3D Time-domain electromagnetic full waveform inversion in Debye dispersive medium accelerated by multi-GPU paralleling

Electromagnetic full waveform inversion in Debye dispersive medium (EFWI-D) is a promising technique to reconstruct the inner structure and electrical properties of the medium such as soil, rock and biological tissues. Same as conventional full waveform inversion, EFWI-D requires high computational cost, especially in the 3D case. To reduce the long computation time, we design and implement the EFWI-D algorithm in time domain using multiple GPU cards. The inversion method is based on the L-BFGS optimization algorithm, which can increase the convergence of the misfit function, while the auxiliary differential equation (ADE) method is employed for modeling the Debye dispersive medium by using exponential time differencing (ETD) finite-difference time-domain (FDTD) approach. Moreover, a multi-stream strategy is performed in the workflow to improve the computation performance. Numerical results illustrate the improvement of the computational performance and the preliminarily feasibility of the proposed inversion algorithm.

电磁全波形反演（EFWI-D）在迪拜色散介质（Debye dispersive medium）中的应用，是一项具有广阔前景的技术，旨在重构介质的内部结构和电学性质，如土壤、岩石和生物组织。与传统的全波形反演方法相同，EFWI-D亦需高昂的计算成本，尤其在三维情形下。为缩短长计算时间，本研究设计并实现了基于时域的多GPU卡EFWI-D算法。该反演方法基于L-BFGS优化算法，能够提升误拟合函数的收敛性，同时采用辅助微分方程（ADE）方法，通过指数时间差分（ETD）有限差分时域（FDTD）方法对迪拜色散介质进行建模。此外，在工作流程中执行多流策略，以提升计算性能。数值结果表明，该方法显著提高了计算性能，并初步验证了所提反演算法的可行性。