Application of Machine Learning Methods in Arrival Time Picking of P Waves from Reservoir Earthquakes

Reservoir earthquake characteristics such as small magnitude and large quantity may result in low monitoring efficiency when using traditional methods. However, methods based on deep learning can discriminate the seismic phases of small earthquakes in a reservoir and ensure rapid processing of arrival time picking. The present study establishes a deep learning network model combining a convolutional neural network (CNN) and recurrent neural network (RNN). The neural network training uses the waveforms of 60 000 small earthquakes within a magnitude range of 0.8-1.2 recorded by 73 stations near the Dagangshan Reservoir in Sichuan Province as well as the data of the manually picked P-wave arrival time. The neural network automatically picks the P-wave arrival time, providing a strong constraint for small earthquake positioning. The model is shown to achieve an accuracy rate of 90.7% in picking P waves of microseisms in the reservoir area, with a recall rate reaching 92.6% and an error rate lower than 2%. The results indicate that the relevant network structure has high accuracy for picking the P-wave arrival times of small earthquakes, thus providing new technical measures for subsequent microseismic monitoring in the reservoir area.     

基于数据增广和CNN的地震随机噪声压制

卷积神经网络(Convolutional Neural Network,CNN)是一种基于数据驱动的学习算法,简化了传统从特征提取到分类的两阶段式处理任务,被广泛应用于计算机科学的各个领域.在标注数据不足的地震数据去噪领域,CNN的推广应用受到限制.针对这一问题,本文提出了一种基于数据生成和增广的地震数据CNN去噪框架.对于合成数据,本文对无噪地震数据添加不同方差的高斯噪声,增广后构成训练集,实现基于小样本的CNN训练.对于实际地震数据,由于无法获得真实的干净数据和噪声来生成训练样本集,本文提出一种直接从无标签实际有噪数据生成标签数据集的方法.在所提出的方法中,我们利用目前已有的去噪方法从实际地震数据中分别获得估计干净数据和估计噪声,前者与未知的干净数据具有相似纹理,后者与实际噪声具有相似的概率分布.人工合成数据和实际数据实验结果表明,相较于F-X反褶积,BM3D和自适应频域滤波算法,本文方法能更好地压制随机噪声和保护有效信号.最后,本文采用神经网络可视化方法对去噪CNN的机理进行了探索,一定程度上解释了网络每一层的学习内容.     

利用三维高斯射线束成像进行地震定位

常规的地震定位方法通常需要拾取地震记录的初至,当初至不明显或被较高水平的噪声淹没时精度较低.本文采用基于三维高斯射线束的偏移成像方法对震源进行定位,较好地解决了该问题.通过三维高斯射线束对台站记录进行偏移归位,并将各台站成像结果的交点作为地震能量释放的中心位置;当各台站成像结果不能交于一点时,采用三维空间高斯滤波方法可实现震源位置的自动获取.提出的变网格计算方案极大地减少了计算量,显著地提高了成像精度和计算效率.利用首都圈地震台网数据,对涿鹿、滦县以及房山三个地震事件进行试算,结果表明:基于变网格三维高斯束偏移成像的地震定位方法自动化程度很高,而且具有较好的抗噪能力,特别适合处理低信噪比资料的地震定位问题.     

一种地震P波和S波初至时间自动拾取的新方法

地震P波、S波初至时间的拾取是地震波分析的一项基础性工作.本文提出了一种新的地震波初至时间自动拾取的方法:首先,把地震波的三分量时程曲线变换为一组空间向的能量变化率时程曲线;然后对能量变化率时程曲线进行STA/LTA(Short Time Average/Long Time Average,短时间的均值/长时间的均值)处理,拾取地震P波和S波的大致初至时间;最后提出采用一种二次方自回归模型对初至附近的能量变化率曲线进行二次方自回归处理,精确拾取出P波和S波的初至时间.本文采用了10组芦山地震的记录数据和150组汶川地震的记录数据对此方法的可靠性进行了检验.以人工拾取结果为参考,此方法具有很高的准确率和稳定性,同时,相比于常用的STA/LTA方法和AIC(Akaike Information Criterion,Akaike信息准则)方法,此方法在计算时间效率方面稍微逊色,但是对S波初至时间的拾取精度和可靠性更高.此方法丰富了地震P波、S波初至时间的自动拾取方法.     

