Fast segmentation algorithms for long hydrometeorological time series

Aksoy H., Gedikli A., Ünal N. E., Kehagias A.

HYDROLOGICAL PROCESSES, vol.22, no.23, pp.4600-4608, 2008 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 22 Issue: 23
  • Publication Date: 2008
  • Doi Number: 10.1002/hyp.7064
  • Journal Name: HYDROLOGICAL PROCESSES
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus
  • Page Numbers: pp.4600-4608
  • Istanbul Technical University Affiliated: Yes

Abstract

A time series with natural or artificially created inhomogeneities can be segmented into parts with different statistical characteristics. In this study, three algorithms are presented for time series segmentation; the first is based on dynamic programming and the second and the third—the latter being an improved version of the former—are based on the branch-and-bound approach. The algorithms divide the time series into segments using the first order statistical moment (average). Tested on real world time series of several hundred or even over a thousand terms the algorithms perform segmentation satisfactorily and fast.

A time series with natural or artificially created inhomogeneities can be segmented into parts with different statistical characteristics. In this study, three algorithms are presented for time series segmentation; the first is based on dynamic programming and the second and the third-the latter being an improved version of the former-are based on the branch-and-bound approach. The algorithms divide the time series into segments using the first order statistical moment (average). Tested on real world time series of several hundred or even over a thousand terms the algorithms performs segmentations satisfactorily and fast. (c) Copyright 2008 John Wiley & Sons, Ltd.