Ensaios em Alta Tensão da UFPA. ISO/IEC 17025: Implantação de um Sistema de Gestão da Qualidade e Calibração de Equipamentos. In: VIII Simpósio Brasileiro de Sistemas Elétricos 2020 (SBSE 2020). Santo André – SP.

ESQUEMA DE TARIFAÇÃO DINÂMICA DE ENERGIA PARA PROGRAMAS DE RESPOSTAS À DEMANDA CONSIDERANDO A INTEGRAÇÃO DE FONTES DE ENERGIA RENOVÁVEIS

The complexity of identifying all the factors that are related to the performance of schools on the Basic Education Development Index (IDEB) is enormous. In this study, three databases were analyzed in order to identify several factors that correlate with low performance in state schools in the state of Pará. This study used educational data mining techniques to first select variables with structural characteristics in the teaching environment, comparing schools with higher and lower performance in IDEB identifying possible relationships with school dropouts. Then, the Randon Florest (RF) algorithm was used to select the most important variables that directly or indirectly impact the IDEB index. After the selection phase, the variables were submitted to the Linear Regression (LR) algorithm. The results point out that the income of students is related to the average family income in the analyzed municipalities. Next, variables related to parents' income were used to identify possible relationship between parents' schooling and students' performance. Finally, the analysis ends with the analysis of the impact of the Municipal Human Development Index (HDI) on the variables related to the students' grade, qualification, and experience of the teachers in the school environment. The results reveal that there is a correlation between the index and student learning in the classroom. On the other hand, better indexes in IDEB are directly related to the adequacy of the curriculum to the subject taught, as well as good working conditions for teachers. This study sought, through the MDE, to unveil different factors that are related to the low performance of state schools in the state of Pará. It shows that the socioeconomic situation of families who have children of school age is not only detrimental to the performance of these students, but also to their permanence in the classroom. Therefore, there must be a greater engagement of schools to stimulate the interest of students to persist in the school environment. In addition to subsidies from the government so that teachers can fully exercise their role in society.

PLANEJAMENTO DE TRAJETÓRIA PARA ROBÔS MÓVEIS AUTÔNOMOS APLICANDO UM ALGORITMO DE OTIMIZAÇÃO POR ENXAME DE PARTÍCULAS COM PICOS DE DIVERSIDADE

This dissertation aims to provide a metaheuristic approach to drone array optimization applied to coverage area maximization of wireless communication systems, with unmanned aerial vehicle (UAV) base-stations. For this purpose, two types of networks utilizing UAVs have been analyzed: a standard Wi-Fi network operating at 2.4 GHz, and a low-power wireless area network (LPWAN), both considering medium to high-density forest environments.LPWAN are systems designed to work with low data rates but still keep, or even enhance, the extensive area coverage provided by high-powered networks. The type of LPWAN chosen herein is LoRa, which operates at an unlicensed spectrum of 915 MHz, and requires users to connect to gateways in order to relay information to a central server – in this case, each drone in the array has a LoRa module installed to serve as a non-fixated gateway. In order to classify and optimize the best positioning for every UAV in the array, three concomitant bioinspired optimization methods have been chosen: the cuckoo search (CS), the flower pollination algorithm (FPA) and the bat echolocation algorithm (BA). All of these methods have a search space distribution based on Lévy / Mantegna flights, and present distinct performance results for both drone array network cases. Positioning optimization results are then simulated and presented via MATLAB, first for the Wi-Fi network and later for a high-range IoT-LoRa network. An empirically adjusted propagation model with measurements carried out on the UFPA campus was developed to obtain a propagation model in forested environments. Finally, drone positioning utilizing the propagation model corrected with measurements is compared with the positioning using the classical theoretical model, showing that the corrected model is more efficient in representing the forest environment than the classical model usually used in recent publications.

To meet the steep increase in the consumption of users and equipment connected to a mobile network, several works and researches are being proposed and developed. In Brazil, the arrival of 5G technology is expected from 2022, and which will use the 3.5 GHz frequency band. Since there are few studies available in the literature, regarding the channel behavior, its characterization of the channel is of important relevance. Thus, in this dissertation, a study carried out on the channel analysis for the 3.5 GHz frequency in a large indoor environment is presented, this scenario is in a lobby of the Val de Cans International Airport, in Belém do Pará. The measurement campaign were performed for Line-Of-Sight (LOS) and by means of channel probing the small-scale channel dispersion parameters are extracted. These parameters such as mean delay, delay spread, the coherence band of the channel and the power profile and delays were also verified. The signal was also investigated through Floot-Interception, Close-In models and their variations are applied and analyzed to evaluate the path loss for co-polarization (V-V and H-H) and cross-polarization (V-H and H-V). The methodology applied on a large scale proved to be adequate with the data for other types of environments and other frequencies found in the literature.

PREDIÇÃO DE SÉRIES TEMPORAIS DA COVID-19: UMA AVALIAÇÃO DOS PREDITORES MLP, LSTM E ARIMA

A common problem in large data sets is missing information, whether due to failure of capture sensors, loss in transport, or another situation that culminates in data loss. Given this situation, the researcher often disregards the missing data, removing them from the set. However, this exclusion can generate inferences that are not valid, especially if the data that remains in the analysis are different than those excluded. To deal with this problem in Structural health monitoring (SHM) data sets, this work makes use of Recurrent Gated Units (GRU) and Long Term Memory (LSTM) neural networks to perform the imputation task. Of missing data. In a step before imputation, the artificial amputation of the data was performed, assuming the Missing Completely at Random (MCAR) missing data mechanism, in percentages of 25, 50, and 75%. The imputation techniques were evaluated using the Average Absolute Percentage Error (MAPE) metric. Subsequently, the damage detection step was applied, as imputed bases were submitted to the Mahalanobis Square Distance (MSD) and Kernel component analysis (kPCA) algorithms to obtain the detected T1 and T2 error rates. From the results obtained, it was possible to observe that the use of LSTM in data imputation achieved better results than GRU in all amputation rates.  Better  performance can also be noticed in the selection of damage detection, where the imputed bases by LSTM achieve better T1 and T2 error detection results.

Uma metodologia de classificação de descargas parciais (DPs) com base em contagem espectral de pulsos é proposta e apresentada neste trabalho. Os dados de contagem espectral são processados usando uma técnica proposta (PD Spectral Pulse Counting Mapping - PD-SPCM), que leva a um mapa de descargas parciais resolvidas em frequência (Frequency-Resolved Partial Discharges - FRPD). O mapa proposto é então usado para detecção e classificação de DPs. Neste trabalho, FRPDs de corona e ranhura são apresentados em bandas de frequência de até 500 MHz, obtidos a partir de medições de laboratório realizadas usando duas barras estatóricas de hidrogeradores. Os sinais eletromagnéticos das DPs foram capturados usando uma antena patch e um analisador de espectros. As DPs corona e ranhura foram escolhidas para análise porque uma pode ser classificada equivocadamente como a outra pelo fato delas poderem apresentar mapas semelhantes de DP resolvidas em fase (PRPD) e podem ocupar as mesmas bandas espectrais. Além disso, as DPs corona e ranhura podem ocorrer simultaneamente. Os resultados obtidos mostram que as DPs corona e ranhura podem ser devidamente identificadas utilizando a metodologia desenvolvida, mesmo quando ocorrem simultaneamente e nas mesmas bandas espectrais. Isso é possível porque, como é demonstrado experimentalmente, DPs corona e ranhura têm níveis apreciáveis de contagem espectral de pulsos em bandas particulares do espectro de frequência.

The impact of epidemics throughout history has been devastating, posing a threat to public health and requiring immediate and effective action by authorities. The consequences are even greater when they are disseminated in underdeveloped or developing countries, resulting in scenarios with high rates of underreporting. Investigation, control, prevention and guidance measures are necessary in an epidemiological scenario. Investigating the behavior of past epidemics using historical time series data can help to define patterns and assist in more realistic predictions of new or known diseases. In this scenario, this thesis presents strategies for the analysis of epidemic underreporting, based on time series, applied in countries with social inequalities, with precarious conditions of health services, inadequate diagnostic network, which have flaws in the dissemination of data related to disease. The strategies are based on time series forecasting models, generated by machine learning techniques, used to predict the expected behavior of epidemic occurrences, revealing outliers, in order to create scenarios closer to reality and provide subsidies to public authorities for decision-making. As a way of showing the effectiveness of the proposal, the COVID-19 pandemic in Brazil is used as a case study.

This work presents a set of hybrid metaheuristics, based on the use of evolutionary strategies in conjunction with the classical and quantum particle swarm optimization algorithms under a multi-swarm environment with master-slave topology. In this context, new heuristics are used to control cluster agglomeration and mutation of particle replications (Fuzzy Mamdani inference machine), as well as the use of local search methods with convergence monitoring based on the global best history. The algorithms are called Fuzzy Competitive Evolutionary Multi-Swarm Optimization with Convergence Monitoring (FCEMSO-CM) and Fuzzy Competitive Quantum-Behavior Evolutionary Multi-Swarm Optimization with Convergence Monitoring (FCQEMSO-CM). In order to validate the results, twenty non-restrictive benchmark problems and four engineering problems will be used in several scientific publications: Welded Beam Project (WBD); Minimzation of Weight Tension/Compression Spring (MWTCS); Speed Reducer Design (SRD); Design of Pressure Vessel (DPV). The algorithms are implemented in CUDA massive parallel computing architecture, providing a more adequate data distribution in relation to the organization of the swarms, as well as the significant decrease of the processing time. With the application of the evolutionary strategies in the PSO and QPSO algorithms, as well as the heuristics of agglomeration control and local search based on convergence monitoring, the solutions proposed in this document offer several advantages, such as the improvement in search capacity and control of the convergence rate.

Through this work, we sought to investigate the performance of optical fiber communication systems with the aid of signal regeneration techniques, such as 2R (Re-amplification and Reshaping) and 3R (Re-amplification, Re-shaping and Re-timing) based om the non-linear effects of Kerr type: self-phase modulation (SPM), cross-phase modulation (XPM) and four-wave mixing (FWM). For this, the commercial software OptiSystem was used to design three communication systems with Mach-Zehnder interferometer (MZI) architecture composed of an acoustic optical filter (AOF) and highly non-linear photonic crystal fiber (HNL-PCF). The objective was achieved through the interactive search for the solution of the nonlinear Schrödinger equation (NLSE) in which the split step Fourier method (SSFM) was used, where in the simulations the power of the input signal, the link length and the frequency difference between the input and the pump signal were used as figure of merit for the analysis of the performance of the systems proposed for the transmission rate of 10 Gb/s. The results were found as a function of eye diagrams, pulse shapes and optical input and output spectra of the systems, by comparing the values of the quality factor (Q-Factor), the bit error rate (BER), eye height, timing jitter and optical signal-to-noise ratio (OSNR) of the degraded and regenerated signals. Therefore, it can be said that through these projects it was possible to carry out the 2R and 3R all-optical regenerations in a satisfactory manner and with great performance values. In short, this thesis was written and organized using three of the articles published in journals, but with the necessary adaptations, with the chapters, as follows: In chapters 1 and 2, the 2R and 3R regeneration systems based on MZI with AOF and HNL-PCF were presented, respectively, and in chapter 3 the all-optical regeneration systems with hybrid modulation and with on-off keying (OOK) modulation with return to zero (RZ) and nonreturn-to-zero (NRZ) encoding. And finally, in chapter 4, the general conclusion, future works, publications and submissions in journals and conferences, as well as the appendix with prints of articles published in journals as first author and co-author. 

This work researches and presents the results of investigations on the channel estimation problem in millimeter-wave MIMO wireless systems considering low-resolution analogto-digital converters (ADC). Three channel estimation models based on Deep Learning (DL) are presented: the matched estimation model, the mismatched model, and the Deep Transfer Learning (DTL) based channel estimation. DL models use Convolutional Neural Networks (CNN) to extract attributes from the databases generated by Raymobtime methodology for Beijing and Rosslyn scenarios. In the experiments performed in this thesis, “pilots” signals were transmitted and used as input to the DL models, which returned the estimated channel matrix as output. In total, six different experiments were carried out. The first experiment made a comparison between the matched, mismatched and DTL models alternating the scenarios of Beijing and Rosslyn as data source. The second compared different channel generation procedures including models that performed a post-processing in the ray-tracing (RT) simulations, those that incorporated MIMO antenna arrays in the RT simulations, and those that used random parameters to generate channels. The third experiment did a detailed investigation about the channel generation procedure with random parameters. The fourth considered data augmentation technique to train channel estimation models. The fifth compared model variations considering multimodal data available in Raymobtime. The sixth and last experiment compared the model proposed in this thesis with the model based on conditional Generative Adversarial Networks (cGAN). The results of these experiments highlight that the model based on DTL has a lower computational cost compared to the one using DL. Regarding the Raymobtime methodology, incorporating MIMO antenna arrays within RT simulations produces more favorable scenarios for the channel estimation task. Considering the results obtained, as a general conclusion of the thesis, it is observed the great importance of the methodology adopted to evaluate the channel estimation techniques that are based on machine learning. For example, the way channel matrices are generated greatly influences the results of experiments, along with the definition of training and test sets.

