论文部分内容阅读
弹性波场的复杂性使得反演问题非线性增强,容易陷入局部极值,需要采用合理的多尺度反演策略降低非线性。在逐频组多尺度反演的基础上引入第二级别的反演策略,即基于阻尼波场的层剥离方法,能够改善反演过程的稳定性。针对频域全波形反演计算效率低、内存占用大的问题,采用双级并行算法:(i)利用多波前大规模并行直接解法(MUMPS)软件包,多节点并行实现波场正演;(ii)基于MPI实现频组内各频率并行计算梯度和步长等,使得多尺度反演算法在提高精度的前提下,保证了计算效率,提高了算法的实用性。Overthrust模型的数值实验表明,本文反演算法能够在有效改善反演稳定性的前提下,高效地获得精度较高的反演结果。
The complexity of the elastic wave field makes the inversion problem non-linearly enhanced and easily plunges into the local extreme. Therefore, a reasonable multi-scale inversion strategy is needed to reduce the nonlinearity. The second level inversion method is introduced based on multiscale inversion by frequency group, that is to say, the layer separation method based on the damping wave field can improve the stability of the inversion process. Aiming at the problem of low computational efficiency and large memory occupation in the full-waveform inversion of frequency domain, a two-stage parallel algorithm is adopted: (i) multi-node parallel wave field forward is implemented by using MUMPS software package; (ii) Based on MPI, the gradient and the step length of each frequency in the frequency band are calculated in parallel, so that the multi-scale inversion algorithm can ensure the computational efficiency and improve the practicability of the algorithm under the premise of improving the accuracy. The numerical experiments of Overthrust model show that the inversion algorithm in this paper can efficiently obtain the inversion results with high accuracy under the premise of effectively improving the stability of inversion.