Streamflow Forecast Research Of A Modified Hybird Model M-EMDSVM

In recent decades,the characteristic of strtamflow becomes non-linear,non-stationary and complexity with the influence of human activities and climatic varication for streamflow growing.It is always a hot and difficult issue to achieve a high accuracy forecast for streamflow.It is of great significance for water resource development and utilization to modify forcast models to enhance the forecast accuracy.Therefore,the main goal of this study ils to develop an accurate forecast model for monthly streamflow in Jialing River Basin and Han River Basin.Firstly,the characteristic of streamflow in Jialing River Basin and Han River Basin has been analysed.Secondly,the analysis of monthly streamflow forecast in Jialing River Basin and Han River Basin have been carried out to provide technical support for the Jialing River to Han River project.The main results of this study are as follow:(1)The characteristic of periodicity,trend,break of streamflow have been analysed using hydrological statistical method and nonlinear theory at Lueyang hydrological station in Jialing River Basin and Huangjinxia reservoir,Sanhekou reservoir in Han River Basin.(2)In this paper,the Autoregressive moving average model(ARMA),Artificial neural network(ANN),Support vector machine(SVM),EMD-SVM and modified EMD-SVM have been built and the advances,disadvantages and applicability of each model has been analysed.(3)The single models of ARMA,ANN,SVM were employed to forecast the monthly streamflow at Lueyang,Huangjinxia and Sanhekou stations,and the performance of SVM was best among the three models with values of NSE were 56.07%,50.49%and 50.40%,respectively.It shows that SVM model has a better performance for nonlinear and non-stationary streamflow series than ARMA model and ANN model.(4)A hybrid model EMD-SVM has been employed to forecast monthly streamflow at Lueyang hydrological station in Jialing River Basin and Huangjinxia reservoir,Sanhekou reservoir in Han River Basin.The performance of EMD-SVM was better than SVM with the values of NES were 80.47%,76.13%and 68.80%,respectively,which proved that the hybrid model can improved the forecasting accuracy,effectively.(5)In this paper,a linear extremum extension method has been raised to improved EMD.Then,a modified hybrid model M-EMDSVM has been built and employed to forecast monthly streamflow at Lueyang hydrological station,Wuhou hydrological station in Jialing River Basin and Huangjinxia reservoir,Sanhekou reservoir in Han River Basin.The values of NSE of M-EMDSVM were 90.55%,91.88%and 94.26%,respectively,and the values have been improved 13%,21%and 37%respectively than that of EMD-SVM model.It proved that the linear extremum extension method can resolve the boundary effect problem of EMD,effectively,and can further improved the forecast accuracy of M-EMDSVM model.(6)The forecast accurate of ARMA,ANN,SVM,EMD-SVM and M-EMDSVM models has been analyzed using RMSE,MAPE and NSE.Combining the values of RMSE,MAPE and NSE and scatter diagram confirmed the high accurate of M-EMDSVM model for monthly streamflow at Lueyang hydrological station,in Jialing River Basin and Huangjinxia reservoir,Sanhekou reservoir in Han River Basin.