Day-ahead price forecasting of electricity markets by combination of mutual information technique and neural network

In the new competitive electricity markets, accurate forecast of electricity prices is valuable for both producers and consumers. Due to the volatility of electricity price signal and limited available information, there is an essential need to accurate and robust forecasting methods for the price prediction. In this paper a data mining technique, mutual information, is proposed for the feature selection of price forecasting. Then, by means of the selected features, a neural network (NN) predicts the next values of the price signal. The whole proposed method (MI+NN) is examined on the day-ahead electricity market of PJM. The obtained results are compared with the results of some other price forecast methods and especially the other feature selection techniques. This comparison indicates the validity of the developed approach.