Performance Comparison of Support Vector Regression, Random Forest and Multiple Linear Regression to Forecast the Power of Photovoltaic Panels

Abstract

With the significant development and expansion of renewable energies, production sources have varied and the network has become more difficult to manage. Therefore, predicting the electricity generated by renewable sources has become critical. In this perspective, machine learning, as part of artificial intelligence, appears to be one of the best ways to achieve this aim. Machine learning techniques can control the variations in renewable energy output and therefore, facilitate their integration into the energy mix. Thus, one of the major goals of this research is to perform a comprehensive comparison of three popular machine learning techniques, including multiple linear regression, support vector regression and random forest, for the hourly prediction of the power produced by photovoltaic solar panels. Residual analysis is performed to visually test the investigated regression models. The results revealed that random forest achieved the best prediction accuracy in the testing phase with R2=96% and RMSE=0.39 KW.