Previsão de geração de energia hidráulica no Brasil: um estudo de caso usando redes neurais artificiais e regressão linear

Abstract

Electricity is fundamental for the economic development of a country. It is precisely for this reason that the electric power sector is one the fastest-growing sectors in the country. Brazil has one of the most diversified energy matrices in the world, characterized by a hydrothermal system and a predominance of renewable energy sources. Hydroelectric energy is one of the most widely used options globally for electricity generation. In 2021, in Brazil, the energy supply by source corresponded to 56,8% hydro, 12,8% natural gas, 8,2% biomass, 3% petroleum derivatives, 3,9% coal and derivatives, 10,6% wind, 2,2% nuclear, and 2,47% solar. Thus, the diversity of each country’s matrix implies different energy planning strategies, as each energy source has its limitations. Forecasts, the ability to assess the history of electricity generation, and project a response can benefit the electricity sector. They are crucial tools in assisting the decision-making process, providing information that promotes efficiency and security in the supply of electric power. In this context, this work aimed to compare the performance of models, Multiple Linear Regression (MLR) and Multilayer Perceptron Artificial Neural Networks (MLP ANN), using the WEKA software, in predicting hydroelectric generation in Brazil. To perform this comparison, a dataset of hydroelectric generation was used. The database presented a historical series of hydroelectric energy generation from January 1996 to August 2022. The methodology consisted of developing prediction models based on Multiple Linear Regression (MLR) and Multilayer Perceptron Artificial Neural Networks (MLP ANN) algorithms using the WEKA software. The results showed that, for a 6-month horizon, the MLP model outperformed MLR, considering the MAE, MAPE and RMSE metrics. The conclusion was that the MLP model demonstrated notable generalization potential within the prediction horizon.