A novel water quality data analysis framework based on time-series data mining. (1st July 2017)

• Record Type: Journal Article
• Authors: Deng, Weihui; Wang, Guoyin
• Is Part Of: Journal of environmental management. Volume 196(2017)
• Page Start: 365
• Page End: 375
• Publication Date: 2017-07-01
• Subjects: Time-series data mining -- Cloud model -- Water quality analysis -- Similarity measure -- Anomaly detection -- Pattern discovery; Environmental policy -- Periodicals; Environmental management -- Periodicals; Environment -- Periodicals; Ecology -- Periodicals; 363.705
• DOI: 10.1016/j.jenvman.2017.03.024
• Languages: English
• ISSNs: 0301-4797

Abstract

Abstract: The rapid development of time-series data mining provides an emerging method for water resource management research. In this paper, based on the time-series data mining methodology, we propose a novel and general analysis framework for water quality time-series data. It consists of two parts: implementation components and common tasks of time-series data mining in water quality data. In the first part, we propose to granulate the time series into several two-dimensional normal clouds and calculate the similarities in the granulated level. On the basis of the similarity matrix, the similarity search, anomaly detection, and pattern discovery tasks in the water quality time-series instance dataset can be easily implemented in the second part. We present a case study of this analysis framework on weekly Dissolve Oxygen time-series data collected from five monitoring stations on the upper reaches of Yangtze River, China. It discovered the relationship of water quality in the mainstream and tributary as well as the main changing patterns of DO. The experimental results show that the proposed analysis framework is a feasible and efficient method to mine the hidden and valuable knowledge from water quality historical time-series data. Highlights: A novel general water quality data analysis framework based on TSDM is proposed. The framework consists of implementation components and common tasks. Two-dimensional normal cloud is adopted to represent the time series. Similarity search, anomaly detection, and pattern discovery are introduced. A case study of dissolved oxygen time series analysis is presented.