Research On Monthly Inflow Runoff Prediction Model Of Danjiangkou Reservoir

Danjiangkou Reservoir is located in the middle and upper reaches of Hanjiang River,spanning Hubei and Henan provinces.It is the water source of South-to-North Water Diversion Middle Project in China.The water supply of Danjiangkou Reservoir has effectively alleviated the water shortage of northern China.Since accurately prediction of Danjiangkou reservoir inflow is highly affect the reservoir flood control and water supply,it is crucial to study on prediction of Danjiangkou reservoir inflow.This paper takes Danjiangkou Reservoir as the research object,analyzes the inflow runoff characteristics and evolution law of Danjiangkou Reservoir.On this basis,a variety of monthly runoff prediction models including multiple linear regression model(MLR),artificial neural network model(ANN),random forest regression model(RFR),support vector regression model(SVR),support vector regression model based on improved gray wolf optimization algorithm(IGWO-SVR)and wavelet coupled IGWO-SVR model are established,the research details are listed as follow:(1)The characteristics of annual and monthly runoff series of Danjiangkou reservoir are quantitatively studied by using statistical analysis method.The running average method,Pearson correlation test,Spearman correlation test and Kendall correlation test are used to test the trend of runoff series.The results show that the annual runoff series is not significantly decreasing at a significant level α=5% condition.Mann-Kendall test and Pettitt test are used to test the abrupt change point of runoff series.The results show that the possible abrupt change point years are 1991 and 2013.(2)Based on the analysis of the characteristics and evolution laws of the Danjiangkou reservoir inflow,four monthly runoff prediction models,including multiple linear regression model(MLR),artificial neural network model(ANN),random forest regression model(RFR)and support vector regression model(SVR),were established.The results of case analysis show that SVR model is better than other models in forecasting monthly runoff series of Danjiangkou reservoir.(3)To overcome the drawback of GWO,an improved Grey Wolf Optimization algorithm(IGWO)is proposed by introducing the backward learning strategy based on chaos theory,nonlinear adjustment strategy of convergence factor,dynamic weight strategy and probability disturbance strategy.The effectiveness of the improved Grey Wolf Optimization algorithm is verified by six test functions.Improved Grey Wolf Optimization algorithm is used to optimize the parameters of SVR model;the results of case analysis show that the prediction accuracy and stability of IGWO-SVR model have been improved to some extent.(4)To fixed the poor performance of IGWO-SVR model at the peak runoff prediction,the wavelet analysis theory is introduced to establish the wavelet coupled IGWO-SVR monthly runoff prediction model,and the original monthly runoff series are decomposed into four subseries and reconstructed.The results of case analysis show that the wavelet coupled IGWO-SVR model at the peak runoff prediction is pretty well.Moreover,the NSE of the model in the training period and test period can reach more than 0.9,which indicate the better prediction performance.