Estimation of Thomsen＇s anisotropic parameters from walkaway VSP and applications

Estimation of Thomsen＇s anisotropic parameters is very important for accuratetime-to-depth conversion and depth migration data processing. Compared with othermethods, it is much easier and more reliable to estimate anisotropic parameters that arerequired for surface seismic depth imaging from vertical seismic profile （VSP） data, becausethe first arrivals of VSP data can be picked with much higher accuracy. In this study, wedeveloped a method for estimating Thomsen＇s P-wave anisotropic parameters in VTImedia using the first arrivals from walkaway VSP data. Model first-arrival travel times arecalculated on the basis of the near-offset normal moveout correction velocity in VTI mediaand ray tracing using Thomsen＇s P-wave velocity approximation. Then, the anisotropicparameters 0 and e are determined by minimizing the difference between the calculatedand observed travel times for the near and far offsets. Numerical forward modeling, usingthe proposed method indicates that errors between the estimated and measured anisotropicparameters are small. Using field data from an eight-azimuth walkaway VSP in TarimBasin, we estimated the parameters 0 and e and built an anisotropic depth-velocity modelfor prestack depth migration processing of surface 3D seismic data. The results showimprovement in imaging the carbonate reservoirs and minimizing the depth errors of thegeological targets.     

相空间地震波初至时刻拾取

提出了一种新的地震波初至时刻拾取的方法,即将原始时间序列信号映射到相空间当中,通过其相空间图的特征进行初至时刻的拾取。相对于非常耗时的传统方法,本方法使得运算速度提高,结果更加精确稳定。     

Seismic phase picking in China Seismic Array using a deep convolutional neuron network*

Seismic phase picking is the preliminary work of earthquake location and body-wave travel time tomography. Manual picking is considered as the most accurate way to access the arrival times but time consuming. Many automatic picking methods were proposed in the past decades, but their precisions are not as high as human experts especially for events with low ratio of signal to noise and later arrivals. As the increasing deployment of large seismic array, the existing methods can not meet the requirements of quick and accurate phase picking. In this study, we applied a phase picking algorithm developed on the base of deep convolutional neuron network (PickNet) to pick seismic phase arrivals in ChinArray-Phase III. The comparison of picking error of PickNet and the traditional method shows that PickNet is capable of picking more precise phases and can be applied in a large dense array. The raw picked travel-time data shows a large variation deviated from the traveltime curves. The absolute location residual is a key criteria for travel-time data selection. Besides, we proposed a flowchart to determine the accurate location of the single-station earthquake via dense seismic array and phase arrival picked by PickNet. This research expands the phase arrival dataset and improves the location accuracy of single-station earthquake.     

基于样本增强的卷积神经网络震相拾取方法

为了快速、高效地从地震数据中识别地震事件和拾取震相,本文利用基于样本增强的卷积神经网络自动震相拾取方法,将西藏林芝地区L0230台站3个月数据作为训练集,该区内另外6个台站连续1个月的波形数据作为测试集,采用高斯噪声、随机噪声拼接、随机挑选噪声、随机截取地震事件等4种样本增强的方法扩增训练集,以提高自动震相拾取技术的准确率。结果显示:样本增强前模型在测试集上的地震事件识别准确率为80%,样本增强后提升至97%,表明样本增强有效地提高了模型的泛化性能和抗干扰能力;在0.5 s误差范围内,震相自动拾取准确率高于81%,在1.0 s误差范围内,准确率高于95%;利用基于样本增强的卷积神经网络震相拾取方法能够检测出人工拾取震相中误标和漏检的震相。      

