A data mining method for selecting input variables for forecasting model of global solar radiation

This paper proposes a method for selecting explanatory variables for global solar radiation forecasting. The weather conditions affect generation output of renewable energy significantly. As a result, smart grids increase the uncertainties caused by the power injections of photovoltaic (PV) and/or wind power generation. To smooth smart grid operation, it is necessary to select the predicted input variables for the forecasting model. In this paper, how to select explanatory variables in the forecasting model of PV generation is discussed because Japan gives much higher priority to PV generation in the framework of future energy policy. The global solar radiation is one of the most important variables in dealing with PV generation output forecasting. This paper focuses on the relationship between the global solar radiation and its explanatory variables. The proposed method makes use of the CART (Classification and Regression Trees) algorithm of data mining method to select the explanatory or input variables in the forecasting model of global solar radiation. CART has the function to give priority to explanatory variables through an index called Variable Importance. The proposed method was applied to real data of global solar radiation in Tokyo, Japan.