A novel optical-electrical-thermal coupled model by BP neural network algorithm: Applied to high-precision prediction for key performance parameters of photovoltaic modules

Existing theoretical models for photovoltaic (PV) modules exhibit inherent limitations, including the incomplete integration of dynamic environmental factors and disconnect between the maximum power point tracking (MPPT) mechanism and physical model. To address these issues, we developed a novel optical-electrical-thermal coupling model (G-model) based on a BP neural network algorithm. This model employs BP neural networks to predict maximum output power ( P m ) of PV modules and the transmittance ( τ g ) of the glass cover, integrates the calculation model for total solar irradiance on inclined surfaces ( G T ), and a heat transfer model, enabling rapid iterative solutions under complex operating conditions. Experimental results demonstrate that the correlation coefficients ( R ) for the training, validation, and test sets of the BP neural network-based prediction models for P m and τ g were all close to 1.0000, indicating a strong agreement between predicted and measured values. The prediction errors were tightly clustered around zero, confirming the high accuracy of the proposed models. The average prediction errors of the G-model for backsheet temperature ( t ted ) and output power ( P ) were 1.22 % and 2.50 %, respectively, which were 62.20 % to 96.10 % lower than those of the reference models (P-model and Q-model). Nine interaction terms — including wind speed ( v ) with horizontal plane direct irradiance ( G b ) and wind direction ( θ ) with dust accumulation ( m ) — exhibited significant interactive effects (P-value