Improving photovoltaic power forecasting accuracy: a comparative study of hybrid models and PVGIS

Accurate photovoltaic (PV) power forecasting is essential for ensuring reliable grid integration, particularly in regions with limited ground-based meteorological data. This study develops a hybrid multi-stage model chain designed for clear-sky PV power forecasting in Karaj, Iran, characterized by a cold semi-arid (BSk) climate. A 90 W polycrystalline module with a 27° tilt was evaluated using MERRA-2 atmospheric inputs and high-resolution on-site measurements for validation. Six global horizontal irradiance (GHI) models, five plane-of-array irradiance (POAI) models, and five module-temperature models were assessed. The Solcast, Perez, and PVsyst models demonstrated the highest accuracy within their respective stages. The combined model chain initially achieved an RMSE of 15.09 % and MBD of 5.12 %. After applying linear correlation adjustments, accuracy improved to an RMSE of 12.38 % and MBD of − 4.87 %. Comparative analysis showed that the optimized chain significantly outperformed the PVGIS-ERA5 and PVGIS-SARAH3 datasets, which exhibited RMSE values exceeding 85 % for the test location. The results highlight the importance of step-by-step validation and model selection under site-specific atmospheric conditions. The proposed framework offers a practical baseline for PV power prediction in semi-arid climates and may be extended in future work through machine-learning integration and multi-climate evaluation.