基于波场优化匹配的高精度全波形反演方法研究

发布时间：2018-06-06 22:04

本文选题：全波形反演 + 频率域　；参考：《吉林大学》2016年博士论文

【摘要】：随着油气资源勘探与开发难度的加大,对地震勘探成像的精度要求也越来越高,常规的速度建模方法已经不能满足地震成像和储层预测对速度模型的分辨率和精度的要求。全波形反演方法通过将拟重建速度模型所对应的正演模拟记录与实际观测地震记录进行匹配,根据匹配误差和判别准则利用合理的优化算法对速度模型不断迭代更新,最后得到能够准确描述地下介质速度分布的速度模型。由于全波形反演方法充分利用了叠前地震记录中的运动学和动力学信息,因此,由该方法重建的速度模型能有更高的分辨率和精度。由于全波形反演算法一般是基于Bom近似尺度的,在解决强非线性问题时有很大局限性,因此全波形反演对初始模型和低频信息的依赖性很强。当反演的初始模型质量较差或地震资料缺少足够的低频信息时,在进行波场匹配时就会出现周波跳跃等现象造成严重的匹配误差,无法取得预期的高精度速度建模结果。因此,构建更加精确的初始速度模型,补偿地震记录的低频信息以减少周波跳跃现象,是目前全波形反演领域的主要攻关研究课题。本文深入讨论了全波形反演在理论上及实际应用中存在的初始模型依赖和周波跳跃等问题,并针对这些问题进行了研究,提出了相应的解决方法,形成了较为有效的解决方案和策略。在全波形反演中,多尺度方法是增加算法稳定性,提高波场匹配程度的常用方法之一。结合Curvelet变换的多尺度特性,本文提出了一种新的多尺度方法,减小了全波形反演对初始模型的依赖,提高了模拟波场和观测波场的匹配程度。在从低频到高频反演的频率域全波形反演中,在反演的不同阶段,根据反演频率的大小以及初始模型的精度高低,利用Curvelet变换,分别提取观测数据不同尺度的数据参与反演：即当初始模型精度较低时,首先使用观测地震数据的大尺度数据参与反演,随着反演频率和反演精度的增加,逐渐加入精细尺度的数据,使模拟数据能够更好的与观测数据进行匹配,以减小全波形反演对初始模型的依赖。本文对比了该多尺度全波形反演方法和常规全波形反演方法在使用较差初始模型时的反演结果,验证了该多尺度全波形反演方法对减小初始模型依赖的有效性。周波跳跃现象是全波形反演难以在实际中应用的难题之一。当模拟地震数据和观测地震数据之间的相位差大于半个周期时,就会产生波场匹配不准的现象,即周波跳跃现象。低频信息在全波形反演中有着至关重要的作用,可以减少周波跳跃现象发生的可能。本文针对全波形反演周波跳跃现象产生的原因,借鉴数字图像相关方法计算图像“变形”位移的思想,提出了波场相位相关时移的全波形反演方法。在反演之前,先对模拟地震数据和观测地震数据的相位信息进行互相关运算,根据互相关结果对模拟地震数据进行时移处理,减小两者之间的相位差,使其满足全波形反演对反演数据的要求,然后再进行全波形反演。通过模型测试表明,在地震数据缺少足够的低频信息时,利用该方法可以避免反演陷入局部极小,得到满足精度要求的反演结果。针对全波形反演对反演数据中低频信息和信噪比的严格要求,本文将波场优化匹配全波形反演的方法应用于上下缆地震采集数据的处理中。通过合并处理后的上下缆采集数据,可以有效的压制地震鬼波,扩展地震资料的频带宽度。因此,上下缆合并数据的宽频特征可以使全波形反演的反演深度以及反演的稳定性有较大提高。通过模型测试,对比了应用上下缆数据的全波形反演结果与应用常规单缆数据的全波形反演结果,验证了应用上下缆数据对改善全波形反演效果的有效性。
[Abstract]:With the increasing difficulty in the exploration and development of oil and gas resources, the precision of seismic imaging is becoming more and more high. The conventional velocity modeling method has not met the requirements of seismic imaging and reservoir prediction for the resolution and accuracy of the velocity model. According to the matching error and criterion, the velocity model is constantly updated according to the matching error and criterion. Finally, the velocity model can accurately describe the velocity distribution of the underground medium. The full waveform inversion method makes full use of the kinematic and dynamic letters in the pre stack seismic record. Therefore, the velocity model reconstructed by this method can have higher resolution and precision. Since the full waveform inversion algorithm is generally based on the Bom approximation scale, it has great limitations in solving the strong nonlinear problems. Therefore, the full waveform inversion has a strong dependence on the initial model and the low frequency information. When the seismic data is lacking enough low frequency information, there will be a serious matching error when the wave field matches the wave field matching. It is impossible to obtain the expected high precision velocity modeling results. Therefore, a more accurate initial velocity model is constructed to compensate the low frequency information of the seismic record to reduce the cycle hopping phenomenon. In this paper, the main problems in the field of waveform inversion are discussed. This paper discusses the problems of the initial model dependence and the cycle hopping in the theory and the practical application of the full waveform inversion, and studies these problems, and puts forward the corresponding solutions and forms a more effective solution and strategy. The multiscale method is one of the commonly used methods to increase the stability of the algorithm and improve the matching degree of the wave field. In this paper, a new multiscale method is proposed in this paper, which reduces