基于地震波形特性的震源类型识别研究

Seismic wave is the unique and reliable mean for investigating the interior of the Earth, carrying with it plentiful and intricate information of Earth interior about source type and wave propagation path.Source types mainly mean the origins of seismic events: natural event or explosion event. There are two approaches for earthequake/explosion recognition: geophysical approach and signal processing then pattern recognition approach. For small magnitude event recognition, criteria based upon the differences of geophysical measures have irrational performance for generalization. However,signal processing then pattern recognition methods are more promising, and nowdays are the most active research topics. Basing upon the preparatory results of our current researches, the project of this proposal prepares further expanding and refining algorithms for seismic wave feature extraction, focusing on: exploring appropriate time-frequency techniques for feature extraction and intensively investigating the characteristics of primary several seconds of measured seismic wave signal; improving pattern recognition algorithms which suitable to small sample-size, multiple and non-consistent features and complex distribution structure sample set of acquied features; asserting and confirming the reliability and applied conditions of various features; affirming the impact of wave time windows length on recognition rate of different source type; formulating the quantitative relationships among P-wave group, S-wave group and surface wave group of different source type, especially destructive strong earthquakes. This investigation on event source type recognition, will be helpful for early warning and monitoring of strong and destructive earthquake, and also beneficial for the reinforcement and reconnaissance of nuclear test, especially the small magnitude underground nuclear explosion.

地震波是人类认识地球内部结构的可靠载体, 同时携带着震源、波传播路径等地球内部信息。震源类型，主要指产生地震的原因:天然事件或人工爆炸事件。地震爆炸识别有两种途径：地球物理途径和信号处理及模式识别途径。对于小当量事件的识别，依赖于地球物理特征差异性的判据推广性能不佳，而信号处理及统计模式识别途径是当前最活跃的研究方向。本项目拟在已开展研究中取得的初步结果基础上，进一步研究地震波形特征提取方法，着重研究地震波形时频分析特征提取方法及破坏性大地震最初几秒波形的特性；改进现有的模式识别算法，使之适合于小样本、多特征、非一致且样本结构分布复杂的地震波形特征样本集；探讨各不同特征的可靠性和适用条件；研究波形时间窗大小对震源类型识别率的影响；研究不同震源类型，尤其是破坏性强震的P波群、S波群与面波群之间的定量关系。震源类型识别研究，有助于破坏性构造地震的监测预警，有助于小当量核试侦察。

地震波是监测遥远地区地下核爆炸试验的唯一手段，是维护国家利益，保卫世界和平的重要手段。本项目研究主要目的是由地震波识别区分天然地震和人工爆炸，以及尝试探讨由先期到达的部分波形尽早探知当前事件震源是否为破坏性强地震。震源类型识别判据已有P波初动方向、震源深度、体波震级与面波震级比、是否存在特征面波Lg等，波形特征通常取地震波频率的高低、小波系数、转角频率等。本项目首先研究了谱比值、波形复杂度、综合比值和自相关系数4个特征，采用非重叠学习集和检验集样本划分方法，这4个特征的正确识别率可以达90%。还研究了以S波最大衰减率、S波平均衰减率和拐角频率这3种特征，正确识别率92%。接着研究了特征美尔倒谱系数（MFCC）特征、线性预测倒谱系数（LPCC）特征和基于希尔伯特-黄变换（HHT）的特征，对于单源事件且干净清晰的地震波形，MFCC和LPCC均比HHT识别效果好（均取12维特征，采用高斯混合模型，正确识别率分别可达95%，97.5%，90%）。一个事件通常有100个以上的观测点，大多数观测点测得的地震波形信噪比往往不高，这些含噪波形或多源观测波形的MFCC、LPCC和HHT的识别效果均不理想（正确识别率小于80%）。然后研究了采用经验模态分解(EMD)分离非平稳、非线性地震信号为多个内模函数(IMF), 从分解后的前3个IMF提取出3个特征--自相关函数均值、倒谱方差、谱幅比特征，此3个特征的正确识别率稳定在97%。研究了利用分形维自动标定P波初至时刻、P波最大值时刻、S波最大值时刻和有效地震波形终止时刻，由此形成地震S波衰减率和地震频谱形态特征，结合S波衰减率和地震频谱形态特征的正确识别率对实验数据而言达100%。最后，还研究探讨了：奇异性特征（正确识别率87%），奇异性与地震震级之间的关系，对称点模式（SDP）图阵和基于SDP图阵的变异系数特征（正确识别率可达92%）等。