Study On Combined Prediction Model Of Medium And Long-term Runoff Based On Model Screening

Water resources are an indispensable key factor for the development of human society.Accurate mid-and long-term runoff forecasts play a vital guiding role in the rational development and utilization of water resources and the improvement of comprehensive utilization of water resources.However,with the development of social economy,the accuracy of traditional runoff prediction methods can no longer meet the actual needs of the society,and it is urgent to find a method to improve the accuracy of runoff prediction.At present,the combined forecasting method provides another way to improve the accuracy of runoff forecasting,and it has become a hot issue in the study of runoff forecasting.The basic idea of combined forecasting is to combine the forecasting results of each model in a certain way,and comprehensively utilize the advantages of each forecasting model,so as to achieve the purpose of improving forecasting accuracy and stability.For the combination forecasting model,the key question is how to choose the single forecasting model participating in the combination,and how to assign appropriate weights to the selected forecasting model.Regarding these two issues,scholars at home and abroad have carried out a lot of research,but they have not formed a unified conclusion.Moreover,the current model selection methods and weight determination methods both have a certain optimization space,which is worthy of further study.Therefore,this article has carried out some exploratory researches on these issues,the main research contents are as follows:Aiming at the problem of how to effectively select the single prediction model participating in the combination,a single model stepwise screening method based on the entropy weight-grey correlation-TOPSIS method is proposed.This method combines the traditional TOPSIS method with the grey relational analysis method,replaces the traditional Euclidean distance with grey relational degree,and constructs the comprehensive closeness of each single predictive model and the ideal model.And sort the models according to the degree of comprehensive closeness,and use the model ranked first as the basic predictive model.Then by adding other individual models to the basic model to establish a combined forecasting model,the individual forecasting models are gradually screened.This effectively avoids the blindness of directly selecting the single predictive model participating in the combination and the one-sidedness of using only a single index to select the model.The current combination forecasting model established with a single criterion cannot improve other criterion indicators,and has certain limitations.To solve this problem,a non-linear optimization combination model based on different criteria and support vector machines(SVM)is established.Firstly,based on the selected single prediction model,the linear combination prediction model is established with the minimum average absolute error(MAE),minimum average relative error(MRE),and minimum root mean square error(RMSE)as the optimization criteria.The results show that the prediction accuracy of these three linear combination prediction models is higher than that of the single prediction model.However,these three combined models have not improved other error criteria,and there is still room for improvement and optimization.In order to take into account the advantages of the three linear combination prediction models,the SVM method is used to combine them non-linearly,and a non-linear optimization combination model based on SVM is established.And compared with several other forms of combined forecasting models,the results show that the SVM-based nonlinear optimized combination model can further improve the performance of forecasting.The weight value assigned by the traditional runoff combined prediction model to each individual prediction model is fixed,which is not consistent with the actual situation and will affect the prediction effect of the runoff combined prediction model.In response to this problem,this paper introduces the Induced Ordered Weighted Average(IOWA)operator into the runoff combination model,and establishes a basic IOWA operator combination prediction model with the minimum sum of squares of prediction errors as the optimization criterion.It can assign the corresponding weight value to the single-term model according to the prediction accuracy of the single-term prediction model at each time point,which overcomes the shortcomings of the traditional combined prediction model and improves the accuracy of the prediction.Considering that the combined forecasting model based on the error sum of squares may encounter the error "magnification" effect,two robust correlation indicators,the gray correlation degree and Theil inequality coefficient,are introduced.A combined prediction model of IOWA operator based on correlation index was established,which further improved the accuracy of prediction and improved the performance of prediction.