Research On Forecasting Method Of Monthly Runoff Based On Variational Mode Decomposition And Bayesian Neural Network

The runoff process is an important part of the hydrological cycle on the earth.The accurate prediction of the runoff inflow is of great significance to the water flow scheduling,water resources planning and management of the basin.However,due to the complexity of the runoff process and the influence of human activities,it is very difficult to accurately capture the variation law of monthly runoff in a changing environment,and it is only possible to predict the monthly runoff time series.However,due to the relatively single information contained in the monthly runoff time series,it is often very difficult to construct a highly accurate forecast model of the monthly runoff time series.Therefore,in order to realize the effective analysis and research of monthly runoff time series,this thesis combines Variational Mode Decomposition(VMD)and Bayesian Neural Network(BNN)to design a model that can realize accurate prediction of monthly runoff time series,and a case application study based on the monthly runoff time series of two hydrological stations in Xianyang and Huaxian in the Weihe River Basin.The specific research work is mainly divided into the following two parts:(1)In view of the difficulty in capturing the internal information features of the monthly runoff time series and the insufficient learning of the prior information in the sample data by the traditional neural network model,a monthly runoff time series prediction model based on VMD and BNN was designed.Based on the good robustness of VMD to noise and the characteristics of accurate decomposition of time series signals,the monthly runoff time series is regarded as a time series signal,and the VMD method is used to decompose the monthly runoff time series into multiple relatively stable intrinsic mode functions(Intrinsic Mode Function,IMF)sequence,each IMF is predicted separately.At the same time,the segmentation model that forecasts all IMFs separately in the previous decomposition ensemble model modeling has been improved.By designing a parallel BNN(Parallel Bayesian Neural Network,PBNN)prediction module based on variational inference,the integrated forecast for all IMFs is realized.Improves the integrity of the predictive model.The VMD-PBNN monthly runoff time series prediction model is constructed by combining VMD and PBNN,which makes the monthly runoff time series prediction results achieve higher accuracy,and the hydrological prediction accuracy level is upgraded from B to A.(2)In view of the problem that the previous decomposition ensemble model cannot adaptively select the embedding dimension of the modal components in the prediction process,the IMFs obtained by the VMD is carried out by using the phase space reconstruction technology(Phase Space Reconstruction,PSR),and the VMD-PSR processing unit is designed to integrate all the reconstructed phase spaces of the IMF,so the adaptive selection of the embedded dimension of the prediction module is realized.Combined with the PBNN prediction module,the VMD-PSR-PBNN monthly runoff time series prediction model is constructed.The prediction accuracy of the monthly runoff time series is further improved.