基于多元索引后继树的时间序列数据流挖掘

本文以地电前兆观测数据流为研究对象,提出了一种新颖的基于多元索引后继树的时间序列数据流挖掘模型。该模型的主要特点是使用一种新颖的位置索引方法,通过一遍扫描创建描述一个序列的所有元素的绝对位置的多元索引后继树,使用索引匹配方法以模式增长的方式生成频繁模式。理论分析与实验表明,该方法简单、高效,具有很好的实用价值。同时,通过对大量持续的地电前兆观测数据流进行分析处理,挖掘隐藏在其中的反映地电参数正常变化规律及趋势的频繁模式,为发现地电前兆异常识别提供依据,从而辅助进行地震预报。

Time-Series Data Stream Mining Based on the Multi-Index Successive Tree

Studying the geoelectric precursory data streams,a new time-series data stream data model based on the Sequence Multi-Index Successive Trees(SMIST) is put forward,which uses index to create SMIST through scanning the sequence by one time,and then uses index matching and pattern growth to generate frequent patterns.The results in both theory analyses and experiments show that this algorithm is so simple and direct,efficient,and has a practical value.In order to discover the hidden regularity of the geoelectric knowledge,a large number of geoelectric precursory data streams are processed and analyzed.The newly-found regular patterns and the trend of geoelectric parameters can provide a basis for detecting abnormal precursors,which can be used in earthquake prediction.