基于ELM-GA的复杂土石坝渗透系数反演模型及其应用

Inversion Model of Permeability Coefficient for Complex Earth-rock Dam Based on ELM-GA

土石坝分区多、地质条件复杂,确定现场实际的坝体和坝基的渗透系数一直是其渗流研究的热点和难点。针对土石坝渗透系数多参数反演问题,利用正交设计法构造渗透系数组合,通过有限元分析建立学习样本,借助极限学习机(ELM)的高度非线性映射能力,建立了渗透系数与水头之间的映射关系,利用遗传算法(GA)搜索确定区域内各分区渗透系数,建立了基于人工智能的ELM-GA反演分析模型,最后利用工程实测资料进行了验证。结果表明,将反演所得的渗透系数用于渗流分析时,观测点压力水头计算值与实测值相对误差均在0.2%之内,结果合理可信,精度满足工程要求。

The earth and rockfill dams have many zones and complex geological conditions,so it is always a hot and difficult problem to determine the actual permeability coefficient of dam body and foundation.For the multi-parameter inversion of permeability coefficient of earth and rockfill dam,the orthogonal design method was used to construct the combination of permeability coefficient,and the learning samples were established through the analysis of finite element.The mapping relationship between the permeability coefficient and the water head was established by using the highly nonlinear mapping ability of the ELM.By using GA to search and determine each permeability coefficient of each partition in the region,the inversion ELM-GA model based artificial intelligence was established.Finally,the inversion results were verified by using the measured data of engineering.The results show that the relative error between the calculated value and the measured value of the pressure head at the observation point is within 0.2% when the permeability coefficient obtained from the inversion is used for seepage analysis.The results are reasonable and reliable,and the accuracy meets the engineering requirements.