时间序列的区域周期模式及挖掘算法  

Mining Regional Periodic Patterns in Time Series Data

作　　者：郭静 陈欣 何杰[2,3] 谭志国 GUO Jing;CHEN Xin;HE Jie;TAN Zhi-guo(Department of Software and Computer,Chongqing Engineering Institute,Chongqing 400056,Chin;College of Computer,National University of Defense Technology,Changsha 410073,China;College of Electronic Science and Engineering,National University of Defense Technology,Changsha 410073,China;Department of Information and Communication,Officers College of PAP,Chengdu 610213,China)

机构地区：[1]重庆工程学院软件与计算机学院,重庆400056 [2]国防科技大学计算机学院,长沙410073 [3]武警警官学院信息通信系,成都610213 [4]国防科技大学电子科学与工程学院,长沙410073

出　　处：《小型微型计算机系统》2018年第10期2180-2185,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61170286)资助;重庆市基础科学与前沿技术项目(cstc2016jcyjA0539)资助

摘　　要：时间序列中周期行为的挖掘是众多领域研究的理论基础.针对时间序列中只在区域部分存在的周期行为,提出了一种新型周期模式概念,称之为区域周期模式.在对区域周期模式进行形式化描述的基础上,进一步提出了三种拥有不同求解目标的区域周期模式挖掘算法.算法1结合已有的部分周期模式挖掘算法和暴力迭代过程进行求解,虽能保证得到完整解,但由于过大的计算量,使其不具备实践应用价值;算法2基于类Apriori原则和三种剪枝策略,不仅能保证完整解,还具备较高的运算效率;算法3通过一阶区域周期模式的密集度推荐计算区域,大幅度地压缩了计算代价,为重要区域周期模式的快速挖掘提供了便捷方法.最后利用公开数据集测试和评估了3种算法的求解完整性和计算性能,验证了所提出算法的设计目标.Mining periodic patterns in time series data is the theoretical basis of many fields. We propose a novel concept about periodic pattern, called regional periodic pattern, which can describe the patterns that only activated in parts of a time series. After formalizing this regional periodic pattern, we further propose three regional periodic pattern mining algorithms for different goals. The first algorithm aims at finding all possible qualified patterns based on existing partial periodic patterns mining algorithm and brute-iterative method. Although it can guarantee the complete results, it does not have pragmatic value because of massive computation cost. The second algorithm is based on apriori-like principle and three pruning strategies. It can not only ensure the complete results, but also have high computational efficiency. The third algorithm can recommend computation ranges for mining process by using intensity of 1-patterns, which can much further reduce the computation cost for mining important regional periodic patterns in time series. At last, the result completeness and computing performance of three algorithms are tested and evaluated using a public dataset. Experimental result demonstrates that the goals of all three algorithms are satisfied.

关 键 词：时间序列 区域周期模式 部分周期模式 Apriori原则 模式挖掘 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]