Previsão de cargas elétricas através do backpropagation estocástico
Carregando...
Arquivos
Data
2021-05-03
Autores
Mota, Camilla Nayara Santos
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
A previsão de cargas elétricas vem se tornando uma atividade fundamental para o funciona mento e operação dos sistemas elétricos de potência, visando fornecer energia elétrica aos con sumidores de maneira segura, eficaz e com menor custo de geração. O modelo proposto de sistema previsor de cargas elétricas, fundamentado em uma rede neural Perceptron, via algo ritmo backpropagation estocástico, visa obter um algoritmo que proporcione a convergência da rede neural de maneira que ela consiga escapar dos mínimos locais. Portanto, propõe-se um sistema previsor de cargas elétricas de natureza computacional que tem como objetivo supe rar os problemas apresentados pelo algoritmo backpropagation convencional. Trata-se de um método inovador, visto que não foi encontrado na literatura artigos que fizessem a abordagem da aplicação de previsão de cargas elétricas através do algoritmo backpropagation estocástico proposto. O algoritmo backpropagation estocástico consiste na atualização de pesos de maneira randômica. Os resultados obtidos na comparação se baseiam no mesmo banco de dados histó ricos de uma companhia de setor elétrico brasileiro. O modelo proposto se mostra promissor, eficiente e uma válida ferramenta para previsão de cargas elétricas.
The electric load’s prediction has become a fundamental activity for the functioning and electrical power systems operation, aiming to provide energy to consumers safely, on effective way and to reduce the engender cost. The proposed model of the electric charge forecasting system, based on a Perceptron neural network, via stochastic backpropagation algorithm, aims to obtain an algorithm that provides the convergence of the neural network in such a way that it manages to escape the the local minimums. Therefore, it is proposed a predictor system of electrical charges of computational nature which aims to overcome the problems presented by the conventional backpropagation algorithm. This is an innovative method, as no article found in the literature that approached the application of forecasting electric charges through the proposed stochastic backpropagation algorithm. The stochastic backpropagation algorithm consists of updating in an affective way, and weights randomly. The results obtained in the comparison are based on the same historical database of a company in the Brazilian electric sector. The proposed model is promising, efficient and a valid tool for predicting electrical charges.
The electric load’s prediction has become a fundamental activity for the functioning and electrical power systems operation, aiming to provide energy to consumers safely, on effective way and to reduce the engender cost. The proposed model of the electric charge forecasting system, based on a Perceptron neural network, via stochastic backpropagation algorithm, aims to obtain an algorithm that provides the convergence of the neural network in such a way that it manages to escape the the local minimums. Therefore, it is proposed a predictor system of electrical charges of computational nature which aims to overcome the problems presented by the conventional backpropagation algorithm. This is an innovative method, as no article found in the literature that approached the application of forecasting electric charges through the proposed stochastic backpropagation algorithm. The stochastic backpropagation algorithm consists of updating in an affective way, and weights randomly. The results obtained in the comparison are based on the same historical database of a company in the Brazilian electric sector. The proposed model is promising, efficient and a valid tool for predicting electrical charges.
Descrição
Palavras-chave
Backpropagation, Estocástico, Previsão de cargas elétricas, Stochastic, Electric load prediction