基于局部采样MCMC方法的时移探地雷达反演

马尔科夫链蒙特卡洛方法(MCMC)是一种启发式的全局寻优算法,可以用来解决概率反演的问题.基于MCMC方法的反演不依赖于准确的初始模型,可以引入任意复杂的先验信息,通过对先验概率密度函数的采样来获得大量的后验概率分布样本,在寻找最优解的过程中可以跳出局部最优得到全局最优解.MCMC方法由于计算量巨大,应用难度较高,在地球物理反演中的应用尚处于起步阶段.作者将MCMC方法应用到时移探地雷达(GPR)反演中,并结合双差法对时移的目标区域进行准确反演.在双差法时移反演的过程中,作者对目标区域使用拓展的Metropolis算法进行局部采样,减小了计算量的同时,进一步提高了目标区域的反演精度.最后对合成的数据进行了反演测试,对目标区域的反演误差进行了分析,说明了基于局部采样的MCMC反演方法的有效性,目标区域反演结果特征明显,反演效果好.

Time-lapse inversion of Ground Penetrating Radar(GPR)based on local sampling MCMC method

Markov Chain Monte Carlo method(MCMC)is a heuristic global optimization method which can be used to solve the inversion problem.The inversion based on MCMC method does not rely on the accurate initial model,and can introduce any complex prior information.A large number of posterior probability distribution samples can be obtained by sampling the prior probability density function.In the process of finding the optimal solution,we can jump out the local optimal solution to get the global optimal solution.Because the huge amount of calculation,MCMC method is still in the initial stage in geophysical inversion.In this paper,MCMC method is applied to the time-lapse inversion of Ground Penetrating Radar(GPR),and the double difference method is used to invert the time-lapse target area accurately.For the double difference time-lapse inversion,the author uses the extended Metropolis algorithm to sample the target area,which reduces the calculation time and improves the inversion accuracy of the target area.Finally,the local sampling MCMC method is applied to the synthetic GPR data,and the inversion error of the target area shows that the MCMC inversion method based on local sampling is effective.The inversion result of the target area has obvious characteristics and good inversion effect.