储层表征

次声成像描绘出更清晰的储层图片

本文提出了一种新的数据驱动分析来定位低频地震源,称为近次声或次声源。将这些次声信号与微震信号相结合,可以更好地表征和监测受刺激的储层体积。

互联网技术网格。 与蜘蛛网的抽象背景。 数据量大。 3D 渲染。
来源:Olena Lishchyshyna/Getty Images

当流体以高速率注入地层时,地下岩石会破裂并振动,产生极低频的能量,就像一个巨大的低音炮一样。地下次声信号可以在不同地质环境(例如火山、间歇泉和海洋岩浆)中的流体驱动体积过程中产生。本文分享了一个数据驱动的工作流程,用于定位和监测增产储层体积(SRV)中的次声和近次声源(2-80 Hz)。

为此,工作流程分析了在 1.5 km 深度的中尺度(~10 m)水力压裂现场实验中获得的水听器测量中的低频信号。这种方法揭示了 SRV 中使用传统微震成像无法看到的岩石变形部分。将这些次声信号与微震信号相结合,可以更好地表征和监测 SRV。

新开发的用于新型次声成像的数据驱动工作流程对于定位和监测地下流体路径非常有价值,从而极大地改善地热和碳氢化合物资源的开发。

水力压裂现场实验场地
本文提出了一种新的数据驱动分析来定位低频地震源(2-80 Hz),称为近次声或次声源。该研究进一步研究了次声源与微震活动和模拟天然裂缝的关系。Chakravarty 等人提供了该研究的更多详细信息(2022)

该研究使用 EGS 协作项目生成的水听器数据,该项目是一个独特的中尺度、仪器密集的测试站点,位于南达科他州利兹,由美国能源部资助。EGS 协作实验 1 是一项中尺度(约 10 m)水力压裂现场实验,位于南达科他州利兹市 Homestake 金矿,深度为 1.5 km。中规模现场研究的目的是通过受刺激岩石体积的密集地球物理仪器来详细描述水力压裂的特征(Kneafsey等人,2020年Chakravarty和Misra,2022年)。图 1更详细地描述了中等规模的现场研究。

次声_Fig1.jpg
图1- ( a) 南达科他州Leads Homestake 矿区EGS 合作实验1 的试验台布局。插图显示了位于 1500 m 深度的井布局示意图。红线和绿线分别代表沿最小水平应力方向钻探的注入井和生产井。(b) 水力增产方案。(c) 水听器OT02测量的次声(2-80 Hz)信号。左侧的黄色和洋红色垂直线显示注入的开始和第一次微震的出现。另一组线标志着最后一次微震事件和流体注入的结束。(d) 网格示意图,显示次声源位置的网格节点(绿点)和传感器位置(黄色三角形)。

水力压裂现场实验过程中产生的次声能量
图2显示了中等规模现场研究中水力压裂过程中在次声和近次声频段(8-20Hz)检测到的能量。在图2中,次声能量与注入的流体体积相关。两个正交水听器串 E1-OT 和 E1-PDB,每个由 12 个水听器组成,记录了流体注入过程中的次声和次声发射。由于E1-OT井上的水听器更接近流体驱动变形,因此E1-OT水听器上的入射次声信号强度更大。E1-OT(图2,红色)水听器测量的次声能量比远处的弦E1-PDB(图2,蓝色)大五个数量级。请注意,随着实验的进行,两个水听器串的累积次声能量的梯度变得逐渐平滑。5 月 24 日,表明粘/滑类型裂缝扩展的脉冲能量释放占主导地位,其中能量释放呈离散爆发形式,导致所有传感器的累积能量曲线中出现强脊。流经裂缝管道的流体会产生长周期次声震颤,表明能量持续释放。这种转变表明从流体驱动的裂缝扩展到流体流过裂缝管道的状态变化。

次声_Fig2.jpg
图2“5月24日(左)、5月25日第1部分(中)和5月25日第2部分(右)注液速率对次声/近次声能量释放的依赖性。水听器阵列E1-OT(红色)垂直于断裂带并与其相交,而水听器阵列E1-PDB(蓝色)与断裂带近平行且比E1-OT更远。

