| China is a mountainous country with frequent occurrence of flood disaster,every year because of local heavy rainfall and flood disaster death account for more than70%of the total number of flood disasters death.Therefore,how to improve the prediction and forecasting levels and enhance the flood forecasting period is of great importance.This paper mainly constriucts the RBF neural network prediction model of rainfall runoff in mountain wate,rshed,and takes the upper reaches of Chongyang river in WuYiShan Hydrometric Station as an example to provide the basis for flood control and disaster reduction decision-making.The lmain contents are as follows:(1)The flood propagation time of sub basin is determined by discretization of basin unit.The basin is divided into 7 sub basins.Excluding the influence of WuYiShan Hydrometric Station,the area weights of Daan,Yangzhuang,Kenkou,Wubian,Lingyang,Langu precipitation station in the remaining six sub-basins are 0.20,0.10,0.20,0.22,0.13,0.15.Accorcding to the centroid method,the subbasin area is compared with the actual rainstorm flood propagation time,comprehensive analysis shows that the flood propagation time of 6 sub-basins is 3h,lh,3h,2h,4h,3h respectively.At the same time,the design flood peak flow of each frequency is calculated,and the results and causes of flood are analyzed rationally.(2)Twenty-one flood samples were collected and divided into 12 training samples,3 monitoring samples and 6 testing samples.The appropriate RBF network learning algorithm was selected,the proposed forecast period T is 1h.Considering T and the rainfall process of six precipitation station whose Weighting and logarithmic transformation were carried out in the upper reaches of Wuyishan Hydrological Station,including the Da'an,Yangzhuang,Kenkou,Lingyang,Langu and Wubian,along with the cumulative rainfall process as input variables,the flow process at the outlet section of Wuyishan Hydrometric Station is output,and a model(model I)of RBF neural network prediction of rainfall runoff in the drainage basin is established.the effect of model prediction is relatively poor,with the overall pass rate of55.6%and the average certainty coefficient of0.586.(3)The monitoring sample was added to the training sample,the 15 floods were used as training samples,and the 6 flood samples were taken as the test samples to improve the models.Considering ?i,max{?i?,min{?i} and T,the time-period rainfall process of the above six rainfall stations was used,which was weighted and logarithmically transformed,together with the flow process at Wuyishan hydrological station at time,min{?i},min{?i}+1,max{?i}-1,max{?i},as input and the export section flow process at Wuyi Hydrometric Station as output to establish RBF model.The result of model prediction is great,the average passing rate of single item and comprehensive index is 100%with the average certainty coefficient of 0.952.(4)On the basis of Model ?,RBF,model is constructed by constructing radial basis functions of Gaussian function exponential form and Log S-function form respectively.The single item and comprehensive qualification rate of the two function prediction schemes reached 100%,and the average certainty coefficient is over 0.900.(5)The two water source Xin'anjiang model was used to simulate the 6 floods in the upper reaches of Chongyang River Basin,the comprehensive qualification rate is 88.9%,and the average coefficient is 0.771.(6)The above model is compared,the error is analyzed and assessed,and Model ? is determined to be the best prediction model for the prediction of the upstream catchment of Chongyang river,providing the basis for the flood control dispatching of the basin. |
PDF Full Text Request