基于核密度估计的时间序列多模式发现  

Multi-patterns discovering for time series based on kernel density estimation

作　　者：陈晓云[1] 

机构地区：[1]福州大学数学与计算机科学学院,福建福州350108

出　　处：《山东大学学报（工学版）》2011年第5期46-52,共7页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(61070020);福建省新世纪优秀人才支持计划资助项目(XSJRC200711)

摘　　要：现有时间序列频繁模式或异常模式发现算法均以发现二者之一为目的,且多数采用硬距离阈值策略。为达到同时发现频繁模式和异常模式的目的,提出时间序列集频繁模式和异常模式发现算法FAP。FAP算法采用高斯核密度作为模式的支持度量,使用最小密度熵方法选择高斯核函数的带宽参数,进而避免了硬距离阈值策略的不足,实现了时间序列多模式发现。因为现实中的时间序列数据集规模较大且常含有噪声,所以FAP采用Haar小波变换压缩原始时间序列并过滤噪声。用来自UCR的数据集测试FAP算法,实验数据集的频繁模式与异常模式均能被正确发现。Existing discovery algorithms of frequent patterns or anomaly patterns can only find one of the two patterns from time series data,and most of them use the hard-distance threshold strategy.The FAP algorithm for mining frequent patterns and anomaly patterns at the same time was presented.The FAP algorithm adopted Gaussian kernel density as the support measure of the pattern,which uses a minimum density entropy approach to select a bandwidth parameter of the Gaussian kernel density function and avoids the defect of the hard-distance threshold strategy,so as to achieve multi-patterns discovering for time series.Because real time series datasets are often large or have noise,Haar wavelet transform was applied to compress the original time series and filter the noise.The FAP algorithm was tested on the time series from UCR data set,and the experimental results showed that the frequent patterns and anomaly patterns could be correctly found from these time series.

关 键 词：时间序列 频繁模式 异常模式 核密度估计 熵 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]