Aplicação de inteligência computacional na resolução de problemas de sistemas elétricos de potência
Carregando...
Data
2017-11-13
Autores
López Sepúlveda, Gloria Patricia
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Nesta tese são utilizados algoritmos de Inteligência Computacional para resolver quatro problemas da área de sistemas elétricos de potência, com o intuito de automatizar a tomada de decisões em processos que normalmente são realizados por especialistas humanos ajudados de métodos computacionais clássicos. Nesta tese são utilizados os algoritmos de aprendizado de máquina: árvores de decisão, redes neurais artificiais e máquinas de vetor de suporte, para realizar o processo de aprendizado dos sistemas inteligentes e para realizar a mineração de dados. Estes algoritmos podem ser treinados a partir das medições disponíveis e ações registradas nos centros de controle dos sistemas de potência. Sistemas Inteligentes foram utilizados para realizar: a) o controle centralizado Volt-VAr em modernos sistemas de distribuição de energia elétrica em tempo real usando medições elétricas; b) a detecção de fraudes nas redes de distribuição de energia elétrica realizando um processo de mineração de dados para estabelecer padrões de consumo que levem a possíveis clientes fraudadores; c) a localização de faltas nos sistemas de transmissão de energia elétrica automatizando o processo de localização e ajudando para que uma ação de controle da falta seja realizada de forma rápida e eficiente; e d) a coordenação de carga inteligente de veículos elétricos e dispositivos de armazenamento em tempo real utilizando a tecnologia V2G, nos sistemas de distribuição de energia elétrica a partir de medições elétricas. Para o problema de controle centralizado Volt-VAr os testes foram realizados em um sistema de 42 barras, para o problema de carregamento de veículos elétricos e de dispositivos de armazenamento os testes foram realizados usando um sistema de 34 barras, e os outros dois problemas foram testados com dados reais fornecidos por empresas do setor elétrico colombiano. O software WEKA, versão 3.8.0, foi utilizado para gerenciar os três algoritmos de aprendizado de máquina através do treinamento e validação dos algoritmos Multilayer Perceptron/Backpropagation para as redes neurais artificiais, o J48/C4.5 para as árvores de decisão e o SMO/PoliyKernel para as máquinas de vetor de suporte. Através dos resultados obtidos é possível comprovar o potencial dos Sistemas Inteligentes e da Mineração de Dados no desenvolvimento de algoritmos de automação para os quatro problemas da área de sistemas elétricos de potência.
In this thesis Computational Intelligence algorithms are used to solve four problems of the area of power electrical systems, in order to automate decision making in processes that are usually performed by human experts aided by classical computational methods. In this thesis the machine learning algorithms are used: decision trees, artificial neural networks and support vector machines to carry out the learning process of Intelligent Systems and to perform Data Mining. These algorithms are trained from the available measurements and actions recorded in the control centers of the systems. Intelligent Systems were used to perform: a) the centralized control Volt-VAr in modern systems of distribution of electrical energy in real time using electrical measurements; b) detection of fraud in electricity distribution networks by performing a data mining process to establish patterns of consumption that lead to possible fraudulent customers; c) fault location in electric power transmission systems by automating the localization process and helping to ensure that a fault control action is performed quickly and efficiently; and d) coordination of intelligent charging of electric vehicles and storage devices using V2G technology in real-time, in electric power distribution systems using electrical measurements. For the centralized control problem Volt-VAr was tested in 42-node distribution system, for the problem of loading electric vehicles and storage devices the tests were performed using a system of 34-node, and the other two problems were tested with actual data provided by companies in the Colombian electric sector. The WEKA software, version 3.8.0, was used to manage the three machine learning algorithms through the training and validation of the Multilayer Perceptron / Backpropagation algorithms for the artificial neural networks, the J48 / C4.5 for the decision trees and the SMO/PoliyKernel for vector support machines. Through the results obtained it is possible to prove the potential of Intelligent Systems and Data Mining in the development of automation algorithms for the four problems of the area of electrical power systems
In this thesis Computational Intelligence algorithms are used to solve four problems of the area of power electrical systems, in order to automate decision making in processes that are usually performed by human experts aided by classical computational methods. In this thesis the machine learning algorithms are used: decision trees, artificial neural networks and support vector machines to carry out the learning process of Intelligent Systems and to perform Data Mining. These algorithms are trained from the available measurements and actions recorded in the control centers of the systems. Intelligent Systems were used to perform: a) the centralized control Volt-VAr in modern systems of distribution of electrical energy in real time using electrical measurements; b) detection of fraud in electricity distribution networks by performing a data mining process to establish patterns of consumption that lead to possible fraudulent customers; c) fault location in electric power transmission systems by automating the localization process and helping to ensure that a fault control action is performed quickly and efficiently; and d) coordination of intelligent charging of electric vehicles and storage devices using V2G technology in real-time, in electric power distribution systems using electrical measurements. For the centralized control problem Volt-VAr was tested in 42-node distribution system, for the problem of loading electric vehicles and storage devices the tests were performed using a system of 34-node, and the other two problems were tested with actual data provided by companies in the Colombian electric sector. The WEKA software, version 3.8.0, was used to manage the three machine learning algorithms through the training and validation of the Multilayer Perceptron / Backpropagation algorithms for the artificial neural networks, the J48 / C4.5 for the decision trees and the SMO/PoliyKernel for vector support machines. Through the results obtained it is possible to prove the potential of Intelligent Systems and Data Mining in the development of automation algorithms for the four problems of the area of electrical power systems
Descrição
Palavras-chave
Aprendizado de máquina, Árvores de decisão, Carregamento de veículos elétricos, Controle centralizado Volt-VAr, Detecção de fraudes, Inteligência computacional, Localização de faltas, Máquinas de vetor de suporte, Mineração de dados, Redes neurais artificiais, Sistemas inteligentes, Artificial neural networks, Intelligent systems, Centralized control Volt-VAr, Computational intelligence, Data mining, Decision trees, Fault location, Fraud detection, Loading of electric vehicles, Machine learning, Support vector machines