Comparação de desempenho entre modelos lineares e redes neurais artificiais para previsão da demanda de energia elétrica

Abstract

The socioeconomic development of countries like Brazil, is directly related to the amount of electricity consumed by them. In order for these countries to continue developing, plans and strategies about the energy sector must be developed, thus, imbalances between the supply and demand of electricity can be avoided. With the objective of providing information to assist in the decision-making process of the electric power sector in Brazil, this work will propose as a methodology the comparison between performance of demand forecasting models, namely: Exponential Smoothing (SE); Auto Regressive (AR); Integrated Auto-regression of Moving Averages (ARIMA); and Multilayer Perceptron Neural Network (MLP). For validation of predictive models, a time series of electricity consumption in Brazil will be collected. As a result, the MLP neural network outperformed traditional statistical models, confirming its power of universal approximation of functions.