禁核试核查地震数据处理系统中关键技术研究/Research of the Technologies in the Seismic Data Processing

2017-10-10 05:39:06

data 算法 method processing AIC



1996年9月全面禁止核试验条约(CTBT)在联合国正式开放供各国签署,为了有效监督条约的履行,CTBT条约制定了一套完整的核查体系,主要由国际监测系统IMS、全球通讯基础设施GCI、现场视察OSI和位于维也纳的国际数据中心IDC组成,能够对全球的违约活动进行实时监测。作为CTBT条约的签约国,我国在履行条约义务的同时,必须开展禁核试核查技术的研究,这对于保护我国的国家利益不受侵犯具有重要意义。
本论文重点研究了禁核试核查数据处理系统中的关键技术。研究工作结合国家数据中心测试平台的建设,分析了IDC的运行模式和数据处理技术,解决了连续数据接收系统CDS调试中的关键技术,对IDC地震数据自动处理中的主要算法进行了研究。在事件识别中,分析了微差爆炸信号识别的时频分析方法。
主要研究成果如下:
(1)IMS监测台站按照连续数据传输协议CD1.0发送数据,数据中心的连续数据接收系统CDS负责接收和转发数据。在没有实际数据来源的情况下,为了调试CDS系统,设计了一套IMS地震台站数据模拟发送系统,分析并掌握了连续数据传输协议CD1.0,开发了读取CD1.0格式地震数据的解压缩算法,能够按照实际IMS地震台站的运行模式,产生并发送CD1.0格式的连续地震数据,为CDS系统的测试提供了可靠的保障。
(2)数据质量检测是地震数据自动处理中的第一步,影响数据质量的问题数据主要有毛刺、限幅和数据丢失。数据质量检测的目的是能够在自动处理的开始阶段检测到问题数据,通过及时处理,防止后续处理出现错误解释。问题数据的检测中,毛刺的检测要复杂些,IDC采用的幅值比较法只能检测幅值较大的毛刺,很容易漏检幅值较小的毛刺。针对这一问题,基于平稳小波变换和非线性能量检测算法,提出一种新的毛刺检测算法,可以准确地检测记录中不同幅值的毛刺。
(3)STA/LTA算法是信号自动检测中的经典算法,这种算法中检测阈值的分布范围在 之间,合适的检测阈值不但要根据实验反复调试,而且要在误检率和漏检率之间取得平衡。针对这一问题,从模式识别的角度出发,提出一种基于支持向量机的信号检测算法。讨论了新算法中数据预处理和模式特征提取的方法,以及支持向量机中核函数的选择问题。利用实际地震数据,分析了这种算法的检测性能,结果表明这种算法简化了检测阈值的选择,在准确检测信号的同时,其误检率相对于STA/LTA算法可以降低约85%,并且具有较强的抗噪性能。
(4)IDC使用三个不同的BP神经网络来区分三分量地震记录中的震相,实际应用中,不同监测台站的神经网络要单独进行训练,增加了数据处理的工作量,并且不利于实时处理。针对三分量地震记录的特点,基于小波变换和偏振分析理论,提出一种多尺度多参数的震相识别算法,能够在自动处理中简单而有效地判别震相类型。
(5)AIC算法是信号初至估计的主要方法,这种算法的不足是计算结果受记录信噪比的影响较大,并且后续震相的初至估计误差较大。针对这种方法的不足,提出两种改进的算法,即:基于AIC算法和信号幅值比的混合方法,以及基于广义分维的地震信号初至估计算法。实验表明,这种基于AIC算法和信号幅值比的混合方法的信号初至估计结果和三分量AIC算法的基本一致。通过对中国数字地震台网乌鲁木齐台记录到的23次天然地震中P波、S波和Lg波进行初至估计,与人工分析结果相比,P波初至估计误差的均方误差为0.71秒,后续震相(S波、Lg波)初至估计误差的均方误差为1.64秒,优于传统AIC算法的估计结果。
考虑到AIC算法的一些固有缺陷,如频率滤波造成的相移以及建立AR模型带来的计算复杂性,提出一种基于广义分维思想的地震信号初至估计算法,根据Mandelbrot-Richardson曲线截距的变化,可以准确检测到地震信号的初至。对于大批量的数据处理,这种算法在节省计算资源方面具有一定的优势。
(6)基于短时傅立叶变换的时频分析是识别微差爆炸信号的主要方法,由于这种方法的时频分辨率很低,有时无法正确地反映信号的特征。提出一种基于AR模型的时频分析方法,在保留短时傅立叶变换优点的同时,具有非常高的时频分辨率。从理论上分析了实信号的离散WVD会产生频域混叠和零频处的交叉干扰,这种基于AR模型的时频分析方法不存在上述问题。处理的数据量较大时,新方法在提高计算速度、节省计算存储量和保持较高的时频分辨率方面具有明显的优势。实际数据的处理表明,这种基于AR模型的时频分析方法能够更加清楚地反映出微差爆炸信号的时频特征。


