Study On Rainfall And Runoff Forecasting Model Of BP Neural Network In Mountainous Watershed

The hydrometeorological and topographical conditions in the mountainous watershed are complex,and different levels of flood disasters occur every year.It has a great impact on the national economy and social harmony and stability.Therefore,it is of great significance to do well in flood forecasting in mountainous watershed to reduce the loss of flood disasters and promote the development of national economy.Based on the reality of flood control and disaster reduction in the watershed,this paper applies the principle of BP neural network to the upstream watershed of Chongyangxi in the mountainous area of northern Fujian,and establishes the BP neural network model to forecast rainfall and runoff.The main contents are as follows:(1)The watershed boundary is extracted from the watershed DEM data,and the watershed is divided into seven unit sub-basins by Tyson polygon method.The area weights of the corresponding seven sub-basins of the seven rainfall stations controlled by Wuyishan,Yangzhuang,Wubian,Da'an,Hangkou,Lingyang and Shibuya are 0.08,0.20,0.19,0.15,0.13 and 0.18,respectively.Based on the measured data,the time of flood transmission in the sub-basin of the unit is determined,the designed flood discharge of each frequency is deduced,and the flood characteristics of the watershed are analyzed.(2)The data of 14 flood discharge processes in the upper reaches of Chongyangxi watershed are selected as training samples.The BP neural network of momentum gradient method is established by taking the rainfall data of Da'an,Yangzhuang,Wubian,Kengkou,Lingyang and Langu stations as input and the flow of one hour before Wuyishan hydrological station as output.Network rainfall runoff forecasting model and LMBP neural network rainfall runoff forecasting model(model1 and model 2).The rainfall-runoff model is validated by using 7 flood process data in the upper reaches of Chongyangxi watershed.The average deterministic coefficients of the two models are 0.983 and 0.975,respectively.The comprehensive qualified rate is 100%,and the accuracy meets the code requirements.(3)Using the above 14 flood samples,the flow of six rainfall stations in the upper reaches of the basin and the flow of one hour and two hours before the Wuyishan hydrological station are taken as input,and the flow of Wuyishan hydrological station is taken as output.Elastic gradient algorithm and Loak-Ribiere conjugate gradient algorithm are used to establish BP neural network rainfall runoff prediction model(model 3,model 4).Similarly,seven flood samples are used to validate the model.The average deterministic coefficients of the two models are 0.985 and 0.989,respectively.The comprehensive qualified rate is 100%,and the accuracy of the two models meets the code requirements.(4)Xin'an River Three-source model is applied to the upper reaches of chongyang river,and the above seven flood samples are simulated.The simulation results are inferior to the above 4 models,with an average deterministic coefficient of0.912.The comprehensive qualified rate is 71.4%.(5)The five models are compared and analyzed.The errors are analyzed and evaluated in five aspects: the average value of absolute relative error of peak discharge,the average value of absolute relative error of process forecast,the average value of absolute relative error of runoff depth,the average value of absolute error of peak-current time and the average deterministic coefficient.The model 3 can provide the basis for flood control and disaster mitigation in the upper reaches of Chongyang River basin.