Comparative study of nonparametric and parametric PV models to forecast AC power output of PV plants

Abstract

In this paper, a comparison between two approaches to predict the AC power output of PV systems is carried out in terms of forecast performance. Each approach uses one of the two main types of PV modeling, parametric and nonparametric, and both use as inputs several forecasts of meteorological variables from a Numerical Weather Prediction model. Furthermore, actual AC power measurements of a PV plant are used to train the nonparametric model, to adjust the parameters of the different PV components models used in the parametric approach and to assess the quality of the forecasts. The approaches presented similar behavior, although the nonparametric approach, based on Quantile Regression Forests, showed smaller biased errors due to the machine learning tool used.