CRIAÇÃO DE UM MODELO DE CLASSIFICAÇÃO DE TWEETS EM PORTUGUÊS RELACIONADOS A CRIMES UTILIZANDO MÁQUINA DE VETORES DE SUPORTE

Esta tese de doutorado desenvolve uma análise de sobretensões transitórias muito rápidas (VFTOs), experimentadas por subestações isoladas a gás (SIGs), que ocasionam falhas de chaves seccionadoras (CSs) e disjuntores (DJs). Um modelo de múltiplas centelhas, para equipamentos de manobras em SIG, foi desenvolvido e proposto para analisar todo o processo de geração de VFTO. As simulações foram implementadas utilizando um programa de transitórios eletromagnéticos, com modelos de circuitos equivalentes para análise de alta frequência. Os parâmetros das VFTOs obtidas nas simulações computacionais, como por exemplo, componentes oscilatórias, valor de pico, tempo de subida e número de oscilações, foram consistentes com as medições de VFTOs realizadas no ensaio em campo, validando a modelagem dos componentes implementados no estudo e permitindo assim uma avaliação consistente da falha. A severidade das VFTOs obtidas por simulação computacional e observadas no DJ, foi considerada alta devido terem apresentados parâmetros consideravelmente maiores que a referência. Os resultados das simulações computacionais, indicaram que as falhas ocorreram em uma tensão abaixo da tensão suportável de impulso atmosférico com uma quantidade elevada de centelhas. Após uma análise completa da falha, considerando os resultados das simulações computacionais, registros de oscilografias e exames visuais, foi possível concluir que o fechamento manual da CS produziu centelhas excessivas e estresses anormais no dielétrico da região do campo elétrico não homogêneo do DJ, impactando desta forma na redução da rigidez dielétrica do gás SF₆ até a geração de arco elétrico de alta intensidade entre o anticorona e o invólucro.

This thesis develops an analysis of very fast transient overvoltages (VFTOs) experienced by gas-insulated substations (GISs) that caused failures in disconnect switches (DSs) and circuit breakers (CBs). A multi-spark modelfor GIS maneuvering equipment was proposed to analyze the entire VFTO generation process. The simulations were implemented using an electromagnetic transient program, with equivalent circuit models for high frequency analysis. The VFTOs parameters obtained in the computer simulations, such as oscillatory components, peak value, rise time and number of oscillations, were consistent with the VFTOs measurements performed in the field test, validating the modeling of the components implemented in the study and thus allowing for a consistent assessment of failure. The severity of the VFTOs obtained by computer simulation and observed in the CB was considered high because they presented parameters considerably higher than the reference. The results of the computer simulations indicated that the failures occurred at a voltage below the withstand voltage of an atmospheric impulse with a high number of sparks. After a complete failure analysis, considering the results of computer simulations, oscillography records and visual examinations, it was possible to conclude that the manual closing of the DS produced excessive sparks and abnormal stresses in the dielectric in the region of the CB inhomogeneous electric field, impacting this form in the reduction of the dielectric strength of the SF₆ gas to the generation of a high intensity electric arc between the anticorona and the casing.

Segue o artigo publicado em periódico A1 - Qualis CAPES:

Analysis of VFTO during the failure of a 550-kV gas-insulated substation (MOREIRA et al., 2020). Publicado na revista Electric Power Systems Research (ISSN: 0378-7796 / Qualis CAPES A1 (Engenharia IV): Fator de Impacto 3,211: JCR 2019);

Multiple-input and multiple-output (MIMO) techniques using millimeter wave (mmWave) prop-agation are essential to achieve the bit rates expected for 5G and 6G systems. This technology evolution relies on good knowledge about the propagation channels. Ray-tracing emerges as a good option among the deterministic channel modeling approaches, avoiding the high cost of performing measurement campaigns with channel sounding at such high frequencies. This dissertation presents improvements on the simulation methodology called Raymobtime, which combines two different software: a vehicle traffic simulation (SUMO) and a ray-tracing simu-lation (Remcom’s Wireless InSite), in order to generate realistic channels that guarantees some MIMO channel requirements as the spacial-time consistency, when there are smooth channel variations between closely separated users, or when they move. The improvements in the Ray-mobtime methodology developed in this dissertation include a channel characterization of cre-ated datasets, using parameters as angle and delay spread. Moreover, the work analyses a strategy to speed up simulations: using a generic setup with omnidirectional antennas in the simulation, with a post-processing stage applying the geometric model channel. The presented results show the impact of this approach in metrics such as channel capacity and channel estimation using deep learning.

In this work we study different microstrip antennas for the 3.5 GHz band of 5G. The first difference between the antennas is the use of silica (εr=4.4) as a substrate for antennas 1 to 3 and bismuth niobate doped with vanadium pentoxide ( εr  =47.8) for the antennas 4 to 6. The use of bismuth Niobate caused the miniaturization of antennas, as foreseen in the literature, being a desirable characteristic for 5G. In addition, periodic structures on the antenna substrate were modeled in antennas 2 and 5, using the plane wave expansion method (PWE), in order to improve their radiation characteristics. Similarly, we use Mushroom-like type periodic structures in antennas 3 and 6, in order to improve some results of the antenna parameters such as gain. All antennas were modeled in GNU Octave software and simulated in Ansys HFSS and CST Studio 2019 software. The best results for silica antennas were antenna 3 with 7.7 dB gain, 0.11 GHz bandwidth and -25 dB of reflection coefficient. Among Bismuth Niobate antennas, the one that obtained the best results for this application was antenna 6 with 3.6 dB of gain, 0.08 GHz of bandwidth and -20 dB of reflection coefficient.

The technological advances introduced by drastically changing the way we interact with the “things” present in the world around us, where IoT networks comply with an important role with the revolutions. In view of communication as a central point that allows all "things" to be part of this connected world, together to form a network of "Internet of Things". Wireless communication provides the benefits of mobility, making it easy to add more devices to the network and making it easy to provide any object the ability to connect to the internet. Currently, there are several wireless communication technologies developed for different needs, ranging from short range to medium and long range, through the LoRaWAN protocol, has become the focus of several and research worldwide. The LoRa network server has the ability to manage the transmission definition of each node individually to maximize battery life and network resources, through the Adaptive Data Rate, however the system needs some favorable channel conditions and measures to execute the Data Rate configuration inherent to the methodology, carried out after the confirmation of the 20 successful uplinks, that is, this configuration cannot be used in situations, where endnodes and / or gateaways are in a mobility situation. Therefore, this work is based on field studies, at the frequency of 915MHz, and aims to implement and test a fuzzy logic control, which performs a Data Rate configuration in LoRa technology in a mobile situation, in intelligent ways through taking decision and in order to optimize several aspects such as the coverage area, power supply, channel conditions, time in the air between others.

Two new types of graphene-based devices operating in the Terahertz (THz) region are suggested and analyzed in this work, being two four-port directional couplers. The first coupler has graphene waveguides tilted sideways at a 60° angle. The second directional coupler already has waveguides forming an angle of 90° between them, thus being called a directional coupler with parallel-orthogonal waveguides. The cross-section of the components has a four-layer structure, composed of graphene, silica (SiO2), silicon (Si) and a thin layer of polysilicon. The operating principle of the devices consists of the propagation of surface polariton plasmon waves (SPP's) in graphene waveguides. To carry out the analysis of the scattering matrices of the devices, we used Group Theory. For the numerical calculations we used the COMSOL Multiphysics software, such that variations of physical and geometric parameters were made to analyze how they influence the device's operation. The directional coupler with 60° laterally inclined waveguides had a signal split of -3.8 dB at the center frequency of 3.75 THz with isolation and reflection better than -20 dB at a bandwidth of 14%. The directional coupler with parallel - orthogonal waveguides presented a signal division of -4 dB at the center frequency of 4.4 THz with isolation and reflection around -17 dB and -20 dB, respectively. Device frequency bands can be dynamically controlled by changing the voltage applied between the graphene layer and the polysilicon layer.

ABORDAGEM COMBINADA USANDO FLORESTA ALEATÓRIA DE CLUSTERIZAÇÃO PARA AVALIAR PADRÕES DE DESCARGA PARCIAL EM HIDROGERADORES

This work describes a sequence of procedures for the implementation of digital controllers applied to the Parrot® AR Drone 2.0 in order to obtain two controllers that will be evaluated: a Bioinspired Controller and a controller designed through Stochastic Augmentation by Generalized Minimum Variance Control (SAGMVC). The Bioinspired Controller is acquired by means of identification via Artificial Neural Networks (ANNs) of the human control profile, which is performed by manual commands via joystick. In turn, the SAGMVC controller is designed using the synaptic weights obtained by ANNs of a bioinspired controller, which can be compared to the polynomial parameters of an RST controller. Thus, it is intended to observe the performance of controllers based on the human capacity to control systems, analyzing, at the same time, how much such systems can be improved. The main objectives pursued by the study are: the consolidation of the Stochastic Augmentation procedure as a method for improving linear controllers; in the improvement of controllers via Artificial Neural Networks; in the dissemination of optimal controllers, focusing on robustness and minimizing the energy that will be wasted; and in the design of controllers for Unmanned Aerial Vehicles. To assess the fulfillment of these objectives, the response of the controlled system, the control signal, the performance indices (ISE, ISU, TVC, IAE) and the gain and phase margins, using simulated and experimental indoor tests, were generated. From the results obtained — dividing the drone control into four dynamics: pitch, yaw, roll and altitude — it is evident that the main objectives were fulfilled, that is, the projected controllers showed improvements in performance indices, in the transient response profile and in the energy expenditure of the system. Furthermore, it was validated the hypothesis that it is possible to design an ANN Bioinspired controller whose synaptic weights are similar to parameters of RST controllers, and it was also inferred that SAGMVC showed significant improvements in system performance, compared to both simple manual control via joystick and Bioinspired control.

With the exponential increase in heterogeneous wireless networks today, there has been a growing interest from the academic community for issues related to handover problems. The main objective of this paper is to evaluate the quality of service and performance of a device with a dual interface that connects simultaneously to two heterogeneous networks with no competition in the exchange of packets between them. It is a proposal to solve the mitigation of handover impacts. The tool used for evaluation was the Network Simulator 2. The results showed a better use of the band in comparison to the scenario using a traditional mobile device.

Descargas atmosféricas diretas em redes aéreas de distribuição causam, inevitavelmente, sérios danos ao isolamento, levando frequentemente ao desligamento parcial ou total do sistema elétrico. A instalação de para-raios pode ser muito eficaz e é comumente usada para minimizar esse problema; no entanto, considerando que as redes de distribuição de energia geralmente apresentam um número muito grande de nós elétricos, o uso maciço de para-raios pode não ser economicamente viável. Dessa maneira, esta tese propõe uma metodologia para alocar pára-raios que podem reduzir significativamente o número de pára-raios a serem instalados, mas ao mesmo tempo mantendo um nível de proteção adequado para a rede de distribuição. A metodologia proposta, denominada Cruzamento de Descargas Diretas (CDD), analisa a criticidade da rede com base em dois fatores principais, que são as magnitudes das sobretensões e o número de flashovers provocados por descargas atmosféricas, e define uma função de desempenho usada para indicar o local recomendado para a instalação dos para-raios. Os estudos de simulação foram realizados usando duas redes de distribuição teste e o software ATP para demonstrar a eficácia da solução proposta, o que é confirmado pelos resultados apresentados.

The Internet of Things (IoT) paradigm connect objects by software, sensors, and several other applications to exchange information with other devices, people, and wireless systems. A Smart Campus scenario uses IoT to enable technologies that improve service performance and well-being for university campus users. A wireless communication network capable of connecting and transmitting information from devices and sensors with IoT application servers is necessary to make this scenario feasible. Low Power Wide Area Networks (LPWAN) have proven to be efficient for connecting intelligent devices. LoRa (Long Range) technology is a type of LPWAN operating in the unlicensed frequency range of 915 MHz, whose communication occurs through gateways that act as exchange points between the sensors and the central server. Optimizing the placement of these gateways and the quantity needed is essential for planning IoT networks and ensuring the quality of coverage. We propose a methodology that uses the Evolutionary Particle Swarm Optimization (EPSO), a meta-heuristic based on evolutionary and swarm intelligence, to determine the LoRa optimal gateway placement for Federal University of Pará – Campus Guamá (UFPA) Smart Campus. The UFPA propagation model, adjusted for the scenario, predicts the received signal strength and is used as an input parameter by EPSO. Four thresholds of signal reception sensitivity generated the optimal placement solutions, and we performed the solutions evaluations through the Pareto front. The stability of the meta-heuristic in finding solutions was confirmed using the Monte Carlo method. The results showed threshold sensitivity strong influence in gateways quantity and coverage area, resulting in a relevant parameter for the planning IoT-LoRa networks, which also implies in a propagation model consistent choice for the deployment scenario. The threshold sensitivity -104 dBm was more suitable for UFPA Smart Campus because forests separate the campuses, and EPSO selected centralized points with sufficient coverage radius for each campus. Based on these results, the optimal placement methodology can assist in the planning and implementation IoT-LoRa networks, reducing possible errors and, consequently, costs in a Smart Campus network design.

