Combined probabilistic forecasting method for photovoltaic power using an improved Markov chain

Abstract

A novel combined probabilistic forecasting method based on an improved Markov chain for photovoltaics (PVs) to enhance the accuracy of probabilistic PV power forecasting is presented. First, a Markov chain (MC) forecasting structure combining precise factors is proposed that considers more influence factors beyond the statistical information of historical data as compared with conventional MCs. Rough set theory is then used to refine the major factors to quantify the influence of those factors. Furthermore, a k-nearest neighbours algorithm is used to select similar samples for building an accurate forecasting model. Based on similar samples, the changing PV trends are more obvious than when using whole historical samples, thus further improving forecasting accuracy. Finally, the effectiveness and superiority of the proposed method are verified by comparing results from simulations with the results from competing methods for two cases using datasets from DESERT KNOWLEDGE AUSTRALIA Solar Centre and GEFCom2014. The simulation results show that the proposed method can provide probabilistic forecasting results with better performance, also, the proposed method can be adapted to various forecasting scenarios.