研究强震孕震过程的几种有序集群方法

本文用五种有序集群方法研究了华北地区和宁夏地区强震的孕震过程。我们得到:五种方法均能模拟地震活动交替的高潮、低潮现象;Fisher方法和有序点群分析方法(采用离差平方和增量作为类间距离时)的效果均较好,表明在各类中诸时间段的结构主要是团状分布而不是链状分布,因此选择预报方法时考虑地震活动的这一特点是重要的;对分移动窗口法和非参数拟合优度法的结果,表明在相对平静期和显著活动期之间确实存在着客观的谷值;图论方法则有助于进一步分出显著活动期中易发生强震的较小峰值。五种方法中以Fisher方法的误识率最低,而非参数拟合优度法则能帮助确定较为合理的分类。此外对资料预先平滑也可改善结果。 

SEVERAL ORDER CLUSTERING METHODS FOR THE STUDY OF SEISMOGENIC PROCESS OF STRONG EARTHQUAKES

In this paper, the seismogenic process of strong earthquakes of North China and the Ningxia region is studied by using five order clustering methods. We found that all five methods can simulate the phenomena of alternatively high and low seimic activities; the better effects of both Fisher's method and order hierarchical clustering method (using incremental surn of squares of deviations as the between-class distance) show that the structures of time intervals in classes are mainly of cluster distribution and are not of chain distribution, thus it is important that this character of seismic activity should be considered in choosing the method of prediction. The results of Webster's method and nonpariametric goodness of fit show that there are centainly objective valle-yes between relatively queit periods and remarkably active periods. Then graph theory further helps us to distinguish the smialler peaks in which strong earthquakes are apt to occurr.The error of Fisher's method is smallest among these five methods, while noupara-metric goodness of fit can help make more reasonable classification. In addition, pre-srnoothing of data can also improve the results.