Comparative Analysis of Machine Learning Models for Photovoltaic Forecasting

As solar energy continues to increase its share in the electrical mix, accurate forecasting of photovoltaic (PV) output will play an increasingly critical role in grid integration. This paper presents a comparative study of four machine learning models, K-Nearest Neighbors (KNN), Random Forest, XGBoost, and Linear Regression, in addition to physical and statistical methodologies for forecasting PV power. Real-field data is used in this work to enhance the trustworthiness and importance of the findings. It also shows the influence of certain input parameters, such as air temperature, runtime hours, and dust, on the performance of the models. Also, the assessment of all models is done using several evaluation criteria, which include mean absolute error (MAE), root mean squared error (RMSE), <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}$</tex> squared (R2) and mean percentage error (MPE), in order to guarantee that there is no bias in the evaluation. The result shows XGBoost as the best-performing model.