论文部分内容阅读
利用地震的空间位置分布,提出了一种可以重构断层网络活动部分的三维结构的模式识别新方法。该方法为所谓动态群聚(或称k均值)方法的推广,是利用震源位置关于它们的中心的全局方差最小化准则将数据点集划分为群集。这种新方法比原来的k均值方法更好,因为它为了把数据集划分为断层形状的各向异性群集而考虑了每个群集全空间位置的协方差张量。给定地震事件目录,就可以输出符合数据空间结构的最优集的平面段。每个平面段在它的位置、大小和取向上都有明显区别。主要的可调参数为地震定位精度,它固定了分辨率,即拟合的剩余方差。分辨率决定了描述地震目录所需要的断层段数目:分辨率越高,重构断层段的结构越精细。该算法成功地重构了合成地震目录的断层段。把该方法用于1992年6月28日发生在南加州兰德斯地震余震序列子集的真实地震目录,重构的断层面段与地质图上已知的断层或者长期目录中明显出现的盲断层完全吻合。讨论了该方法将来的进一步改进及其在断层带内结构的多尺度研究的潜在应用。
Based on the spatial distribution of earthquakes, a new pattern recognition method is proposed to reconstruct the three-dimensional structure of the active part of the fault network. This method is a generalization of the so-called dynamic clustering (or k-means) method, which divides data point sets into clusters using the global variance minimization criterion with respect to their center at source positions. This new method is better than the original k-means method because it takes into account the covariance tensor of the full space position of each cluster in order to divide the dataset into anisotropic anisotropic clusters. Given a catalog of seismic events, it is possible to output plane segments that conform to the optimal set of data space structures. Each plane segment has a significant difference in its position, size and orientation. The main adjustable parameter is the seismic location accuracy, which fixes the resolution, that is, the fitted residual variance. The resolution determines the number of fault segments required to describe the seismic catalog: the higher the resolution, the finer the structure of the reconstructed fault segment. The algorithm successfully reconstructs the fault segments of synthetic earthquake catalog. This method was applied to the real seismic catalog that occurred on June 28, 1992, in a subset of the aftershock sequence of the Lancaster Earthquake in Southern California, and the reconstructed fault segments and the known faults on the geological map or the blind The fault completely coincides. The further improvement of this method and its potential application in multiscale studies of fault-in-band structures are discussed.