Electricity Price Forecasting Based on Multi-models Combined by Evidential Theory
Da Liu · Proceedings of the CSEE · 2008
The combined weights in traditional combined method for electricity forecasting are obtained with calculating the historical forecasting errors, with no considering of the environmental factors. Five models from support vector machine (SVM), artificial neural networks (ANN), and time series forecasting techniques were selected to forecast the electricity price from different views. Four models from ANN and SVM were selected as experts to evaluate the credit of forecasting results of the five above models, with historical forecasting errors and environmental influence. The credit were combined to calculate the weights with Dempster-Shafer (DS) evidential theory. The final forecasting was obtained by the weighted forecasting. The experiment of California power utilities validates the proposed method.