高精度速度建模的反射波波形反演方法与应用研究

The key step of pre-stack depth migration is the velocity model building. The existent tomographic methods in image domain cannot obtain adequately accurate models, which are also time-consuming because of the manual event picking. Theoretically, full waveform inversion (FWI) is the most accurate subsurface parameter inversion method at present. Not only to overcome the dependence on low frequency data and large offset information in conventional FWI, but also to avoid the cycle skipping and the need for true amplitude migration in conventional reflection full waveform inversion (RFWI), based on the success of previous correlation-based reflection full waveform method using two-way wave equation, we try to develop a new multi-stage and multi-step RFWI method in this proposal. The success of this project relies on reflection sensitivity kernel decomposition and the construction of different reasonable objective functions. The object of this project is to develop a high-precision, automatic and efficient background velocity model building method for middle and deep parts in the case of 2D and 3D situations. Specifically, we first proceed correlation-based reflection full waveform method using one-way wave equation, to estimate the initial model efficiently. Then we use the pure phase matched RFWI method to improve the resolution of inversion results. At last, we develop RFWI method to estimate the background velocity and the model perturbation simultaneously. To better address the key issues of seismic imaging, all the developed model building methods are based on partitioned multi-nodes parallel CPU/GPU mode to deal with the great computation and I/O burden in three dimentional RFWI.

叠前深度偏移的关键是速度建模，目前普遍使用的成像域层析反演建模方法，精度不高，人工拾取也影响了使用效率，全波形反演（FWI）在理论上是目前精度最高的参数反演方法。为了克服常规FWI依赖低频及大偏移距信息的缺陷，规避传统反射波全波形反演（RFWI）的“跳周”现象和对真振幅偏移的依赖，我们在成功地以双程波为引擎实现基于相关的RFWI的基础上，本课题拟通过波形反演核函数分解及合理目标函数构建，分阶段、分步骤地实施RFWI，从而为建立高精度中深层速度模型而发展一套自动、高效的二维及三维RFWI方法。具体地，使用单程波方程进行相关目标函数的RFWI以快速建立初始速度模型，利用纯相位目标函数的RFWI以提高速度反演精度，在此基础上发展同时估计背景速度与模型扰动的RFWI方法，并通过多核异构的CPU/GPU协同分块并行模式解决三维RFWI内存需求大和计算效率低的问题，更好地解决地震成像中的速度建模问题。

反射波波形反演是地下弹性参数的高精度建模手段，本次研究完成的二维及三维基于单程波方程的反射波波形反演方法，避免了双程波RWI方法中容易在界面处产生高波数假象，更有利于背景速度的反演。该方法无需人工干预，从而为高效、全自动的背景速度模型建立提供了一种有效方法技术，该方法运用到东海的实际资料中取得了良好的效果。. 另外，围绕声波和弹性波的反演和成像，发展了一系列的新方法、新技术，例如：初至波各向异性走时反演方法、初至波各向异性波形反演方法、变密度两参数声波波形反演方法、变密度两参数声波最小二乘偏移成像方法、弹性波三参数波形反演方法、弹性波变密度最小二乘偏移成像方法、基于波形反演的微地震定位方法、改进的伴随状态法初至波走时层析成像方法、基于波动方程的初至波走时包络及波形联合反演方法、反射波和棱柱波联合最小二乘逆时偏移成像方法、在较大的速度误差情况下的最小二乘逆时偏移成像方法，等等。这些方法技术为地球内部结构反演和油气勘探研究提供多种有效的方法技术工具。