The research and development of wireless communication systems are accompanied by the characterization of the channel on which the signal will be transmitted. In outdoor environments there are several factors that influence the propagation of the signal, such as large buildings, the characteristics of the environment: relief, topography and vegetation, among others, causing losses due to diffraction, reflection and scattering. In the design of wireless communication systems, knowledge of the characteristics of radio propagation in the propagation environment is essential. Thus, in the literature there are several works aimed at the analysis and construction of propagation models. These models assist in the design of wireless network coverage, making it a useful tool in the design of link systems. Numerous propagation prediction methods are available, among them there is the double slope model, known in the literature as Dual Slope and the model with multiple slopes, known in the literature as Multi-Slope. One of the challenges of these models is to obtain the segmentation distance of the route, in addition to the number of segmentation points for the model with multiple segments. Several methodologies have been proposed to solve these challenges and, in this work, a methodology will be presented to optimize this search with the aid of a genetic algorithm. The genetic algorithm is one of the main research topics in Evolutionary Computing, which in turn is a branch of Computational Intelligence research, proposing a new approach to solving problems inspired by natural selection. These algorithms simulate natural population survival and reproduction processes, comprised of a set of search and optimization techniques inspired by the natural evolution of the species. In this algorithm, the favorable characteristics configure patterns that are strengthened in the successors. In turn, unsuitable traits are rarely passed on to descendants so that they can disappear from the population. The objective function of the genetic algorithm assesses how efficient the segmentation distance proposed by the individual in the population is. In view of the results obtained, it is possible to conclude that the individual who has the best accuracy allows the decay coefficients most adjusted to the segments to be obtained. Thus, the methodology proved to be efficient for obtaining the reference distance in noisy channel modeling.

A expansão dos serviços de banda larga e das aplicações de TIC representa um desafio significativo para as operadoras de banda larga, especialmente quando se considera os requisitos relacionados ao custo total de propriedade dessas tecnologias. Nos últimos anos, o processo de expansão avançou significativamente nos últimos anos, mas ainda representa um desafio a ser superado, dada a necessidade de prestar serviços de baixo custo às populações de áreas rurais e remotas. Questões relacionadas à geografia do local, baixa renda dos residentes e falta de infraestrutura pública levam a uma relação desvantajosa entre as receitas potenciais para as operadoras e os custos mais altos de implantação da infraestrutura de rede. Embora a literatura relacionada indique diversos esforços de pesquisa direcionados ao fornecimento de conectividade, faltam abordagens que considerem características específicas de tais regiões ou tenham serviços e aplicativos de rede adaptados às necessidades dessas comunidades. Assim, este trabalho propõe um framework técnico e econômico para a implantação de redes banda larga em áreas rurais e remotas, realizando o dimensionamento do custo total de propriedade da rede, considerando despesas de capital e operacionais da rede. O escopo da proposta também contempla a aplicação de técnicas de análise de viabilidade econômica para auxiliar a tomada de decisão por meio da compreensão do impacto dos investimentos financeiros realizados e dos lucros esperados pelas operadoras de banda larga. Além disso, propomos a utilização de indicadores socioeconômicos para projetar o potencial impacto social no desenvolvimento dessas regiões. Elaboramos um estudo de caso para demonstrar o funcionamento do framework proposto e seus principais componentes. Considerando dados reais de um município amazônico, mostramos, ao reduzir os custos de implantação, que é possível reduzir o custo de assinatura de serviços de banda larga para usuários finais, contribuindo assim para o processo de democratização do acesso aos serviços digitais.

“Desenvolvimento de Software para Automação da Modelagem de Redes Elétricas de Grande Porte no Alternative
Transient Program”.

ANÁLISE DE DESEMPENHO DE ESTRATÉGIAS CONVENCIONAIS E EMERGENTES PARA CONTROLE DE TENSÃO EM REDES DE DISTRIBUIÇÃO DE BAIXA TENSÃO COM PRESENÇA DE MICROGERAÇÃO FOTOVOLTAICA

A demanda de tráfego em sistemas de comunicação sem fio tem se tornado uma questão importante nas últimas décadas. Projeta-se uma tendência cada vez maior para os próximos anos, com a expectativa de que o tráfego de dados explosivo se materialize e os usuários móveis imponham novos requisitos de qualidade de serviço. Essa crescente demanda de tráfego, combinada com cenários de rede heterogênea cada vez mais complexos e densos (HetNet), apresenta cada vez mais desafios para as operadoras de rede móvel em termos de serviço, cobertura, balanceamento de carga e qualidade de serviço. Considerando o mecanismo de associação tradicional com base na potência máxima recebida, as HetNets tendem a permanecer desequilibradas, tornando um desafio atender aos requisitos de tráfego dos usuários móveis. Além disso, esta proposta de tese propõe uma abordagem baseada em aprendizado de máquina para orquestrar conjuntamente a associação de usuários e mecanismos de balanceamento de carga em HetNets. A abordagem proposta consiste em um otimizador de duas fases, onde ao invés de tentar maximizar a taxa de downlink alcançável por usuário, indicadores chave de desempenho como a satisfação da UE e perfis de QoS prioritários são considerados. A primeira fase acopla algoritmos de inteligência de enxame com métodos de agrupamento dinâmico para detectar e agrupar usuários com condições de canal ruins. Além disso, a segunda fase explora a adoção de técnicas de aprendizagem por reforço profundo para ajustar o processo de associação de usuários, para fornecer melhores condições de tráfego aos UEs. Esta fase considera as cargas das estações base e a relação sinal-interferência-mais-ruído (SINR) do equipamento do usuário, para influenciar de forma distribuída o comportamento da rede e otimizar o equilíbrio da rede. Ao considerar o uso de ferramentas de simulação, essa abordagem orientada a objetivos pode representar uma solução promissora para fornecer melhores níveis de satisfação do usuário a longo prazo.

DESENVOLVIMENTO DE UMA BATEDEIRA DE AÇAÍ EM CORRENTE CONTÍNUA E MONITORAMENTO DE SUA APLICAÇÃO EM UMA EDIFICAÇÃO RIBEIRINHA NA AMAZÔNIA SUPRIDA POR SISTEMA FOTOVOLTAICO ISOLADO

The planning and analysis of the fifth-generation mobile networks (5G) considering phenomena such as the tidal effect, is extremely important for the efficiency and quality of services provided to users. The tidal effect can be characterized as the space-time fluctuation of data traffic, inevitable in different types of networks, generating areas of intense resource overload in places/periods of constant migration of people and use of mobile equipment. To study the impacts of this phenomenon in the face of the high demand for data from 5G networks, this work presents an analysis in the Manhattan (New York) scenario through real data from mobile subscribers collected from social networks based on geolocation LBSN (Location-Based Social Network). Factors for access availability, resource allocation and load provisioning in the network were considered, aggregating the main components of the C-RAN (Centralized Radio Access Network) and D-RAN (Distributed Radio Access Network) in the proposed model. Considering the impact of the problems caused by the tidal effect, as well as the high traffic demand expected by the 5G, it was evident in the studied scenario the importance of improvements in planning, in the choice of network access architectures used, in the models for scaling/provisioning network, as well as dynamic resource allocation schemes at times of varying traffic density. The results showed that the scaling and provisioning of Small Cells (SCs) can be performed to meet the changing demands of the network, applying dynamic resource allocation schemes in times of low and high traffic density, enabling a reduction of up to 10% in quantitative number of SCs, increasing the throughput of the data flow by 3.3%, reducing the probability of blocking by 3.6%. The results also demonstrated that the analysis of the tidal effect was able to determine the hotspots in the scenario where the demands demanded greater coverage resources, connections in the SCs and/or Macro Cells (BS), as well as the areas with the highest rates of disconnected users.

The efficient protection of electric power distribution networks against lightning discharges is a crucial problem for distribution electric utilities. To solve this problem, the great challenge is to find a solution for the installation of surge arresters at specific points in the electrical grid and in a sufficient quantity that can ensure an adequate level of equipment protection and be within the utility´s budget. As a solution to this problem using ATP (Alternative Transient Program), this paper presents a methodology for optimized surge arrester allocation based on genetic algorithm (GA), with a fitness function that maximizes the number of protected equipment according to the financial availability for investment in surge arresters. As ATP may demand too much processing time when running large distribution grids, an innovative procedure is implemented to obtain an overvoltage severity description of the grid and select only the most critical electric nodes for the incidence of lightning discharges, in the GA allocation procedure. Results obtained for the IEEE-123 bus electric feeder indicated a great reduction of flashover occurrence, thus increasing the equipment protection level.

In this work, graphene nanowires are applied to patch antennas (the antenna patch is composed of graphene material). The radiation characteristics of these antennas are analyzed in the terahertz band. Antenna variations were designed to observe the influence of certain components in relation to the performance of the proposed antenna, such as pbg substrate, nanowire edges with armchair structure, nanowire edges with zigzag structure, pbg graphene and dielectric substrate. The first variation is the addition of graphene nanoribbons (armchair and zigzag) on the left and right edges of the graphene patch, in order to evaluate the influence of these structures on a patch antenna. Subsequently, air holes are inserted in the silica substrate in order to reduce the effect of surface waves and to increase bandwidth, the insertion of a graphene PBG substrate with dielectric material is also analyzed. Octave and HFSS software were used for the modeling and simulation of the antenna and its variations. The results obtained (loss of return, impedance, radiation diagram, gain and current density) were good when the structure was modeled only with silica substrate and the best nanoribbon contributed to the performance of the graphene patch antenna was that of the structure in zigzag. However, when the periodic substrate was added the antenna radiation characteristics improved considerably compared to antennas without the substrate with air holes.

In this dissertation, we present a proposal for a surface plasma resonance sensor (SPRsensor) in the surface plasma coupled emission configuration (SPCE) based on two layers ofgraphene. The numerical analysis of the sensor is performed by the Finite Element Method,where the near and far fields in the sensor structure are calculated. The model consists of amultilayer structure: Air/ Graphene/ SiO2/ Graphene/ Prism BK7, with an infinitesimalhertzian dipole operating in the Terahertz range with λ ≈ 250μmsimulating the analytes (which behave like re-irradiating particles) to be detected in the sensor’s output response.Graphene layers are modeled as current surface densities Js0 with conductivity described bythe Kubo Formula. At first, a comparative approach is made between the model proposedin the THz range and a reference model in the optical range. Then, two parametricanalysis of the near and far field are performed: one as a function of the chemical potential (μc = 0 eV, 0.3 eV, 0.4 eV, 0.5 eV e 0.6 eV) applied to the graphene layers. In the secondanalysis, the dipole’s orientation is set at θ = 0°, 45° e 90° while the dipole’s heightregarding to the sensor’s surface is changed by h ≈ 8 μm, 40 μm and 80 μm for each ofthe orientations, where we verify the coupling of TM waves in the prism region and itsimprovement in the sensor’s directivity and sensitivity, implying that graphene improvessensors in the SPCE configuration.

The fifth-generation (5G) cellular networks incorporate a large variety of technologies in order to address very distinct use cases. Assessing these technologies and investigating future alternatives are complicated when one relies only on simulators. 5G testbeds are an important alternative to simulators and many have been recently described, emphasizing aspects such as cloud functionalities, management and orchestration. This dissertation presents a 5G mobile network testbed with a virtualized and orchestrated structure using containers, which focuses on integration to artificial intelligence (AI) applications. The presented testbed uses open-source technologies to deploy and orchestrate the virtual network functions (VNFs) to flexibly create various mobile network scenarios, with distinct fronthaul and backhaul topologies. Distinctive features of the testbed are its relatively low cost and the support to using AI for optimizing the network performance. The dissertation explains how to deploy the testbed structure and reproduce the presented results with the provided code. AI-based radio access network~(RAN) slicing and VNF placement are used as examples of the testbed capabilities.

Twitter has become a great source of research data for knowledge discovery, being a social network that disseminates, in addition to personal information, it is possible to share opinions and information about events in general. Considering this, the present study aims to develop a classifier in order to analyze and identify criminal and non-criminal tweets. The Recurrent Neural Network Long Short-Term Memory (RNN – LSTM) was tested for this classification process. For the execution of the algorithm, it was necessary to create a database, since there are no studies or databases available related to the classification of news tweets in crime or not, they were collected from newspaper accounts in the city of Belém and classified manually , containing 6000 records. The results obtained with the use of RNN - LSTM proved to be quite satisfactory for the chosen domain. Based on the results of this research, we observed the effectiveness of the RNN classifier (LSTM) reached the F1 score value of 98%, accuracy of 81% and coverage of 83%. With the results obtained, it can be analyzed that the Recurrent Neural Network (LSTM) is the efficient and satisfactory classifier model for the classification process of tweets related to crimes.

Population growth and new consumer needs, among other factors, have lead to growing energy demand, without a concomitant increase in energy generation. This way, reduction and rationalization of energy consumption, especially by residential users, have become a global concern generating a need for developing techniques for efficient management and distribution of the available energy. Non-Intrusive Load Monitoring (NILM) techniques have provided valuable information about energy consumption for power generation companies as well as consumers. Such information is important  for making decisions related to sustainable use of energy resources. This study proposes an automated system based on Artificial Neural Network for performing some of the NILM tasks. A stacked neural network was developed to extract features of power signals of appliances to identify those in operation during a given period. This information is then used to disaggregate individual appliance loads through the total aggregate signal, and consumption is calculated through numerical integration. The system was tested using real data from two databases about appliances with On/Off, multi-level, and variable consumption patterns collected in low frequency. The performance metrics, resulting from identification and disaggregation tasks, demonstrate the efficiency of the proposed system.

Fronthaul networks have evolved substantially to meet the demands of fifth-generation (5G) cloud radio access networks. In particular, the industry has converged to Ethernet as the main transport technology, with fronthaul-specific encapsulation protocols running on top of it. Since traditional Ethernet is asynchronous and does not support timing distribution, clock synchronization protocols such as the IEEE 1588 precision time protocol (PTP) have gained a fundamental role in the operation of the 5G fronthaul. This work thoroughly analyzes the challenges for PTP-based synchronization in this context, with particular focus on the so-called partial timing support (PTS) scenario, where some network elements do not actively contribute to PTP’s operation. The fundamental problem of PTS is that the network imposes a harsh environment for accurate clock distribution, primarily due to the packet delay variation. As a result, non-conventional strategies have to be exploited to process PTP timing signals. This work explores various estimation strategies for this scenario. Furthermore, it describes an experimental Ethernet fronthaul setup based on field-programmable gate arrays (FPGAs), which was developed to evaluate the synchronization algorithms. In the end, it also investigates practical 5G use cases for synchronized timing signals that were implemented in the testbed.

