Desarrollo de un modelo predictivo basado en inteligencia artificial para la optimización de la gestión energética en sistemas solares híbridos basados en energía solar

Abstract

Abstract: This study addresses the need to improve the proactive management of the hybrid solar system at Universidad Técnica Particular de Loja. Its objective is to develop and validate an artificial intelligence-based predictive model to support energy scheduling through simulation. Historical photovoltaic power generation and electricity demand data from January to June 2025 were processed at an hourly resolution, and a day-ahead forecasting framework was developed. A 24-hour persistence model, Random Forest, and xgboost were compared; subsequently, the predictions were integrated into simulated rule-based dispatch and linear optimization scenarios. Random Forest achieved the best test performance: for electricity demand, it obtained an RMSE of 1142.96 Wh and an R² of 0.9204, whereas for photovoltaic power generation, it achieved an RMSE of 1164.14 Wh and an R² of 0.7776. The simulation recorded self-consumption close to 100%, self-sufficiency ranging from 39.24% to 44.02%, and a maximum planned cost reduction of 0.521%. It is concluded that the model is methodologically viable for energy planning, although the economic benefit remains limited under the evaluated conditions.