Coupled Model Medium And Long-term Runoff Prediction And Its Correction Study In Chengbi River Basin

Changes in watershed runoff affect issues such as reservoir operations and water allocation,which in turn affect the economic development of society.With climate change and increasingly intense human activities,the study of runoff changes has become a matter of great interest to hydrologists.Meanwhile,due to the rise of artificial intelligence and the maturity of multidisciplinary fusion technology,machine learning and coupled model prediction research become research hotspots for runoff prediction.This paper takes the Chengbi River basin as the research object,and takes the measured monthly flow depth at the dam head station as the research data,and proposes a runoff prediction model method adapted to the basin,which can provide a reference for the efficient use of water resources in the basin.The main research content and results of this paper are as follows.(1)The Mann-Kendall method,the linear trend method,the M-K trend method,and the Morlet wavelet analysis method were used to conduct sudden change analysis,trend analysis,and period analysis of monthly runoff depth,monthly maximum runoff depth,and monthly minimum runoff depth at the dam head hydrological station in the basin,respectively.The results show that: the monthly runoff depth at the Bashou hydrological station show abrupt change points between 2015 and 2017 during a total of 492 months in 41 years;Monthly extreme runoff depths show abrupt change points during 2014-2015;The abrupt change point in the monthly minimum runoff depth occur during1987-1988.The monthly runoff depth shows an increasing trend,with the largest increase among the three series;The monthly maximum runoff depth shows an increasing trend with the second largest increase;Monthly minimal runoff depth shows an increasing trend with the smallest increase.Monthly runoff depth has 4 scale periods;Monthly maximum runoff depth has 4 scale periods;The monthly minimal value runoff depth has 3 scale periods.(2)Prediction study of monthly runoff depth in watersheds based on three single models: BP model,SVM model and Elman model.The results show that:in the three prediction models,BP model,SVM model and Elman model,the NSE is 0.4880,0.4459 and 0.4282,respectively.Among them,the BP model predicts the best results.(3)Study of monthly runoff depth in watersheds using PSO-BP,PSO-SVM and PSO-Elman models based on particle swarm optimization algorithms.The results show that: among the coupled prediction models of PSO-BP model,PSO-SVM model and PSO-Elman model,the NSE values are 0.4924,0.5017 and 0.6194,respectively,and the PSO-Elman model has the best prediction.(4)The nine pre-treatment methods coupled with the BP model,SVM model and Elman neural network model were used to build a coupled model based on three pre-treatment methods,EMD,EEMD and EWT,and these nine models were used to study the monthly runoff depth of the watershed.The results show that: Among the three models coupled with the BP model,the EWT-BP model has a better prediction,with a forecast grade of A;Among the three models coupled with the SVM model,the EWT-SVM model has a better prediction with a forecast rating of A;Among the three models coupled with the Elman model,the EWT-Elman model has better forecasting results with a forecast grade of B.(5)Study of monthly runoff depth in the basin using EWT-PSO-BP coupled model,EWT-PSO-SVM coupled model and EWT-PSO-Elman coupled model.The results show that: The EWT-PSO-Elman coupled model has the best prediction.(6)The EWT-PSO-BP model,EWT-PSO-SVM model and EWT-PSO-Elman model were calibrated for prediction effectiveness using quantile mapping method.The results show that: The corrected models are all optimized in all three metrics compared to the pre-corrected models.Among the three model corrections,the EWT-PSO-Elman model has the best result after correction,and the forecast grade reaches A,which can improve the forecast accuracy well.