Bayes决策理论下的速度谱智能化解释及建模方法  

Intelligent interpretation and modeling method of velocity semblance based on Bayesian decision theory

作　　者：伍国富 肖明图[2] 王华忠[1] 凌越[2] 赵玉合[2] WU Guofu;XIAO Mingtu;WANG Huazhong;LING Yue;ZHAO Yuhe(School of Ocean and Earth Science,Tongji University,Shanghai 200092,China;Research Institute of Exploration and Development-Northwest,PetroChina,Lanzhou,Gansu730020,China)

机构地区：[1]同济大学海洋与地球科学学院,上海200092 [2]中国石油勘探开发研究院西北分院,甘肃兰州730020

出　　处：《石油地球物理勘探》2023年第3期590-597,625,共9页Oil Geophysical Prospecting

基　　金：国家重点研发计划变革性技术关键科学问题重点专项“多元信息联合驱动的地震成像方法研究”(2018YFA0702503);国家自然科学基金项目“特征反射波波动理论层析反演与建模方法研究”(42174135);中国石油前瞻性基础性项目“前陆冲断带深层地震复杂波场传播机理与成像关键技术研究”(2021DJ-0304)联合资助。

摘　　要：高精度的速度建模作为强非线性问题,需要一个比较正确的初始速度模型,而基于CMP道集的初始背景速度扫描估计是最稳健的方法。面对规模巨大的CMP道集,研究智能化的初始背景速度扫描估计方法是有必要的,其核心是合理的速度谱解释。这可以看作是基于速度谱解释人员的先验知识和层位约束信息,在Bayes决策意义下,在高维速度谱数据体中,以风险决策函数值最小为原则、挑选最合理的时间—速度(TV)对。为此,提出了一套以人工交互拾取速度谱逻辑思想为指导的决策框架。首先生成速度谱数据体及类叠加剖面,通过计算相干属性从类叠加剖面提取层位结构;再依据结构信息对速度谱能量团进行K均值聚类,对于每个类别以先验信息和数据空间分析的统计信息为约束,自动迭代搜寻使代价函数最小的TV对;最后通过插值平滑生成速度场,且经过基于统计量约束的质量控制降低了横向不连续性。该方法将层位信息的利用贯穿到从聚类到自动拾取的整个过程,并且将解释人员的先验认识及邻域拾取结果量化为自动拾取时的约束量,体现了速度谱解释的智能化,缩短了速度建模周期。As a strong nonlinear problem,the high precision velocity modeling requires a relatively correct initial velocity model.The initial background velocity sacnning method based on the common middle-point(CMP)gathers is the most robust method.Facing the huge size of CMP gathers,it is necessary to research the intelligent initial background velocity analysis method,and its core is the reasonable velocity semblance interpretation.This can be viewed as a scan for the most reasonable time-velocity(TV)pairs with the smallest value of defined risk decision function in a high-dimensional velocity semblance data based on the prior knowledge of the interpreters and the horizon constraint in the sense of Bayesian decision.Therefore,this paper proposes a decision framework guided by the logic in manual interaction picking.Firstly,the semblance data and“pseudo stack profile”are generated to further extract the structural information from the profile by calculating the coherence properties.Then,K-means clustering is performed on the semblance panel according to the structural information.For each classification,TV pairs that minimize the cost function constrained by priori information and statistical information from the data space are iteratively searched.Finally,the velocity field is formed by interpolation and smoothing,as the lateral discontinuities are reduced by quality control based on statistics constraints.This method implements the utilization of horizon information into the whole process from clustering to automatic picking,and quantifies the priori knowledge of the interpreter and neighborhood picking results into restrained variables in automatic picking.This can reflect the intelligence of velocity semblance interpretation and shorten the velocity modeling cycle.

关 键 词：CMP道集 智能化速度建模 Bayes决策 层位约束 聚类 

分 类 号：P631[天文地球—地质矿产勘探]