基于GPU加速的包络波形反演

摘要

波形反演利用地震资料的相位和振幅信息对速度进行建模,是当前分辨率最高的速度估计方法.波形反演对初始速度场具有较高的要求,当初始速度场和真实的速度场相差较远时,容易产生周波跳跃问题.通常在进行速度反演时,采用多尺度方法减轻波形反演的非线性,然而当实际地震资料中缺少大偏移距和低频数据时,多尺度方法不再适用.此外,计算效率也限制了波形反演的应用.因此,这里提出基于GPU加速的包络波形反演来获取包含长波长信息的初始速度场,并提高计算效率.通过对Marmousi模型的测试,验证了方法的正确性和有效性.%Waveform inversion utilizing phases and amplitudes of seismic data to construct velocity model is the velocity estimation method with the highest resolution.However, it requires relatively accurate initial velocity.When the initial velocity is far from the true velocity, circle-skipping problem arises.Multi-scale method is often used to mitigate the nonlinearity of waveform inversion, but it fails when seismic data don't have long-offset and low-frequency information.In addition, computational efficiency restricts the application of waveform.Thus, we propose GPU-based envelop waveform inversion in this paper to obtain initial velocity field with long wavelength information and improve computational efficiency.And test on Marmousi model verifies the correctness and validity of the method proposed in this paper.