Statistical and Artificial Neural Networks Models for Electricity Consumption Forecasting in the Brazilian Industrial Sector

Felipe Leite Coelho da Silva,Javier Linkolk López-Gonzales,Paulo Canas Rodrigues,Rodrigo Salas,Kleyton da Costa

doi:10.3390/en15020588

Abstract

Forecasting the industry’s electricity consumption is essential for energy planning in a given country or region. Thus, this study aims to apply time-series forecasting models (statistical approach and artificial neural network approach) to the industrial electricity consumption in the Brazilian system. For the statistical approach, the Holt–Winters, SARIMA, Dynamic Linear Model, and TBATS (Trigonometric Box–Cox transform, ARMA errors, Trend, and Seasonal components) models were considered. For the approach of artificial neural networks, the NNAR (neural network autoregression) and MLP (multilayer perceptron) models were considered. The results indicate that the MLP model was the one that obtained the best forecasting performance for the electricity consumption of the Brazilian industry under analysis.

Highlights

In recent years, projections of electricity consumption for the Brazilian industrial sector have been studied, both for short and long term [1]
Martínez-Álvarez et al [12] used datamining techniques to electricity demand forecasting; a comparative study of different timeseries models for energy consumption forecasting of smart buildings in a university campus in the south of Spain [13]; there was a study about energy consumption forecasting to an industrial building using an artificial neural network (ANN) algorithm [14], and artificial intelligence techniques were used for energy demand planning in smart homes [15]
Sulandari et al [16] used singular spectrum analysis, fuzzy systems, and neural networks for electricity load time-series forecasting in Indonesia

Summary

Introduction

Projections of electricity consumption for the Brazilian industrial sector have been studied, both for short and long term [1]. Considering the industrial sector as one of the largest electricity consumers, studies must be carried out to ensure a minimum of predictability for legislators and consumers’ decision-making processes [3] In this context, several models have been used to obtain electricity predictions, such as the regression models using only weather variables for predicting load demand in England and Wales [4]; linear regression models for electricity consumption projections in Italy [5]; the Box and Jenkins models as well as the exponential smothing models for electricity demand in European countries [6]; the neural network models for power load forecast in Brazil [7]; the Bayesian dynamic linear model for short-term forecasting of Brazilian industry electricity consumption [8]; additive semi-parametric models for energy load forecasting in Australia [9]; bottom–up model for electricity consumption in Taiwan’s cement industry [10]; bottom–up approach for electricity consumption forecasting of the pulp and paper sector of the Brazilian industry [1]; and bottom–up stochastic approach for electricity consumption forecasting of a sector of the Brazilian industry [11]. Sulandari et al [18] presented a study comparative with the methods of Singular Spectrum Analysis, fuzzy systems, and neural networks for Indonesian electricity load demand forecasting

Objectives

Methods

Results

Conclusion