Characterization And Prediction Study Of Unsteady Runoff In The Basin

With climate change and human activities,runoff sequences are characterized by non-stationarity and complexity.The study of non-stationary runoff characteristics and prediction in river basins can provide theoretical guidance for the efficient use of water resources in river basins.The thesis is based on the analysis and prediction of unsteady features in the upper basin of Pingtang hydrological station of Chengbi River.(1)Monthly runoff from Pingtang hydrological station in the basin was characterized by trend analysis,abrupt change analysis,cycle analysis and trend continuity analysis.The results show that: in month55 a,runoff from January,March,June,October and December shows an increasing trend,while runoff from the remaining months and annual runoff shows a decreasing trend;March,May and December show a sudden change in runoff,while the remaining months and annual runoff show no sudden change;there are 2-5 scale cycles in runoff from each month and 2 scale cycles in annual runoff;in each month,except for June,runoff from the remaining 11 months and annual runoff will show a decreasing trend in the future;there are 2-5 scale cycles in runoff from each month and 2 scale cycles in annual runoff.Positive persistence characteristics.(2)Two types of single prediction models,support vector machine model(SVM)and neural network model(Elman),were constructed to carry out prediction studies on watershed runoff,respectively.The results showed that: for the single prediction model,the best prediction was obtained by using a 660-month runoff sequence of 55 a as the model data input(NSE=0.61);the SVM,SVM(M),Elman and Elman(M)models(NSE=0.61,0.53,0.57 and 0.50,respectively)were used as the model data input.0.4967),the SVM model predicts the best results.(3)The coupled prediction models based on empirical modal decomposition(EMD),ensemble empirical modal decomposition(EEMD)and empirical wavelet transform(EWT)were constructed to compare the predictions of runoff from watersheds,respectively.The results show that the NSE range of the six coupled prediction models is 0.65-0.86,which is0.08-0.28 higher than that of the single prediction model,which proves that the coupled prediction model based on data pre-processing method effectively improves the accuracy of runoff prediction and can provide technical support for future watershed runoff prediction.