Wang Jian, Yang Dinghui, Jing Hao, Wu Hao
Department of Mathematical Sciences, Tsinghua University, Beijing 100084, China.
Department of Mathematical Sciences, Tsinghua University, Beijing 100084, China.
Sci Bull (Beijing). 2019 Mar 15;64(5):321-330. doi: 10.1016/j.scib.2019.01.021. Epub 2019 Feb 2.
Full waveform inversion (FWI) has been increasingly more and more important in seismology to better understand the interior structure of the Earth. FWI, by taking advantage of both the traveltime and amplitude in the data, provides high-resolution model parameters of the earth which can produce images with high resolution. However, this inversion method conventionally suffers from non-uniqueness due to many local minima of the objective function and large computing costs. In this study, we propose a new FWI method in a semi-random framework by integrating the ensemble Kalman filter and uniform sampling without replacement. Numerical results demonstrate that the new method can achieve high-resolution results and a wider convergence domain. Accordingly, the new method overcomes the disadvantage of conventional FWIs that depend strongly on the initial model.
全波形反演（FWI）在地震学中对于更好地理解地球内部结构变得越来越重要。FWI通过利用数据中的走时和振幅，提供地球的高分辨率模型参数，从而能够生成高分辨率图像。然而，由于目标函数存在许多局部极小值以及计算成本高昂，这种反演方法传统上存在非唯一性问题。在本研究中，我们通过整合集合卡尔曼滤波器和无放回均匀采样，在半随机框架下提出了一种新的FWI方法。数值结果表明，新方法能够获得高分辨率结果以及更宽的收敛域。因此，新方法克服了传统全波形反演严重依赖初始模型的缺点。
