Research on Photovoltaic Power Forecasting Based on SOM Weather Clustering

Abstract

Changeable weather leads to random output from Photovoltaic (PV). Therefore, it is difficult to forecast PV power generation accurately by manual forecasting or traditional forecasting methods. We proposed a new method to improve forecast accuracy based on SOM-BP Neural Network. Firstly, we clustered different weather types by Self-Organizing Maps (SOM) to solve the problem of inaccurate PV power forecasting in the conditions of changeable weather. Secondly, Back Propagation (BP) Neural Network was applied to predict PV power generation considering the result of weather clustering from SOM methods. Finally, we relied on the PV power generation data and meteorological data in Qingpu District, Shanghai, trained the overall model and acquired the forecast results. The results showed that the SOM-BP model had the greatest forecast accuracy through the error comparison analysis among different models.