王迪^1,,
袁三一^1,,,
袁焕²,
曾华会²,
王尚旭¹
1. 中国石油大学(北京)油气资源与探测国家重点实验室, 北京 102249
2. 中国石油天然气股份有限公司勘探开发研究院西北分院, 兰州 730000

基金项目: 国家重点研发计划(2018YFA0702504), 国家自然科学基金(41974140), 中央高校基本科研业务费专项资金(462019QNXZ03), 中国石油天然气集团有限公司-中国石油大学(北京)战略合作科技专项(ZLZX2020-03), 中国石油大学(北京)科研基金(2462020YXZZ008和2462020QZDX003)联合资助


详细信息

作者简介: 王迪, 女, 1996年生, 在读硕士研究生, 主要从事地震速度建模、动校正和地震解释方面的研究. E-mail: foilboil456@qq.com

通讯作者: 袁三一, 男, 1983年生, 研究员, 博士生导师, 主要从事高分辨率地震资料处理、地震反演、智能地球物理勘探和复杂油气藏储层预测等方面的研究. E-mail: yuansy@cup.edu.cn

中图分类号: P631

收稿日期:2020-08-11
修回日期:2021-01-03
上线日期:2021-03-10

Intelligent velocity picking based on unsupervised clustering with the adaptive threshold constraint

WANG Di^1,,
YUAN SanYi^1,,,
YUAN Huan²,
ZENG HuaHui²,
WANG ShangXu¹
1. State Key Laboratory of Petroleum Resources and Prospecting, China University of Petroleum(Beijing), Beijing 102249, China
2. The Research Institute of Petroleum Exploration and Development, Northwest Branch, Lanzhou 730000, China


More Information

Corresponding author: YUAN SanYi,E-mail:yuansy@cup.edu.cn

MSC: P631

--> Received Date: 11 August 2020
Revised Date: 03 January 2021
Available Online: 10 March 2021


摘要
摘要:目前叠加速度的获取主要是通过人工拾取速度谱，存在着效率低、耗时长且易受人为因素影响的缺点.本文提出了一种基于自适应阈值约束的无监督聚类智能速度拾取方法，实现叠加速度的自动拾取，在保证速度拾取精度的同时提高拾取效率.利用时窗方法在速度谱中计算自适应阈值，从而识别出一次反射波速度能量团作为速度拾取的候选区域.然后，根据K均值方法将不同的速度能量团聚类，并将最终的聚类中心作为拾取的叠加速度.最后，依据人工拾取速度的经验，加入了离群速度点的后处理工作，以获得更光滑的速度场.模型和实际地震数据测试结果表明，本文提出的方法总体上与人工拾取叠加速度的精度相当，但明显提升了速度拾取效率，极大缓解了人工拾取负担.
关键词: 无监督/
速度拾取/
自适应阈值/
聚类/
人工智能

Abstract:In seismic interpretation, stacking velocity is mainly acquired by manual picking from velocity spectra, which is time-consuming and highly susceptible to human experience. To improve the picking efficiency, we develop an automatic velocity picking approach based on an adaptive threshold constrained unsupervised clustering. A sliding time window is applied on velocity spectra to seek adaptive thresholds, which contribute to identifying effective energy clusters and construct candidate set from the identified eligible points. The clusters in the candidate set are automatically assigned into different groups by the classical K-means clustering method. The ultimate centroids of each group are marked as the stacking velocities picked automatically. Based on the experience of manual interpretation, we introduce a brief post processing procedure to eliminate velocity outliers and obtain a smooth velocity field. Both synthetic and real data tests demonstrate that our proposed method can significantly relieve the burden of manual labor, improve the efficiency and in the meanwhile retain relative high accuracy.
Key words:Unsupervised/
Velocity picking/
Adaptive threshold/
Clustering/
Artificial intelligent



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=608009f9c23e67607ae103e1