确定性反演协同约束的叠后随机地震反演方法

在前人的工作基础上，提出了确定性反演协同约束的叠后随机地震反演方法。以序贯Gibbs扰动模拟、扩展Metropolises—Hastings （M-H）算法为核心，引入层序地层网格，自适应融入地质统计学、构造以及沉积模式等信息，整个反演过程无需计算地层的局部倾角或进行复杂的坐标转换。针对随机地震反演结果的高频成分仍具有较大的不确定性，造成不同实现展示的储层特征差异较大，结合序贯Gibbs扰动模拟与同位协同克里金，通过引入确定性反演结果的协同约束，进一步限定随机反演候选解空间，从而降低随机地震反演高频成分的不确定性。获得以下认识：①相比确定性反演，随机地震反演可以产生高分辨率的反演结果，其中垂向变差影响随机反演的垂向分辨率，横向变差影响随机反演的横向连续性。②与均匀地震网格相比，由于层序地层网格融入了构造和沉积模式等信息，因此更适合变差函数横向约束随机反演；借助重采样矩阵，在层序地层网格进行抽样模拟，在均匀地震网格进行褶积正演，整个反演过程既满足构造和沉积模式的约束，同时又符合地球物理原理。③通过引入确定性反演的协同约束，可进一步限定候选解的解空间，增强波阻抗随机反演结果与确定性反演结果的相关性，进而降低随机反演高频成分不确定性。实际数据试算表明，通过对比随机反演的四个不同实现，验证了所提算法的有效性。

The methodology of a post-stack stochastic seismic inversion with the co-constraint of deterministic inversion

Based on the previous work, a post-stack stochastic seismic inversion with the co-constraint of deterministic inversion was proposed. This new algorithm centered on sequential Gibbs sampling and an extended Metropolises-Hastings (M-H) algorithm. Owing to the introduction of sequence stratigraphy grids, geostatistics, structures, and sedimentary modes were integrated into the stochastic seismic inversion adaptively, thereby saving the effort of calculating the local dip of the strata or implementing complicated coordinate transformations. The high-frequency components (HFCs) in the stochastic seismic inversion result still featured large uncertainty, which resulted in big differences in reservoir characteristics among different implementations of stochastic seismic inversion. Given this problem, this paper combined sequential Gibbs sampling with collocated cokriging and introduced the co-constraint of deterministic inversion to restrict the candidate solution space and ultimately to reduce the uncertainty of the HFCs in the stochastic seismic inversion result. The following conclusions were obtained: ① Compared with deterministic inversion, stochastic seismic inversion can produce high-resolution results. The vertical resolution of the results is affected by the vertical variation whereas the lateral continuity of the results is affected by the lateral variation. ② Compared with uniform seismic grids, sequence strati-graphy grids are more suitable for stochastic inversion with a lateral variation constraint because of their integration with structures and sedimentary modes. With the help of a resampling matrix, sample simulation is carried out on sequence stratigraphy grids and convolutional forward modeling is carried out on uniform seismic grids. The whole inversion process not only meets the constraints of the structures and sedimentary modes but also conforms to geophysical principles. ③ Due to the introduction of the co-constraint of deterministic inversion, the candidate solution space is further restricted and the correlation between the results of stochastic inversion and deterministic inversion is enhanced. Furthermore, the uncertainty of the HFCs in the stochastic inversion results is reduced. Trial calculations with real data reveal that this new algorithm is verified by a comparison among four implementations of stochastic inversion.