Accurate prediction of module temperature in bifacial photovoltaic (PV) systems equipped with single-axis solar tracking is critical for optimizing energy yield and mitigating thermal stress, particularly in tropical climates. However, existing empirical and data-driven approaches often lack physical interpretability or fail to adequately represent the nonlinear thermal behavior of operating PV modules. This study presents a novel multistage symbolic optimization (MSO) framework for predictive thermal modeling that addresses these limitations. The proposed two-level hierarchical approach first derives a physically interpretable symbolic equation using genetic algorithms (GA), followed by a second corrective symbolic regression stage in which GA and Alpha Evolution (AE) are evaluated as competing optimizers. This work presents a novel application of the AE algorithm to symbolic regression in PV thermal modeling. The methodology is validated using one year of high-resolution (5-min) operational data from a utility-scale bifacial PV plant with solar tracking in Colombia. The MSO–AE model achieved an R 2 of 0. 9439, an RMSE of 3. 18 °C, and an MAE of 2. 01 °C, outperforming the MSO–GA benchmark by 7. 3% in MAE and 6. 5% in RMSE, and surpassing recent single-stage symbolic regression models by 3. 7–20%, while preserving closed-form, interpretable expressions suitable for real-time control applications. A field-derived heating coefficient of 0. 034 °C/ (W/m 2) was identified. Observations show module temperatures exceeding 70 °C between 14: 00 and 16: 00, reducing electrical efficiency to 17. 31%, corresponding to a 12. 1% loss relative to standard test conditions. This thermal degradation resulted in average economic losses of USD 110. 7 per hour during peak periods and an annual energy loss of 6738. 6 MWh, equivalent to USD 336, 929 at a benchmark electricity price of USD 50/MWh. These results define economic thresholds for thermal management investments in bifacial PV systems operating in tropical environments. • Solar heating coefficient of 0. 034 °C/ (W/m 2) quantified for tropical PV. • Multistage symbolic optimization achieves 0. 9439 determination and 3. 18 °C error. • Alpha Evolution reduces MAE by 7. 3% and RMSE by 6. 5% over genetic algorithms. • Interpretable equations enable real-time thermal control in embedded systems. • Thermal losses reach 6739 MWh/year with annual economic impact of 202 k–472 k.
Vargas et al. (Wed,) studied this question.