数据驱动分析来定位次声和近次声源
由于信号的突发性,我们使用了基于互相关的网格搜索方法,该方法通常用于识别局部和区域尺度的震源(Wech和 Creager 2008)。该算法需要输入水听器信号、网格(图 1d)和速度模型。我们设置网格尺寸以匹配水听器网络,并将其在两个方向上延伸 30%,同时对压缩波使用 5.5 km/s 的各向同性速度。滚动窗口定位算法的工作原理是在每个窗口对来自每个站的成对信号进行互相关,并测量它们作为信号位移函数的相似性。将根据相关图计算出的时滞与基于输入速度模型计算出的站对之间的理论时滞进行比较。通过最小化建模目标函数和观测目标函数之间的差异,具有最小失配的网格节点被识别为源位置。

给定合适的各向同性速度模型,确定算法源位置的关键参数是窗口长度和窗口重叠。由于检测初到信号的不确定性,确定突发信号的信号持续时间并非易事。应用短期平均/长期平均 (STA/LTA) 方法通常会导致高估脉冲持续时间。为了估计脉冲持续时间,我们将 STA/LTA ​​滤波器应用于来自 E1-OT 井的水听器数据样本,然后校正高估。获得1秒的平均值作为次声信号的平均脉冲持续时间。根据此信息,选择 1 秒的窗口长度和 0.5 秒的窗口重叠用于后续分析。

后处理以细化次声/近次声源位置
数据驱动的网格搜索算法容易出现误报。因此,应用一系列后处理滤波器来确定可靠的低频源位置。

  • 第一个滤波器将窗口的归一化互相关系数设置为 0.95 的上限和 0.6 的下限。这些边界之外的窗口将被丢弃,因为它们可能包含相关噪声或不相关信号。
  • 第二个滤波器使用在整个实验过程中计算的水听器阵列的相对功率来丢弃归一化波束功率低于本底噪声的定位时间戳。相对功率阈值(0.3)用于区分定位时间戳和非定位时间戳。
  • 第三个滤波器使用自举法对基于互相关的位置执行 20 次迭代。在每次迭代中,随机删除 5% 的互相关图,所得散布被视为位置不确定性的度量。具有最高 10% 散点值的数据点将被丢弃。
  • 最后一个过滤器使用网格搜索算法中获得的失配来丢弃显示最高失配值的数据。删除了 50% 具有最高失配值的数据。尽管丢失了一半的数据,但源位置的空间覆盖范围几乎没有变化,这证明了失配过滤器的有效性。

次声/近次声源与微震源的时空演变
图3说明了新确定的次声/近次声源位置随时间的演变。5 月 24 日早些时候,源头垂直于注入井展开(图 3a)。超过 2245 小时 UTC 时间后,源头会沿着与注入井近平行的线状分布集中,随后的事件遵循相同的方向,但从注入点向北迁移。5月25日第1部分(图3c),源头的分布比以前更加分散,早期源头沿注入点以南的东西走向,后期事件则形成相对稀疏的线趋势向北的注射点。还观察到东西方向的两条近平行线。在5月25日第2部分注入开始时(图3e),次声源落在注入点两侧的先前描述的两条线上,与注入井近垂直,后来的事件与注入近平行排列出色地。

次声/近次声震源(绿色)和微震震源(黑色)的联合分析,如图 2 所示。图 3b、d 和 f 封装了采集仪器可观察范围内的频率(2 Hz 至 15000 Hz)。因此,由流体注入驱动的高频和低频压裂现象都被捕获。联合数据反映了流体注入引起的地下变形,该变形位于连续体上,一端代表裂缝上的高频、小规模剪切滑移,另一端代表低频、大规模空隙体积膨胀或收缩。由此得出的结论是,微震活动和次声信号包含有关由于流体注入而导致的岩石变形的互补信息,并且它们的联合分析可以更完整地呈现地下受刺激裂缝的情况。

