A data-driven and mechanism-explicit combined electrical performance predicting model for photovoltaic modules under engineering conditions

In practical engineering, photovoltaic (PV) module electrical performance, influenced by multiple and dynamic environmental factors, depends on understanding their complex nonlinear relationship for optimal maximum power point tracking (MPPT) target determination. However, existing methods often struggle to adapt to complex environments and fail to adequately model the coupling relationships among key electrical parameters. To address the aforementioned issues, this paper proposes an electrical performance prediction model for PV modules that combines data-driven and mechanism-explicit modeling approaches. This model leverages the strengths of data-driven nonlinear modeling to effectively predict the output characteristics of PV cells across a range of operating conditions. It supports maximum power point (MPP) estimation and MPPT control design by providing accurate predictions of PV electrical characteristics. Analysis via the Spearman correlation coefficient pinpoints input environmental variables linked to electrical parameters. Utilizing the inter-correlation among electrical parameters, LightGBM is employed for single-objective regression, with the Regressor Chain model being employed for integrated sequential prediction. Three PV explicit models, utilizing predicted electrical parameters, depict the output characteristic curves of the PV cell modules across various working conditions. The proposed model was validated using a dataset from polysilicon PV cells measured across various climatic conditions and time frames. Results indicate that the proposed Regressor Chain-LightGBM model accurately predicts PV electrical parameters under dynamic environmental conditions. On the test dataset, the RMSE values for Imp, Isc, Voc, and Vmp are 2.39%, 2.59%, 7.69%, and 9.78%, respectively. The predicted electrical parameters are further integrated into three mechanism-explicit PV models (C1, C2, and C3) to reconstruct the I–V and P–V characteristic curves, achieving coefficients of determination (R2) above 0.96 and up to 0.9997 under certain operating conditions. These results demonstrate the strong predictive accuracy and generalization capability of the proposed method, providing effective support for explicit PV modeling, MPP estimation, and future MPPT control design.