【摘要】 针对基于钻孔数据的传统显式三维地质建模方法存在过程繁琐、模型质量难以保证等缺点,本文提出了一种基于机器学习的隐式三维地质建模方法,将地层三维建模问题转换为地下空间栅格单元的属性分类问题.分别基于支持向量机、BP神经网络等分类算法,实现了钻孔数据的自动三维地质建模.实际建模结果表明,对于有限、稀疏的钻孔数据,支持向量机方法建模准确率较高,建模效率、效果优于显式建模方法.最后通过敏感性分析研究了超参数对建模结果准确率、模型形态的影响,为可控的自动三维地质建模提供了一种新的解决思路.更多还原

【Abstract】 Considering the complex modeling process and difficulty in guaranteeing the model quality of traditional explicit 3 D modeling methods,an implicit 3 D geological modeling method for borehole data based on machine learning was proposed,which transformed the strata 3 D modeling problem into a process of geological attribute classification of the underground spatial grid units. Based on the classification algorithms of support vector machine and BP neural network,automatic 3 D geological modeling from borehole data was realized. The results demonstrate that for sparse and limited borehole data,support vector machine can generally perform better than explicit methods. Finally,the influence of hyper-parameter on modeling accuracy and model shape is studied through sensitivity analysis,which provides a new solution for controllable 3 D geological modeling.更多还原