Comparison Between Hierarchical Time Series Forecasting Approaches for the Electricity Consumption in the Brazilian Industrial Sector

Abstract

In Brazil, the industrial sector is the largest electricity consumer. Therefore, energy planning becomes important for industrial development. Electricity consumption data in the Brazilian industrial sector can be organized into a hierarchical structure composed of each geographic region (South, Southeast, Center‐West, Northeast, and North) and their respective states. This work aims to evaluate the predictive capacity of the bottom‐up, top‐down, and optimal combination approaches used to obtain electricity consumption forecasting in the Brazilian industrial sector. These approaches were integrated with the predictive models of exponential smoothing, Box and Jenkins, and neural networks. The results showed that the bottom‐up approach integrated with the Long Short‐Term Memory (LSTM) model provided the best predictions and outperformed the other hierarchical forecasting approaches with an average MAPE of less than 3%.