As the structural health monitoring (SHM) process evolves from research to practice new challenges arise from their practical implementations. Normal variations in the structure's dynamics caused by operational and environmental conditions are still one of the biggest challenges for real applications of automated monitoring systems. In those cases, the normal variability can mask the existence of damage or blur the severity of damage occurrence. In the context of SHM applications, data normalization is referred to as the process of filtering these normal effects to provide a proper evaluation of the structural health condition and, although the number of approaches for normalization is growing fast, most of them still imposing serious limitations for deployment in real-world monitoring campaigns, mainly related to constraints regarding the data distribution and model parameters, which in many cases still not producing an adequate damage detection performance. On the other hand, a growing number of applications require autonomous non-contact approaches for vibration-based monitoring, pushing efforts into the development of vision-based techniques for modal analysis and modal identification, which can be coupled with machine learning algorithms for real-time dynamics monitoring. Despite some recent and impressive applications of computer vision techniques for SHM, the majority of works still relying on the positioning of speckle or high-contrast markers over the surface of the structure, which can greatly reduce their range of applicability. Therefore, as major contributions, the present thesis addresses the development and proposes novel output-only approaches for data acquisition, modal analysis and damage detection based on computer vision and machine learning techniques, capable of blindly identify high-resolution full-field vibration modes from video measurements only, detect and potentially quantify the damage level in structures of arbitrary complexity. First, machine learning algorithms are proposed for intelligent vibration-based damage detection on accelerometer data. A deep autoencoder is designed to autonomously learn the normal vibration condition of a monitored structure and then identifies damaged conditions from changes in the modal (natural) frequencies. Additionally, a cluster-based technique is introduced to classify different undamaged and damaged conditions of a structure, providing useful insights on the structural behavior. This straightforward clustering procedure automatically discovers the optimal number of clusters representing the main state conditions of the structural system without requiring any manual parameter setting. Second, techniques for blindly identify full-field high-resolution mode shapes and other modal parameters from video measurements only are introduced to performing 2D modal analysis using commercial cameras, with application on bench-scale laboratory structures and for other problems involving 2D wave decomposition. Additionally, it is introduced the first approach to date for efficient and extremely high-resolution 3D structural dynamic modal analysis from dynamic point cloud data acquired using a commercial, low-cost, time-of-flight imager. Solutions to the blind source separation problem are employed to estimate high-resolution 3D mode shapes, modal coordinates, and resonant frequencies using a modified version of the technique employed for the 2D case. It is demonstrated that the proposed approaches have the potential of being general-purpose ones, capable of performing modal analysis and monitoring of amorphous structures. All proposed techniques are validated on laboratory experiments and real-world vibration monitoring datasets to demonstrate their ability to perform the monitoring of varied forms of structures under different conditions and scenarios. Keywords: Video-based dynamics monitoring, Non-contact measurements, Damage-sensitive feature extraction, Stacked autoencoders, Blind source separation, Output-only modal analysis, Damage identification

MODELOS DE RÁDIO PROPAGAÇÃO BASEADOS EM MACHINE LEARNING USANDO PARÂMETROS GEOMÉTRICOS PARA PERCURSO MISTO CIDADE-RIO

A evolução da computação tem possibilitado avanços substanciais em pesquisas relacionadas à engenharia e em projetos industriais. Nestas áreas, o emprego de ferramentas computacionais tem se intensificado para simulação e obtenção de determinados parâmetros do projeto. No entanto, a crescente demanda por precisão e o aumento gradativo da complexidade das estruturas e sistemas, resulta num processo de simulação cada vez mais demorado, pois a avaliação de um único critério pode consumir várias horas, bem como vários dias ou até mesmo semanas. Logo, um método que minimize o tempo de simulação e otimização, pode, assim, economizar tempo e dinheiro. Nesse contexto, a computação bioinspirada (bioinspired computing - BIC), se apresenta precisa e eficiente, onde muitos métodos computacionais tradicionais falham e, consiste em novo mecanismo para suprir tais dificuldades. Assim, neste trabalho, é realizado um estudo acerca de alguns dos algoritmos BIC mais utilizados na atualidade para projeto e otimização de problemas gerais na engenharia e na indústria. Doravante, se vislumbra desenvolver um código de otimização meta-heurístico multiobjetivo que apresente menor custo computacional e, consequentemente, menor tempo para processamento dos dados. Inicialmente, é realizada uma investigação eletromagnética da superfície seletiva de frequência do tipo losango planar estudada, através de simulações computacionais. A análise numérica é feita usando os métodos FEM com o auxílio de software comercial. O processo de síntese consiste em sintonizar a frequência de ressonância da estrutura e a largura de banda para operação em sistemas de comunicação via satélite, mais precisamente na Banda-X, e Ku. A modelagem das estruturas é realizada por uma rede neural artificial e o processo de otimização é realizado por algoritmos meta-heurísticos. Os resultados obtidos por esses códigos são comparados aos simulados pelo software comercial e aos medidos. Observou-se boa concordância entre os resultados simulados e medidos, bem como uma substancial redução no tempo de processamento da estrutura otimizada.

UM FRAMEWORK INTELIGENTE BASEADO EM PREVISÕES DE QOE PARA O BALANCEAMENTO DE CARGA EM REDES 5G

PLANEJAMENTO DE IMPLANTAÇÃO DE REDES MÓVEIS DE NOVA GERAÇÃO CONSIDERANDO ASPECTOS DE RÁDIO E TRANSPORTE

PROPOSTA DE PROCESSO PARA AVALIAÇÃO DE DESEMPENHO DE RECURSOS HUMANOS UTILIZANDO TÉCNICAS DE INTELIGÊNCIA COMPUTACIONAL: UMA ABORDAGEM UTILIZANDO CLASSIFICADORES BASEADOS EM REGRAS E ALGORITMOS DE APRENDIZADO SUPERVISIONADO

Cloud Radio Access Network (C-RAN) is an architecture to decrease implementation costs of mobile networks deployments, in which the traditional basestations are split into the Baseband units (BBUs) and Remote Radio Units (RRU). BBUs are located in datacenters and perform most of the digital signal and network processing. Several BBUs are located in a datacenter, which allows decreasing electricity, connectivity and thermal conditioning costs. RRUs are located at the cell sites and perform a few or no processing, which comprises transmission and reception of signals to interface with users’ equipment. In C-RAN a network segment called fronthaul connects the BBUs and the RRUs, in which, in most of the cases, CPRI protocol is used and allows transporting baseband signals to be transmitted and received baseband signals, resources to synchronize in frequency and time BBUs and RRUs, and control data.

A consequence of the evolution of the mobile networks is the decrease of the cell radius, due to several factors, such as, the transmission in higher carrier frequencies. This process is known as densification of the mobile networks. Besides, next generation mobile networks have more dependence on multiple antenna techniques, and it results in a significant increase in the fronthaul requirements to transport the signals. Because of that, there is a concern the fronthaul technologies evolution do not follow the evolution of the mobile networks. Therefore, there is interest from industry and academy parties on technologies to make the fronthaul cheaper and more efficient. The result is that many works have investigated technologies for the fronthaul.

In this context, this thesis investigates digital signal processing techniques that allows more efficient signals transport in fronthaul, for which the transport of frequency-multiplexed analog signals is studied. This form of transmission contrasts with the transport time-multiplexed digitized signals, which is adopted in current technologies such CPRI. The transport of analog signals has potential to decrease significantly the fronthaul bandwidth required to transport the signals in relation to the time-multiplexed digitized transport. However, analog transmission has a drawback of causing distortion in the signals. Specifically, this thesis investigates digital signal processing techniques to allow efficient multiplexing and demultiplexing to be deployed in the BBU and RRU. These techniques are evaluated in relation to the computational complexity and the distortions. It is also distortions caused by the transmission in fronthaul, for which it is proposed a procedure to estimate the signal-to-noise ration of the fronthaul. The signal-to-noise ratio is applied to calculate a pre-emphasis that is applied to the signals before transmission in fronthaul. The techniques are evaluated in experiments of transmission in an optical link. Results are shown as measurements, distortion estimation, and complexity of the algorithms.

In this work, a small hybrid photovoltaic-wind water pumping system is modeled and analyzed through computational simulations. In the proposed system, photovoltaic generator and wind generator are connected via a two-input DC-DC converter, whose output is directly coupled to a DC motor-pump. The two-input DC-DC converter makes the use of a battery bank optional, and acts on the sources drained currents simultaneously or individually to operate them at desired points of their characteristic curves. The simulations demonstrate good performance of the system, which it is feasible to be implemented in practice and can be used in remote regions or where necessary.

5G networks have arrived and are able to shift paradigms within the areas of connectivity, maintainability, scalability and availability. Their aim is to ensure that users can remain online with their devices at any time and in any place. In light of this, solutions for centralized networks are becoming attractive since they are manageable and low-cost. In these conditions, the use of computing techniques such as fuzzy logic benefits the supervision and planning of networks by optimizing and controlling resources as well as managing the system. However, planning a centralized network tends to raise challenges in the optical sectors that are located between the telecommunications center and the base station. Technologies such as radio over fiber are being examined to meet the demands in the fronthaul market, although they raise many other challenges. For this reason, in this study, we have adopted an intelligent strategy to determine in a balanced way what radio over fiber signals should be allocated in the optical sector. This strategy is implemented using fuzzy logic and may be used in decision making to plan the future of centralized networks. The results show that it is possible to make tradeoffs between the capital expenditure and performance of the system.

MITIGAÇÃO DE VARIAÇÕES DE TENSÃO EM UMA INDÚSTRIA CERVEJEIRA USANDO BANCO DE CAPACITORES FIXOS E CHAVEADOS

Com o crescimento da e-health, os dispositivos vestíveis destacaram-se devido à sua praticidade e conforto na detecção dos dados pessoais. Dispositivos vestíveis em e-Health fornecem acesso de fácil uso, bem como um retorno de informações para o usuário. Em geral, esses dispositivos possuem uma variedade de sensores que capturam várias informações do ambiente e do usuário. As informações mais populares coletadas por smartwatches e pulseiras é sobre a medição de batimentos cardíacos, passos, oxigenação e fotopletismografia (PPG) e eletrocardiograma (ECG). Cada processo com sensores tem ruídos, ondas eletromagnéticas e movimentos que podem interferir ao identificar, analisar e verificar o indivíduo. A partir disso, a filtragem é um passo indispensável em qualquer processo. A identificação é dividida em etapas, e o principal e essencial é a filtragem porque é onde o pré-processamento começa. Esses dispositivos vestíveis dependem de dispositivos móveis para autenticação do usuário. Se o usuário precisar de validação, ele recorrerá a métodos tradicionais em outros equipamentos que possuem sensores de reconhecimento, como íris, rosto ou impressões digitais. Para superar esses problemas, esta dissertação é dividida em duas etapas. A primeira trata do pré-processamento do sinal (captação e filtragem). Segunda etapa trata do modelo de autenticação de pessoas a partir de sinais fotopletismografia (PPG) e eletrocardiograma (ECG)

The design of automatic inspection systems for the inspection of train wagons and railway  components that can cause derailment is a crucial factor for increasing productivity and safety in industrial environments. We propose a method for the detection and tracking of components of mining train wagons using a detector based on Convolucional Neural Networks and Deep Learning. In our experiments, our focus was on detecting the so called wagon's “super structure” , but the technique can be easily extended to detect other components.

Dois modelos FDTD (Finite-Difference Time-Domain) para solos dispersivos e ionizados, respectivamente, são propostos para representar realisticamente as respostas transitórias dos sistemas de aterramento. O modelo para solos dispersivos é desenvolvido inserindo a dependência da condutividade e permissividade elétrica com a frequência na equação de Maxwell-Ampère através do aproximador de Padé para o termo σ+jωε. A partir disso, são obtidas equações de atualização explícitas para o método FDTD. Ainda, para validar o modelo de solos dispersivos, foram realizados testes experimentais em sistemas de aterramento. No modelo FDTD para solos não lineares, são modeladas quatro fases para a ionização do solo: sem ionização, presparking (primeiro estágio da ionização), sparking (segundo estágio da ionização) e desionização. Mecanismos identificados experimentalmente na literatura para ionização da água são associados com as fases de presparking e sparking da ionização do solo. Com base nisso, deduziu-se que a ionização fraca do gás (fase de presparking) presente dentro das bolhas produzidas devido aplicação de correntes elevadas e a ionização da solução aquosa (fase de sparking) ocorre durante o processo de ionização no solo. Uma função matemática é proposta para representar o crescimento suave da condutividade elétrica em um ponto do solo durante a fase de sparking. Também, sugere-se que, quando os efeitos associados ao crescimento limitado da quantidade de cargas livres e suas mobilidades restritas são consideradas, a resistividade residual deve ser adotada não somente na fase de presparking, mas também durante o estágio de sparking. O modelo FDTD para solos ionizados é validado através da reprodução numérica de experimentos publicados na literatura. Os resultados obtidos através dos modelos FDTD desenvolvidos neste trabalho apresentaram excelente concordância com os respectivos resultados experimentais.

