Mutivariable mutual information based feature selection for electricity price forecasting

In a competitive electricity market, the electricity price forecasting provides a useful information for suppliers to develop bidding strategies or to make investment decisions. Customers can also be benefit by scheduling their electricity consumption according to the electricity price. Selecting proper and useful features plays an essential role in improving the forecasting accuracy. In this paper, multivariable mutual information is applied for feature selection to select the more relation between the output price and the candidate input features. Support vector regression (SVR) is applied as the regression system. Experimental results show that the SVR with the proposed feature selection achieves more accurate prediction than ones with other well-known feature selection methods.