the dependence of the full waveform inversion on the initial model, and improves the matching degree of the simulated wave field and the observation wave field, and the inversion from low frequency to high frequency in the Curvelet transform. In the frequency domain full waveform inversion, in the different phases of the inversion, according to the size of the inversion frequency and the accuracy of the initial model, the Curvelet transform is used to extract data from different scales of the observation data to participate in the inversion, that is, when the initial model is low, the large scale data of the observed seismic data are used to retrieve the data. The increase of the inversion frequency and inversion accuracy is gradually added to the fine scale data, so that the simulated data can be better matched with the observed data in order to reduce the dependence of the full waveform inversion on the initial model. This paper compares the multiscale full waveform inversion method and the conventional full waveform back replay method in the use of the poor initial model. It is proved that the multiscale full waveform inversion method is effective for reducing the dependence of the initial model. The cycle hopping is one of the difficult problems that are difficult to apply to the full waveform inversion. When the phase difference between the simulated seismic data and the observed seismic data is more than half the period, the phenomenon of the wave field matching is not allowed, that is, the hopping of the wave field. Phenomenon. Low frequency information plays a vital role in full waveform inversion, and it can reduce the possibility of Zhou Bo jumping phenomenon. In this paper, in view of the causes of the full waveform inversion of Zhou Bo jumping phenomenon, the idea of calculating the "deformation" displacement of the image by using the digital image correlation method is used to draw the full waveform inversion of the phase shift of the wave field. Before inversion, the phase information of the simulated seismic data and the observed seismic data is interrelated, and the phase difference between the simulated seismic data is processed according to the correlation results, and the phase difference between the two is reduced, so that it meets the requirements of the full waveform inversion to the inversion data, and then the full waveform inversion is carried out. The model test table is carried out. In the absence of sufficient low frequency information of seismic data, this method can avoid the local minimum of inversion and get the result satisfying the requirement of precision. In view of the strict requirement of the low frequency information and signal to noise ratio in the inversion data, the method of wave field optimization matching full waveform inversion is applied to the seismic mining of the upper and lower cables. In the processing of data collection, the seismic ghost wave can be suppressed effectively and the bandwidth of the seismic data can be expanded effectively through the combined processing of the upper and lower cables. Therefore, the broadband characteristics of the combined data can greatly improve the inversion depth and the stability of the inversion. The results of the full waveform inversion of the cable data and the full waveform inversion results of the conventional single cable data verify the effectiveness of the application of the cable data to the improvement of the total waveform inversion results.
【学位授予单位】：吉林大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P631.4

【参考文献】