BiLSTM-CNN network applied to the forecast of electricity consumption in agroindustry of the southwest region of Goiás

Abstract

This work proposes using the BiLSTM-CNN hybrid network to predict the electricity consumption of an agroindustry. The agroindustry is located in the southwest region of Goiás. The BiLSTM-CNN network is based on integrating the BiLSTM network (Bidirectional Long Short Term Memory) with the CNN network (Convolutional Neural Network). The database presents a daily series of energy consumption between January/2016 and December/2021, totaling 2153 observations. Prediction models based on Artificial Neural Networks were implemented in Python using the Keras framework. Results obtained from the proposed network and the recurrent networks LSTM, GRU, and RNN were compared using the metrics R² (Coefficient of Determination), RSME (Root Mean Squared Error), MAPE (Mean Absolute Percent Error), and MAE (Mean Absolute Error). It was found, for a 30-day horizon, that the BiLSTM-CNN model performed better. It was concluded that the proposed forecast model could serve as a support tool for managers of the agroindustry under study.