Estimação do consumo de energia elétrica baseado em modelos de previsão de demanda

Abstract

Given human evolution, countries are producing and consuming electrical energy at every moment. The energy input involves social, economic and environmental dimensions that reverberate directly for the well-being of the population and the competitiveness of a country. The market has the need for a precise planning on the supply, consumption and expansion of electricity. In such terms, demand forecasting is essential for an efficient management based on more assertive decision-making, guided by public policy guidance, to forecasting trends, seasonality and future cycles, so that investments in the electrical sector are correctly allocated. Considering the importance and the impacts of efficient planning of electrical energy demand , this work aims to analyze and compare forecast models in order to determine which ones will have the most efficiency when measuring electrical energy consumption in Brazil. For that, a set of data regarding electricity consumption (GWh) was collected from Eletrobras open database, made available by the Institute of Applied Economic Research - Ipeadata (2021), under a defined monthly frequency from January 1976 to July 2021. The performance of the forecasts performed were compared using the accuracy measures: the Mean Error (ME); the Root Mean Square Error (RMSE); the Mean Absolute Error (MAE), the Mean Absolute Percentage Error (MAPE) and the Mean Absolute Scaled Error (MASE). From the analysis of such measures, the four best models were defined. Then, it was found that all models analyzed proved to be very accurate and similar, which defines that all predictions can be used, however, the Neural Network Autoregression model with Forecast Intervals was defined as the most appropriate, with an error percentage of 0.2%. Therefore, the results obtained are relevant to signal the organizations about the possibility of applying low-cost methods with a high degree of assertiveness that will significantly impact on planning and decision-making.