Copper-based technologies have been using twisted-pair cables as transmission lines, i.e., exploiting the TEM (Transverse Electromagnetic) mode to transmit data. However, the use of ever-higher frequencies drastically increase the attenuation of the TEM mode, limiting the reach and the achievable data rate of the system. On the other hand, a recent work propose a new approach for exploiting the transmission capacities of those cables: use their high-order guided modes. These higher-order TE (Transverse Electric) or TM (Transverse Magnetic) modes arise when the wavelengths associated to the operating frequencies are smaller than the cross-sectional dimensions of the cable. When using twisted pair cables as waveguides, the possibility of achieving transmission rates of the order of Terabit per second was envisaged, assuming that the model of the surface guided mode of a single bare-wire, known as the Sommerfeld line, properly describes the attenuation of the higher order guided modes of a twisted pair. However, it is evident that Sommerfeld model is unappropriated since the pairs of a copper cable exhibit a physical structure much more complex. This work investigates the propagation of various guided modes in twisted pair cables, from numerical simulations in HFSS software, considering real constructive characteristics such as pair twisting and conductor and insulator materials. The use of higher-order guided modes on coaxial cables is also evaluated. Based on the results obtained, it is also determined the aggregate data rate of each cable as a waveguide via Shannon’s capacity. The results show that there are around 20 propagation modes available on each cable up to 300 GHz. Data rate results show that, for 10 meters, it is possible to reach more than 1 Tbps on just one twisted pair, up to 3 Tbps on a four-pair twisted cable and approximately 0.5 Tbps in a coaxial cable.

The development of data transmission standards and access technologies has always been linked to the need to supply high data rates. Thus, technologies already implemented, such as the DSL (Digital Subscriber Line) system that uses twisted pairs for transmission, have been adjusting to support this growing data demand. However, twisted pairs, as well as all physical medium of propagation in DSL systems, are used as a transmission line, using the transverse electromagnetic mode (TEM). However, the physical limitations of the twisted pair as a transmission line prevent the system from reaching speeds in the ranges of Terabit per second. So, recently, the use of twisted pairs as a waveguide, as an alternative to the transmission line mode, is already the object of study and brings with it the possibility of further increasing the data rate in this available infrastructure. However, one of the challenges for this implementation is the coupling of the guided modes in the twisted pair, which are very complex structures compared to the most used waveguides. In this scenario, this dissertation aims to evaluate the coupling of the radiated signal between an antenna with radial propagation and the twisted pair cable, analyzing which antenna structure is more compatible with the twisted pair cable. Considering that neither the structure of the twisted pair as the twisted pair cable has a closed analytical model for the behavior of the electromagnetic field in these structures, all the results were obtained from numerical simulations in the HFSS software (High Frequency Structure Simulator), in terms of the antenna scattering parameter and the intensity of the electric field in the twisted pair. With these parameters, it was possible to evaluate the coupling efficiency between the antenna and the waveguide, as well as to evaluate the signal guidance and to predict the transmission rate available to users. In these evaluations, it was highlighted the coupling of the antenna structure with the shielded twisted pair cable, which reached levels of up to 78.85% at frequencies of 0.3 THz, in addition to confirming that despite the use of these cables for data transmission is not as effective as shown in the literature, it is still feasible, since it has been shown from numerical simulations that signal guidance throughout its structure is possible.

O aumento no volume de serviços e aplicações móveis, além do crescimento acelerado das demandas de acesso sem fio, representam desafios significativos para a próxima geração de redes móveis, a quinta geração (5G). Esse aumento no volume de aplicações é reflexo do crescente número de dispositivos que estão conectados à rede, consumindo de maneira desordenada e gerando alta carga de dados. Outro ponto de grande impacto no comportamento do fluxo de dados é a migração diária em larga escala de pessoas nos centros urbanos, causando o chamdo Efeito de Maré. Este promove a flutuação espaço-temporal do tráfego ao longo do dia, causando um problema para o controle e gerenciamento da rede. O fenômeno de marés provoca ainda outras adversidades, dentre elas há a baixa eficiência no uso dos recursos de hardware, o desbalancemento de carga, a subutilização de recursos e a ociosidade de capacidade de rede. Com base nestas informações e considerando o conhecimento dos operadores de serviços sobre mobilidade de assinantes, dados do movimento de usuários da cidade de Nova York foram extraídos através de uma LBSN (Location-Based Social Network). Considerando a alta demanda de tráfego esperada para o 5G e os problemas oriundos do Efeito de Maré na arquitetura de rede atual, nesta dissertaçaõ é proposta uma heurística com duas abordagens de provisionamento de recursos de hardware (uma baseada na taxa de transferência agregada e outra no número de usuários conectados). Os resultados apontam que o provisionamento de rede atendeu à variabilidade do tráfego do cenário utilizado, minimizando a Probabilidade de Usuário Bloqueado, maximizando a eficiência da Unidade de Banda Base (BBU) e quantificando as Small Cells (SCs) necessárias para atender à demanda dos usuários.

The techniques of dynamic systems identification are algorithms of most importance for generating mathematical and computational models capable to represent the dynamic of systems and processes present in many fields of society, such as: industrial processes; automobiles; food production; aerospace vehicles; biological systems and etc. The identification of these systems, which generally have more than one variable of input and output (multivariable systems) and also are nonlinear, it is very important for science and engineering in relation to the development of new control techniques, fault monitoring and prediction of operating state of these mechanisms. Nonetheless, the identification of nonlinear MIMO (Multiple Input Multiple Output) systems is a hard task, as much due the difficulty of implementing the classic algorithms for solve this problem, as the fact that nonlinear systems require complex models for represent their dynamics in satisfactory way. In order to solve this problem, this work proposes a framework capable of performing as much the identification of nonlinear multivariable dynamic systems in fuzzy TSK model, which can represent in simple way the coupling among the variables involved in identification, as the selection of regressor vector used in model. The framework is tested and compared with RNA and a Hammerstein-Wiener model in identification of two nonlinear MIMO industrial plants:  Continuous Stirred Tank Reactor (CSTR); Industrial Dryer. The comparison of these three techniques is made with base in indices of Mean Squared Error () and Variance Accounted For (), further the analysis of residues between the observed and estimated data. The results show that the proposed framework got the best performance, based in the two indices, in 80% of outputs estimation of the two multivariable plants, and also reached the best performance in 60% of residual analysis of plants identification.

Nos últimos anos diversas meta-heurísticas foram utilizadas para solucionar o problema de restauração de Sistemas de Distribuição de Energia (SDE) de forma eficiente. Dentre elas destaca-se o Algoritmo Evolutivo Multiobjetivo por Tabela (AEMT), que juntamente com a utilização da Representação Nó-Profundidade (RNP), trouxeram um grande avanço na área pois a RNP sendo utilizada como estrutura de dados é possível realizar modificações na topologia do SDE sem perder a radialidade e restabelecer o fornecimento de energia para todas as cargas desligadas após a falta. Devido a falta de exploração de outras metaheurísticas que utilizam a RNP como estrutura de dados, este trabalho visa realizar uma análise de desempenho comparando o AEMT com outras três meta-heurísticas, a Busca Tabu, Colônia de Abelhas e Estratégias evolutivas. A análise foi realizada em 3 SDEs com 84, 119 e 135 barras e foram simuladas 3 faltas em cada sistema.

The introduction of new technologies as well as the improvement of materials and processes has contributed to accelerate the growth in the efficiency of the electric induction motors (IM). Currently in the market it is possible to observe the operation of IM classes IE3 and IE4, as well as minor efficiencies, in the same way proposals for IM class IE5 are being developed. This work aims to show the impact of 2nd, 3rd, 5th and 7th order voltage harmonics on the temperature and performance of IE2, IE3 and IE4 induction motor classes, the latter being a hybrid motor of permanent magnets and squirrel cage (Line Start Permanent Magnet Motor, LSPMM's). The measurements were divided into two stages, first feeding each of the motors with undistorted voltages and then entering each of the harmonics individually and in combination in percentages of 2% until reaching 25%. The results showed that better efficiency classes present considerable improvements in relation to consumption, temperature and noise, however, also showing non-linear characteristics, as was the case with the LSPMM. It is also presented as the presence of individual harmonics in electric motors, results in the appearance of other harmonics, according to the percentage of distortion present. Finally, in order to predict the variation of temperature in relation to the percentage of different voltage harmonic distortion, statistical models were created resulting in good approximations for the temperature increase in the presence of voltage harmonics.

The production of primary aluminum is carried out in several factories around the world. With the technological advances in the last decades, the high power of storage and data processing has allowed to solve problems that previously were considered extremely difficult. In this sense, a new paradigm emerges: industry 4.0, which is based on intelligent processes, integrated management, managed energy, high standards and quality. Many of the challenges that permeate industry 4.0 include machine learning techniques that facilitate data interpretation, enabling models to emulate the behavior of the production system with high accuracy, and to find hidden patterns in the huge data set. This PhD qualification proposes models based on machine-learning techniques (Artificial Neural Networks, various learning algorithms) and deep learning (Long-Term Memory Networks) to emulate the behavior of temperature, fluorided alumina and alumina concentration of a furnace reduction of primary aluminum from several inputs, using real data. In addition, the clustering of furnaces with similar behavior was carried out with the aim of increasing the accuracy of the mentioned models. The results indicate that the models based on Long-term Memory Networks are 20 to 25% more accurate than those based on Artificial Neural Networks. Additionally, it was verified that the clusters found do not significantly improve the model's accuracy. Further tests are still needed to confirm this proposition. It is important to mention that an extensive review of the literature was made through a systematic review that considered hundreds of scientific studies that approach modeling in the area of primary aluminum production.

Growing demand for higher data rates and better quality of Internet services has resulted in high investments in mobile network infrastructure by operators. In this context, Centralized Radio Architecture (CRA) is a promising solution that relies on centralizing, sharing, and better-allocating network resources, resulting in reduced deployment and operation costs when properly planned. While the benefits of CRA are numerous, this cost reduction can also be enhanced through the adoption of alternative energy sources. One of the options for this is the use of solar energy through photovoltaic systems, which adopts the sizing of its basic components, such as panels and inverters. However, one of the main barriers to the concrete use of such a system is its high cost of implementation, which cannot be overlooked. Therefore, the proper disposal of such equipment through optimization approaches, considering the energy demand of CRA , represents a challenge to be overcome. To this end, this paper proposes a strategy to minimize the cost of implementing a photovoltaic system by reducing the number of inverters, which is formulated as a Linear Integer Programming (ILP) problem, with the aim of further reducing costs related to CRA. From the results, it is evident that the optimization technique used implies the reduction of the Total Cost of Ownership (TCO) of the photovoltaic system, as well as the environmental sustainability through the reduction of carbon dioxide (CO₂) emissions in the atmosphere

In this work, a study is made for the different cases of optical nanoantennas. At nanoantennas considered are: dipole, dipole-loop, isolated loop and nanorod-loop, initially the transmission mode is analyzed to obtain parameters important for the study of nanoantennas, such as input impedance, coefficient of reflection, radiation efficiency, gain diagram. Then for analysis in the mode nanoantennas are excited by a linearly polarized plane wave, varying the direction of polarization. Where the power received by the load versus frequency is The nearby electric field is investigated for all four cases. The following is the application of an optical nanocircuit, which is composed of a circular nanoantenna of reception connected to a bifilar Optical Transmission Line (OTL) on one side, and a dipole on its terminal. In this case, a plane wave is the source and the near field are investigated, it is also made for analysis comparison mode optical nanocircuit separately for connected dipole and loop cases OTL with a load, the power received by the load versus frequency is calculated varying the polarization direction of the plane wave. And finally to get results closer to realistic applications an aperture probe with a beam is applied gaussian focus on nanocircuit. Numerical analyzes will be performed by the Finite Element Method (FEM).

A new type of four-ports circulator is suggestted and analyzed in this paper. The cross section of the componentes features a three layer structure consisting of graphene, silica and silicon. The plane of the circulator figure consists of a graphene circular resonator and four waveguides connected to it. The graphene resonator is normally magnetized in its plane by an external DC magnetic field. The physical principle of the devide is based on the dipole resonance of the magnetizes  graphene resonator. We investigated the influence of different parameters on the characteritics of the circulator. We use Coupled Mode Theory to prove the numerical calculations obtained through computer simulations, the device obtained approximately - 17 dB isolation and -3 dB insertion losses with 4.55 THz center frequency with 1 T polarization DC magnetic field, where the center frequency can be controlled by altering the fermi energy of graphene.

In this work is presented a hybrid bioinspired optimization technique that associates a General Regression Neural Network (GRNN) with the Multiobjective Bat Algorithm (MOBA), for the design and synthesis of the Frequency Selective Surfaces (FSS), aiming its application in data communication systems by diffusion of millimeter waves, specifically, in the IEEE 802.15.3c standard. The designed device consists of planar arrangements of metallizations (patches), diamond-shaped, arranged over a RO4003 substrate. The FSS proposed in this study presents an operation with ultra-wide band characteristics, its patch designed to cover the range of 40.0 GHz at 70.0 GHz, i.e., 30.0 GHz bandwidth and 60.0 GHz resonance. The upper and lower cutoff frequencies, referring to the transmission coefficient’s scattering matrix (dB), were obtained at the cutoff threshold at -10dB, to control the bandwidth of the device.

In this work, we present a theoretical analysis of wireless optical nanolinks formed by plasmonic nanoantennas, where the antennas considered are Yagi-Uda and cylindrical nanodipoles made of Au. The numerical analysis is performed by the linear method of moments, where the transmission power and the near electric field are investigated and optimized for three nanolinks: Yagi-Uda/Yagi-Uda, Yagi-Uda/dipole and dipole/dipole. Some results are also obtained by the Finite Element Method. The results show that all these case can operate with good transmission power at different frequencies by adjusting the impedance matching in the transmitting antennas and the load impedance of the receiving antennas.

