岩体力学参数反演的代理蜜獾优化方法

针对复杂地下工程岩体力学参数反演时因大量调用数值计算模型导致计算耗时大的问题，提出一种新的仿生优化代理反演方法，即蜜獾优化算法-高斯过程回归-三维快速拉格朗日数值计算(honey badger algorithm-Gaussian process regression-FLAC^3D,HBA-GPR-FLAC^3D)方法。该方法将围岩的实测位移与数值计算结果间的误差作为目标函数，将岩体力学参数作为优化变量，利用全局寻优性能优异的HBA搜索目标函数全局极小值，并采用牛顿优化算法进行当前最优算子邻域的局部寻优，局部寻优中采用GPR代理模型而非基于FLAC^3D计算所构建的目标函数作为算子适应度评价工具。研究表明，与基于单纯仿生优化算法的反演方法相比，在达到相同计算精度条件下，所提出方法的数值模型调用次数显著降低，适用于单次数值计算较为耗时的复杂地下工程岩体力学参数快速识别。

A surrogate Honey Badger Algorithm optimization method for back analysis of rock parameters

Aiming at the problem of time-consuming calculation due to a large number of numerical calculation models used in the back analysis of geomechanical parameters in complex underground engineering, a new bionic optimization surrogate back analysis method (HBA-GPR-FLAC^3D) is proposed. The method takes the error between the numerical calculation results of the surrounding rock and the measured value as the optimization objective function, and the geomechanical parameters as the optimization variables. The HBA with excellent optimization performance is used to search for the global minimum value of the objective function. In the local optimization of the current optimal operator neighborhood, the GPR surrogate model is used as the operator fitness evaluation tool instead of the objective function constructed based on FLAC^3D calculation. The research shows that compared with the back analysis method based on the bionic optimization algorithm, the number of numerical model calls of the proposed method is significantly reduced under the condition of the same calculation accuracy. It is suitable for rock mechanics parameters of complex underground engineering whose single-time numerical calculation is relatively time-consuming.