论文详情
速度建模技术的自动化是走向智能化建模的基础,基于CMP道集的叠加速度分析技术是业界常用的初始速度建模方法,也是整个速度建模流程的起点。“两宽一高”观测系统采集到的地震数据规模大,传统的人工拾取模式难以开展密点速度分析,而一些速度自动拾取方法难以适用于低信噪比数据(尤其是陆上资料),因而无法广泛实际应用。针对当前自动速度分析方法中遇到的困难,提出了一种智能化的叠加速度建模技术策略,其关键是构建一种“合理的”时间速度对(简记为T-V对)的自动筛选过程:首先,在缺乏速度场先验信息的情况下,通过生成“伪叠加剖面”并利用图像处理算法提取其中蕴含的构造特征并获得密点速度分析的种子点;然后,基于相邻CMP道集的统计信息约束叠加速度的横向变化以及层速度的纵向变化;最后,在结构约束下自动筛选最符合地质逻辑和统计趋势的T-V对。通过将处理人员的逻辑与经验转化为速度分析的各种约束条件,实现“合理的”T-V对智能化筛选过程,降低人工成本并缩短速度建模周期,推动叠加速度建模处理流程从自动化走向智能化。二维模拟数据测试结果验证了方法的有效性,二维陆上资料建模结果证明了方法处理低信噪比数据时的稳健性。
An automated velocity estimation is fundamental for intelligent model building.The stacking velocity analysis technique is widely used in the oil industry,and constitutes the first step in the model building workflow.Owing to the Terabyte- or even Petabyte-sized data recorded by a “broadband,wide-azimuth,and high-density” acquisition system,it is difficult to build a stacking velocity model using the traditional human-picking-based workflow,especially if high-density picking is required.Many automated methods have been developed to expedite the velocity analysis.Although proved to be effective and efficient for synthetic data processing,most of these methods are not stable when applied to data with low signal-to-noise ratios (SNRs),such as land data.To address this issue,a fully automated stacking velocity modeling strategy was proposed,which is based on picking “reasonable” time-velocity (T-V) pairs during the automatic selection.First,in the absence of a priori knowledge about the velocity field,a “pseudo stack profile” is generated and image processing algorithms are used to extract seeds that are consistent with geological structures.Then,based on statistical information from adjacent common mid-point gathers,the horizontal variation in stacking velocity and vertical variation in layer velocity are constrained.Finally,the T-V pairs that are most geologically consistent are automatically picked under structural constraints.By translating the logic and experience of the processors into various constraints for velocity analysis,a “reasonable” T-V pair automatic picking process is realized,thus freeing the processors from time-consuming,repetitive,and mechanical operations,thereby reducing labor costs and shortening the model building cycle.Tests on two-dimensional (2D) synthetic data showed that the proposed method can achieve a good modeling accuracy for noiseless data.Moreover,an application to 2D field data demonstrates that the proposed method is robust and effective,and is thus very promising for industrial-scale.
变革性技术关键科学问题重点专项(2018YFA0702503)、国家重点研发计划深海关键技术与装备重点专项(2019YFC0312004)、国家自然科学基金(42074143,41774126)、上海市浦江人才计划资助(20PJ1413500)、中国石化地球物理重点实验室项目(33550006-19-FW0399-0041,33550006-20-ZC0699-0011)共同资助。