机器学习高分辨融合反演在地层对比中的应用——以珠江口盆地开平凹陷开平A构造带为例  

Application of Machine Learning High-Resolution Fusion Inversion in Stratigraphic Correlation:A Case Study of Kaiping A Structural Belt in Kaiping Sag of the Pearl River Mouth Basin

作　　者：吕华星 陈兆明 张振波 姜大朋 李克成 郭伟 Lyu Huaxing;Chen Zhaoming;Zhang Zhenbo;Jiang Dapeng;Li Kecheng;Guo Wei(Shenzhen Branch,CNOOC China Limited,Shenzhen 518000,Guangdong,China)

机构地区：[1]中海石油(中国)有限公司深圳分公司,广东深圳518000

出　　处：《吉林大学学报(地球科学版)》2025年第1期289-297,共9页Journal of Jilin University:Earth Science Edition

基　　金：中海石油深海开发有限公司综合科研项目(KJGG2022-0104)。

摘　　要：开平凹陷已发现的主要含油气层段主要位于古近系恩平组和文昌组,沉积环境主要为辫状河三角洲平原亚相,以分流河道微相为主。由于多期河道横向摆动,地层具有较强的非均质性,导致地层对比难度极大。本文利用机器学习算法支持向量机强大的数据拟合能力获得低频、中频和高频三部分数据的最佳融合频率及权重,得到具有高纵向分辨率和高横向地震分辨率细节特征的反演结果,提升井间地层尖灭位置和地层连通性描述精度。高分辨率v_(P)/v_(S)反演结果表明,该方法能够识别厚度为6 m的储层,具有较高的纵向和横向分辨率,有利于在研究区进行地层对比。The main oil and gas-bearing strata discovered in Kaiping sag are primarily located in the Paleogene Enping Formation and Wenchang Formation.The depositional environment is mainly braided river delta plain subfacies,and the microfacies of underwater distributary channel is dominant.Due to the lateral oscillation of multiple river channels,the strata have strong heterogeneity,making stratigraphic correlation extremely difficult.This paper utilizes the powerful data fitting capabilities of machine learning algorithm support vector machine to determine the optimal fusion frequencies and weights for low,medium,and high-frequency data segments.This process yields inversion results with enhanced vertical resolution and detailed lateral seismic characteristics,thereby improving the accuracy of interwell stratigraphic pinchout positions and the description of stratigraphic connectivity.The high-resolution v_(P)/v_(S)inversion results demonstrate that this method is capable of identifying reservoirs with a thickness of 6 m.It offers superior vertical and horizontal resolution,which is conducive to stratigraphic correlation within the study area.

关 键 词：开平凹陷 古近系 机器学习高分辨融合反演 地层对比 

分 类 号：P631.4[天文地球—地质矿产勘探]