次声_Fig3.jpg
图3-次声/近次声震源与微震震源的时空演化。(a、b) 5 月 24 日;(c、d) 5 月 25 日第 1 部分;(e, f) 5 月 25 日第 2 部分。彩色点显示使用新开发的数据驱动方法检测到的次声/近次声源,而黑点显示同时记录的微震活动。蓝线和红线分别表示注入井和生产井。注水井 E1-I 上的粉色星星标记了注水点。与注入井和监测井近平行的黑线为水听器管柱E1-PDB,近水平线为E1-OT管柱。线上的橙色方块标记了监测井上的水听器传感器。

次声/近次声源的拟议机制        
通过高频和低频信号的联合分析,与仅使用一种方法相比,可以获得更多有关裂缝的信息。流体注入裂隙岩石会导致裂缝扩大或收缩,从而产生机械波,从而产生高频剪切运动和低频纵波。如图 4所示,该图显示了离散裂缝网络中天然裂缝的方向和分布。微震活动的存在表明流体相互作用动员了临界应力裂缝,而次声活动表明加压天然裂缝产生低频压缩运动。高度破碎的岩石导致次声测量显着,微震测量相对较低,而完整的岩石环境导致低频地震可以忽略不计,而高频微震的发生相对较多。此外,与裂隙岩石相比,完整岩石中高频信号的衰减要小得多,从而产生更多高频信号,并且更容易被传感器测量。

次声_Fig4.jpg
图4' ( a)自然裂隙岩石体积中流体注入驱动的次声和微震能量释放的示意图。MEQ 是微震事件。(b) 完整岩石体积中流体注入驱动的次声和微震能量释放的示意图。(c)与微震和离散裂缝网络(DFN)的比较。组合位置云与从 DFN 推断的整体方向非常一致。根据压裂前试验台的DFN模型,裂缝方向和次声源位置之间存在很强的一致性。

结论
在 3 天的刺激中,总共获得了 322 个、818 个和 1,117 个次声源位置。早期阶段的脉冲能量释放对应于裂缝扩展,而刺激后期阶段的更平滑释放对应于管道中流体流动产生的震颤状运动。具有可用信噪比的次声信号仅在相对较高的流体注入速率下产生。次声是突发信号,因此基于阈值的方法的初至拾取变得不准确。因此,应用数据驱动的基于互相关的网格搜索来定位次声源位置。设计并应用了四个过滤步骤来改进源位置。


致谢
本材料基于美国能源部、科学办公室、基础能源科学办公室、化学科学、地球科学和生物科学部门支持的工作,奖励编号为 DE-SC0020675 和德克萨斯 A&M TRIAD 资助。作者感谢 EGS Collab 团队在概念化和执行实验以及托管开源数据方面所做的努力。作者感谢 Texas A&M 高性能研究计算工作人员的支持。


参考文献
Chakravarty, A. 和 Misra, S. 2022。利用三分量加速度计数据进行无监督学习,监测中尺度水力裂缝的时空演化。国际岩石力学和采矿科学杂志https://doi.org/10.1016/j.ijrmms.2022.105046

Chakravarty,A.等人。2022。水力压裂驱动的次声信号——地下工程的新型信号。https://doi.org/10.1002/essoar.10512584.1

Kneafsey,TJ,等人。2020。EGS 合作项目:实验 1 的经验教训。https: //www.osti.gov/biblio/1766466

Wech, AG 和 Creager, KC 2008。Cascadia TTremor 的自动检测和定位。地球物理研究快报https://doi.org/10.1029/2008GL035458

原文链接/jpt
Reservoir characterization

Infrasound Imaging Paints Clearer Reservoir Picture

This article presents a new data-driven analysis to locate low-frequency seismic sources, referred to as near-infrasound or infrasound sources. Combining these infrasound signals with microseismicity signals allows for better characterization and monitoring of the stimulated reservoir volume.

Grid of internet technologies. Abstract background with cobwebs. Large amount of data. 3d rendering.
Source: Olena Lishchyshyna/Getty Images

When fluids are injected into formations at high rates, underground rocks break and vibrate, creating extremely low-frequency energy, like a giant subwoofer. Subsurface infrasound signals can be generated during fluid-driven volumetric processes in different geological settings such as volcanoes, geysers, and oceanic magma. This article shares a data-driven work flow used to locate and monitor the infrasound and near-infrasound sources (2–80 Hz) in the stimulated reservoir volume (SRV).