The most commonly used way to assess the stator isolation condition in hydrogenerators is monitoring partial discharges (DPs). In this work, we present a system for DP pattern classification using a new approach called Probabilistic Self-Organizing Map. Several literature techniques have been combined for preprocessing and pattern visualization. The methodology proposed obtains the separation boundaries on the map that maximize accuracy and automatically determine the probabilities for each type of Dp pattern.

SMARTLVGRID - UMA PLATAFORMA APLICADA À CONVERGÊNCIA SMART GRID DE CIRCUITOS LEGADOS DE BAIXA TENSÃO

The area of physical therapy rehabilitation always aims to improve the quality of life of patients, but the process of physical therapy can be considered tedious and tiring. Considering this problem, the Serious Games are presented as an aid tool in the recovery process of patients undergoing physical therapy treatments, which use virtual reality as a motivating element in the patient recovery process. The present work presents the design and evaluation process of a serious motor rehabilitation game which was evaluated by professionals and patients and presented relevant results in the criteria of ease of learning, efficiency, level of inconsistencies, user satisfaction and ease of memorization. through the use of the System Usability Scale (SUS). The Serious Exergame Utility - Questionnaire (SEU-Q) was also used to evaluate the perceived usefulness of the proposed Serious Game, which helped to identify important aspects related to the benefits of its use.

A Study on Changes in Land Use and Land Cover in Settlement Projects in the Amazon and their Impact on Native Forest Degradation.

Millions of Brazilians are affected by epilepsy. The diagnosis of patients with epilepsy is critical for initiating appropriate treatment. However, the diagnosis relies on visual inspection of neuronal electrical activity, recorded by electroencephalography (EEG), by trained neurophysiologists. Due to this, this process is time consuming and may require days of continuous EEG recording, which makes the diagnosis costly. Hence, the present dissertation proposes a framework for automatic identification of the EEG of epileptic subjects. The method can be applied to short-term records of resting-state EEG. The proposed system combines the use of a feature extracted from the power spectral density of EEG signals and machine learning algorithms. The attribute used is an estimate of functional connectivity between EEG pair of channels, named debiased weighted phase-lag index. The algorithms used for classification were linear discriminant analysis (LDA) and support vector machines (SVM). EEG signals were recorded during the interictal state, i.e., the period between seizures and had no epileptiform activity. In order to test the method proposed, records of 11 epileptic patients and 7 healthy subjects were used. The algorithms used reached their maximum performances, 100% accuracy and unit area under the receiver operating characteristic (AUROC), when a feature vector with 190 attributes was used as input. The results show the effectiveness of the proposed system, given the high segregation capacity of the groups.

The prediction of visual quality is crucial in image and video systems. Image quality metrics based on the mean square error prevail in the field, due to their mathematical straightforwardness, even though they do not correlate well with the visual human perception. Latest achievements in the area support that the use of convolutional neural networks (CNN) to assess perceptual visual quality is a clear trend. Results in other applications, like blur detection and de-raining, indicate the combination of information from different scales improves the CNN performance. However, to the best of our knowledge, the best way to embody multi-scale information in visual quality characterization is still an open issue. Thus, in this work, we investigate the influence of using multi-scale information to predict image distortion. Specifically, we propose a single-stream dense network that estimates a spatially-varying quality metric parameter from reference image. The proposed method achieved a reduction of 36.37% and 69.45% for the number of parameters and FLOPs, respectively, and its performance is compared with a competing state-of-the-art approach by using a public image database.

Today, electricity generation and transmission utilities are paid for the availability of their transmission functions (FTs) and no longer for what is generated and transmitted in the National Interconnected System (SIN). In this context, more and more efforts are made by generating and transmitting utilities so that untimely disconnections do not occur or be avoided. In addition to the loss of revenue when unplanned outages occur in the system, utilities are subjected to a rigid oversight process by regulators agents, such as the National System Operator and the National Electric Energy Agency. Investments in technology are allowing new directions for the operation and maintenance of power equipment, since high-capacity data-processing devices are enabling optimized predictive maintenance techniques based on the protection, control and monitoring of electrical quantities of the electrical system, as well as guaranteeing the speed and security of information. Intelligent Electronic Devices (IEDs) are multiprocessor systems with hardware and software that continuously work with electrical quantity measurements, protection, command, control, monitoring and have robust memories to record lists of events and waveforms of the analog signals in real time. In this work, real cases will be presented in which FTs shutdowns were avoided with the analysis of the lists of events and waveforms, which allows utilities to anticipate the problem, have the best decision making and significantly reduce its financial losses in the operation and maintenance of its electrical system.

Computer Numerical Control (CNC) is a technology made up of several blocks. Among these, lies the Trajectory Planning block, responsible for reference profile generation that are fed to position control loops. The need for Trajectory Planning arises from the mechanical constraints inherent to every plant to which CNC technology is applied. The machine's operational limits must be respected, in order to avoid several issues, such as: loss of precision, early wear of machine's parts and excessive vibration. This paper proposes a novel smooth real-time trajectory generation setup based on an embedded system platform. A real-time snap and jerk bounded control algorithm is proposed, to achieve continuous and smooth feed motion in traditional Numeric Control code file, dealing both with straight lines and arcs. A local motion blending algorithm, applicable to the proposed method, is also presented. The developed algorithm was deployed to a BeagleBone Black, an embedded System-on-Chip, single board computer and tested in a prototype router machine. A comparison between the proposed method against the seven segments and trapezoidal acceleration methods is presented, both in terms of performance and of real-time computing viability. Simulation and Experimental results demonstrate the effectiveness of the proposed method to generate velocity, acceleration, jerk and snap bounded three dimensional trajectories, reducing the RMS error in up to 8.2% and 22.38% when compared to the Seven Segments and to Trapezoidal Acceleration methods, respectively. Assessing the error area on straight angles, the proposed method produced error areas 24% and 80% smaller when compared to the Seven Segments and to Trapezoidal Acceleration methods, respectively.

The concept of direct current distribution minigrids has been gaining ground in academia and industry regarding the development of distribution grid applications with high penetration of distributed energy sources and storage systems. The adoption of a direct current distribution system facilitates the integration of sources such as photovoltaic and wind generation and storage systems such as batteries, as these technologies operate intrinsically in direct current. In this sense, this work presents the development of a direct current distribution nanogrid installed in the laboratory of the Group of Studies and Development of Energy Alternatives (GEDAE), of the Universidade Federal do Pará. The developed grid is composed by three PV generation systems and storage in battery banks and three load banks, distributed over the 200 m grid in a ring topology, on a 24 Vdc bus. Two simulation methodologies were developed and are capable of reproducing the nanogrid’s operational behavior under static and dynamic conditions, allowing the evaluation of the grid performance over a day of operation. Tests are also presented with measurements at strategic points of the grid to evaluate the system behavior under specific operating conditions, being normal or under contingency. The results attest the nanogrid's ability to reliably meet the loads, as long as it respects the limitations of the implemented power generation and storage system. In addition, it was found that the characteristics related to the charge controller topology benefits the power quality for the developed grid size and topology.

Currently, the medical images have a significant impact on the diagnosis of pathologies, consequently, the development of metrics for the evaluation of such images has been increasingin in order to  preserve or improve its original features when subjected to compression, enhancement and digital transmission operations. The complexity in order to performance analysis of a metric includes the correlation of metrics with the subjective data obtained experimentally and also the difficulty of obtaining such data. Therefore, development of new metrics and comparative evaluation of metrics applied to diagnostic images are relevant. In this work, a methodology for the acquisition of subjective data and visual quality assessment metrics Bayesian network based  developed specifically for the context of diagnostic images   are proposed.

Knowledge about the extent of riverbed overflow is extremely necessary for the determination of areas at risk. The City of Altamira-PA, located on the banks of the Xingu River, historically suffers from extreme events of floods that provoke floods, causing great damages to the population. Considering the problem, this paper presents a monthly level prediction system of the Xingu River based on neural networks perceptron of multiple layers. For the development of the system, rainfall data were used in the basin and sub-basins of the Xingu River, and SST information (Sea Surface Temperature) from 1979 to 2016. The satisfactory results demonstrate the great applicability the artificial neural networks to the problem.

The power transformer is one of the most important equipment in the electric power system, allowing the feasibility of connecting the generating centers to the consumer centers, even over long distances. Reliable and continuous operation is of fundamental importance for service maintenance and is subject to various types of disturbances that can lead to failures. In this perspective, studies of the dynamic behavior of transformer windings through computer simulations have been widely used to safely and accurately evaluate their operation. Therefore, this paper presents the methodology for research on a 50 MVA power transformer using the finite element method for steady state and time domain analysis. Thus, the study was performed by means of couplings (electromagnetic-mechanical circuit). In the first analysis (circuit study), the ATPDraw software was used to obtain the behavior of the inrush current and solidarity energization during the transformer bank energization. Therefore, in the ANSYS MAXWELL software, electromagnetic studies were performed. For this, a real 3D model was used (taking into account the characteristics of the core lamination and windings, some of them in disc format). Thus, the results are presented as the behavior of magnetic induction and electromagnetic forces in the equipment windings. Finally, in the ANSYS STRUCTURAL software, structural (mechanical) studies were performed. Also, as before, a close-to-real 3D model was used, presenting as results the behavior of the total deformation in the winding, the mechanical stress suffered and the degree of safety during the occurrence of energization. The steady state studies were considered three operating conditions: nominal condition, solidarity energization and inrush current. For the nominal condition, the equipment's plate data was used, for the energizing condition (solidarity energization and inrush current) the largest amplitude obtained during the simulation was used. It is noteworthy that for the time domain analysis, only the condition of the inrush current was analyzed, both for the high computational cost required and for being the worst condition in the steady state analysis.

This work implements satellite tracking techniques in inclined orbit and compares their performances with other techniques, describing operaton and implementng in a simulaton environment and in a real control system to generate the performance evaluaton of each one. One of the techniques investgated is reinforcement learning. Tracking satellites in inclined orbit is of paramount importance for telecommunicatons using this type of link, allowing automatc communicaton maintenance and extending the service life of satellite services in this situaton. Algorithms widely used in tracking, both in the literature and in commercial equipment, result in estmates or predictons of satellite positon rather than actual positon, and require a study of the specifc characteristcs of the region where the base staton is located. The complexity and investments in the tracking technique vary according to the commitment made in the antenna installaton and control systems, being necessary to compare the existng methods before their implementaton. The work develops the environment necessary to simulate satellite communicaton, from recepton to antenna movement, to analyze the performance of the technique in

Com o crescimento dos dados móveis e as rápidas tendências de urbanização, haverá uma densidade extremamente alta de links de counicação sem fio nas cidades, sendo assim os usuários esperam um ambiente onde possam ter acesso à internet aos seus dispositivos a qualquer hora e local. Devido a problemas no CAPEX/OPEX, uma implantação de small cells não é uma estratégia ecnômica em cenários  de tráfego de dados intenso, desta forma, a utilização de UAVs para melhorar a cobertura e o desempenho da rede torna-se viável. A fim de Qualidade de Serviço da rede, foi proposto um sistema computacional para realizar tomada de decisão que recebe como entrada informações da rede como vazão, taxa de perda de pacote e atraso, e retorna a qualidade de rede para aquele tipo específico de aplicação. O sistema como um todo verifica a necessidade ou não do uso de Veículos Aérios Não Tripulados (UAVs) para melhorar a qualidade da rede e a cobertura da área de maior demanda. De outra forma, os UAVs permanecem na estação base. Através do método proposto houve melhorias na Qualidade de Serviço (QoS) da rede, permitindo uma perda de pacotes e no atraso, e um aumento na vazão.

Este trabalho apresenta os resultados obtidos da pesquisa de aplicação de Redes Neurais Profundas para o problema de diagnóstico de faltas incipientes em transformadores baseado na análise dos gases dissolvidos em óleo (DGA). Dois modelos são propostos utilizando Redes Neurias Autocodificadoras Empilhadas e redes Neurais Convolucionais. Para o desenvolvimento do sistema foi utilizada a base de dados TC10 de equipamentos faltosos inspecionados em serviço e usada para a publicação da norma IEC 60599.

Among the devices that guarantee the generation of electricity through the photovoltaic conversion process the module is the basic (elemental) unit of a photovoltaic system. The literature review revealed that exists “countless” mathematical models used to predict the performance of modules. However, in general, they can be classified into two categories: power models and models based on the equivalent photovoltaics cell/module circuits. Therefore, in this work 14 models were selected, being 11 of power, 2 based on equivalent circuits and 1 which is the composition of both, but all with the objective of estimating the maximum power (P_mp), which reflects the desired operating point on a given system. The evaluation process of the selected models was made from two statistical approaches: the descriptive, also referred to as traditional; and the comparative, which seeks to relate the estimates versus the measured values and admitted as reference. For this, 4 different photovoltaic modules were used, but of the same technology (polycrystalline), experimentally measured using a monitoring system installed in the testing area of the Study Group and Development of Energy Alternatives (GEDAE / UFPa). Statistical evaluations made it possible to identify the most adherent models, both in terms of precision and accuracy. However, it is important to emphasize that this is not enough to determine a single best model, because depending on the application objectives, it may be that the effort is a determining factor, or even more important than its accuracy. Thus, besides the statistical evaluations, at the end of this work an ordinal classification of the models is presented, considering a compromise relationship established between precision, accuracy and effort.

