基于高维多属性决策过程的复杂地表初至波识别与走时检测方法

波动理论初至波走时层析速度反演及建模成为当今陆上地震波成像中关键步骤之一、“两宽一高”地震数据采集技术的逐渐普及使得人工进行海量数据的初至波识别及走时检测变得不再可行、复杂地表区的油气勘探越来越成为常规、各种机器学习算法的出现,所有这些因素都要求人们必须进一步深入探索高精度的初至波识别及走时检测方法。将问题定位为弱初至波掩埋在(较)强噪声中,提出了一套复杂地表区初至波识别及走时检测的自动化智能化处理技术流程,主要步骤包括炮集中与初至波相关的预处理、高维特征空间的构建、多属性加权K均值聚类划分初至波分布区域、多属性约束的马尔可夫决策过程(Markov decision process,MDP)进行初至走时检测等。在MDP理论框架下,选择合理的基准面消除初至波高频道间时差,将三维(或二维)炮集中的初至波场视为一个有机的整体,提取高维特征属性,引入多属性加权K均值聚类划分出初至波分布区域,缩小拾取范围,利用多属性约束构建马尔可夫最优演化算法,充分考虑初至信息之间的横向连续性,进行初至波走时检测。实际数据测试结果表明,该方法能够自动地以较高的精度和稳健性进行初至走时的拾取,在中等复杂度情形下有较好的实用效果。

The wave theory first-arrival travel time tomographic velocity inversion and modeling has become a keystep in onshore seismic wave imaging.With the gradual popularization of “wide-azimuth,high-density,and broadband” seismicdata acquisition technology,it is no longer feasible to manually identify the first break and travel time in massive data.Oiland gas exploration in complex surface areas is becoming more and more routine.Together with the emergence of various machinelearning algorithms,these factors promote the need to further explore high-precision,first arrival identification,and traveltime detection methods and technologies.To resolve the problem of weak first arrivals buried in strong noise,the present studyproposes a set of intelligent automatic processing procedures for first-arrival identification and travel time detection in complex surface areas.The main steps include preprocessing related to first-arrivals during shot collection,construction of high-dimensional feature space,division of first arrival distribution area by multi-attribute weighted K-means clustering,and detection of first-arrival travel time by the multi-attribute-constrained Markov decision process (MDP).Under the framework of MDP theory,first,a reasonable base plane is selected to eliminate the jumping time difference between traces.Then,considering the first arrival wave field in the three-dimensional (or two-dimensional) shot set as a whole,high-dimensional feature attributes are extracted,and multi-attribute weighted K-means clustering is used to divide the distribution area of the first arrivals and narrow the picking range.Finally,the multi-attribute constraints are used to construct the Markov optimal evolution algorithm,which fully considers the horizontal continuity in the information of the first arrivals.The actual data test results indicates that the method can automatically pick up the first arrival travel time with high accuracy and robustness.The proposed method has a good practical effect in the case of moderate complexity.