Prediction of photovoltaic panel cell temperatures: Application of empirical and machine learning models

Abstract

In this study, 25 different empirical models predicting the cell temperatures of PV panels were statistically analyzed and predictions were made using machine learning models. As a result of the correlation analysis, a strong positive correlation was found between the power output (Pm) of PV panels and solar radiation (Is) (PV1: r = 0.93, PV2: r = 0.94), indicating that photovoltaic energy conversion directly depends on solar radiation. In the analyses using error metrics (MAE and RMSE), the performances of wind speed dependent (MW) and independent (M) models were compared. However, since error metrics alone are not sufficient to assess the statistical significance and reliability of the models, t-test and confidence interval analyses were applied. The M5 model in the PV1 panel and the M1 model in the PV2 panel provided the closest predictions to the actual values. When MW models are analyzed, MW1 model for PV1 and MW13 model for PV2 produced the most accurate predictions. Within the scope of machine learning methods, different regression types such as Linear Regression, Non-Linear Regression, Decision Tree Regression and Support Vector Regression were evaluated and their prediction performances were compared. Among the Decision Tree based models, Extra Trees was one of the prominent models with R2 = 0.960 for PV1 and R2 = 0.974 for PV2 in the test phase.