Assessing solar-to-PV power conversion models: Physical, ML, and hybrid approaches across diverse scales

Solar energy, a sustainable and environmentally friendly power resource, has been widely adopted across industrial, urban, and agricultural sectors. With global photovoltaic (PV) installations reaching nearly 446 GW (GW) in 2023, accurate assessment and estimation of PV power output has become critical for enhancing system performance, grid integration and energy management In this study, we evaluate the effectiveness of physical, machine learning (ML), and hybrid models for solar-to-PV power conversion across varying spatial and temporal scales and diverse operational scenarios. For relatively simple cases requiring immediate power output predictions without reliance on historical data, we applied and assessed physical and ML models. In more complex, large-scale PV station scenarios, we analyzed the limitations of physical and end-to-end ML models, identifying significant estimation errors caused by the nonlinear behavior of solar irradiance. To address these challenges, we developed an advanced hybrid model that combines a temporal-aware multi-level mixer (TMM) algorithm with the physical IV model, achieving significantly improved estimation accuracy. By benchmarking the model performance using metrics such as root mean square error (RMSE) and mean bias error (MBE), this work provides a comprehensive analysis of the strengths and weaknesses of different models under varied conditions. The findings offer valuable guidance for model selection and optimization, supporting the development of more reliable and efficient solar energy systems.