Time series forecasting via integrating a filtering method: an application to electricity consumption

Leite Coelho da Silva F.; da Silva Cordeiro J.; da Costa K.; Saboya N.; Canas Rodrigues P.; López-Gonzales J.L. · Computational Statistics (2025) · DOI: 10.1007/s00180-024-01595-x

Resumen

Analysis and forecasting of the industrial sector electricity consumption is important for energy planning and control, in addition to being essential for the developing of a country or region. In this context, electricity consumption projections are highly relevant for the decision-making of companies operating in energy systems with the aim of optimizing such operations. This paper proposes a methodology for forecasting electricity consumption by integrating a time series filtering method with any classical forecasting model. In particular, in order to evaluate the performance of the proposed methodology it was used the Seasonal and Trend decomposition using Loess (STL) decomposition method integrated with several classical statistical models (Holt-Winters, seasonal autoregressive integrated moving average, dynamic linear model, and TBATS) and the artificial neural networks approach (NNAR - neural network autoregression, MLP - multi-layer perceptron, and LSTM - long short-term memory). The methodology was applied to the electricity consumption of Brazil’s cement industry. Based on the MAPE and RMSE precision metrics, the proposed methodology obtained the best performance for the time series under analysis via introducing the filtering method. The LSTM model integrated with the STL decomposition presented the best results among the models considered. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2025.

Instituciones: Universidad Peruana Unión