Use of Multilinear Adaptive Regression Splines and Numerical Weather Prediction to forecast the power output of a PV plant in Borkum, Germany

The development of accurate forecasting methods for renewable energy sources can be an important tool for the integration of such systems in the electricity grid. In this paper we focus on a forecast technique for the production of a photovoltaic plant, one day in advance, with the ultimate target of the optimal management of an energy storage system. The procedure is based on a regression model that takes as input the weather forecasts of the US Global Forecasting Service (GFS) and it is trained and tested on one year of power production data of a 1.3MW plant located in Borkum, Germany. The method used is the Multilinear Adaptive Regression Splines, that allowed the automatic definition of a reasinably simple model for the system and whose regression coefficients can be easily interpreted. The forecasted power obtained by the model proved to have a high correlation with the measured data and relatively low errors even with a limited number of features included in the model and a low number of training samples.

@Article{MM17,
  author       = {Massidda, L. and Marrocu, M.},
  title        = {Use of Multilinear Adaptive Regression Splines and Numerical Weather Prediction to forecast the power output of a PV plant in Borkum, Germany},
  journal      = {Solar Energy},
  volume       = {146C},
  pages        = {141-149},
  month        = {april},
  year         = {2017},
  publisher    = {Elsevier},
  keywords     = {Photovoltaic systems, power production forecast, multilinear adaptive regression splines, numerical weather prediction},
  doi          = {10.1016/j.solener.2017.02.007},
  url          = {https://publications.crs4.it/pubdocs/2017/MM17},
}