Previsão de carga multinodal usando a rede neural ART-ARTMAP fuzzy
Arquivos
Data
2013-07-12
Autores
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Previsão de cargas é uma atividade indispensável para o funcionamento dos sistemas de energia elétrica e contribuem para o planejamento e a operação, visando garantir o fornecimento de energia aos consumidores de forma segura, confiável e econômica. A previsão a curto prazo ajuda a estimar o fluxo de carga, tomar decisões de prevenção de sobrecargas, entre outras ações que são rotineiramente executadas. Para realizar a previsão é necessário identificar os padrões do comportamento de consumo e da sua relação com as variáveis externas do ambiente no sistema. A maioria dos estudos de previsões de cargas é realizada utilizando métodos estatísticos, onde é necessário modelar a carga matematicamente. Apesar de fornecerem bons resultados utilizam métodos complexos é de difícil modelagem. As técnicas de inteligência artificial proporcionam uma nova ferramenta, capazes de modelar uma grande quantidade de dados de cargas e construir a relação entre as variáveis do sistema de forma automática. Dentre essas técnicas inteligentes destacam-se as redes neurais e a lógica fuzzy que são utilizadas para previsão de cargas. Nesta pesquisa, apresenta-se um método de previsão multinodal (em vários pontos de interesse da rede elétrica) de carga elétrica, de curto prazo, utilizando uma rede neural artificial baseada na arquitetura ART (Adaptive Resonance Theory), denominada de rede neural ART-ARTMAP Fuzzy. As redes neurais ART apresentam características fundamentais, a estabilidade e a plasticidade, para treinar e prever de forma rápida e confiável. O método proposto foi implementado na plataforma MATLAB, onde foi possível realizar a previsão de cargas, por exemplo, de um sistema composto por nove subestações. Como forma de avaliar os resultados obtidos pela previsão, foi calculada...
Load forecasting is an essential activity for the operation of electric power systems and contribute to the planning and operation in order to ensure the supply of electricity to consumers in a secure, reliable and economical way. The short-term forecasting aids estimating the load flow, makes decisions to prevent overloads, among other actions that are routinely performed. It is necessary to identify the consumer behavior patterns and the relationship with the external environment variables of the system to perform the forecasting. The estimated load is determined via the following predictors. Many studies of load forecasting use statistical methods, where it is necessary to mathematically model the load. While providing good results in using complex methods it is difficult to model. The artificial intelligence techniques provided a new tool capable of modeling a large amount of data loads and build the relationship between the variables of the system. Among these intelligent techniques there are the neural networks and fuzzy logic which are used to load forecasting. This research presents a method for short-term multinodal electrical load forecasting (at various points of main interest), using an artificial neural network architecture based on ART (Adaptive Resonance Theory) neural network called ART-ARTMAP Fuzzy. ART neural networks have fundamental characteristics, stability and plasticity, to train and provide fast and reliable results. The method was implemented in Matlab platform where it was possible to perform the prediction, for example, for a system composed of nine substations. In order to evaluate the results obtained by the predictions the maximum percentage error of the forecast and the average percentage error were calculated. Historical load data yielded by the Electricity Commission of New Zealand are used to execute the training and... (Complete abstract click electronic access below)
Load forecasting is an essential activity for the operation of electric power systems and contribute to the planning and operation in order to ensure the supply of electricity to consumers in a secure, reliable and economical way. The short-term forecasting aids estimating the load flow, makes decisions to prevent overloads, among other actions that are routinely performed. It is necessary to identify the consumer behavior patterns and the relationship with the external environment variables of the system to perform the forecasting. The estimated load is determined via the following predictors. Many studies of load forecasting use statistical methods, where it is necessary to mathematically model the load. While providing good results in using complex methods it is difficult to model. The artificial intelligence techniques provided a new tool capable of modeling a large amount of data loads and build the relationship between the variables of the system. Among these intelligent techniques there are the neural networks and fuzzy logic which are used to load forecasting. This research presents a method for short-term multinodal electrical load forecasting (at various points of main interest), using an artificial neural network architecture based on ART (Adaptive Resonance Theory) neural network called ART-ARTMAP Fuzzy. ART neural networks have fundamental characteristics, stability and plasticity, to train and provide fast and reliable results. The method was implemented in Matlab platform where it was possible to perform the prediction, for example, for a system composed of nine substations. In order to evaluate the results obtained by the predictions the maximum percentage error of the forecast and the average percentage error were calculated. Historical load data yielded by the Electricity Commission of New Zealand are used to execute the training and... (Complete abstract click electronic access below)
Descrição
Palavras-chave
Redes neurais (Computação), Carga e distribuição elétrica, Sistemas de energia eletrica - Distribuidor de carga, Neural networks (Computer science)
Como citar
ANTUNES, Juliana Fonseca. Previsão de carga multinodal usando a rede neural ART-ARTMAP fuzzy. 2013. 110 f. Tese (doutorado) - Universidade Estadual Paulista Júlio de Mesquita Filho, Faculdade de Engenharia de Ilha Solteira, 2013.