基于GA-BP的大坝渗透系数多目标反演分析方法  

Multi-objective Inversion Analysis Method for Dam Permeability Coefficient Based on GA-BP

作　　者：唐少龙[1] 熊威[1] 万小强[1] 罗梓茗 万思源 汪庆[1] TANG Shao-long;XIONG Wei;WAN Xiao-qiang;LUO Zi-ming;WAN Si-yuan;WANG Qing(Jiangxi Provincial Institute of Water Sciences, Nanchang 330029,China)

机构地区：[1]江西省水利科学研究院,南昌330029

出　　处：《中国农村水利水电》2020年第9期213-216,共4页China Rural Water and Hydropower

基　　金：江西省重点研发计划项目(20181BBG78004);江西省水利厅科技项目(201820ZDKJ12)。

摘　　要：渗透系数是进行工程渗流计算分析的重要参数,针对渗透系数多目标反演问题,构建渗透系数与测点压力水头为训练样本,采用BP神经网络对大坝渗透系数进行反演;针对BP神经网络收敛速度慢、泛化能力差的缺点,通过遗传算法对BP神经网络权值、阈值进行优化,最终形成GA-BP神经网络多目标渗透系数反演模型,并进行实例验证。结果表明:基于GA-BP神经网络反演所得渗透系数用于渗流分析所得观测点压力水头与实测值相对误差最大为3.6%,结果合理可信,并且在收敛速度和精度上优于传统BP人工神经网络。The permeability coefficient is an important parameter for the calculation and analysis of engineering seepage.For the multi-objective inversion problem of permeability coefficient,this paper constructs the permeability coefficient and measuring point pressure head as training samples,inversion of dam permeability coefficient by BP neural network.Optimization of BP neural network weights and thresholds by genetic algorithm,overcome the shortcomings of slow convergence and poor generalization ability of BP neural network.Finally,the GA-BP neural network multi-objective permeability coefficient inversion model is established and verified by examples.The results show that the relative error between observation point pressure head and measured value is 3.6%when the permeability coefficient of GA-BP neural network inversion is used.The result is reasonable and credible,and superior to traditional BP artificial neural network in convergence speed and accuracy.

关 键 词：GA-BP人工神经网络 GA遗传算法 多目标反演分析 正交设计 渗透系数 

分 类 号：TV139.14[水利工程—水力学及河流动力学]