Similaridade comportamental do consumo residencial de eletricidade por rede neural baseada na Teoria da Ressonância Adaptativa

Carregando...
Imagem de Miniatura

Data

2016-08-25

Autores

Justo, Daniela Sbizera [UNESP]

Título da Revista

ISSN da Revista

Título de Volume

Editor

Universidade Estadual Paulista (Unesp)

Resumo

Esta pesquisa será dedicada ao desenvolvimento de uma metodologia com vistas à compreensão e ao exame do comportamento do hábito de consumo de eletricidade residencial, via análise de similaridade, baseado no uso de uma rede neural da família ART (Adaptive Resonance Theory). Trata-se de uma rede neural composta por dois módulos ART-Fuzzy, cujo treinamento é realizado de modo não supervisionado. No primeiro módulo, serão usadas, como entrada, as informações que caracterizam os hábitos de consumo e a situação socioeconômica. A saída do primeiro módulo junto com os dados referentes aos equipamentos eletroeletrônicos da residência compõem a entrada do segundo módulo que, finalmente, produz informações, na saída, relativas ao diagnóstico pretendido, ou seja, a formação de agrupamentos similares (clusters). Todo o processamento da rede neural modular é realizado com dados binários, os quais são gerados a partir de informações quantitativas e qualitativas. As redes neurais da família ART são estáveis e plásticas. A estabilidade refere-se à garantia de sempre produzir soluções, ou seja, não se observa problemas relativos à má convergência. A plasticidade é uma característica que possibilita a execução do treinamento de forma contínua sem destruir o conhecimento adquirido previamente. É um recurso pouco observado nas demais redes neurais disponíveis na literatura especializada. Com essas propriedades (estabilidade e plasticidade), combinada com o processamento de dados essencialmente binários, confere ao sistema neural uma ampla capacidade de produzir objetivos que podem ser facilmente modificados visando atender requisitos preestabelecidos pelos usuários (consumidor, empresa do setor elétrico). Neste sentido, o resultado esperado é a obtenção de informações referentes à similaridade de consumidores, à qual pode-se vislumbrar alguns benefícios, por parte dos consumidores, como melhorar o hábito de consumir energia elétrica, oferecendo também, por meio do conhecimento dos consumidores similares, a obtenção de melhores estratégias de negociação com os fornecedores, principalmente, no caso de sistemas smart grids. Neste novo paradigma do setor elétrico, há uma forte tendência do(s) consumidor(es) escolher(em) livremente a empresas fornecedoras de energia elétrica. Além disso, é discutida uma melhor forma para a realização da previsão de carga em pontos da rede elétrica onde há uma maior incerteza, e.g., nos barramentos mais próximos do consumidor (transformadores etc.), i.e., as incertezas no contexto da previsão de carga total do sistema são aumentadas à medida que se adentra a partir da carga global até chegar ao consumidor final, em especial ao usuário residencial. A base de dados, para a fase de treinamento da rede neural, é construída a partir de informações disponibilizadas por consumidores voluntários via o preenchimento de formulário. Realizada a fase de treinamento, a rede neural adquire um conhecimento incipiente afeito de ser aperfeiçoado ao longo do tempo, quando se implementa o recurso da plasticidade.
This work develops a methodology to understand and analyze the behavior of residential electricity consumption by similarity analysis, based on a neural network of ART (Adaptive Resonance Theory) family. The neural network is composed of two Fuzzy-ART modules whose training are non-supervised. At the first module, the inputs are information that characterize the consumption habits and the socio-economic situation. The output of the first module with the data referred to electro-electronic equipment available at the residence compose the input of the second module, which finally produces information at the output related to the diagnosis proposed, i.e. the formation of clusters. All the neural network processing is realized with binary data, which are generated from quantitative and qualitative information. ART family neural networks are stable and plastic. The stability assures that it always produces a solution, i.e. there is no convergence problem. The plasticity is a characteristic that allows executing the processing continuously without losing the knowledge previously learned. Those advantages are seldom observed in other neural networks available at the specialized literature. Considering these properties (stability and plasticity), combined with the data processing exclusively binary, the neural network is capable to be modified when necessary to attend pre-defined requests by the users (consumers, distributers, etc.). Therefore, the expected result is to obtain information referred to the similarity with consumers, and with this information, the consumers can improve their habits or even negotiating with the producers in case of smart grid systems. This new electrical system paradigm, the tendency is that the consumers can arbitrarily choose the electrical distributers. Furthermore, the work discusses the best way to realize load forecasting in points where there is uncertainty, e.g., on the busses near the consumers (transformers), i.e., the uncertainties considering the global forecasting increase if the information of residences is not considered. The database for the training phase of the neural network was built by a quiz form filled by some volunteer consumers. Afterwards, when finishing the training phase, the neural network acquires knowledge that along time can implement the plasticity resource.

Descrição

Palavras-chave

Redes elétricas, Consumo residencial de eletricidade, Agrupamento de consumidores residenciais, Características socioeconômicas e hábitos de consumo, Redes neurais artificiais, Teoria da ressonância adaptativa, Rede neural ART-Fuzzy, Smart grids, Residential electricity consumption, Clustering of residential consumers of electricity, Socioeconomic characteristics and consumption habits, Artificial neural networks, Adaptive theory resonance, ART-Fuzzy neural network

Como citar