To that end, the work flow analyzes the low-frequency signals in the hydrophone measurements acquired during a mesoscale (~10 m) hydraulic-fracturing field experiment at a depth of 1.5 km. This approach revealed sections of rock deformation in the SRV that were not visible using traditional microseismic imaging. Combining these infrasound signals with microseismicity signals allows for better characterization and monitoring of the SRV.

The newly developed data-driven work flow for a novel infrasound imaging will be valuable for locating and monitoring subsurface fluid pathways, tremendously improving geothermal and hydrocarbon resource development.

Site for the Hydraulic-Fracturing Field Experiment
This article presents a new data-driven analysis to locate the low-frequency seismic sources (2–80 Hz), referred to as near-infrasound or infrasound sources. The study further investigates the relationship of the infrasound sources to microseismicity and modeled natural fractures. More detail of the study is available in Chakravarty et al. (2022).

The study uses hydrophone data generated by the EGS Collab Project, a unique mesoscale, densely instrumented test site located in Leads, South Dakota, and funded by the US Department of Energy. The EGS Collab Experiment 1 is a mesoscale (~10 m) hydraulic-fracturing field experiment situated at the Homestake Gold Mine in Leads, South Dakota, at a depth of 1.5 km. The aim of the intermediate-scale field study is detailed characterization of hydraulic fracturing through dense geophysical instrumentation of the stimulated rock volume (Kneafsey et al. 2020; Chakravarty and Misra 2022). Fig. 1 describes the intermediate-scale field study in more details.

Infrasound_Fig1.jpg
Fig. 1—(a) Testbed layout of EGS Collab Experiment 1 in Homestake Mine in Leads, South Dakota. Inset shows the schematic of the well layout, situated at a depth of 1500 m. Red and green lines represent injection and production wells, respectively, drilled along the minimum horizontal stress direction. (b) Hydraulic stimulation protocol. (c) Infrasound (2–80 Hz) signal measured by hydrophone OT02. Yellow and magenta vertical lines on the left show the start of injection and appearance of first microseismicity. The other set of lines mark the last microseismic event and the end of fluid injection. (d) Schematic of the grid showing grid nodes (green dots) and sensor locations (yellow triangles) for infrasound source location.

Infrasound Energy Generated During the Hydraulic-Fracture Field Experiment
Fig. 2 shows the energy detected in the infrasound and near-infrasound frequency band (8–20 Hz) during the hydraulic stimulation at the intermediate-scale field study. In Fig. 2, the infrasound energy is correlated to the injected fluid volume. Two orthogonal hydrophone strings, E1-OT and E1-PDB, each consisting of 12 hydrophones, recorded the infrasound and infrasound emission during fluid injections. As the hydrophones on the well E1-OT are closer to the fluid-driven deformation, the incident infrasound signal intensity is greater on the E1-OT hydrophones. The infrasound energy measured by E1-OT (Fig. 2, red) hydrophones is five orders of magnitude greater than the distant string E1-PDB (Fig. 2, blue). Note that, as the experiment proceeds, the gradient of cumulative infrasound energy for both the hydrophone strings becomes progressively smoother. Impulsive energy release indicative of stick/slip type of fracture propagation is dominant on 24 May, wherein the energy release is in discrete bursts, resulting in strong ridges in the cumulative energy curves from all sensors. Fluid flow through fracture conduits generates long-period infrasound tremors, indicative of long-duration energy release. This transition indicates a regime change from fluid-driven fracture propagation to fluid flow through fractured conduits.

Infrasound_Fig2.jpg
Fig. 2—Dependence of fluid-injection rate on infrasound/near-infrasound energy release on 24 May (left), 25 May part 1 (center), and 25 May part 2 (right). Hydrophone array E1-OT (red) is perpendicular to the fractured zone and is intersected by it, whereas the hydrophone array E1-PDB (blue) lies subparallel to the fractured zone and farther away than E1-OT.

