Application of Graphical Modelling to Selecting Input Variables for Solar Radiation Forecasting

This paper proposes an efficient method for selecting input variables of solar radiation forecasting. The proposed method makes use of Graphical Modeling that clarifies the dependence of output variables on input ones. It is useful to estimate a causal relationship between variables though the direct graph. Recently, photovoltaic systems are widely spread in power systems from a standpoint of reducing CO2 emissions. However, it is often pointed out that they offer unstable generation conditions due to meteorological conditions. To overcome the problem, an accurate model is required for the forecasting method. In this paper, a Graphical Modeling method is used to determine the appropriate input variables of ANN in solar radiation forecasting. It is important to calculate realistic correlation between input and output variables by excluding superficial correlation that brings about erroneous results. The proposed method is successfully applied to real data.