The Comprehensive Nuclear-Test-Ban Treaty (CTBT) was opened for signature at United Nations Headquarters in September 1996. To verify compliance, a verification regime has been established consisting of International monitoring system, Global Communication Infrastructure, On-site Inspection and International Data Center, et al. In order to enhance our verification abilities and protect our privileges, the verification technology must be studied before the Treaty entering into force.
Some important technologies in data processing system for nuclear test ban verification were studied in the dissertation. Coopertated with the construct of test system of national data center, the study work analyzed the operation mode and data processing methods, resolved some important problems for testing continuous data subsystem CDS, analyzed these main algorithms of IDC automatic data processing. In the study of event identification the method of time-frequency analyse for discriminating ripple fired explosions were analyzed.
The main contribution of this dissertation is summarized as follows.
(1) The monitoring data of IMS stations are transmitted based on protocol CD1.0, and data
are received and forwarded by continuous data subsystem CDS in data processing center. In order to testing CDS without real monitoring data, a set of data transmitting system simulating IMS seismic stations were designed, a kind of data uncompressing algorithm based on protocol CD1.0 was developed, which could read and transmit continuous seismic data in format of CD1.0.
(2) In automatic processing the first step is data quality check, which be used to find or
repair problem data that may lead to wrong results for rest processing. Spikes, repeated amplitude values and dropouts are considered to be problem data, in which the detection of spikes is more complex. Amplitude comparison is the main method used in IDC to detect spikes, the drawback of the method is that small spikes could not be detected efficiently. Therefore, A novel algorithm based on stationary wavelet transform and nonlinear energy operator was proposed, which could detected spikes in various amplitude accurately.
(3) STA/LTA is a traditional method in automatic signal processing, in which the threshold is
range from 0 to , the optimal threshold must be experiment with lots of data to get a balance between false detection ratio and miss detection ratio. In order to resolve the deficiency, a novel algorithm based on support vector machine was proposed, in addition, the way of preprocessing and feature extracting, the selection of kernels function for support vector machine were also discussed. Experiments with real records showed that the novel algorithm could detect signal accurately with lower false alarm even in the records of lower signal to noise ratio.
(4) Three different BP neural networks are used in IDC to identify seismic phases in three
component seismograms, which increase computation load of data processing in real applications due to each seimic station must use different neural networks. Based on wavelet transform and polarization analyse, a novel algorithm of multiple scales and multiple parameters for phase identification was proposed, which could identify seismic phases efficiently in automatic processing.
(5) The algorithm AIC is the main method for estimating first break of seismic signals, but
the estimation results is too sensitive for signal-to-noise of seismograms, especially for onset estimation of later signals. Two modified algorithms, i.e. a hybrid method based on AIC and signal amplitude variety, an arrival time estimation based on generalized fractal, were proposed. Experiments showed that performance of the first modified method was same as that of AIC algorithm in three component records. The arrival times for 23 seismic P-phases and later phases recorded at WMQ station in CDSN are picked by this new method. In comparison of manual results, the root mean square error of P-phase is 0.71 second, and the root mean square error of later phase is 1.64 second, which is better than the result obtained by traditional AIC algorithm.
While considering some inherent deficiencies of AIC algorithm, such as phase shift for frequency band filter and computation complexity for construction of AR model, the second modified method are present. The change of intercept of Mandelbrot-Richardson curve can be used to estimate arrival time of seismic signal accurately. The second method may save computation resources for real data processing.
(6) STFT spectrogram is the main time-frequency analysis method used in identification of
ripple fired explosions. However, for the time-frequency resolution of STFT Spectrogram is very poor, some important features of ripple fired explosions can not be extracted directly from STFT spectrogram. A novel method of time-frequency analysis based on Auto-regressive model (AR) is presented, which inherits merits of STFT Spectrogram and has very good time-frequency resolution. In this paper we analyzed the reasons why the discrete Wigner-Ville-Distribution (WVD) of real-valued signal sampled at the Nyquist rate has spectral aliasing, whereas the new method has no such problems. When data for processing are very large, the new method may has excellent performance for promoting velocity of calculating, saving storage and keeping high time-frequency resolution. In addition, the applications of the new method were also illustrated for identifying ripple-fired explosions.