Data-Driven Analysis To Locate Infrasound and Near-Infrasound Sources
Because of the emergent nature of the signals, we used a grid search approach based on cross-correlation, which has been commonly used for identifying sources of tremors on local and regional scales (Wech and Creager 2008). The algorithm requires the input of a hydrophone signal, a grid (Fig. 1d), and a velocity model. We set the grid dimensions to match the hydrophone network and extended it by 30% in both directions, while using an isotropic velocity of 5.5 km/s for the compressional wave. The rolling window location algorithm works by cross-correlating pairwise signals from every station at each window and measuring their similarity as a function of signal displacement. The time lag calculated from the correlogram is compared with the calculated theoretical time lag between station pairs based on the input velocity model. By minimizing the difference between the modeled and observed objective functions, the grid node with the minimum misfit is identified as the source location.

Given a suitable isotropic velocity model, the key parameters determining the source locations of the algorithm are the window length and window overlap. Determining the signal duration of emergent signals is nontrivial because of uncertainty in detecting first arrivals. Application of short-term average/long-term average (STA/LTA) methods usually lead to overestimating the pulse duration. To get an estimate of the pulse duration, we applied the STA/LTA filter to a sample of hydrophone data from well E1-OT and then corrected for the overestimation. An average value of 1 second was obtained as the average pulse duration of the infrasound signals. With this information, a window length of 1 second and window overlap of 0.5 second is chosen for subsequent analysis.

Post-Processing To Refine the Infrasound/Near-Infrasound Source Locations
The data-driven, grid search algorithm is prone to false positives. Hence, a series of post-processing filters is applied to determine reliable low-frequency source locations.

  • The first filter sets an upper bound of 0.95 and a lower bound of 0.6 on the normalized cross-correlation coefficients of the windows. Windows outside these bounds are discarded because they may contain correlated noise or uncorrelated signal.
  • The second filter uses the relative power of the hydrophone array computed throughout the experiments to discard located timestamps with normalized beam power below the noise floor. A threshold value of relative power (0.3) is used to differentiate located and nonlocated timestamps.
  • The third filter uses bootstrapping to perform 20 iterations for the cross-correlation-based locations. In each iteration, 5% of the cross-correlograms are removed randomly and the resulting scatter is considered a measure of location uncertainty. The data points with the highest 10% of scatter values are discarded.
  • The last filter uses the misfits obtained in the grid search algorithm to discard data showing the highest misfit values. 50% of the data with the highest misfit values are removed. Despite losing half the data, the spatial coverage of the source locations shows little change, demonstrating the effectiveness of the misfit filter.

Spatiotemporal Evolution of Infrasound/Near-Infrasound Sources Compared With Microseismic Sources
Fig. 3 illustrates the evolution of the newly identified infrasound/near-infrasound source locations over time. During the early time on May 24, the sources spread out perpendicular to the injection well (Fig. 3a). Beyond 2245 hours UTC, the sources become concentrated along a lineament that is subparallel to the injection well, and subsequent events follow the same direction but migrate northward from the injection point. In Part 1 of May 25 (Fig. 3c), the sources have a less diffuse distribution than before, with early-time sources along an east/west trend south of the injection point and later events forming a relatively sparse line trend to the north of the injection point. Two subparallel lineaments in an east/west direction are also observed. At the beginning of injection in Part 2 of May 25 (Fig. 3e), the infrasound sources fall on the previously described two lineaments on either side of the injection point, being subperpendicular to the injection well, and later events align subparallel to the injection well.

The joint analysis of infrasound/near-infrasound sources (green) and microseismic sources (black), as shown in Figs. 3b, d, and f, encapsulates frequencies on the observable bounds of acquisition instrumentation (2 Hz to 15000 Hz). As a result, both high- and low-frequency fracturing phenomena driven by fluid injection are captured. The joint data reflects fluid-injection-induced subsurface deformation that lies on a continuum, with one end representing high-frequency, small-scale shear slippage on fractures and the other end representing low-frequency, large-scale void volume dilation or contraction. This leads to the conclusion that microseismicity and infrasound signals contain complementary information about rock deformation because of fluid injection, and their joint analysis renders a more complete picture of the stimulated fractures in subsurface.

