基于地表参数扩展模型的波形反演方法研究

发布时间：2018-03-08 13:28

本文选题：波形反演　切入点：梯度分解　出处：《中国石油大学(华东)》2015年硕士论文　论文类型：学位论文

【摘要】：全波形反演方法(Full Waveform Inversion)利用地震波的振幅和相位信息,通过迭代更新模型参数使实际地震记录和合成记录差值最小,得到准确可靠的地下模型参数。它具有高分辨率、高精度的反演特点,可以给叠前时间、深度偏移提供高精度的速度场,为岩性判断和油气藏识别提供可靠的速度资料。但是,当采集地震数据缺失低频和远偏移距信息时,常规的全波形反演方法只能恢复地下模型的短波长成分。本文以时间域声波波形反演方法为基础,采用共轭梯度下的FWI并行反演流程,通过模型试算和效果分析,证明了常规的全波形反演可以在初始速度模型比较准确时得到高精度的反演结果,但是当初始速度模型的低频信息不准确时就不能收敛到全局最优。本文首先给出了一种利用梯度分解和重组来提高全波形反演中长波长更新量的方法,该方法将全波形反演梯度分解为层析项和偏移项,梯度的层析项主要更新模型参数的长波长分量,由传播方向相同的波场互相关得到,而偏移项更新模型参数的短波长分量,由传播方向相反的波场互相关得到。本文利用坡印廷矢量进行波场分离进而将梯度分解,在将两项重组后得到层析项加强的新梯度。数值测试结果表明这种基于波场分离的层析波形反演方法可以加快收敛速率而且可以改善反演效果。然而基于梯度分解的层析波形反演方法只能加重低频信息,在低频信息缺失时仍不能得到较好的反演结果。最后本文对扩展波形反演方法进行了研究,通过模型扩展可以将数据域波形反演和成像域速度分析联系起来,以此为基础建立扩展波形反演目标泛函,从而结合这两者的优势。模型扩展有两种方式,面向地表和面向地下,面向地表的模型扩展具有更快的计算效率,但是精度有限,面向地下的模型扩展精度高,但是成本较高,本论文采用第一种扩展方式,也就是基于地表扩展的扩展模型反演并将其与基于梯度分解的波形反演方法相结合。模型测试表明,炮域扩展波形反演方法可以很好的提取低波数模型信息,模型试算表明可以在常规方法不能收敛时收敛到正确结果。
[Abstract]:Full Waveform inversion method (full Waveform inversion) utilizes amplitude and phase information of seismic wave, and iteratively updates model parameters to minimize the difference between actual seismic records and synthetic records, and obtains accurate and reliable underground model parameters, which have high resolution. The characteristics of high precision inversion can provide high precision velocity field for prestack time and depth migration, and provide reliable velocity data for lithology judgment and oil and gas reservoir identification. The conventional full waveform inversion method can only restore the short wavelength components of the underground model. Based on the time-domain acoustic wave inversion method, the FWI parallel inversion flow under conjugate gradient is adopted in this paper. It is proved that conventional full-waveform inversion can obtain high precision inversion results when the initial velocity model is more accurate. However, when the low frequency information of the initial velocity model is not accurate, it can not converge to the global optimum. In this paper, we first present a method to improve the medium and long wavelength update of the whole waveform inversion by using gradient decomposition and recombination. In this method, the whole wave inversion gradient is decomposed into the tomography item and the migration term. The long wavelength component of the model parameter is mainly updated by the gradient tomography term, which is obtained from the cross-correlation of the wave field with the same propagation direction, while the migration term updates the short wavelength component of the model parameter. In this paper, we use the Poynting vector to separate the wave field and decompose the gradient. After recombination of the two terms, a new gradient with enhanced chromatographic terms is obtained. The numerical results show that this method based on wave field separation can accelerate the convergence rate and improve the inversion effect. However, based on the gradient fraction, the inversion effect can be improved. The computed waveform inversion method can only aggravate the low frequency information. When the low frequency information is missing, the inversion results can not be obtained. Finally, the extended waveform inversion method is studied in this paper. The data domain waveform inversion and the imaging domain velocity analysis can be linked by model expansion. Based on this, the target functional of extended waveform inversion is established, which combines the advantages of these two methods. There are two ways to extend the model, which are surface oriented and underground oriented, and surface oriented model expansion has faster computational efficiency, but the accuracy is limited. The expansion accuracy of the underground model is high, but the cost is high. In this paper, the first expansion method is adopted, that is, the extended model inversion based on the surface expansion is combined with the waveform inversion method based on gradient decomposition. The model test shows that, The extended waveform inversion method in gun-domain can extract the information of low wavenumber model very well, and the model table shows that the model can converge to the correct result when the conventional method can not converge.
【学位授予单位】：中国石油大学(华东)
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：P631.4

【相似文献】