非稳态地震数据反射系数反演方法研究与应用

发布时间：2018-11-27 18:51

【摘要】：传统的反射系数反演基于褶积模型,理论上其要求输入的地震数据是稳态的(即未考虑地层品质因子Q的衰减与频散效应)。如果地震波在传播过程中其Fourier频谱(振幅谱和相位谱)不变化,称这种地震波是稳态的。相反,地震波在传播过程中其频谱发生变化(振幅衰减和相位频散),称这种地震波是非稳态的。由非稳态地震波组成的数据称为非稳态地震数据。由于地层的粘弹性,实际地震数据是非稳态的。对非稳态地震数据进行传统反射系数反演须先补偿地层Q滤波效应,通常选用反Q滤波。然而反Q滤波的振幅补偿存在固有的数值不稳定问题,且这一问题并没有从本质上解决。因此,本文提出一种从非稳态地震数据直接反演反射系数的方法,称其为:非稳态稀疏反射系数反演(Nonstationary Sparse Reflectivity Inversion,NSRI)。NSRI的主要优点有:通过将地层Q滤波算子集成到褶积模型中,其避免了反Q滤波振幅补偿固有的数值不稳定问题。NSRI同时避免了时变反褶积类方法所面临的时变子波估计问题。NSRI不仅将传统反射系数反演拓展至可处理非稳态地震数据,而且兼具传统反射系数反演的功能。在建立NSRI反演方程组时,本文只用了输入地震数据信噪比高的部分谱,并采用压缩感知领域基追踪算法来优化求解l_1-l_2约束目标函数,以得到稀疏反射系数解。合成数据实验表明:NSRI可从非稳态数据中直接获取反射系数解,而无需对输入的非稳态数据先做反Q滤波。在数据信噪比低时,NSRI仍可获得合理的反射系数解。当输入的Q值存在适度误差时,并不会造成NSRI结果的剧烈变化。当输入地震子波的主频和相位存在适度误差时,NSRI仍可获得合理的反演结果。接着本文采用物理模拟数据测试了NSRI。物理模型的制作比例为1:5000(即物理模型1毫米代表实际5米),速度比例约为1:1。物理模型由强吸收衰减介质组成。对于物理模拟数据,知道真实的反射系数解,但并不知道地震波在物理模型中的传播机理以及物理模型介质的Q衰减频散机理,因此物理模拟数据对NSRI来说是盲测试。物理模型介质的Q值是在超声频率下测量衰减与频散数据,并用参考样品法和谱比法计算所得。尽管原始数据中包含多次波、绕射波、边界反射等干扰,为避免破坏原始数据的Q滤波效应,本文并没对原始数据做预处理。NSRI对物理模拟数据分析证实:所推导的非稳态地震数据模拟公式是正确的;实验室测量的物理模型介质的Q值是可信的。NSRI的功能和优点得到证实。由NSRI反演反射系数同震源子波褶积重构的Q补偿结果要优于增益控制反Q滤波结果。实际地震数据测试进一步证实NSRI的功能和优点。
[Abstract]:The traditional reflection coefficient inversion is based on convolution model, which requires the seismic data to be steady (that is, the attenuation and dispersion of formation quality factor Q) are not taken into account. If the Fourier spectrum (amplitude spectrum and phase spectrum) of a seismic wave does not change during its propagation, the seismic wave is said to be stable. On the contrary, the frequency spectrum of seismic wave changes (amplitude attenuation and phase dispersion) during its propagation, which is called unsteady. Data consisting of unsteady seismic waves are called unsteady seismic data. Due to the viscoelasticity of the formation, the actual seismic data are unstable. In the traditional reflection coefficient inversion of unsteady seismic data, the formation Q filter effect must be compensated first, and the inverse Q filter is usually used. However, the amplitude compensation of inverse Q filter has the inherent numerical instability problem, and this problem has not been solved in essence. Therefore, a direct inversion method of reflection coefficient from unsteady seismic data is presented in this paper, which is called unsteady sparse reflection coefficient inversion of (Nonstationary Sparse Reflectivity Inversion,. The main advantages of NSRI). NSRI are: by integrating the Q filter operator into the convolution model, It avoids the numerical instability problem inherent in amplitude compensation of inverse Q filter. NSRI also avoids the time-varying wavelet estimation problem faced by time-varying deconvolution method. NSRI not only extends the traditional reflection coefficient inversion to deal with unsteady seismic data, but also avoids the problem of time-varying wavelet estimation. It also has the function of traditional reflection coefficient inversion. In order to obtain the sparse reflection coefficient solution, only the partial spectrum with high SNR of the input seismic data is used to establish the NSRI inversion equations, and the compressed sensing domain basis tracking algorithm is used to optimize the solution of the l_1-l_2 constrained objective function. The experimental results show that NSRI can directly obtain the reflection coefficient solution from the unsteady data, without the need of inverse Q filter for the input unsteady data. When the signal to noise ratio is low, the NSRI can still obtain a reasonable reflection coefficient solution. When there is a moderate error in the input Q value, the NSRI result will not change dramatically. When there are moderate errors in the main frequency and phase of the input seismic wavelet, NSRI can still obtain reasonable inversion results. Then the physical simulation data is used to test the NSRI.. The physical model is made at a ratio of 1: 5000 (that is, the physical model is 1 millimeter for the actual 5 meters), and the speed ratio is about 1: 1. The physical model consists of a strong absorption attenuation medium. For the physical simulation data, the true reflection coefficient solution is known, but the propagation mechanism of seismic wave in the physical model and the Q attenuation dispersion mechanism of the physical model medium are not known, so the physical simulation data is blind to NSRI. The Q value of physical model medium is obtained by measuring attenuation and dispersion data at ultrasonic frequency and using reference sample method and spectral ratio method. Although the original data contains multiple waves, diffraction waves, boundary reflection and other disturbances, in order to avoid destroying the Q filtering effect of the original data, In this paper, the original data are not preprocessed. The analysis of physical simulation data by NSRI confirms that the derived formula of unsteady seismic data is correct; The Q value of the physical model medium measured in the laboratory is credible. The functions and advantages of NSRI are confirmed. The Q compensation result of the reconstruction of reflection coefficient and source wavelet convolution from NSRI inversion is better than that of gain control inverse Q filter. The actual seismic data test further confirmed the function and advantage of NSRI.
【学位授予单位】：中国石油大学(北京)
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P631.44

【相似文献】