Research On Multi-model Comprehensive Forecasting Method With Variable Weights

The prediction of system response is a complex process that is affected by many factors.In order to solve the system response prediction problem,a better strategy is to combine several suitable models for prediction.The weights of the combined forecasting methods are mostly deterministic values,and the combined forecasting models with fixed weights sometimes do not reflect the actual situation well.The variable-weight combined forecasting method improves the reliability and accuracy of the forecast by associating the weighted value with the variable value.This paper further explores the issues of solving the weight function and the estimation of the confidence interval in variable-weight combined forecasting.The main content of this article is as follows:Firstly,model-combined forecasting,model uncertainty,and several common methods for solving weights in combined forecasting are introduced.The existing problems in combined forecasting are analyzed,and the advantages of variable weight combined forecasting method are addressed.The key issue of variable-weight model combined forecasting is how to solve the weight function.In addition,the confidence interval of variable-weight combined forecasting needs to be estimated.Secondly,two new models of variable-weight combined forecasting are proposed.The first is that a variable-weight combined prediction model based on the fitting method.By creating new optimization problems,based on the corresponding relationship between data-weight values obtain the optimal weight function.Through examples,the accuracy and reliability of the proposed model prediction methods are shown.The second is that a variable-weight combined prediction method based on the bootstrap method and maximum entropy.The bootstrap method for solving the confidence interval has the advantage of that it is independent of sample distribution.The maximum entropy method can obtain the optimal weight function within the confidence interval of a given weight.The example results show that this model has higher prediction accuracy than the traditional variable-weight combined prediction model.Finally,there are deficiencies in the estimation of the confidence interval for the variable-weight combined forecasting method.Two new estimation methods for predicting confidence intervals of variable-weight combined forecasting are proposed.The first is that confidence interval estimation method of variable-weight combined forecasting based on the bootstrap method.The second is that the method of estimating the confidence interval of variable-weight combined prediction model based on the adjustment factor method.It is used to express the difference between the variable-weights combined and the single model for the prediction results,and then the confidence interval is estimated.Examples are given to show the effectiveness and feasibility of the new methods for confidence interval estimation.