Hausdorff分数维识别地震道初至走时

地震波初至走时的识别在地震勘探、人工地震层析成像以及全球地震层析成像方法研究中起重要作用．初至走时拾取的精度在很大的程度上影响地震层析成像及演的精度．本研究以提高地震波初至走时拾取的精度及定量化程度为目标,利用计算地震道时间序列分数维的方法,实现了地震波初至走时的自动拾取．本文以分形理论为基础,进行了地震道时间序列Hausdorff分数维的计算．计算结果表明地震道时间序列的分数维在初至到达前后具有不同的数值,其变化点能够定量指示出初至走时的位置．本文还给出了利用该方法对实测数据进行初至走时拾取的实例．     

地震层析成像技术在岩体完整性测试中的应用

地震波层析成像借鉴了医学上X射线断面扫描的基本原理,利用地震波穿过地质体后走时及能量的改变等物理信息,通过数学处理重建地质体内部图像,从而得到所研究地质体的岩性及构造分布。本文利用这种方法,在一个钻孔中利用电火花震源激了弹性波,在另一个钻孔布设多个检波点同时接收,拾取弹性波初至时间,将接收到的数据利用SIRT方法进行反演迭代计算,最终形成一个弹性波速度谱图,然后利用岩土体的弹性波速度差异推断岩体完整性分布。与其它测试方法比较,该方法分辨率高,空间位置准确,在工程物探、岩土工程勘察中具有较好的应用前景。     

基于深度学习的地震检测模型在区域台网的泛化性研究

将识别地震的深度学习算法PhaseNet应用于四川台网和首都圈台网,对该模型的泛化能力进行了测试和评估.首先利用2010年1月至2018年10月首都圈台网199个地震台站记录的29 328个事件(ML0～ML4)所对应的126761段事件波形,以及2019年4-9月四川及邻省部分台网227个地震台站记录的16595个事...     

A field experiment of walkaway distributed acoustic sensing vertical seismic profile in a deep and deviated onshore well in Japan using a fibre optic cable deployed inside coiled tubing

A two-dimensional walkaway vertical seismic profiling survey using distributed acoustic sensing was conducted at an onshore site in Japan. The maximum depth and the deviation of the observation well were more than 4,000 m and 81 degrees, respectively. Among the several methods for installing fibre optic cables, we adopted the inside coiled tubing method, in which coiled tubing containing a fibre optic cable is deployed. The signal-to-noise ratio of the raw shot gather was low, possibly due to poor coupling between the fibre optic cable and the subsurface formation resulting from the fibre optic cable deployment method and the existence of considerable tubewave noise. Nevertheless, direct P-wave arrivals, P–P reflections and P–S converted waves exhibited acceptable signal-to-noise ratios after careful optimization of gauge length for distributed acoustic sensing optical processing and the application of carefully parameterized tubewave noise suppression. One of the challenges in current distributed acoustic sensing vertical seismic profile data processing is the separation of P- and S-waves using only one-component measurements. Hence, we applied moveout correction using two-dimensional ray tracing. This process effectively highlights only reflected P-waves, which are used in subsequent subsurface imaging. Comparison with synthetic well seismograms and two-dimensional surface seismic data confirms that the final imaging result has a sufficiently high quality for subsurface monitoring. We acquired distributed acoustic sensing vertical seismic profile data under both flowing conditions and closed conditions, in which the well was shut off and no fluid flow was allowed. The two imaging results are comparable and suggest the possibility of subsurface imaging and time-lapse monitoring using data acquired under flowing conditions. The results of this study suggest that, by adopting the inside coiled tubing method without drilling a new observation well, more affordable distributed acoustic sensing vertical seismic profile monitoring can be achieved in fields such as CO₂ capture and storage and unconventional shale projects, where monitoring costs have to be minimized.     

