WSEAS Transactions on Circuits and Systems
Print ISSN: 1109-2734, E-ISSN: 2224-266X
Volume 25, 2026
Electric Load Demand Forecasting using Neural Networks (LSTM), Autoregressive Methods (ARMA), and Statistical Methods (Monte Carlo)
Authors: , , , ,
Search Articles
Abstract: This article presents a comparative study of three short-term electricity demand forecasting approaches: the Monte Carlo statistical method, the autoregressive moving average (ARMA) model, and the long short-term memory (LSTM) neural network. Historical electricity demand data from the San Rafael substation in Cotopaxi, Ecuador, for the period 2020-2023 were used. The Monte Carlo method generated a synthetic series with an accuracy greater than 70% in terms of R2, but it exhibited deviations throughout the predictions. The ARMA model, while showing a prediction accuracy greater than 80% in terms of R2 During weekdays, deteriorated over longer prediction periods. The LSTM network demonstrated superior predictive capacity, reaching values below 10% for MAPE and NMSE, an RMSE below 100 kW, and a coefficient of determination. R2 above 90%, across all prediction horizons (24, 48, and 72 hours). Finally, the LSTM network offers better performance for the planning and energy management of electric power systems.
Keywords:
Electric load demand forecasting, LSTM network, ARMA model, Monte Carlo method, Time series analysis, Deep learning
Pages: 210-220
DOI: 10.37394/23201.2026.25.19