Weight Changeable Combination forecast method based On ACA

The weighted coefficient in combination forecasting model is constant in most of current studies. In fact, if the weighted coefficient is assumed to be constant, the effectiveness and accuracy of combination forecasting model may be unsatisfied. In addition, samples near the time now should be greater contribution to prediction. Taking into the account of the above reasons, a new combination forecast method with changeable weight considering the importance of sample is proposed. ACA is used to search for the best sample weight, and then PLS is used to confirm the weight function of combination forecast model. An empirical study is given to test the effectiveness of this new method.