摘要
投影寻踪聚类(PPC)是投影寻踪和非分层聚类方法的综合,它同时尝试分类数据和寻找聚类结构的低维有代表性的特征。对于本项研究,RichardJ.BoltonRichardJ.B提出一个基于正交标准变量(考虑样本的大小)的投影寻踪指数和判断投影寻踪聚类的最优类别数的方法对新疆地区地震活动作聚类分析,模拟聚类结构。找出不同区域的有效的参量组合结构,判断后续聚类结构,对未来地震大小做出预测。通过对由不同的时间窗长和滑动步长的地震活动性自变量的PP聚类分析,发现短期预测效果不好,而时间窗长分别为12个月和18个月,相应的步长为6个月和12个月的PP聚类效果较好:柯坪地区、乌恰地区的投影寻踪聚类分析的内符检验对应率达60%以上。通过对比分析,认为取时间窗长为18个月的结果更好些。这与自变量的选取有关——地震活动性参量主要是反映中期震兆信息。
Projection pursuit cluster (PPC) is the colligation of projection pursuit and unstratified cluster method. In the time, it was attempted to stratify the datum and search low-dimensional representative characteristics of clustering structure. Richard J. Bolton Richard J.B had advanced the method of projection pursuit index and the best sort number of projection pursuit cluster based on the normal standard variable, and author use it to analyze the earthquake activities of Xinjiang region clusteringly, simulate the clustering structure. Found the best parameters' combination structure of different regions, determined the latter clustering structure, and predicted the future earthquakes magnitude. Author analyzed the independent variables PP cluster of different time window length and moving step length, found that the short term prediction efficacy isn't good, when the time window length is 12 months and 18 months, the step length is 6 months and 12 months, the projection pursuit cluster efficacy is good; the corresponding ration of protest tests of the projection pursuit clustering analysis beyond 60% in Keping region and Wuqia region. Author thought the effect is good when the time window length is 18 months by contrasting analysis, it have relation with the independent variables— earthquake activities' parameters reflect the medium term earthquake precursory information mainly.
出处
《内陆地震》
2005年第1期8-15,共8页
Inland Earthquake
基金
新疆地震科学基金资助课题
199905.
关键词
投影寻踪
聚类分析
投影寻踪聚类
新疆地区
Projection pursuit
Clustering analysis
Projection pursuit cluster
Xinjiang region
