地学前缘 ›› 2025, Vol. 32 ›› Issue (2): 311-331.DOI: 10.13745/j.esf.sf.2024.5.31

• 非主题来稿选登 • 上一篇    下一篇

超深层断缝带岩溶缝洞体储层结构分析与智能预测

李凤磊1,2,3,4(), 林承焰1,2,3,*(), 王蛟4, 任丽华1,2,3, 张国印1,2,3, 朱永峰5, 李世银5, 张银涛5, 关宝珠5   

  1. 1.深层油气全国重点实验室(中国石油大学(华东)), 山东 青岛 266580
    2.山东省油藏地质重点实验室, 山东 青岛 266580
    3.中国石油大学(华东)地球科学与技术学院, 山东 青岛 266580
    4.山东石油化工学院, 山东 东营 257061
    5.中国石油塔里木油田勘探开发研究院, 新疆 库尔勒 841000
  • 收稿日期:2024-01-05 修回日期:2024-04-10 出版日期:2025-03-25 发布日期:2025-03-25
  • 通信作者: *林承焰(1963—),男,教授,博士生导师,主要从事储层地质学和油气藏描述、沉积学、油气地质与勘探。E-mail: lincy@upc.edu.cn
  • 作者简介:李凤磊(1984—),男,博士研究生, 研究方向为储层地质学及油气藏描述。 E-mail: lfl_winter@163. com
  • 基金资助:
    国家自然科学基金项目(42002144);中石油重大科技项目(ZD2019-183-006)

Structure analysis and intelligent prediction of carbonate fractured-vuggy reservoirs in ultra-deep fracture zone

LI Fenglei1,2,3,4(), LIN Chengyan1,2,3,*(), WANG Jiao4, REN Lihua1,2,3, ZHANG Guoyin1,2,3, ZHU Yongfeng5, LI Shiyin5, ZHANG Yintao5, GUAN Baozhu5   

  1. 1. State Key Laboratory of Deep Oil and Gas, China University of Petroleum (East China), Qingdao 266580, China
    2. Shandong Key Laboratory of Oil Reservoir Geology, Qingdao 266580, China
    3. School of Geosciences, China University of Petroleum (East China), Qingdao 266580, China
    4. Shandong Institute of Petroleum and Chemical Technology, Dongying 257061, China
    5. Research Institute of Petroleum Exploration and Development,Tarim Oilfield Company, PetroChina, Korla 841000, China
  • Received:2024-01-05 Revised:2024-04-10 Online:2025-03-25 Published:2025-03-25

摘要:

塔里木盆地超深层断缝带岩溶缝洞体储层主要集中于断裂带附近,垂向深度大、水平宽度小,对储层识别精度要求高。常规缝洞体识别主要针对其振幅响应特征采取各种技术手段,只能提供储层走向和储集体大致范围。本文提出一种基于二维Res-UNet残差网络和标签数据优化相结合的断缝带缝洞储集体精细解释方法,通过提高二维训练样本更新频率和优化效率,修改Res-UNet残差网络结构,达到提高储层预测精度的目的。本次研究中,将卫星图像、无人机扫描、现场踏勘和地质雷达等信息与地震资料初步解释成果相结合,明确了断裂、裂缝和锥形溶洞为主的地质模型。通过深度域三维地震数据体子波分析、成像测井、声波测井等提取地层和储层速度,选择20、25和35 Hz共3个主频雷克子波,制作了不同储层宽度正演模型,拟合出基于频率和地层速度的储层宽度与其地震响应之间的关系经验公式,通过钻穿储层的水平井轨迹、直井远探测数据和测井数据等,提取实际资料中的储层宽度与地震响应关系完成验证。基于这一公式,结合研究区目的层振幅属性,提取目的层的储集体宽度范围。同时,根据三维地震剖面分析,得出研究区储集体组合特征,通过分析、设计、验证和重新设计等试验流程,针对研究区正演模型设计最优的子波、偏移孔径和采样间隔等参数。建立起可实时更新的二维训练样本库,并利用合成缝洞数据和实际缝洞数据相结合的方法对缝洞标签进行迭代优化。基于这一样本库,开展二维Res-UNet残差网络训练与预测,解决深层网络的退化问题,最终实现了缝洞体结构的精细预测。

关键词: 塔里木盆地, 超深断控岩溶缝洞储层, 深度学习, 残差网络, 样本库优化

Abstract:

Ultra-deep carbonate fractured-vuggy reservoirs, typically located near fault zones, are characterized by significant burial depth, large vertical extent, and narrow horizontal distribution, making seismic identification highly challenging. Conventional fracture-vuggy identification techniques primarily rely on amplitude response characteristics, which only provide the general orientation and approximate extent of the reservoir. To address these limitations, a refined interpretation method for fractured-vuggy reservoirs was proposed based on a two-dimensional Res-UNet residual network and optimized label data. This method enhances the update frequency and optimization efficiency of 2-D training samples by modifying the Res-UNet residual network structure, thereby improving the accuracy of reservoir prediction. Geological models of faults, fractures, and karst caves were constructed using a combination of satellite imagery, UAV scanning, field reconnaissance, geological radar, and seismic interpretation results. Wavelet analysis of 3-D seismic data in the depth domain, along with imaging and acoustic logging, was performed to extract formation and reservoir velocities. Wavelets at frequencies of 20 Hz, 25 Hz, and 35 Hz were selected to create forward models for different reservoir widths. An empirical formula was derived to establish the relationship between reservoir characteristics and seismic response based on frequency and formation velocity. Verification of the empirical formula was conducted using horizontal well trajectories, vertical well remote detection data, and logging data to extract the relationship between actual reservoir characteristics and seismic responses. This formula was then applied to amplitude attributes of the target layer in the study area to estimate the reservoir width range. The reservoir combination characteristics of the study area were identified, and a comprehensive detection workflow—incorporating analysis, design, verification, and redesign—was established through the analysis of 3-D seismic profiles. Optimal parameters, including wavelet frequency, migration aperture, and sampling interval, were designed for the forward model of the study area. A 2-D training sample database capable of real-time updates was developed, and the fracture-vuggy labels were iteratively optimized by integrating synthetic fracture-vuggy data with actual data. Using this database, the Res-UNet residual network was trained and applied to address the degradation problem commonly encountered in deep networks. This approach enabled the fine-scale prediction of fracture-vuggy structures, significantly improving the resolution and accuracy of reservoir interpretation.

Key words: Tarim Basin, fault-controlled fracture-vuggy reservoirs, deep learning, Res-UNet residual network, sample library optimization

中图分类号: