Previsão do consumo de energia elétrica em uma agroindústria: um estudo de caso usando redes LSTM e CNN-LSTM

Abstract

This work aims to develop and compare models of Artificial Neural Networks (ANN) to predict the consumption of electricity in an agroindustry. The database, made available by the Electrical Energy Commercialization Chamber (CCEE), presents a historical series, of electricity consumption, in the period between January 2020 and January 2021, representing 6816 hourly observations. Models, based on the LSTM and CNN-LSTM architecture, were implemented, in Python language, using the Keras framework. Results obtained from the models were compared using the metrics MAPE (Mean Absolute Percentage Error), MAE (Mean Absolute Error) and RMSE (Root Mean Squared Error). We verified, for a horizon of 24 hours, that the CNN-LSTM model presented better performance.