Previsão do consumo de eletricidade em uma instituição de ensino superior: um estudo de caso usando o Weka

Abstract

Electric power has become part of society's basic needs. Its sustainable consumption is the responsibility of each and every consumer. Predicting this consumption can be an awareness tool that provides relevant information for energy saving. In this context, this study aimed to compare time series models in the prediction of electricity consumption of a higher education institution. The database, made available by UTFPR - Campus Medianeira, presented a historical series of energy consumption from December 2020 to May 2022. The methodology consisted of developing prediction models, based on Linear Regression (LR) algorithms, Artificial Neural Networks of the Multilayer Perceptron (MLP) type and Regression Support Vector Machine (SVR), using the WEKA software. The results showed that, for a four-day horizon, the MLP model presented better performance, considering the MAPE, MAE and RMSE metrics. The conclusion was that, despite great variability in the data due to the pandemic, it was still possible to obtain good results using the MLP model.