王华忠
同济大学海洋与地球科学学院, 波现象与智能反演成像研究组, 上海 200092
基金项目: 国家重点研发计划深海关键技术与装备重点专项（2019YFC0312004），国家重点研发计划变革性技术关键科学问题重点专项（2018YFA0702503），国家自然科学基金（41774126，42074143），上海市浦江人才计划资助（20PJ1413500），南方海洋科学与工程广东省实验室（湛江）资助项目（ZJW-2019-04）和中国石化地球物理重点实验室基金（33550006-19-FW0399-0041，33550006-20-ZC0699-0011）资助
详细信息
作者简介: 罗飞, 男, 1990年生, 博士在读, 主要从事地震波传播理论及速度建模研究. E-mail: luofei19901217@126.com
中图分类号: P631 收稿日期:2020-04-01
修回日期:2021-02-26
上线日期:2021-06-10
Automatic first break picking based on Constrained Markov Decision Processes (CMDPs)
LUO Fei,WANG HuaZhong
Wave Phenomena and Intelligent Inversion Imaging Group(WPI), School of Ocean and Earth Science, Tongji University, Shanghai 200092, China
MSC: P631
--> Received Date: 01 April 2020
Revised Date: 26 February 2021
Available Online: 10 June 2021
摘要
摘要:随着地震数据采集技术的进步，地震数据量日益增加，全自动、高精度的地震初至走时拾取技术受到了更加广泛的关注.本文将初至拾取看作特征空间内带约束的Markov决策过程，在奖励函数空间，按一定准则全局寻优获得积累奖励值最大的路径，从而达到在高维空间自动拾取初至信息的目的.同时，状态值函数中包含与距离相关的折扣因子γ，使Markov决策过程拾取初至能够考虑地震数据的横向连续性，并且回避地震数据中的坏道信息.在此基础上，本文方法进一步引入受空间几何信息约束的动作（Actions）和转移概率（Transitions Probability），从而降低了对起始状态和折扣因子选取的难度，让地震数据初至走时拾取更加准确和自动化.实际数据测试结果表明，在初至能量较弱（信噪比较低）情况或浅层存在相邻较近复杂波形时，本文提出的约束Markov算法仍能准确地进行初至走时的自动拾取，并且具有一定的质量监控能力，让拾取结果更有物理意义.
关键词: 机器学习/
特征属性/
空间结构约束/
Markov决策过程/
初至自动拾取
Abstract:Picking first-breaks is an important step in seismic processing. The large volume of the seismic data calls for automatic and objective picking. In this paper, we formulate the first-breaks picking as Constrained Markov Decision Processes (CMDPs) in a feature space. By designing reasonable criteria, global optimization is carried out in a reward function space to determine the path with the largest cumulative reward value, so as to achieve the purpose of automatically picking up first arrival information in the high-dimensional space. At the same time, the state value function contains a distance-related discount factor γ, which enables the Markov decision process to pick up the first-arrival continuity to consider the horizontal continuity of the seismic data and avoid the bad track information in the seismic data. On this basis, the method of this paper further introduces reasonable actions and transition probability constrained by spatial geometric information, thereby reducing the difficulty of selecting the initial state and discount factor, and making the seismic data picking up more accurate and automatic. Tests on real seismic data show that this method can automatically pick up first arrival information accurately and has a certain QC ability, especially when the first arrival energy is weak (the signal-to-noise ratio is low) or there are adjacent complex waveforms in the shallow layer.
Key words:Machine learning/
Feature attributes/
Spatial structure constraints/
Markov decision process/
First arrival automatic picking
PDF全文下载地址:
http://www.geophy.cn/data/article/export-pdf?id=74e06a5c-3882-4eb7-a054-54b1746a0eeb
