基于多种群遗传算法和动态时间归整的智能层序地层划分对比方法——以渤南洼陷古近系下部为例  

Intelligent identification of sequence stratigraphy interfaces constrained by multi-population genetic algorithm and dynamic time warping: a case study of Lower Paleogene strata in Bonan subsag

作　　者：刘天娇 贾海波 余继峰 韩超[1] 冯乔 LIU Tianjiao;JIA Haibo;YU Jifeng;HAN Chao;FENG Qiao(College of Earth Science and Engineering,Shandong University of Science and Technology,Qingdao,Shandong 266590,China;Key Laboratory of Sedimentary Mineralization and Sedimentary Minerals in Shandong Province(Shandong University of Science and Technology),Qingdao,Shandong 226590,China)

机构地区：[1]山东科技大学地球科学与工程学院,山东青岛266590 [2]山东省沉积成矿作用与沉积矿产重点实验室(山东科技大学),山东青岛266590

出　　处：《中国科技论文》2023年第1期71-80,102,共11页China Sciencepaper

基　　金：国家自然科学基金资助项目(42102126)。

摘　　要：针对深层-超深层因古生物化石种类少、取心困难、标志层不稳定、层系埋藏深、地震品质差而存在的层序地层划分与对比困难的问题,提出综合利用测井、录井资料为输入参数,采用多种群遗传算法(multi-population genetic algorithm, MPGA)和动态时间归整(dynamic time warping, DTW)相结合进行渤南洼陷古近系下部层序地层的划分与对比,并以多种常规频谱分析(最大熵频谱分析(integrated prediction error filter analysis, INPEFA)、连续小波变换(continuous wavelet transform, CWT)、时频分析等)对智能划分结果进行验证。基于这种智能识别,以渤南洼陷深部地层古近系下部沙四段-孔店组为例,在渤南洼陷西部邵古1井、邵古2井古近系下部识别出沙三段/沙四段(E_(s3)/E_(s4))界面、沙四段/孔店组(Es4/Ek)界面、孔店组(Ek)底部等层序界面,将邵古1井和邵古2井古近系下部划分出4个三级层序界面并进行连井层序对比。通过与多种频谱分析方法验证结果对比发现,综合利用基于Fisher最优分割理论的MPGA对单井层序地层进行精细划分,结合DTW技术进行连井层序对比,可提高深层及超深层层序地层划分对比的准确性和可靠性。所提出的单井地层自动划分与连井对比方法能摆脱人为主观因素对层序地层划分的干扰,实现基于机器学习的层序地层智能识别,在各类岩石的层序地层对比中具有良好的应用前景。Due to problems such as few paleontological fossils, difficulty in coring, unstable marker layers, deep burial and poor seismic quality, there are difficulties in sequence stratigraphic division and correlation. Therefore, this study proposes a comprehensive utilization of wireline logging and mud logging data as input parameters, using a combination of multi-population genetic algorithm(MPGA) and dynamic time warping(DTW) to divide and correlate the sequence stratigraphy of the Lower Paleogene in Bonan subsag. Meanwhile, the intelligent partition results were verified by a variety of conventional spectrum characteristics(integrated prediction error filter analysis(INPEFA),continuous wavelet analysis(CWT), time-frequency analysis, etc.). Based on this intelligent recognition, taking the Lower Paleogene E_(s4)-Ek formation in the deep strata of Bonan subsag as an example, the Paleogene E_(s3)/E_(s4)interface, E_(s4)/Ek interface, Ek bottom interface and possible sequence stratigraphic interfaces within the strata were identified in the lower Paleogene of wells Shaogu 1 and Shaogu 2 in the western Bonan subsag. The Paleogene of well Shaogu 1 and well Shaogu 2 are divided into four third-order sequence boundaries and the sequence comparison of wells is carried out. Comparing with the verification results of multiple spectrum analysis methods, the results demonstrate that the comprehensive use of MPGA based on the Fisher optimal segmentation theory to finely divide the single well strata, combined with the DTW to compare the sequence of connected wells, can improve the accuracy and reliability of the sequence stratigraphic division and correlation of deep and ultra-deep strata. The proposed method can get rid of the interference of human subjective factors on sequence stratigraphic division correlation, and realize sequence stratigraphic intelligent recognition based on machine learning, which has good application prospect in sequence stratigraphic correlation of various rocks.

关 键 词：层序地层划分对比 渤南洼陷 古近系下部 多种群遗传算法 动态时间归整 频谱特征分析 

分 类 号：P539.2[天文地球—古生物学与地层学]