Systems and methods for structural clustering of time sequences

• US 7,369,961 B2
• Filed: 03/31/2005
• Issued: 05/06/2008
• Est. Priority Date: 03/31/2005
• Status: Expired due to Fees

• Current Assignee: GlobalFoundries, Inc.
• Original Assignee: International Business Machines Corporation
• Inventors: Yu, Philip S., Castelli, Vittorio, Vlachos, Michail
• Primary Examiner(s): Barlow, Jr.; John E.
• Assistant Examiner(s): KUNDU, SUJOY K
• Application Number:
  US11/096,485
• Publication Number:
  US 20060224356A1
• Time in Patent Office: 1,132 Days
• Field of Search: 702/176, 702/182, 702189-190, 702/167, 707 6- 7, 707/10, 707/200, 707/202
• US Class Current: 702/176
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 2218/08 Feature extraction

Abstract

Arrangements and methods for performing structural clustering between different time series. Time series data relating to a plurality of time series is accepted, structural features relating to the time series data are ascertained, and at least one distance between different time series via employing the structural features is determined. The different time series may be partitioned into clusters based on the at least one distance, and/or the k closest matches to a given time series query based on the at least one distance may be returned.

Claims

1. A method of performing structural clustering between different time series, said method comprising the steps of:

accepting time series data relating to a plurality of time series;

ascertaining structural features relating to the time series data;

determining at least one distance between different time series via employing the structural features; and

partitioning the different time series into time-invariant clusters containing at least one of the time series based on the at least one distance;

wherein the clusters are stored in a computer memory;

wherein said ascertaining step comprises;

computing all structural features; and

automatically selecting a number of most relevant features; and

wherein said step of automatically selecting a number of most relevant features comprises;

selecting a threshold; and

retaining features having value larger than the threshold; and

said step of selecting a threshold comprises selecting a threshold which serves to discard features having values attributable to statistical variations via;

computing a resampling estimate of the distribution of feature values attributable to statistical variations;

selecting a value of probability of type 1 error; and

selecting as a threshold a value that guarantees the selected value of probability of type 1 error for a distribution equal to the resampling estimate of the distribution.

2. The method according to claim 1, further comprising the step of determining common periodicities corresponding to each of the clusters.

3. The method according to claim 1, further comprising the step of predetermining a number of structural features to compute.

4. The method according to claim 1, wherein:

said ascertaining step comprises ascertaining frequency content relating to the time series data; and

said ascertaining step further comprises implementing a Discrete Fourier Transform.

5. The method according to claim 1, wherein:

said ascertaining step comprises determining an orthogonal transformation relating to the time series data;

the orthogonal transformation of the data comprising a Discrete Wavelet Transform.

6. The method according to claim 1, further comprising the steps of:

selecting at least one of said structural features; and

said determining is performed via employing the at least one structural feature selected.

7. The method according to claim 6, wherein said selecting step is performed by a user.

8. The method according to claim 6, wherein:

said selecting step is performed automatically; and

said selecting step comprises;

identifying candidate structural features; and

verifying the candidate structural features.

9. The method according to claim 8, wherein:

said identifying step comprises;

computing a periodogram of the different time series; and

identifying peaks of the periodogram; and

said verifying step comprises;

computing an autocorrelation; and

selecting identified peaks of the periodogram that lie on hills of the autocorrelation.

10. The method according to claim 1, wherein said ascertaining step comprises:

computing all structural features; and

automatically selecting a number of most relevant features.

11. The method according to claim 10, wherein said step of automatically selecting a number of most relevant features comprises:

selecting a parameter k corresponding to a number of structural features to keep; and

retaining the k features that contain the highest amount of periodic content.