Physics-Informed Neural Network Modeling of Passive Photovoltaic Cooling Using Experimental Benchmark Data

The efficiency of photovoltaic (PV) systems decreases as module temperature rises under high solar irradiance, leading to reduced power output and accelerated material degradation. In this study, a physics-informed neural network (PINN)–based predictive framework is developed to model the performance of passively cooled PV panels using previously published outdoor experimental data as a validated physical reference. Experimental datasets corresponding to three passive cooling configurations phase change material (PCM), aluminium fins, and a hybrid PCM–fin system reported earlier under identical operating conditions are employed as benchmark inputs for model training and validation. The proposed PINN explicitly incorporates a thermodynamically consistent temperature–efficiency relationship into the learning process, enabling physically constrained prediction of PV power output as a function of irradiance and temperature. The trained models demonstrate high predictive accuracy across all cooling configurations, with test-set coefficients of determination of approximately 0.99, 0.97, and 0.98 for the PCM, fin, and hybrid systems, respectively. When compared with a conventional artificial neural network trained under identical conditions, the PINN reduces the root-mean-square prediction error by approximately 12–18% and exhibits improved stability under previously unseen operating conditions. Overall, the results show that physics-informed learning provides a reliable and interpretable approach for modelling photovoltaic performance under passive cooling conditions. By leveraging validated experimental benchmarks rather than introducing new measurements, the proposed framework enables effective use of limited data and offers a scalable tool for comparative performance assessment and design exploration of passive PV cooling strategies in high-temperature environments.