A seismic multi-approach method for characterizing contaminated sites

A geophysical campaign to characterize the subsurface of a contaminated site down to a depth of several tens of meters was carried out under the HYGEIA-CEE project. On this site, seismic techniques were combined to image the geological structures; i.e. seismic reflection, P-wave tomography and spectral analysis of surface waves. Because these techniques consider different wave components in the processing, they can be expected to provide complementary information concerning the site lithology. The special feature of this experiment is the fact that the same seismic acquisition device, consisting of a mobile central unit, a drop-weight seismic source, and a sensor line of gimbal mounted geophones, was used for each of the techniques. Two perpendicular seismic lines were set up in the field for testing two geophone spacings. Three processing procedures, one each for the seismic reflection, P-wave tomography and spectral analysis of surface waves, were developed for producing seismic images from the P-wave reflectivity, the first P-wave arrivals and the dispersion of Rayleigh waves, respectively. The images show good complementarity in terms of investigation depth. The results are also in good agreement with available borehole data: the sandy layers seem to be related to low velocities, since the high velocities are better explained by the presence of clayey and gravelly intervals. The contribution and the limits of this seismic multi-approach method is discussed.     

时窗能量特征法拾取微地震波初始到时的可行性研究

微地震（MS）波初始到时的自动拾取是MS监测数据处理的关键技术之一,也是实现MS震源自动定位的技术难点.本文在MS震源定位结果反演与推断的研究基础上,对不同类型MS波的到时点特征进行了分析与描述,并对不同时窗长度下能量特征值的变化规律进行了研究,提出了控制时窗移动范围和确定时窗长度自适应参数的具体方法,利用建立的MS波初始到时点特征的模式识别库,对拾取的到时进行模式归类、定量评价和匹配,提高了自动拾取结果的可靠性.研究结果表明,对典型的信噪比高的MS波,到时自动拾取的结果与手工拾取的结果基本一致;对无量纲大振幅的MS波,到时自动拾取结果的可靠性要高于手工拾取,对信噪比低和到时点不清晰的MS波自动拾取的可靠性较低.     

基于深度卷积神经网络的地震震相拾取方法研究

地震震相拾取是地震数据自动处理的首要环节,包括了信号检测、到时估计和震相识别等过程,震相拾取的准确性直接影响到后续事件关联处理的性能,影响观测报告的质量.为了提高震相拾取的准确性,进而提高观测报告质量,本文采用深度卷积神经网络方法来解决震相拾取问题,构建了多任务卷积神经网络模型,设计了分类和回归的联合损失函数,定义了基于加权的分类损失函数,以三分量地震台站的波形数据作为输入,同时实现对震相的检测识别和到时的精确估计.利用美国南加州地震台网的200万条震相和噪声数据对模型进行训练、验证和测试,对于测试集中直达波P、S震相识别的查全率达到98%以上,到时估计的标准偏差分别为0.067s,0.082s.利用迁移学习和数据增强,将模型用于对我国东北地区台网的6个台站13000条数据的训练、验证和测试中,对该数据集P、S震相查全率分别达到91.21%、85.65%.基于迁移训练后的模型,设计了用于连续数据的震相拾取方法,利用连续的地震数据对该算法进行了实际应用测试,并与国家数据中心和中国地震局的观测报告进行比对,该方法的震相检测识别率平均可达84.5%,验证了该方法在实际应用中的有效性.本文所提出的方法展示了深度神经网络在地震震相拾取中的优异性能,为地震震相和事件的检测识别提供了新的思路.     

Deep learning for P-wave arrival picking in earthquake early warning

Fast and accurate P-wave arrival picking significantly affects the performance of earthquake early warning(EEW)systems.Automated P-wave picking algorithms used in EEW have encountered problems of falsely picking up noise,missing P-waves and inaccurate P-wave arrival estimation.To address these issues,an automatic algorithm based on the convolution neural network(DPick)was developed,and trained with a moderate number of data sets of 17,717 accelerograms.Compared to the widely used approach of the short-term average/long-term average of signal characteristic function(STA/LTA),DPick is 1.6 times less likely to detect noise as a P-wave,and 76 times less likely to miss P-waves.In terms of estimating P-wave arrival time,when the detection task is completed within 1 s,DPick′s detection occurrence is 7.4 times that of STA/LTA in the 0.05 s error band,and 1.6 times when the error band is 0.10 s.This verified that the proposed method has the potential for wide applications in EEW.     

