MACHINE LEARNING APPROACHES FOR IMPROVING THERMAL EFFICIENCY OF SOLAR HYBRID SYSTEMS

Abstract

                                                                          Abstract
Hybrid photovoltaic–thermal (PV/T) collectors deliver both electricity and heat, offering higher solar energy utilization compared with single-purpose devices. Their performance, however, is governed by complex interactions between climate, flow dynamics, and system geometry, which makes accurate modeling and optimization challenging. This study introduces an integrated framework that couples thermodynamic and exergy analysis with advanced machine learning (ML) models—ANN, SVM, Random Forest, XGBoost, and ANFIS—to predict and enhance PV/T behavior. A combined dataset of simulated and experimental values was employed for training and validation. Among all models, XGBoost achieved the best performance (R² = 0.998, RMSE < 0.01), demonstrating superior predictive capability. To optimize operational parameters, the ML surrogates were incorporated into metaheuristic algorithms (GA, PSO, GWO), which identified operating regimes with ~15% higher thermal efficiency and ~27% higher exergy efficiency while preserving electrical output. Sensitivity analysis confirmed irradiance and mass flow rate as the dominant factors. These results confirm ML-assisted optimization as a robust pathway for designing next-generation PV/T systems.