COMPARATIVE ANALYSIS ON THE PREDICTION PERFORMANCE OF STATISTICAL MODELS AND NEURAL NETWORKS. A CASE STUDY BASED ON ENERGY CONSUMPTION FROM RENEWABLE SOURCES

This article aims to forecast energy consumption from photovoltaic (PV) sources by implementing and comparing three different predictive models to improve interpretability, accuracy, and computational efficiency. The models selected in this approach are: ARIMA (Auto Regressive Integrated Moving Average), a statistical model for linear dependencies and trends, SARIMA (Seasonal ARIMA), an extension of ARIMA for seasonal fluctuations; and NAR (Nonlinear Auto Regressive Neural Network), a machine learning model for nonlinear relationships. The performance of these techniques is tested and confirmed using a dataset of real measurements related to the energy from photovoltaic sources consumption, monitored at 15-minute intervals in 2024. The results of the predictive performance of the three models are analysed for day-ahead and week-ahead forecasting horizon. The added value of the proposed approach consists in proving how improved forecasting directly contributes to energy efficiency and loss reduction in PV systems, by enabling better energy management and demand-response strategies.