Integration of Graphical Modeling with Fuzzy Clustering for Casual Relationship of Electric Load Forecasting

This paper proposes a new method for selecting input variables in short-term electric load forecasting models. It is known that input and output variables do not follow the Gaussian distribution in load forecasting. In this paper, a hybrid method of graphical modeling (GM) and deterministic annealing EM (DAEM) clustering is presented to clarify causal relationship between the explained one-step-ahead electric load and the explanatory variables. GM is effective for estimating the relationship between variables with the Gaussian distribution. The DAEM algorithm is used to decompose non-Gaussian data into clusters of Gaussian data so that GM is applied to Gaussian data in clusters. The proposed method is successfully applied to the real data.