小样本数据下三维地应力反演分析

对于地下矿山等工程，初始地应力场的有效获取意义重大，但由于地质情况复杂，实测数据不足等原因，小样本数据下的三维地应力反演分析不易进行。本文在进行现场应力测量的基础上，提出了一种以边界荷载法为基础，结合反向传播神经网络（back propagation neural network，BP）的Flac3D地应力反演改进方法。以建个元高速公路五老峰隧道为示例，首先通过建立反演区域的三维地质模型，通过边界荷载法试算各边界条件的大致范围，再利用BP神经网络的非线性特征建立反演参数与地应力之间的函数关系，最后进行正算反推求解反演最佳边界条件。实测值与反演值进行对比，误差在可接受的范围内，证明了该方法在复杂地质体地应力反演中的可靠性与实用性。

Three Dimensional In-situ Stress Inversion Analysis With Small Sample Data

For underground mines and other projects, the effective acquisition of in-situ stress is of great significance. However, due to complex geological conditions and insufficient measured data, it is difficult to carry out three dimensional in-situ stress inversion analysis with small sample data. Based on in-situ stress measurement, an improved method of Flac3D in-situ stress inversion is proposed in this paper. This method combines boundary load method and back propagation neural network. Firstly, taking the expressway tunnel as an example, the three-dimensional geological model of the inversion region was established, and the approximate range of each boundary condition was calculated by the boundary load method. Then, the functional relationship between inversion parameters and ground stress was established by using the nonlinear characteristics of BP neural network. Finally, the optimal inversion boundary condition was solved by forward calculation and backward calculation. The measured value is compared with the inversion value, and the error is within the acceptable range, which proves the reliability and practicability of the method in the inversion of in-site stress.