摘要
隧道施工期超前探测对于避免突涌水灾害的发生具有重要作用,为满足隧道三维电阻率超前探测快速化解译与成像的要求,本文提出了一种基于GPU并行的蚁群算法与最小二乘方法相结合的混合反演算法.该方法结合线性反演与非线性反演的优点,利用蚁群算法全局搜索能力强的优点为最小二乘反演提供较优的初始模型,以克服最小二乘算法容易陷入局部最优的缺点,提高了隧道三维电阻率反演成像的精度.同时,基于蚁群算法的天然并行性,提出了CUDA环境下的GPU并行策略,实现了三维电阻率反演的快速化成像.其次,开展了基于GPU混合反演的数值算例,与传统最小二乘线性反演进行了对比,基于GPU并行计算的混合反演计算效率得到了显著提高,对含水构造的位置、形态有较好的反映,压制了三维反演的多解性.最后开展了物理模型试验,结果表明基于GPU混合反演探测的低阻异常体与实际含水构造的位置较为相符,发现基于GPU加速的混合反演方法在提高探测精度与加快反演速度方面具有显著优势,为三维电阻率混合反演方法在隧道超前探测实际工程中的应用奠定了基础.
Ahead prospecting during tunnel construction is of vital importance for avoiding geo-hazards like water inrush. In order to meet the requirement of imaging and fast interpretation in 3D resistivity ahead prospecting in tunnels, this paper provides a joint inversion algorithm based on a GPU parallel ant colony algorithm and the traditional last-square inversion method. Through the combination of the linear and non-linear inversion, the global search ability of Ant Colony Optimization （ACO） could provide a better initial model for the least-square inversion. Thus one can prevent it from falling into a false local minimum while having a fast convergence by the least-square inversion and improve the imaging accuracy of the in-tunnel 3D resistivity inversion. Moreover, considering the inherent parallelism of the ant colony algorithm, a GPU parallel strategy under CUDA is provided for the fast imaging of 3D resistivity inversion. Secondly, compared with conventional least-square linear inversion, the numerical simulation using this GPU joint inversion indicates that the GPU joint inversion algorithm can significantly improve the computational efficiency and present a better identification of the position and spatial shape of the water-bearing structure. It can also suppress the non-uniqueness of least-square linear inversion. In the end, the result of the physical model test shows that the low-resistivity anomaly detected by the GPU joint inversion is coincident with the position of the actual water-bearing structure. It can efficiently suppress the non-uniqueness, improve the detection accuracy and accelerate the inversion speed. Moreover, it helps to lay the foundation of the practical application of 3D resistivity joint inversion ahead prospecting in tunneling.
出处
《地球物理学报》
SCIE
EI
CAS
CSCD
北大核心
2017年第12期4916-4927,共12页
Chinese Journal of Geophysics
基金
国家重点基础研究发展计划(973计划)项目(2014CB046901
2015CB058101)
国家重大仪器设备研制专项(51327802)
国家自然科学基金(51479104
41502279
51739007)
国家重点研发计划(2016YFC0401801
2016YFC0401805)
山东省重点研发计划(2016GSF120001)共同资助
关键词
隧道含水构造
三维电阻率超前探测
GPU并行计算
混合反演
蚁群算法
模型试验
Water-bearing structure in tunnel 3D resistivity ahead prospecting GPU parallel computing Joint inversion Ant colony algorithm Model test
