海洋多要素长时间序列的motif规则树构建方法研究  

Research on construction method of motif rule tree for multi-element marine long time series

作　　者：赵丹枫[1] 黄雁玲 黄冬梅 林俊辰 宋巍 ZHAO Danfeng;HUANG Yanling;HUANG Dongmei;LIN Junchen;SONG Wei(Department of Information Technology,Shanghai Ocean University,Shanghai 201306,China;Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海海洋大学信息学院,上海201306 [2]上海电力大学,上海200090

出　　处：《海洋通报》2020年第5期558-566,共9页Marine Science Bulletin

基　　金：国家重点研发计划(2016YFC1401902)。

摘　　要：对海洋数据进行挖掘能够有效地预测海洋灾害事件。海洋监测数据具有时序长、间隔短、多要素间强关联的特点,对长时间序列进行直接分析挖掘速度慢、效率低,现有方法大多采用符号化时间序列方法,但可能导致部分信息丢失且破坏要素间的关联性。本文定义了时间序列motif,用于发现时间序列中重复出现的,先前未知的局部信息,解决了符号化导致的信息丢失的问题,实现了时间序列motif的精确快速提取。通过构建motif规则树,实现了海洋多要素时间序列间强关联规则的挖掘。最后,给出关联规则评价参数,同随机游走数据对比后,证明了本文方法的有效性。Data mining of marine data can effectively predict marine disaster events.Ocean monitoring data has the characteristics of long time series,short intervals,and strong correlation among multiple factors.Direct analysis and mining of longterm sequences will result in slow speed and low efficiency.Most of the existing methods use symbolic time series methods,but may cause some information Lost and disrupted relationships between features.This paper defines a time series motif,which is used to find repetitive and previously unknown local information in the time series,solves the problem of information loss caused by symbolization,and realizes accurate and fast extraction of time series motif.The mining of strong association rules between marine multi-element time series is realized.Finally,the evaluation parameters of association rules are given,and the comparison with random walk data proves the effectiveness of this method.

关 键 词：预测 海洋多要素时间序列 强关联规则 motif规则树 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]