Infrasound_Fig3.jpg
Fig. 3—Spatiotemporal evolution of infrasound/near-infrasound sources compared with microseismic sources. (a, b) May 24; (c, d) May 25 Part 1; and (e, f) May 25 Part 2. Colored points show infrasound/near-infrasound sources detected using the newly developed data-driven approach, while black points show the simultaneously recorded microseismicity. Blue and red lines indicate injection and production wells, respectively. Pink star on the injection well E1-I marks the injection point. Black line subparallel to injection and monitoring wells is hydrophone string E1-PDB, and subhorizontal line is string E1-OT. Orange squares on lines mark the hydrophone sensors on the monitoring wells.

Proposed Mechanisms Governing Infrasound/Near-Infrasound Sources        
Through joint analysis of high- and low-frequency signals, more information can be obtained about fractures compared with using only one method. Fluid injection into fractured rock can cause cracks to expand or contract, creating mechanical waves that generate both high-frequency shear motion and low-frequency P-waves. This is illustrated in Fig. 4, which shows the orientation and distribution of natural fractures in a discrete fracture network. The presence of microseismicity indicates fluid interactions mobilizing critically stressed cracks, while infrasound activity suggests pressurized natural fractures that generate low-frequency compressional motion. The highly fractured rock leads to significant infrasound measurement and relatively low microseismic measurement, while the intact rock environment results in negligible low-frequency seismic and relatively greater occurrence of high-frequency microseismic. Moreover, the attenuation of high frequency signals is far less in intact rock compared with fractured rock, resulting in more high-frequency signals being generated and more easily measured by sensors.

Infrasound_Fig4.jpg
Fig. 4—(a) Schematic representation of the fluid-injection driven infrasound and microseismic energy release in a naturally fractured rock volume. MEQs are microseismic events. (b) Schematic representation of the fluid-injection driven infrasound and microseismic energy release in an intact rock volume. (c) Comparison with microseismic and discrete fracture network (DFN). The combined location cloud shows strong agreement with overall orientation inferred from the DFN. Based on the DFN model of the testbed before fracturing, a strong agreement exists between the fracture orientations and infrasound source locations.

Conclusions
A total of 322, 818, and 1,117 infrasound source locations were obtained for the 3 days of stimulations. Impulsive energy release at earlier stages corresponded to fracture propagation, while a smoother release at later stages of stimulation corresponds to tremor-like motions generated from fluid flow in conduits. Infrasound signals of usable signal-to-noise ratio are produced only at relatively high fluid-injection rates. The infrasound is emergent signal, so first arrival picking from threshold-based methods is rendered inaccurate. Therefore, a data-driven cross-correlation-based grid search was applied to locate the infrasound source locations. Four filtering steps were designed and applied to improve the source locations.


Acknowledgements
This material is based on work supported by the US Department of Energy, Office of Science, Office of Basic Energy Sciences, Chemical Sciences Geosciences, and Biosciences Division, under Award Number DE-SC0020675 and the Texas A&M TRIAD funding. The authors acknowledge the efforts of the EGS Collab team for conceptualizing and executing the experiments and hosting the open-source data. The authors are grateful to Texas A&M High Performance Research Computing staff members for their support.


References
Chakravarty, A., and Misra, S. 2022. Unsupervised Learning From Three-Component Accelerometer Data To Monitor the Spatiotemporal Evolution of Mesoscale Hydraulic Fractures. International Journal of Rock Mechanics and Mining Sciences. https://doi.org/10.1016/j.ijrmms.2022.105046.

Chakravarty, A. et al. 2022. Hydraulic Fracturing-Driven Infrasound Signals—a New Class of Signal for Subsurface Engineering. https://doi.org/10.1002/essoar.10512584.1.

Kneafsey, T. J., et al. 2020. The EGS Collab Project: Learnings from Experiment 1. https://www.osti.gov/biblio/1766466.

Wech, A.G., and Creager, K.C. 2008. Automated Detection and Location of Cascadia TTremor. Geophysical Research Letters. https://doi.org/10.1029/2008GL035458.