Study On Manas River Runoff Evolution Characteristics And Its Mediumand Long-Term Forecast Model

Affected by climate change and human activity,the runoff series shows the characteristics of unsteady and non-linear,generally assumed runoff time series in the runoff prediction is smooth,but affected by climate change and human activities in recent years,most of runoff series show that the complex inconsistency,to reduce the prediction error caused by,how to effectively forecast of runoff series,The construction of higher precision forecast model is the focus of hydrological research.Out of mountain pass in manas river runoff time series and the meteorological elements from 1956 to 2014 as the research object,by using the methods of mathematical statistics out of mountain pass to the manas river runoff space-time change rule and the temperature,precipitation,comprehensive analysis and on this basis,build the combination forecast model based on empirical mode decomposition simulated analysis of runoff time series,The applicability of the model in Manas River was explored to capture the potential regularities of runoff.According to the main results of this study,we can understand the historical variation trend of temperature,precipitation and runoff at the exit pass of Manas River.Based on this,we can conduct scientific management and further research in the study area.The main contents and achievements are as follows:(1)Runoff evolution characteristics of Manas River59 years of manas river runoff time series analysis,known years distribution of runoff is extremely uneven,the boundary has the obvious and the plentiful mainly concentrated in May to September,altogether five months in runoff accounted for 84% in the year and runoff in the largest appeared in July,the manas river easy appear dry year and last longer,the longest wet can reach 8 years;And the duration of the wet period is relatively short,which is prone to dry water events.The sliding t test is used to test the abrupt variation of runoff time series,which proves that the runoff changed dramatically in 1995.Mana-Kendall test showed that the runoff had a trend of significant increase.The annual runoff series of the Manas River showed a cycle of 2-6 years,10-30 years and 30-60 years.During the 59 years from 1956 to 2014,the centers of the high-water years were determined as 1966 and 1996,and the centers of the low-water years were determined as 1952,1980 and 2010.It can be determined that 45 years is the primary period and the secondary main period is 16 years to control the cycle change of Manas River.(2)Vector autoregressive combination(EMD-ARIMA)prediction model based on empirical mode decompositionThrough the empirical mode decomposition of monthly runoff data from 1956 to 2014,it is shown that runoff time series has different variation characteristics under different frequencies,and the trend term of the rising trend of runoff in Manas River is obtained.Runoff can be degraded according to different time scales by empirical mode decomposition,and a relatively stable component and a residual term representing the trend of runoff change can be obtained.The runoff time series of Manas River outlet pass is decomposed into four IMF components and a trend term,which have been verified to be stable sequences.The runoff time series was brought into the single model and the combined model for runoff prediction.The runoff simulation accuracy R value of the ARIMA model directly applied to the monthly runoff was 0.91 and the pass rate was47%.The R value of EMD-ARIMA is 0.96,and the pass rate is 72%.The simulation accuracy of EMDARIMA combination is higher than that of single ARIMA runoff simulation,indicating that the EMDARIMA model has more advantages in the process of runoff prediction than the ARIMA model.(3)The prediction model based on climate factor neural network(GRNN)Multiple linear regression method,Spearman correlation coefficient method and average influence value method were used to select atmospheric circulation factors as input items of the neural network model.The prediction results of the GRNN model after screening the predictors have good performance in the process fitting.It can be concluded that compared with the input factors of precipitation and air temperature,the addition of atmospheric circulation factors can improve the runoff prediction accuracy of the single GRNN model,and the qualified rates of the three screening methods can be increased by 5%,10% and 7%,respectively.(4)GRNN model and EMD-ARIMA combined prediction model based on predictor screeningIn order to further improve the accuracy of the model,for the high frequency components of IMF1,IMF2 and IMF3 with poor performance in the ARIMA model,the optimized atmospheric circulation index is taken as the input factor of the GRNN model and runoff as the output factor for prediction.The predicted values of IMF4 and trend term R are combined in accordance with the results of ARIMA.Through the combined prediction of GRNN model and EMD-ARIMA model after screening the atmospheric circulation index,all indexes have been optimized,which is mainly reflected in the improvement of qualified rate by more than30%,and the significant reduction of error,RMSE and MARE by 39% on average.(5)Model evaluation and forecast result analysisThrough the comparative analysis of the results of nine runoff prediction models,it is found that EMD decomposition can improve the eligibility rate of ARIMA model by 25%,but for the high frequency components IMF1,IMF2 and IMF3,the relative error of ARIMA model reaches more than 70%,and the prediction results are poor.The accuracy of GRNN model can be effectively improved after screening the predictors,among which the MIV method is the most suitable for Manas River,and the final response relationship of atmospheric circulation to the runoff evolution of Manas River can be obtained.The GRNN model combined with EMD-ARIMA has the highest pass rate,and the TOPSIS model also has the highest score.