一种优化的自底向上时间序列分段算法  

An Optimized Bottom-Up Time Series Segmentation Algorithm

作　　者：孙焕良[1] 邱邦华[1] 魏溯华[2] 

机构地区：[1]沈阳建筑大学信息与控制工程学院,辽宁沈阳110168 [2]沈阳建筑大学科技产业处,辽宁沈阳110168

出　　处：《沈阳建筑大学学报（自然科学版）》2007年第6期1049-1052,共4页Journal of Shenyang Jianzhu University：Natural Science

基　　金：辽宁省自然科学基金(20052006);辽宁省教育厅攻关计划(05L354)

摘　　要：目的正确衡量分段算法的优劣,提高自底向上算法的分段精度.方法分析现有分段评价标准存在的不足,综合考虑压缩比和精度,提出相同压缩比下的拟合总误差越小算法相对更优的分段评价标准.通过去除原自底向上算法初始分段两两连接的偶数限制,提出新的自底向上算法.结果测试显示新的评价标准能有效避免错误评判.新的自底向上算法的拟合总误差比现有算法减少了一半以上.结论新的评价标准可以更准确地区分算法的优劣,比现有标准更合理.新的自底向上算法具有更高的精度,整体优于原算法.Segmentation is the basis of the time series data mining, its advantages directly influence the analysis＇ result. In order to evaluate segmentation algorithm＇s quality correctly, we proposed a new segmentation evaluation criterion that an algorithm which has the smaller sum of error under the same compression is better. The new criterion considered both of the compression and the fidelity. We also introduce a novel optimized Bottom-Up time series segmentation algorithm. The new algorithm removes the classic Bottom-Up algorithm＇s even limitation. Experiments on the real-life datasets show that the new criterion is more reasonable than the existing criteria and the novel optimized Bottom- Up algorithm is superior to the classic one.

关 键 词：时间序列 数据挖掘 分段评价标准 自底向上 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]