A Two-Step Learning-by-Examples Method for Photovoltaic Power Forecasting

In this paper, an innovative machine learning (ML) approach for the prediction of the output power generated by photovoltaic (PV) plants is presented. Toward this end, a two-step learning-byexamples (LBE) strategy based on support vector regression (SVR) is proposed to learn the complex relation among the heterogeneous parameters affecting the energy production of the power plant. More specifically, the first step is aimed at down-scaling the weather forecasts from the standard air temperature and the solar irradiance to the local module temperature and the plane-of-array (POA) irradiance. Then, the second step predicts the output power profile given the down-scaled forecasts estimated at the previous step. The advantages and limitations of the proposed two-step approach have been experimentally analyzed exploiting a set of measurements acquired in a real PV plant. The obtained results are presented and discussed to point out the capabilities of the proposed LBE method to provide robust and reliable power predictions starting from simple weather forecasts.