径向基函数神经网络在地应力场反演中的应用

初始地应力场是地下洞室设计所需的基本指标。基于江边电站引水隧洞沿线区域的地质资料，利用快速拉格朗日分析程序FLAC3D建立了该区域的数值计算模型，使用了随深度变化的侧压力系数以及应力和位移的混合边界条件，通过模拟引水隧洞沿线不同岩性的岩石、断层破碎带和蚀变带进行正分析计算。根据工程现场实测点的主应力数据，基于径向基函数（RBF）神经网络原理，反演了计算区域的岩体力学参数和初始地应力场。其计算结果与实测地应力值基本吻合，满足精度要求，说明反演结果与工程实际相符，所使用的反演方法是合理的。

Application of radial basis function neural network to geostress field back analysis

Initial geostress field is a basic index for underground engineering design.According to geological data of Jiangbian hydropower station diversion tunnel area,a numerical model of this area is established by fast Lagrangian analysis of continua in three dimensions(FLAC3D).The mixed boundary conditions and lateral pressure coefficients which change with depth are applied to simulate different kinds of rocks,fault fracture zone and alteration zone along the diversion tunnel,so as to conduct positive analysis.According to the measured principal stress data of the engineering site,based on the principle of radial basis function(RBF) neural network,back analyses of rock mechanical parameters and initial stress field are made.The calculated results are consistent well with the measured results and meet accuracy requirements.The results show that the back analysis results are in line with the actual project;and back analysis method is reasonable.