Machine learning algorithms for photovoltaic system power output prediction

Abstract

Accurate photovoltaic (PV) production forecasting is necessary for the optimal integration of this technology into existing power systems and is important for both grid and plant operators. The purpose of this work is to assess the performance of different machine learning models for predicting the power output of PV systems. Specifically, a variety of methods were explored including artificial neural networks (ANNs), support vector regression (SVR) and regression trees (RT), with varied hyper-parameters and features. The output power prediction performance of each model was tested on actual PV production data-sets acquired over a period of a year and compared against an existing persistence model (PM). The comparative analysis between the different optimally devised models showed that the ANN outperformed the other models, achieving the lowest prediction mean absolute percentage error (MAPE) and normalised root mean square error (nRMSE) of 0.6 % and 0.76 %, respectively. The prediction capabilities of the SVR and RT models can be considered equivalent for this case since the nRMSE results were 1.13 % and 1.33 %, respectively. Finally, all the models achieved higher relative improvement over the PM since the skill score (SS) ranged from 86 % to 92 %.