Inversion of subsurface physical parameters using modem optimization methods

利用地震或微震资料反演地下物性参数是地震工程中常用的工作之一。通常采用最小二乘法进行反演。然而，由于失配函数的非线性特性需要最小化，很难找到最优模型。近年来，运筹学领域发展了几种不使用失配曲面梯度信息的现代优化方法。在本研究中，我们探讨了启发式搜索方法在波浪数据反演中的适用性。首先，采用遗传算法、模拟退火和禁忌搜索三种启发式搜索方法反演浅层横波速度剖面的瑞利波相速度；与线性化最小二乘反演不同，它们不需要导数计算，也不需要初始模态，只需要正演模拟计算。通过数值实验比较了三种启发式算法的性能。在没有任何特定初始模型的情况下，这三种算法都可以从合成数据中重建具有或不具有速度反转的井浅横波剖面。遗传算法和模拟退火算法具有较快的收敛速度。然而，模拟退火可以找到最小的不拟合模型。在禁忌搜索中，不匹配的减少是最缓慢的。接下来，我们将三种方法应用于垂直钻孔阵列地震记录的频谱比反演，以估计s波速度和q值。我们可以观察到与上述相同的结论。将模拟退火法应用于电阻率测深数据和微微H/V谱数据的联合反演。通过联合反演，我们可以同时估计电阻率和横波剖面。最后，将神经网络技术应用于初始纵波和横波发生时间的自动估计。我们可以从估计的旅行时间数据中估计出东京地区由于沉积层造成的旅行时间延迟。少

Inversion of subsurface physical parameters from earthquake or microtremor data is one of the often used works in Earthquake Engineering. Usually least-square method is applied in the inversion. However, we have difficulties to find an optimal model because of nonlinear characteristics of misfit function to be minimized. Recently several modem optimization methods without use of gradient information of misfit surface are developed in the field of operation research. In this study, we investigated applicability of heuristic search methods in inversion of wave data.First, three heuristic search methods, Genetic Algorithm, Simulated Annealing and Tabu Search are implemented to invert Rayleigh wave phase velocity for shallow S-wave velocity profiling. Unlike linearized least-squares inversion, they do not require derivative calculation, or an initial mode, only the forward modeling calculation. The performances of the three heuristic techniques are compared with numerical experiments. With … More common paramerization and search limits, the three algorithms can reconstruct well shallow S-wave profiles with and without a velocity reversal from the synthetic data without any specific initial models. The Genetic Algorithms and the Simulated Annealing show a fast convergence of the misfits. However, the Simulated Annealing can find models with the smallest misfits. It is also found that the decrease of the misfits is the most gradual in the Tabu Search. Next we applied the three methods to inversion of spectral ratios of earthquake records from a vertical bore hole array to estimate S-wave velocity and Q-values. We can observe the same conclusions as obtained above. Furthermore the simulated annealing is used to a joint inversion of resistivity sounding data and microtrernor H/V spectral data. We can estimate resistivity and S-wave profiles simultaneously from this joint inversion. Finally, neural network technique is applied to automatic estimation of onset times of initial P-and S-waves. We can estimate travel time delays due to sedimentary layers in Tokyo area from the estimated travel time data. Less