城市地下空间致灾构造三维电阻率非规则阵列探测与智能反演成像理论

Many engineering disasters such as water inrush, collapses, damage to buried pipelines occur during the exploitation of urban underground space. And the main reason is that detection results of disaster causing structures are inaccurate by existing technologies due to the complicated urban environment. This project focuses on the theoretical difficulties in the data acquisition, inversion and interpretation in 3D resistivity detection, and high water content geological structure, buried pipelines are taken as the main detection targets. Theoretical and experimental research methods are used in this project. Firstly, for the information acquisition, unstructured model is used to simulate the electric field in complicated urban environment, deep study of non-contact induction measurement is taken, and irregular array observation mode is proposed. Secondly, for the data inversion, modeling and learning algorithms of deep learning are analyzed. Particle swarm optimization, differential evolution, linear inversion joint, Bayesian regularization, and GPU parallel methods are used to optimize the inversion, then 3D intelligent resistivity inversion based on deep learning is established. Thirdly, for the data interpretation, disaster causing structures are recognized and characteristic parameters are achieved by the image classification. The relationship between physical structure and abnormal structure in inversion results is established, then man-machine interactive interpretation and recognition method of disaster causing structures and quantitatively representation method are proposed. The intended results of this proposed project have important theoretical significance and great application value to the disaster prevention in the exploitation of shallow urban underground space.

城市地下空间开发过程中经常出现透水塌陷、既有管线破坏等工程灾害，主要原因是城市环境极为复杂，现有技术对致灾构造探测结果不准确。本项目针对城市复杂环境三维电阻率探测信息获取、数据反演、解译识别存在的理论难题，以高含水率地质构造、既有管线为主要探测对象，采用理论与试验并重的方法开展研究。首先，针对信息获取，以三维非结构化模型研究城市复杂环境地电场特征，突破非接触感应测量方式，提出任意布极的非规则阵列观测模式；然后，针对数据反演，研究深度学习建模与学习算法，采用粒子群优化、微分进化、线性联合、贝叶斯正则化、GPU并行的策略进行反演优化，形成三维电阻率深度学习智能反演成像方法；最后，针对解译识别，以图像分类方法进行致灾构造识别与特征参数提取，建立实物与探测结果异常构造的映射关系，形成致灾构造人机交互式解译与定量表征方法。研究成果对城市浅层地下空间开发灾害预防具有重要的理论意义和重大的应用价值。

城市地下空间开发过程中经常遭遇透水塌陷、既有管线破坏等灾害，为解决城市复杂环境导致致灾构造探测不准确的难题，本项目提出开展三维电阻率非规则阵列探测与深度学习反演成像研究。针对城市复杂环境，建立低阻异常体目标区域局部加密的非结构化计算模型，实现了非规则测线电极布置。针对非规则阵列观测装置，以局部优化算法为基础，通过引入优度函数优选观测装置集，建立了三维电阻率探测任意布极条件下观测装置优化方法。构建了三维电阻率探测深度学习神经网络ERI-MCNN，通过对预测模型与原始模型拟合的损失函数施加深度-距离加权约束，提高了探测深部区域反演效果。建立了多种低阻异常组合的样本数据集，进行了三维电阻率探测深度学习反演成像数值试验，开发了三维电阻率探测深度学习反演GPU并行加速算法。经过工程现场验证，形成了三维电阻率非规则阵列探测与深度学习智能反演方法，对解决城市复杂环境致灾构造探测难题具有重要意义。