Recent advances and cost savings in optical devices have spurred a high interest in sensors based on fiber optics applied to measure physical and mechanical parameters. Several reseraches have demonstrated the great advantages of optical sensors such as low cost, compatibility, immunity to eletromagnetic interference, applicability and its versatility as temperature sensors, vibration, magnetic field, curvature, among others compared to convencional sensors. Wide versatility of optical sensors enable curvature measurement with wide aoolicability as fluid flow sensor and Health Structural Monitoring monitoring (SHM).

This work aims to study the propagation of optical power in the structure based on coer diameter mismath technique (CDM), Singlemode-Multimode-Singlemode-Multimode-Singlemode (SMSMS), with curvature aiming at application as a low cost alternative, easy fabrication and acceptable sensitivity for flow measurement in pipes. The two multimode sections present in the sensor act as a coupler and re-coupler of core and cladding modes, and the singlemode section, in te middle, acts as a sensing element. The sctructure was analyzed by experimental measurements and numerical simulations. Experimental analyzes were performed for curvature meaurement and for flow measurement. In both the sensor generates interference patterns when it is bent. Numerical modeling was performed using the finite difference beam propagation method using software BeamProp 9.0 from the company Rsoft TM. Results demonstrate that aplicability as curvature sensors and flow measurement is feasible.

This work investigates and proposes methodologies for detecting faults in industrial plants equipped with control systems operating in closed loop. The feedback control reduces the sensitivity of the closed-loop system to plant variations. Thus, the detection of faults in systems operating in closed loop is challenging due to. In this work, we propose the use of a tool to estimate the open mesh transfer function from closed loop data, in order to estimate parametric models for fault investigation. The technique investigated is an adaptation of the Two-Stage Method, which unfolds the process of closed-loop identification in two sequential stages of open-loop identification. In the first step, an approximation of the complementary sensitivity function for the closed loop system is identified. Subsequently, an approximation of the sensitivity function is estimated from the estimates obtained for the complementary sensitivity function. By using parametric identification techniques, an estimate of the uncorrelated noise input signal is obtained. This allows to obtain a reasonably accurate model of the open-loop plant. Results of computer studies are presented and discussed in order to highlight the strengths and weaknesses of the proposed strategy.

Modern communication systems can exploit the increasing number of sensor data currently used in advanced equipment and reduce the overhead associated with link configuration. Also, the increasing complexity of networks suggests that machine learning (ML), such as deep neural networks, can effectively improve 5G technologies. The lack of large datasets make harder to investigate the application of deep learning in wireless communication. This work presents a simulation methodology (RayMobTime) that combines a vehicle traffic simulation (SUMO) with a ray-tracing simulator (Wireless InSite), to generate channels that represents realistic 5G scenarios, as well as the creation of LIDAR sensor data (Blensor). The created dataset is utilized to investigate beam-selection techniques on vehicle-to-infrastructure using millimeter waves on different architectures, such as distributed architecture (usage of the information of only a selected vehicle, and processing of data on the vehicle) and centralized architectures (usage of all present information provided by the sensors in a given moment, processing at the base station).

The results indicate that deep convolutional neural networks can be utilized to select beams under a top-K classification framework. It also shows that a distributed LIDAR-based architecture provides robust performance irrespective of car penetration rate, outperforming other architectures, as well as can be used effectively to detect line-of-sight (LOS) with reasonable accuracy.

A introdução de tecnologias como Carrier Aggregation (CA), Massive Multiple Input Multiple Output (MIMO) e Coordinated Multipoint (CoMP), visando melhorar a capacidade dos sistemas móveis de banda larga, aumenta o desafio para implantação do Mobile fronthaul devido à limitação da capacidade da infraestrutura, para suportar altas taxas de transmissão. Uma abordagem usada para lidar com a limitação da capacidade do frontahul é a compressão do sinal transmitido. Várias técnicas vêm sendo desenvolvidas para compressão do sinal no fronthaul e, a maioria dessas técnicas comprimem o sinal transmitido em banda base. Neste trabalho é desenvolvida uma técnica de compressão, para cenários específicos dos sistemas Rádio-sobre-Fibra digital configurados para transmissão do sinal em frequência intermediária. Esta técnica usa informações sobre o estado de canal de rádio (Channel state information), para controlar a compressão do sinal no fronthaul. Os resultados das simulações com a técnica desenvolvida demonstram a sua capacidade para reduzir o erro de quantização, permitindo a transmissão de sinais com modulação de 64 QAM, usando resoluções menores (até 4 bits por amostra) no quantizador. Portanto, taxas de compressão de 2:1, nos dados úteis, e um EVM próximo a zero são alcançados. Além disso, a técnica de compressão desenvolvida supera as técnicas μ-Law e A-Law, demonstrando tempos de execução dentro do limite de latência esperado nos sistemas 5G (1ms). Por fim, o desempenho da rede fronthaul é avaliado experimentalmente em um enlace ótico de 20 km, considerando cenários com e sem compressão do sinal.

ANÁLISE DE SISTEMAS NÃO-LINEARES E SÍNTESE DE OPERADORES INVERSOS POR SÉRIES DE VOLTERRA DIAGONAIS

Regulação de tensão e frequencia em microredes ilhadas com veículos elétricos e geração distribuida utilizando otimização por enxame de partículas

In this work we are analyzing the simulations of two microfita antennas. The first is an antenna using the ceramic material Bismuth Niobate (BiNbO4) doped with Vanadium Pentoxide (V2O5) on the substrate. The antenna patch was designed with indented power line to facilitate matching of impedances and the substrate with air holes was placed just below the patch to further decrease the losses. The second is a nanoanthene with Graphene Patch in the Terahertaz range and PBG (Photonic Band Gap) substrate with triangular mesh, and holes in the following height configurations h1, h2 and h3. At time h1 the substrate is fully drilled, while at heights h2 the holes will be made top to bottom of the substrate and the height h3 is the antenna with substrate drilled from the bottom up to the middle of the substrate. Therefore three antennas are created in these geometries using a triangular hole network. The arrangement of the holes in the dielectric substrate constitute the PBG structure, to increase the performance and efficiency of these antennas, extinguishing surface waves in the substrate of microfite antennas. The adopted geometry also improves antenna parameters such as efficiency and bandwidth. The commercial software HFSS and CST were used for the simulations of the antennas. After the numerical simulation steps the results of the parameters of these devices were obtained. The first antenna (periodic lattice with ceramic substrate) obtained a return loss of -36.21 dB, at a resonance frequency of 10.26 GHz, with a bandwidth of 2.18 GHz. In the simulations of the antennas of microfita with Patch of Grafeno the antenna h3 obtained double transmission band with chemical potential of graphene of 0,3 eV.

Centralized radio access networks are present as a potential alternative for next generation of wireless networks, because they are able to provide high data rates and allow the reduction of structural and operational costs in the network. The centralized architecture implements the concept of fronthaul, and enables the challenge to increase the capacity of data transmission in these links. In this way, the study of digital signal compression techniques presents itself as an alternative to reduce the cost of implementing centralized radio access networks.This work investigates the use of vector quantization methods in the compression of complex samples of baseband LTE signals. We propose the use of Genetic Algorithms in the training of sub-optimal codebooks for the vector quantization process in order to reduce the errors imposed in this process and consequent increase in fronthaul capacity. Results showed that the proposed compression algorithm allows reduction of fronthaul data rates associated with acceptable errors. It has been shown to be possible data rate compression factors of 3 to 12 times, with errors of approximately 1 % to 2%, respectively, proving the effectiveness of codebook training process in LTE signals present in the downlink of centralized radio access networks.

SISTEMA INTELIGENTE DE BALANCEAMENTO DE FASES EM REDES DE BAIXA TENSÃO PARA UNIDADES CONSUMIDORAS MONOFÁSICAS

Non-reciprocal components are an indispensable part of many microwave and optical systems. In the future, THz technology will also require these components. Existing publications show that the bandwidth of graphene based circulators in the THz region can be 10% to 20% with the use of very complicated structures. The suggested circulator consists of a graphene junction with a concave pattern and three waveguides symmetrically connected to it. Graphene is supported by SiO₂/ Si layers. The circulating behavior is based on the non-symmetry of the graphene conductivity tensor that appears due to magnetization by a DC magnetic field normally applied to the plane of the graphene. We discuss the main parameters that define bandwidth and its influence on bandwidth. Circulators have record bandwidth that is twice as high as those published. We have shown that the circulator Y may have a bandwidth of 42% in the frequency range (2.75 THz to 4.2 T Hz), with insulation better than -15 dB and insertion losses greater than -2 dB, provided by the polarization DC magnetic field 1.5 T and the chemical potential of graphene 0.15 eV. For the two 4-port circulators we achieved a bandwidth of 44% for the same physical parameters.

It is a general consensus that increasing the integration of wind power in the total generation mix means operating and planning methodologies, as well as operational standards, must be revisited. The main reason is that the number of random variables and system complexities considerably increases with the addition of wind power. The main idea of this thesis is to discuss operating reserve issues from short- and long-term perspectives when a significant portion of the generation system is composed of wind power. By modeling and evaluating the generation and load uncertainties, the paper analyzes the complicated relationship between all these variables. Moreover, the thesis proposes a discussion of the probability distribution functions of the reserve needs in order to evaluate the impact on operating reserve capacity. It essentially means that load and wind power forecasting errors and forced unit generating outages will be represented, which characterize the main operating system obstacles associated with the performance of the operating reserve. In fact, under or over forecasting involving system load or system production can lead to different effects on system’s balance, mainly when the portion of wind power is significant in the total system generation. The experiments are conducted on the modified configuration of the IEEE-RTS 96 and on the planning configuration of the Portuguese Generation System.

With the expansion of data demand, current access technologies may soon become obsolete. It is estimated that from 2020 with the arrival of 5G technology and fifth generation fixed broadband, rates of at least 10 Gbps should be provided to end users. In addition, criteria such as low latency, flexibility and energy savings are key factors. In this work, the application of the coaxial cables in the frontaul of 5G networks and in Gigabit G.fast access networks also known as G.mgfast, in order to find high capacity and energy efficiency are evaluated. In this work, it will be shown that rates of up to 40 Gbps can be achieved in an RG06 coaxial cable with support for 140 antennas meeting 3GPP transmission criteria. In addition, we will see that data rates of up to 10 Gbps can be reached in RG59, RG06 and RG11 cables in G.mgfast networks, but bearing in mind that there is an optimum relationship between the rate and transmitted power. In addition, an optimization algorithm applied to coaxial cables as well as a type of analogue fronthaul are also proposed.

This research presents an innovative methodology to predict the potential for hydropower generation using artificial intelligence techniques. A comparative analysis between the deep learning technique called Group Method of Data Handling (GMDH) and artificial neural networks with different optimization algorithms were applied for evaluation according to the case study for the future Jatobá plant, a plant in the Tapajós river basin, state of Pará, Brazil. The mean monthly rainfall in the subbasins of the Tapajós River was used as input to feed the developed models, which convert the volume of rain into energy. The non-physical models employed showed excellent skill and good efficiency to simulate this naturally complex and non-linear process, according to the statistical parameters of evaluation. Simulations were also carried out to evaluate the impact of climate change on hydropower generation for the future Jatobá power plant. The GMDH model presented a better performance among the others, highlighting its behavior during the dry season, critical period for the hydropower generation and that defines the steady energy of the enterprise. The success of this approach will reduce uncertainties and subsidize preliminary studies for plant implementation, as well as simulate scenarios to support planning, reduce costs and generate synthetic data for time series of power generation covering periods without observational field data.

This work presents the proposal of two automatic systems to aid in the detection of anomalies in heart beats and medical decision support. The systems were developed for the identification of rhythmic arrhythmia and morphological arrhythmias from signals obtained from an electrocardiogram (ECG). Both systems are based on a competitive structure of Convolutional Autoencoders (CAE), and each network was trained to reconstruct the signals presented at its entrance. For the case of the rhythmic classifier, the system was developed from the use of the ECG signals, without undergoing a feature extraction process, and for the case of the morphological classifier, the system was based on the QRS complex extracted from the ECG signal. For the development and testing of the systems, the database MIT-BIH Arrhythmia of ECG signals was used. An accuracy of 88.9% was achieved for the Rhythmic Classifier and 81.73% for the Morphological Classifier, in the case in which the evaluation basis is considered. The results obtained demonstrate the applicability of the proposed competitive structures to the arrhythmia classification problem.

In this work, two graphene antennas are proposed. The first antenna is a graphene device with dynamical control of radiation pattern. This device is formed by a graphene dipole with two coplanar graphene parasites. Working in the frequency range 1.94−2.13 THz, this antenna has four operation states that can be selected by setting specific chemical potentials by external electric field on the graphene elements. In state 1, a dipole-like radiation diagram is produced. States 2 and 3 are characterized by specific directional radiation patterns. By selecting the operation state 4, the antenna is switched off. When a silicon dioxide substrate with thickness of 1 micrometer is considered, chemical potentials are adjusted so that the device operates in 1.94−2.15 THz range, with negligible changes of its four radiation patterns with respect to the case without substrate. The second antenna consists of a loop graphene antenna with reconfigurable operation band. This antenna consists of a ring graphene element excited by a graphene-based emitter photomixer. By modifying the chemical potential of graphene, the device can operate at multiple frequencies in THz band. The maximum of the antenna total efficiency occurs near to its second resonance frequency, in which the antenna presents a dipolelike radiation pattern. The fractional bandwidth and the operation band of this device can be changed from 21.74% (2.05−2.55 THz) to 30.64% (2.57−3.5 THz). It is shown that the loop antenna, in its second resonance frequency, operates in a way that is compatible to the model of array of small dipoles.

