基于BP神经网络的人机联合率定SWMM研究

为解决雨洪管理模型(SWMM)在率定过程中参数复杂、过程繁琐等问题。以西宁市某地块为例建立SWMM模型,利用Morris筛选法进行参数灵敏度分析,并根据灵敏度分析的结果进行人工率定;另外利用BP神经网络对模型进行率定,并结合参数灵敏度对其进行优化。对3种率定方案进行分析,结果表明:水文水力模块参数的相对灵敏度基本一致,其中灵敏度较大的参数为子汇水区面积(Area)、不透水率(Imperv)和不透水区洼地蓄积量(Destore-Imperv),并且不同降雨条件下模型参数的灵敏度存在差异。经过优化后的BP神经网络参数率定方法的模型模拟效果最好,纳什系数最大。结合灵敏度优化BP神经网络的人机联合率定方法一方面能提高BP神经网络率定的准确性,另一方面又能提高传统人工率定的效率。

Research on the Man-machine Joint Calibration SWMM Based on BP Neural Networks

In order to solve the problems of stormwater management model(SWMM)in calibration process,such as complex parameters and tedious process.This paper takes a certain block of Xining city as an example to establish SWMM model.Morris screening method is used to analyze the sensitivity of parameters,and artificial calibration is carried out according to the results of sensitivity analysis.In addition,BP neural network is used to calibrate the model,and parameter sensitivity is combined to optimize it.Through the analysis of three calibration schemes,the results show that the relative sensitivity of hydro-hydraulic module parameters is basically same,among which the more sensitive parameters are sub-catchment Area,Imperv and Destore-imperv.Moreover,the sensitivity of model parameters is different under different rainfall conditions.The optimized BP neural network parameter calibration method has the best simulation effect and the Nash coefficient is the largest.On the one hand,the method which combines with sensitivity to optimize BP neural network can improve the accuracy of BP neural network calibration,on the other hand,it can improve the efficiency of traditional artificial calibration.