Electricity consumption forecasting in Brazilian northeastern region

Abstract

Electric power systems collect large volumes of data that can provide valuable information on energy consumption. Electric utilities can use this historical consumption data to assist in the decision-making process with regard to energy production, through estimating the expected energy consumption. In this work, the energy consumption forecasting problem was modeled as being univariate with a temporal step forward. The algorithms: Naive (Persistent, Mean and Median), SARIMA, MLP, CNN and LSTM were used; and a greedy search of its hyperparameters was performed in order to find the best configuration associated with each algorithm. In addition, for comparison and choice of the best forecasting algorithm, the MAPE metric and the modified Deibold-Mariano hypothesis test were used. For the proof of concept of the methodological proposal, energy consumption data from the Northeast region of Brazil between the years 2004 to 2019 were used.