论文详情
针对传统时间域相位编码全波形反演的炮间串扰问题,提出了一种基于随机炮分配的相位编码时间域全波形反演策略,并实现了基于多卡GPU的加速。基于四卡GPU加速的模型反演实验表明：在GPU和相位编码技术的共同加速下,时间域全波形反演的计算效率可得到显著提升,同时由于各卡间的随机炮分配,基于随机炮分配的相位编码时间域全波形反演策略能更有效地压制串扰噪声,因此其比传统的相位编码全波形反演具有更高的反演精度和收敛效率。
To deal with the crosstalk problem of the traditional time-domain,a strategy of phase-encoded full waveform inversion (FWI) with random shots allocation is proposed and implemented based on the acceleration of multi-card GPU in the paper.Numerical experiments based on a four-card GPU show that,under the acceleration of phase encoding technique and GPU,the computational efficiency of time-domain FWI can be improved greatly.Meanwhile,by the random allocation of shots in each card,this strategy can suppress the crosstalk more effectively.Thus,the time-domain phase-encoding FWI with random shots allocation has higher inversion precision and convergent rate than the traditional method.
国家自然科学基金项目(41574105)资助。