复杂地质条件下坝区初始地应力场二次反演分析研究报告

李海波　裴启涛　刘亚群

摘要：为准确获得南水北调西线工程加塔坝区初始地应力分布规律，提出综合反映工程区复杂地质条件及地层剥蚀过程的地应力场二次反演方法。首先，主要考虑工程附近大区域内的地形地貌、断层褶皱及河谷的发育演化史等因素，建立坝区大尺度计算模型，采用遗传神经网络法及FLAC3D计算程序对坝区初始地应力场进行一次反演；然后，考虑坝区附近主要地质构造，建立小尺度精细模型，通过从一次反演中提取小范围模型边界上的应力值进行拟合，初步获得精细模型的非线性边界条件，并采用遗传神经网络法对边界参数进行优化，对初始地应力场进行二次反演。研究结果表明：一次反演计算获得的初始地应力场在局部构造附近与实测值相差较大；二次反演考虑局部地质构造的影响，同时结合一次反演的计算成果，各测点的应力计算值与实测值吻合更好。

关键词：地应力场；反演；地层剥蚀；非线性边界；遗传神经网络

Abstract：In order to accurately obtain the distribution rule of initial geostress field in Jiata dam area in west route of South-to-North water transfer project，a new two-stage back analysis method of initial geostress field which considers the complex geological conditions and the ground denudation process is presented. Firstly，considering mainly the influence of topography，faults，folds，and developmental history of river valleys，a large-scale calculation model for the dam area is established. And the genetic neural network method as well as FLAC3D is applied to the first-stage back analysis of initial geostress field. Then，considering the main geological structures in the vicinity of the dam area，a small-scale refined model is created. The preliminary non-linear boundary conditions of the refined model can be obtained by fitting the stress values which are extracted from the large-scale model of the first-stage back analysis. After that，calculation for the second-stage back analysis of geostress field is conducted by optimizing the boundary parameters based on genetic neural network method. It is shown that the difference between the calculated results and the measured data is large in the vicinity of the local geological structures during the first-stage back analysis. And the calculated results are much closer to measured data in the second-stage back analysis by considering the small geological structures as well as the results in the first-stage back analysis. Therefore，the proposed method in this paper provides great reference to similar projects.

Keywords：geostress field；back analysis；ground denudation；non-linear boundary；genetic neural network