The Single Model Chosen And Parameters Optimization For Combined Forecasting Model

The combined model is an integrated model that makes full use of various fore-casting models, which was presented by Bates and Granger. The combined forecasting theorem made a great progress in the past forty years. The results shown that the com-bined model indeed provides more stable and reliable forecasting results than the single models. Therefore, it with important practical significance to study the combined model.Although there is a great progress in forecasting theorem study, its research needs to strengthen. Because most of the current studies focused on the weight coefficients and objective functions chosen, very few papers study other topics of combined model. The further study is significant to perfect combined forecasting theorem and promote the forecasting developments.This paper presents the single model selection for the linear combined forecasting model. The results shown that the forecasting precision can be improved by the single model selection, so it provides a new method for improving the forecasting precision of the combined model. For the linear combined model, the author point out that significant test is necessary to prove the combined forecasting model is reasonable. Therefore, it makes up the deficiency of the combined forecasting model.The mean of the single model is adopted initially for the combined model. Then, some researchers applied the game theory principles to seek the optimal weight coeffi-cients, such as Shapley value. With the development of intelligent optimization algo-rithm, it has found more application domains. Therefore, the author proposed that the weight coefficients can be found by the chaotic particle swarm optimization algorithm. And the results proved that the chaotic particle swarm optimization algorithm is an effec-tive method for the determination of weight coefficients. The proposed method is applied to the load forecasting and the results are satisfactory.