I. Introduction
Time series streams have attracted many researchers' interest recently, due to the wide applications, such as sensor network monitoring, electrocardiogram detection, online stock analysis. Compared with the traditional achieved data, time series streams have their own special characteristics: (1) Data element's arrival is on-line. (2) System can not control the arrival order. (3) Time series stream are potentially unbounded in size. (4) Once an element has been processed, it is discarded and can not be retrieved[1]. In order to achieve real-time response, one-pass scan and low time complexity are usually required. In data streams, we usually need monitor the interested patterns so as to make correct decisions. For example, in ECG detection, given some predefined patterns such as Normal or Ischemia patterns, ECG tester can get a set of realtime patients' data, the doctor can make correct decision according to which patterns the ECG data are similar to.