摘要
离群检测的目的在于找出隐含在海量数据中相对稀疏而孤立的异常数据模式。由于高维数据的特殊性,传统的离群挖掘算法往往不适合发掘高维空间中的离群点。本文将蚁群算法用于改进超图模型,提出了一种新的离群检测算法——AHHDOD算法,在检测出离群数据模式的同时给出离群点的归属。经检验,该算法能有效收敛于最优解,且其时间复杂度更低。最后,将该方法应用于矿难预警检测中,能对可能出现的危机状况给出预警提示。实验证明,该方法取得的预警结果是可信的和可接受的。
The aim of outlier detection is to find out abnormal data patterns concealed in abundant data sets which were sparse and isolate. Mine disaster occurred much more frequently in our country, so it is urgent to take out an effective method to prevent mine disasters and guarantee miner's life and property. In this paper, we presente a new method - AHHDOD, it could not only find out the abnormal data patterns, but also can give the attribution of them. Finally, this method was put into use in the mine disaster forewarning system. The results proved that its convergence and complexity is better than other existed algorithms, and its forewarning result is credible and acceptable.
出处
《系统工程》
CSCD
北大核心
2008年第11期116-122,共7页
Systems Engineering
基金
国家自然科学基金资助项目(90510010)
教育部博士点基金资助项目(20050287026)
关键词
蚁群算法
超图
高维数据
离群检测
矿难预警
Ant Colony Algorithm
Hypergraph
High Dimensional Data
Outlier Detection
Mine Disaster Forewarning