Download presentation
Presentation is loading. Please wait.
Published byTracy Cummings Modified over 8 years ago
1
1 Novel Online Methods for Time Series Segmentation Xiaoyan Liu, Member, IEEE Computer Society, Zhenjiang Lin, andHuaiqing Wang IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. 20, NO. 12, DECEMBER 2008 Adviser ︰ Yu-Chiang Li Speaker ︰ Yu-Hui Wu Date ︰ 2009/12/03
2
2 Outline Introduction Related Work Novel Online Segmentation Algorithms ︰ FSW and SFSW Complexity Analysis Experiments Conclusions
3
3 Introduction Time series have increasing importance Finance Medicine Manufacturing Approximate representation of the time series is one of the solutions Discrete Fourier Transform Discrete Wavelet Transform Singular Value Decomposition Piecewise Aggregate Approximation Piecewise Linear Approximation (PLA)
4
4 Introduction Piecewise Linear Approximation (PLA) Widely used Simplicity Segmentation problem Representation quality Computing efficiency
5
5 Related Work Segmentation problem can be stated in the following form:
6
6 Related Work Segmentation methods Online Offline complete sequences Top-down algorithms unsegmented sequence introduce one cutting point
7
7 Related Work Bottom-up algorithms n-1 segments Each point is a segmenting point SW(sliding windows) algorithms Online method and low computation complexity Lacking a global view
8
8 Novel Online Segmentation Algorithms Reduces the time complexity Maximum vertical distance (MVD) FSW(Feasible space window) SFSW(Stepwise feasible space window)
9
9 Novel Online Segmentation Algorithm Maximum Vertical Distance and Segmentation Criterion
10
10 Novel Online Segmentation Algorithm SW method can be reduced by using this segmentation criterion
11
11 Novel Online Segmentation Algorithm FSW method Candidate Segmenting Point (CSP) Serch for the farthest CSP
12
12 Novel Online Segmentation Algorithm
13
13 Novel Online Segmentation Algorithm SFSW method SW method lacking an overall view
14
14 Novel Online Segmentation Algorithm Intuitions behind the stepwise strategy
15
15 Novel Online Segmentation Algorithm Stepwise segmenting process
16
16 Complexity Analysis The worst case of FSW
17
17 Complexity Analysis SWAB Bottom-Up and SW SWAB is a small constant factor SW Time to compute a single segment Time complexity or
18
18 Complexity Analysis FSW Finding one segmenting point is K segments in time series T SFSW Time complexity of produce segment Time complexity of SFSW
19
19 Complexity Analysis
20
20 Experiments
21
21 Experiments
22
22 Experiments
23
23 Conclusions Successfully reduces the computational complexity of the classic SW method from to FSW and SFSW methods always generate far fewer segments than the other two methods
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.