Previsão de consumo residencial e industrial de energia elétrica na região sudeste: um estudo de caso usando redes neurais e regressão linear

Abstract

Electrical energy, an indicator of a population's quality of life, plays a relevant and prominent role in the world economy. Brazil, in this scenario, stands out for being among the largest global economies and the southeast region for being the population and industrial center of the country. In this context, this work compared time series models from the WEKA software to predict residential and industrial electrical energy consumption in the southeast region of Brazil. The databases, made available by the Energy Research Company – EPE, present historical series of energy consumption, residential and industrial, in the period between January/2004 and December/2022, totaling 228 instances. Time series forecast models, based on MLP and Linear Regression neural networks, were built in the WEKA software. The MAPE, MAE and RMSE metrics were used to compare the results of the prediction models. Experimental results demonstrated, for a 4-month horizon, that the MLP Neural Network presented the best performance.