Geração de energia hidrelétrica: uma análise comparativa de métodos de previsão estatísticos e de inteligência artificial aplicados a dados da região sul do Brasil

Abstract

Due to Brazil's vast hydroelectric potential, hydroelectric power is the country's main source of energy production. In addition, generating energy from the power of rivers has environmental and economic advantages. Furthermore, in the literature, hydraulic energy forecasting is gaining prominence in the scientific community, but the use of classic forecasting models is being little used. In this context, the aim of this study is to evaluate the performance of the AutoRegressive Integrated Moving Average (ARIMA), Support Vector Regression (SVR) models and a combination of both models using the simple arithmetic mean to forecast hydroelectric power generation in the states of Paraná, Santa Catarina and Rio Grande do Sul, Brazil. For the study, time series with 363 monthly observations were extracted from the National Electricity System Operator, from June 2013 to June 2023. The Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test was used to identify stationarity, and the ACF and PACF functions were used to analyze autocorrelation in each time series. ARIMA, SVR and both models were used to forecast horizons of 1, 3, 6 and 12 months ahead. To assess the accuracy of the forecasts, the performance error measures Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) were used. The results show that, in general, the SVR model performed better in terms of RMSE for forecasting hydroelectric power generation up to 12 months ahead, whose forecast error, based on the test set, was 967.76 MWmed, 795.71 MWmed, and 811.28 MWmed, for the states of Paraná, Santa Catarina, and Rio Grande do Sul. With a forecast range of 95% probability, the minimum and maximum forecasts for hydroelectric power generation from July 2023 to June 2024 will be 678.17 MWmed and 5489.3 MWmed, 297.28 MWmed and 4242.08 MWmed, 337.96 MWmed and 2459.77 MWmed, for Paraná, Santa Catarina, and Rio Grande do Sul respectively. This study makes it possible to plan power system operators in order to verify the capacity of this resource to individually supply the growing energy demand in the southern region of Brazil.