Modelo de regresión simbólica basado en la optimización de Harris Hawks para la predicción de la temperatura en módulos fotovoltaicos bifaciales

Abstract

Introduction: Accurate temperature prediction in bifacial photovoltaic (PV) modules is crucial for optimizing energy efficiency and system longevity. This study presents a symbolic regression model optimized using the Harris Hawks Optimization (HHO) algorithm and compares its performance with a Genetic Algorithm (GA)-based symbolic model and statistical methods, using real-world data from a 26.6 MW bifacial PV plant in Colombia.

Objective: To develop an interpretable symbolic regression model to predict the temperature of bifacial PV modules with solar trackers, using solar radiation and solar time as input variables.

Method: Four models were designed and compared: multiple linear regression (MLR), gradient descent-enhanced MLR, symbolic regression with GA, and symbolic regression with HHO. A one-year dataset with 5-minute resolution was used. Correlation and normality analyses were conducted, and model performance was assessed using RMSE and R² metrics.

Results:  The gradient descent-enhanced MLR model showed the best performance (RMSE: 4.92; R²: 0.86), followed by the SR-GA model (RMSE: 7.14; R²: 0.71). The SR-HHO model exhibited faster convergence and better performance with smaller datasets, though it showed lower accuracy with larger data volumes (RMSE: 13.91; R²: 0.09).

Conclusions: Symbolic models are effective for interpreting thermal behavior in bifacial PV modules. HHO is computationally efficient with small datasets, while GA provides more stable performance with large datasets. A hybrid approach combining both algorithms is recommended to improve predictive performance.