为精准刻画三维地质体模型内部空间分布特性，深入探究其复杂结构与物质赋存规律，以超前地质信息力学参数可视化解译为目的，提出一种基于粒子群优化（particle swarm optimization, PSO）算法与支持向量回归（support vector regression, SVR）的普通克里金插值法（ordinary Kriging, OK），并使用一种半均匀网格三维地质建模方法。首先，根据钻孔数据，结合岩性描述、深度信息以及取样位置对三维地质建模区域进行地层划分； 然后，针对地质层面的钻孔稀疏数据，采用普通克里金插值法实现空间属性的最优无偏估计，同时利用支持向量回归模型对高程残差进行校正，并通过粒子群优化算法实现参数优化； 最后，通过三维地质建模算法建立超前地质解译区域内的三维复杂地质体模型。采用该方法可以有效表征地层互层、尖灭等特殊地质现象，满足超前地质信息力学参数解译后内部属性信息三维映射精度的要求。
Given the demands of highly concealed and complex construction and operation, digital intelligent management and control of engineering projects has become an inevitable trend. To accurately characterize the internal spatial distribution of a three-dimensional (3D) geological model, detect its complex structures and material storage rules, and realize visualization interpolation of advance geological information, an ordinary Kriging (OK) interpolation method based on particle swarm optimization (PSO) and support vector regression (SVR) is proposed using a semiuniform grid 3D geological modeling method. First, stratigraphic division of the 3D geological modeling area is performed based on borehole data, lithological descriptions, depth information, and sampling locations. Then, based on sparse borehole data at the geological level, the OK interpolation method is adopted to achieve optimal, unbiased estimation of spatial attributes. Concurrently, an SVR model is used to correct elevation residuals with parameter optimization realized through PSO. Finally, a 3D geological modeling algorithm is employed to construct a 3D complex geological body model in the advance geological interpretation area. This approach effectively characterizes special geological phenomena, such as stratum interbedding and pink-out, while satisfying 3D mapping accuracy requirements for internal attribute information following mechanical parameter interpretation of advance geological information. It lays a foundation for reconstructing 3D geological and tunnel surface models at the engineering regional scale using multisource information and geological constraints and provides visual models for multiregion parallel-solving algorithms tailored to tunnel construction.
