An Improved Bayesian-based Approach for Short Term Photovoltaic Power Forecasting in Smart Grids

Abstract

Smart grid behaviour is characterized by significant uncertainties due to the time-varying nature of powers generated by random energy sources and of load demands. These uncertainties introduce several technical problems in smart grid planning and operation and new issues have to be addressed. In this context, an important role is played by probabilistic methods aimed to forecast random power productions and load demands. This paper improves a method recently proposed in the literature to perform a very short-term forecast of the active power produced by photovoltaic systems. The method, based on the Bayesian theory, is enriched by modifying the auto-regressive time-series model to take into account the dependence of the clearness index on some meteorological variables. Moreover, sets of sample data of the clearness index and of the involved meteorological variables extracted from measurements on the basis of 15 minutes is chosen to improve the forecast of the photovoltaic active power. Numerical applications are presented to give evidence of the obtained improvements in the power forecast when the cloud cover and the above sample are considered.