Application of fuzzy cognitive maps to electricity consumption prediction

The purposes of this research are to find a model to forecast the electricity consumption in a household based on fuzzy cognitive map (FCM) prediction capabilities. The data analysis has been performed with three different learning algorithms based on the fuzzy cognitive map model which are (a) the multi-step gradient method (MGM), (b) the real coded genetic algorithm (RCGA) and (c) the structure optimization genetic algorithm (SOGA) as a new improvement of the RCGA able to handle the model's complexity investigated in our study. The suitable forecasting methods and two different forecasting periods were chosen by considering the smallest value of MSE (Mean Square Error) and RMSE (Root Mean Square Error), respectively. The proposed FCM-based prediction algorithms were compared with known statistical methods such as ARIMA and artificial intelligent methods such as ANNs and neuro-fuzzy models. The experiments conducted in this study showed that the multi gradient-based algorithm for FCM learning was the most efficient with respect to the accuracy of prediction expressed by the well-known used errors in the two selected forecasting periods.