The use of distributed generation (DG) and distributed energy sources (DER) close to the consumer centers has been gaining prominence around the world. The high penetration level of DG will shift the operation paradigm of electric energy systems, affecting power quality delivered to consumers. The active power injection from DG can cause several consequences to distribution systems, being overvoltage issue the most impacting of them. The reverse power flow from the DG units can cause voltage rise issue on distribution networks and consequently the DG shutdown. During unity power factor (pf) operation of DG, the distribution operators control overvoltage by active power curtailment. In this scenario emerges the GD hosting capacity concept, which consists of estimating the maximum power injected by GD without compromising voltage quality of the distribution system. This dissertation estimates the DG hosting capacity on MV distribution systems using genetic algorithms, considering as main limiting factor the overvoltage issue. Performance tests were performed on IEEE 33-bus and IEEE 69-bus systems

Vehicular Ad hoc Networks (VANETs) play an important role in the efficiency of road traffic by improving safety and acting as a facilitator of services for passengers, drivers and public safety officers. Recent improvements in the routing protocols and topologies used in vehicular networks have contributed to improvements in scalability, reliability and the quality of the information-sharing experience. Vehicles can cooperate with each other to stream videos of accidents or disasters and provide visual information of the monitored area with great precision. This Ph.D thesis proposes a Collaborative Routing Protocol for Video streaming VANETs (CRPV) using the service of fog storage to minimize the sharing of content. The routing table is based on an indicator that is generated by combining the speed, location and recording angle parameters of each vehicle involved in vehicular collaboration to reduce the unnecessary exchange of video data in vehicle-to-vehicle communications. The results of the simulations show that the proposed model performs favorably when compared to other routing protocols with respect to the availability of end-to-end communication and Quality of Experience.

The applications that consume high bandwidth and energy consumption have been increasing considerably fast in mobile networks. However, the mobile devices do not offer support to battery capacity for the intensive / continuous use of such applications. In addition, mobile networks currently have a high degree of heterogeneity and comprise a wide variety of networks with different technologies, such as LTE, Wi-Fi and WiMAX. Therefore, it is necessary the tradeoff to ensure that QoE is provided to users in this scenario, as well as an energy efficiency strategy designed to extend the battery life of mobile devices. This thesis proposes an intelligent architecture based on fuzzy logic, capable of providing support to decision making to save the energy of mobile devices within an integrated LTE and Wi-Fi network. The simulated experiments show the benefits and feasibility of the proposed solution.

This work presents and discusses the implementation of classic and robust pole cancellation controllers to recover the current after the application of shorting type faults, besides other analyzes are made, such as the impact that the short causes in the voltage Vabc , current Iabc and as the passive filter developed to mitigate the higher energy harmonics of this system (which are of the 5th and 7th order) act on this voltage and current respectively.
The system chosen is based on the Cigrée Benchmark (Sood, 2004) and represents an important industrial system, which is the furnace for the casting of aluminum ingots, used in multinational companies, among them, Hydro Alunorte. to be a contribution to industrial companies, especially in our region.
The developed controllers first used the classic IP pole-cancellation methodology and the other one aiming for a greater reliability margin in its use, since it is designed to be a robust controller and that responds to the uncertainties of the plant that are caused by the set of elements of electronics, such as inductors and resistors.
Therefore, the plant was linearized and from the transfer function of the interval plant, a robust controller was developed using the numerical methods of Kharitonov and Bhattacharyya, which relate the set of interval polynomials and their transfer functions to the formation of a polyhedron where one can choose the controller from the gain and phase margins established by the designer, thus increasing the reliability of the designed controller.
The results obtained in the simulation tests showed the good performance of the designed controller, with a fast recovery of the current, after the applied fault. This allowed for the important result of providing a safe operation of the industrial plant, where without the use of the controller could lead to problems such as accidents, production stoppage and loss of stability.

The implementation of the 5G standard will make the current mobile network architectures evolve towards C-RAN configurations, which are characterized by concentrating processing on a base station, from where the signal is distributed to remote antennas. To maintain uniform coverage, these systems rely on a dense network of low-power antennas scattered throughout buildings. This approach increases the complexity of the network’s Multi Input Multi Output (MIMO) system, which may hamper certain measurements involving equipment at both ends of the link. This work presents a method for the estimation of Far End Crosstalk (FEXT) and Insetion Loss (IL) using only one end of the link in order to avoid synchronization problems present in complex MIMO systems. Compared to other methods with similar proposals, the presented technique combines a simpler approach to a lesser degree of dependence on dual loop measurements, besides complementing techniques that can accomplish these measurements, but in a restricted range of frequencies.

Vehicular Ad-hoc NETworks (VANETs) allow users, services, and vehicles to share information, and will change our life experience with new autonomous driving applications. Multimedia will be one of the core services in VANETs, and are becoming a reality in smart environments, ranging from safety and security traffic warnings to live entertainment and advertisement videos. However, VANETs have a dynamic network topology with short contact time, which leads to communication flaws and delays, increasing packet loss and decreasing the Quality of Experience (QoE) of transmitted videos. To cope with this, neighbor vehicles moving on the same direction and wishing to cooperate should form a platoon, where platoon members act as a relay node to forward video packets in autonomous VANETs. This master’s dissertation introduces a Game Theory approach for Platoon-based driving (GT4P) for video dissemination services in urban and highway VANET scenarios. GT4P encourages the cooperation between neighbor vehicles by offering reward (money or coupon) for vehicles participating in the platoon. In this sense, GT4P establishes a platoon by taking into account vehicle direction, speed, distance, link quality, and travel path, which reduces the impact of vehicle mobility on the video transmission. Simulation results confirm the efficiency of GT4P for ensuring video transmissions with high QoE support compared to existing platoon-based driving protocols.

Smart Grid communications networks are considerably different from the traditional communication systems used to access the Internet when considering users, applications, Quality of Service and, especially, the impacts/losses due to malfunctions. Such data networks are generally owned and used exclusively by electrical system operators and require a high financial investment. Therefore, this paper presents an economic analysis to compare different possibilities of data network deployment for the Smart Grid. The results showed that 5G in comparison to other technologies obtained the best evaluation for the implementation of the communication of a data network applied to the Smart Grid, since the data of Quality of Service and the results obtained in the Total Cost of Ownership showed that in the medium and long term the 5G has its lower cost compared to other technologies used. Using a network configuration with 150 Femtocells and 2 Macrocells, Quality of Service obtained in regular and restricted transmission mode was 100% for uplink and 99.4% for downlink.

The present work proposes a radio propagation model for the Amazon region called Mixed Path. The techniques used for Mixed Path model are Geometrical Optics (GO) and the Uniform Theory of Diffraction (UTD).  Only ten rays are considered the main contributors to calculate the total electric field. Increasing the number of rays does not improve the accuracy of Mixed Path model since the scenario is for receivers located in long distances. Then slope diffraction or multiple reflections means a low electric field that does not contribute significantly to the total electric field.  The parameters of Mixed Path model such as electrical constants, antennas height, buildings height among others, are analyzed in order to know the influence of them in the received electric field. Measured data in the central frequency of 521 MHz of a Digital Television station in the city of Belem of Pará are used to validate Mixed Path model. This city is located in the Amazon region of Brazil and presents mixed routes formed by city, river, and forest. Because digital television has a wide coverage and reception flexibility, Mixed Path was designed for receivers at the user’s level for the service of Mobile Digital Television (M-DTV) and for fixed receivers on the roofs of homes for Home digital television (H-DTV). Finally, the proposed model and other models in the literature are compared with the data measured for M-DTV, being Mixed Path the model with the lowest RMS error.

Associative Neural Network, Classification of mental fatigue, Cognitive performance

This works presents a method of controlling the radiation diagram of a graphene dipole by the chemical potential. The dipole analyzed has rectangular geometry with power suplly by voltage source in the center, where each arm of the dipole is maintained to a different chemical potential. The geometry is analyzed by the two-dimensional moments method, with an equivalent surface impedance. Calculations of the main radiative properties of the antenna are presented as a function of the chemical potential. The results show that the greater the different between the chemical potentials, the greater the displacement of the main lobe of the radiation diagram.

This paper provides a performance evaluation of load flows methods to distribution power grids, considering them passives and actives. To that, the methods used were the following: Newton-Raphson, Fast decoupled with axis rotation, Fast decoupled with complex normalization and current summation. The performance evaluation of such methods were made considering the iteration number and the time processing that each method needed to converge, other criterion applied was the accuracy level of each method, through of the comparison of the voltages values taken by the Newton Raphson simulations related to the other 3 methods. Also, it was made in this work, comparisons of the performance of each method related to the iteration number on varying the r/x ratio of the test system feeders of  two buses, furthermore was verified the influence of the bases angles choices in the iteration number of the modified decoupled methods.

The use of control techniques are of a great matter to keep competitive performances, for electrical and industrials systems, having in mind as close as possible behavior setpoint tracking to a desired operation setpoint, aiming a reduced deviation and oscillation. In this work, are investigated fractional order automatic control techniques to improve the performance of industrial systems. Two fractional-order control techniques are studied, The Fractional Order Lead-Lag (FOLL) based on the frequency response method and applied to the improvement of power system stabilizers (PSS). Control laws are implemented in the form of digital control in an embedded system, based on DSPIC. The compensator’s performance is evaluated by performing several experimental tests on a 10 kVA reduced scale power system located at the UFPA Electrical Engineering Laboratory. Pulse in the generator voltage reference variation tests at are carried out at various power operation points of the microgenerator system, in addition to the robustness analysis of the system using a robust plot tool from the Bode diagram known as RBode. Second, there an investigation of a fractional pole placement technique (FOPP) that takes into account temporal response criteria of three terms fractional order systems, which in the case of this work are the overshoot and settling time, applied in a coupled tanks system, and a DC/DC Buck converter, where the FOPP technique is compared with two other techniques, the classical pole placement technique (IOPP), and a technique tuning of FOPID controllers based on Gain Margins and Phase Margins (GMPM). The results are corroborated for the two systems performing simulations in Matlab/Simulink Environment. The results show a reduction of at least approximately 15% in the ITAE and ISE indexes related to the dynamic performances of the systems addressed in this study, related to the controlled variable, with the insertion of the fractional controllers based on both the FOLL topology and the FOPP and GMPM techniques, compared to the values obtained from these controller indices tuned by conventional whole order techniques.

Currently, the transmission system operation is overwhelmed by the large amount of information coming from the most different monitoring systems, which must analyzing this information to keep the system in acceptable operating conditions, according with Brazilian rules for power sector. In this sense, this present thesis proposes an on-line system for the diagnosis of faults in transmission lines. Such a method is predicated on the monitoring and analysis of the transmission line’s leakage current. More specifically, the system is composed of modules capable of self-adapting to scheduled improvements. The developed methodology specifically addresses the diagnostic module, which the characteristics of the faulty harmonic spectrum of the leakage current are extracted, and thereafter identifies the most prominent fault in a multi-event scenario. To extract leakage current signals characteristics, the analytical redundancy was used, which, from laboratory and field data, to determine the normal behavior of the LT, to elaborate the LT model in normal operation and with the anomalies. With normal and faulty leakage current is accomplished the feature extraction, which uses suitable algorithms based on features obtained from state of art and data from laboratory. After choosing the extraction algorithm that has the best performance for multiple faults, are proposed classifiers to determine the most prominent fault in the transmission line. The classifier design took into account that the system needs to adapt to the changes occurred in the transmission line, emboding the knowledge about the system, since this is very dynamic.

The research in condensed matter physics with insulators and superconductors topological has contributed greatly to the characterization of the surface properties and modes zero in nanowires. We investigate theoretically, through the recursive Green’s function approach, the electron transport through the T-shaped quantum dot (PQ) with a single level and spinless, connected to a zigue zague chain and coupled to a p-wave superconductor. This model is an extension of the Kitaev chain for a network triangular of finite-size with for three, four, and five sites. We find that the Majorana zero modes can be tuned through the coupling parameters of the device and the linear conductance show both the Majorana Bound States (MBS) in topological phase and in the general topological phase maximally robust. This more realistic model allows the detection of MBS through of the control of the parameters governing the electronic tunneling and can be helpful for relevant experiments.

In recent years given the technological advancement of the communication media, and the increasing of users demand who wants a high quality of these services offered to them, the companies have increased the number of Radio Base Stations in cities, where many of these are located in environments with high housing density. Considering that each antenna or set of antennas present in these stations have an electromagnetic field of radiofrequency and transmit radiation, the concern with the population living in the adjacency of the transmitting antenna is the studies object of systems than regulate the companies that offer radiofrequency services, as well as, is the object of studies that aim not only to discover the effects of the contact with the ionizing and non-ionizing radiation present in these fields, but also to find out if the standard established for the regulation of services is being fulfilled. In this work presents the stages of development of a module, which consists of an extension added to the simulator for planning mobile communication networks (SIMPLARCOM). The module proposed allows, through the Virtual Reality environment (VR), to build and configure different scenarios, as well as the parameters of the transmission antenna, to provide an environment for non-invasive tests to evaluate non-ionizing radiation exposure; and identify potential insecure areas for housing, providing information for aid in decision-making regarding the relocation of transmitter antennas and aiming to decrease the ERP (effectively radiated Power) radiated by these Antennas. The module considers the guidelines present in resolution Nº. 303, published by the National Telecommunications Agency (ANATEL). In the results obtained is possible, navigate through the constructed scenario and check the value of the received power, the field intensity, the operation frequency, the antenna being analyzed and whether a certain point in the scenario is or is not receiving radiation at according to the threshold permitted by ANATEL.