基于波形相似特征的微地震事件初至拾取及全局校正

微地震事件初至拾取是井下微地震监测数据处理的关键步骤之一.初至误差的存在会使微地震震源定位结果产生较大偏差,进而影响后续的压裂裂缝解释.通常初至拾取过程对所有的微地震事件选择相同的特征函数并采用一致的拾取参数进行统一处理,然而当事件的能量、震源机制、传播路径以及背景噪声等存在明显差异时,所得初至拾取结果差别显著.为了提高微地震事件初至拾取标准一致性,本文提出基于波形相似特征的初至拾取及全局校正方法.该方法首先利用互相关函数对每个事件内的各道记录进行时差校正,得到初始初至信息并形成叠加道,再对所有事件的叠加道进行全局互相关得到事件间初至相对校正量,最终初至结果可以通过各个事件的初始初至信息与其相对校正量相加得到.方法将所有微地震事件初至结果作为一个整体处理,从而能够克服常规方法初至拾取标准一致性差的缺陷.实际资料处理结果表明,相比于常规方法,该方法可以有效提高事件初至拾取和定位结果的一致性.     

基于加权K均值聚类的多属性初至拾取方法

为提高初至拾取方法的准确性和自适应能力,将变异系数加权K均值聚类算法引入初至拾取中。首先提取均方根振幅、相邻道相关性、线积分、振幅谱主频等多种地震属性;然后针对地震属性进行加权K均值聚类,自动识别初至所在时窗;最后结合相位校正法,实现时窗内初至波起跳时间的拾取。在此基础上通过实际数据测试,并与长短时窗能量比法、反向传播神经网络方法对比,验证了本文方法的有效性与可行性。结果表明,基于加权K均值聚类的多属性初至拾取方法能较快速、准确地拾取低信噪比数据的初至,并且无需人为判断时窗,从而提高了拾取的自适应能力。      

A real time processing system of seismic waves using personal computers — Function and characteristics

Conclusions The real-time processing system of CTSN performs following: A/D conversion; automatic event detection; event data saving; automatic measure of P and S arrivals; event location and print out the calculated results. It is corrected at ny moment by using the off-line system. Since December 1993, this system has been operating normally in the CTSN. More than 4 000 earthquakes have been recorded in the system. It has high accuracy in automatic picking P and S arrivals. The location of the earthquakes determined by on-line system are close to those given in published catalogues which are determined by manual procedure. This system can finish locate event in three minutes. It also gives satisfactory epicenter locations for distant events by inputting manually S arrivals in the off-line system. The operation of this system had brought the technical superiority of the CTSN. It not only reduces the labor intensity and simplifies the working procedure, but also makes our research facility into the superior ranks in this field of our country. In conclusion, the real-time processing system of seismic wave provides technical support for the daily requirements of monitoring seismic activity as well as a lot of digital waveform data used seismic research in Sichuan area. This subject is sponsored by the Scientific and Technical Committee of Sichuan Province.     

地震检测与震相自动拾取研究

针对微震事件易受噪声干扰等特点,本文将STA/LTA方法和基于方差的AIC方法（var-AIC）相结合,在震相到时初步拾取的基础上,使用台站的德洛内（Delaunay）三角剖分及台站间最大走时差约束来减少噪声干扰的影响. 利用到时进行地震定位之后,根据台站预测到时,在设定的时间窗内对地震震相进行更精细的分析. 特别是针对微震事件信噪比低的特点,设计了基于偏振分析的拾取函数,根据窗内STA/LTA方法和var-AIC方法的拾取结果自动选择合适的值作为震相到时. 最后,对西昌流动地震台阵2013年304个单事件波形数据的分析处理和检验结果表明,本文方法较传统方法具有更高的地震事件检测能力和更高的震相拾取精度.