节点文献

微地震位置和震源机制的快速波形反演及搜索引擎算法的研究

Fast Elastic Full Waveform Inversion and Search Engine for Microseismic Location and Focal Mechanism

【作者】 张雄

【导师】 张捷;

【作者基本信息】 中国科学技术大学 , 固体地球物理, 2016, 博士

【摘要】 徽地震监测已经被广泛应用于矿场,地热,和油气行业。比如,它是页岩气开发中用来成像注水压裂的裂缝分布的非常好的工具,微地震位置和震源机制等可以为现场工程师提供重要的信息来评估注水压裂的效力等等。本文主要目的在于利用全波形匹配的方法来推断微震位置和震源机制解,分别从两个不同的方法出来研究这个问题。一方面,从波形反演的角度来同时获得微震的位置和震源机制解。我们首先介绍一种基于梯度的方法来同时反演震源位置和震源机制,为了克服这种基于梯度的方法的局部极小值的问题,我们同时提出一种基于一种全局最优化算法的快速波形反演算法。另一方面,从快速搜索的观点来研究这个问题,我们提出应用计算机科学行业的搜索引擎的概念来解决微震波形的匹配问题。本文中主要研究的两个方法可以总结如下:1,微震位置和震源机制的快速弹性波全波形反演算法给定一个速度模型,我们首先在可能的震源位置网格点上计算好格林函数库。反演过程中我们计算一个基于离散和预先准备好的格林函数库的近似的目标函数。由于提前计算好了格林函数,合成波形的计算变得快捷,从而使得我们能够利用一种全局最优化算法,邻域算法,来实现同时反演震源位置和震源机制。该方法中的目标函数应用了包络相关的概念来匹配高频的波形数据。在事件检测后,该方法并不需要拾取P和S波走时,我们利用波形残差和相关系数来评价波形的拟合程度。我们利用合成和实际数据例子测试了该方法,在输入数据中包含噪音和速度模型存在误差的情况下,合成数据例子表明能够较好的恢复出真实模型。我们同时测试了452个实际数据微震事件,并且和走时网格搜索算法进行了对比,快速波形反演得到的微震位置分布能够和走时网格搜索算法结果一致。2.微震搜索引擎算法类似于互联网搜索引擎,该方法能够在1s内同时估计出微震事件的位置和震源机制,来检测注水压裂过程。对于给定的采集系统和速度模型,我们首先在所有可能的位置网格点上计算所有可能的微震事件波形,从而建立一个搜索数据库。然后通过计算机快速搜索技术,多个随机K维树方法,根据数据库中波形数据的振幅和相位信息来排列并建立一个索引。当一个微震事件发生时,和输入波形近似的最佳波形能够通过匹配数据库的特征很快地找到。该方法不仅仅返回一个最佳解,而是类似于互联网搜索中的一个解集,因此我们可以利用得到的解集来进一步研究结果的置信度和解析度。同样类似于互联网搜索引擎,微震搜索引擎不需要其他输入参数和处理经验:这样,对于任何用户结果都会一样。我们同样利用合成数据和实际数据例子验证了该方法,结果显示微震搜索引擎有很大的潜力应用于微地震的实时监测问题。

【Abstract】 Passive microseismic monitoring has been widely used in many fields such as mining, geothermal, and gas/oil industries. For example, it is a valuable tool for mapping fractures in shale gas development. The microseismic event locations and focal mechanisms provide important information for the site engineers to assess the effectiveness of the hydraulic fracturing operations. In this study, we aim at inferring both the microseismic event locations and focal mechanisms simultaneously by matching the microseismic full waveform data through inversion methods or search engine method. We firstly introduce a gradient based method to simultaneously invert the event location and focal mechanism. To deal with the local minimum problem, we present a fast elastic full waveform inversion method based on a prepared Green’s function database. We also borrow the search engine concept from the computer science industry to solve the location and focal mechanism problem in one second. The main methods presented in this study can be summarized as following:1. Fast elastic full waveform inversion method for microseismic location and focal mechanism.Given a velocity model, we first calculate synthetic Green’s functions in the possible location grids and create a database. We then convolve the source with the Green’s function to generate synthetic waveform for calculating an approximate objective function based on the discrete and precomputed Green’s function database. Fast computation of synthetic seismograms with the Green’s function database allows using Neighborhood Algorithm to determine the global minimum in a computationally efficient manner. The method applies an envelope crosscorrelation approach to match high-frequency waveforms, and it requires event detection but no accurate time picking needed. We use both waveform residual and crosscorrelation concept in the objective function. The method is efficient to find the best matched waveform to the input with the Neighborhood Algorithm applied although there are noises in the input waveform or error in the velocity model. We also apply the method to 452 microseismic events obtained from two stages during hydraulic fracturing operations in the Barnett shale.2. Microseismic search engine for the estimation of the source location and focal mechanism.Similar to a web search engine, we develop a microseismic search engine that can estimate both an event location and the focal mechanism in less than a second to monitor the hydraulic fracturing process. The method is extended from a real-time earthquake monitoring approach for seismological applications. We first calculate the full waveforms of all possible microseismic events over a 3D grid with a known velocity model for a given acquisition geometry to create a database. We then index and rank all of the seismic waveforms in the database by following the characteristics of the phase and amplitude of the waveform through a computer fast search technology, specifically, the Multiple Randomized K-Dimensional (MRKD) tree method. When a microseismic event occurs, the approximate best matches to the entry waveform are found immediately by comparing the characteristic features between the input data and the database. The method returns not just one but a series of solutions, similar to a web search engine. Thus, we can obtain a solution space that delineates the resolution and confidence level of the results. Also similar to a web search engine, the microseismic search engine does not require any input parameter or processing experience; thus, the solutions are the same for any user. We demonstrate the method with both synthetic and real data, and the method shows great potential for the routine real-time monitoring of microseismic events during hydraulic fracturing.

  • 【分类号】TE357.6;P631.4
  • 【被引频次】4
  • 【下载频次】722
  • 攻读期成果
节点文献中: 

本文链接的文献网络图示:

本文的引文网络