Application of DS evidence theory in combined price forecasting

The method of combined forecasting establishes multiple forecasting models, and calculates the final results by summing up the weighted results of every model. However, it is hard to determine the weights of the combination. The weights of the combination are determined by the history predicted errors in the traditional method of combination. Since it does not take into account of the external environmental influencing factors in forecasting a point price, the historical predicted errors have impact on the forecasting result. Actually, the forecasting precision of different methods is influenced by the environmental factors directly. This paper discusses how to apply DS evidence theory into combined price forecasting with multiple models. The main framework of research is as follow: firstly, modeling data is used to train multiple models, which produces the model and training errors applied in price forecasting; secondly, training errors are converted into credibility of each model in predicting the point. Then, as an output value the credibility is outputted, while the environmental factors of this moment are input variables. As a result, credibility training and forecasting model is built up; thirdly, the environmental factors are regarded as objected input, and the built credibility forecasting model is used to forecast the credibility; fourthly, several models are used to forecast the price respectively; fifthly, DS evidential reasoning method is used to synthetically calculate conformity the credibility of each model, and the combined weight of each model is obtained; sixthly, the forecasted price of multiple models is combined with the predicted weight and the final price is obtained. Finally, a study case serves to analyze and compare combined price forecasting based on DS evidence theory with equally weighted average combined forecasting and variance- covariance optimized combined forecasting and the results illustrate that this model is valid.