Variation Characteristics And Forecast Of Runoff In The Upper Reaches Of The Yangtze River Based On Extreme-point Symmetric Mode Decomposition

Runoff change characteristics and runoff forecast are the prerequisites and foundations for the rational development and effective use of water resources.The accuracy of runoff forecast is directly related to maximizing the comprehensive benefits of optimal allocation and reasonable utilization of water resources in the river basin.In view of the fact that runoff series are non-stationary series with large-scale cyclic and non-linear trend aliasing,the direct prediction accuracy is low.Extreme-point.Symmetric Mode Decomposition(ESMD)method has studied the changing characteristics and forecasting problems of runoff sequences.In terms of runoff sequence changes,first,the ESMD method was used to decompose the annual,monthly and daily runoff sequences of eight stations in the upper reaches of the Yangtze River into steady modal components and trend residuals at different time scales,which effectively screened large cyclic and non-linear trends.Then,the runoff sequence is obtained by using the fast Fourier transform(FFT)periodogram,the trend remainder under the best adaptive global mean curve(AGM),and the frequency and amplitude time-varying diagram of ESMD time-frequency analysis.Multi-time scale periodic changes,non-linear trends and abrupt changes,and grasp the changing characteristics of runoff sequences.Finally,the results are compared with Morlet wavelet transform,Mann-Kendall(M-K)trend test,Hurst index,Mann-Kendall(M-K)mutation test and sliding T test.The results show that there are periodic changes in the time scale of the eight main and tributary rivers in the upper reaches of the Yangtze River.Except for the increase in runoff at Zhuxi Station,the runoff sequences of other hydrological stations have a decreasing trend.The flow will continue to decrease;due to the combined effects of climate change and human activities,the mutation times of the hydrological stations in the main tributaries will also be different.In terms of runoff prediction,first,the ESMD-BP neural network combined forecasting model was established using the ESMD method to smooth the processing technique and an error back propagation network(Back Propagation Neural Network,BP neural network)that can approximate any nonlinear mapping.Then,using the ESMD method smoothing processing technology,combined with Elman Neural Network(ENN)nonlinear approximation speed and good dynamic characteristics,an ESMD-ENN combined forecasting model was established.Finally,the two combined forecasting models were applied to the annual,monthly,and daily runoff forecast of Cuntan Station and Wulong in the upper reaches of the Yangtze River,and a single BP neural network and a single ENN were used for comparative analysis.The results show that the two combined forecasting models are "decomposition,prediction,and reconstruction" modes,which combine the advantages of adaptive analysis of ESMD data,smoothing processing,and nonlinear approximation of neural networks,which can improve the accuracy of runoff prediction.Compared with the ESMD-BP neural network combined forecasting model,the ESMD-ENN combined forecasting model has an additional correlation layer,which increases the ability to process dynamic information and has higher prediction accuracy.Finally,this paper uses four distribution models:normal distribution,t location-scale distribution,stable distribution,and logistic distribution to fit the runoff forecast errors,and to optimize the best model for describing runoff forecast errors.By superimposing the forecast error simulated by the optimal model with the forecast value of runoff and correcting the forecast,the forecast accuracy and level can be effectively improved,and the forecasted runoff process will be closer to the actual incoming water.Furthermore,it provides a reliable basis for medium and long-term and short-term operation and decision-making of the reservoir.