Burst Detection by Analyzing Shape Similarity of Time Series Subsequences in District Metering Areas

Abstract

The paper proposes a burst detection method that relies on shape similarity analysis of time series subsequences (i.e., slices of time series). Subsequence libraries are constructed using flow (or water demand) data. Increase-rate distance is used to evaluate the shape similarity between subsequences, and abnormal subsequences are those that have low shape similarity with others. An abnormal subsequence searching algorithm first is used to remove abnormal subsequences, and the remaining subsequences are used to form reference libraries. Then the shape similarity between newly collected subsequences and reference libraries is evaluated to detect bursts. In the detection, a modified version of the abnormal subsequence searching algorithm can reduce the number of false alarms by finding the don’t-care segment in subsequences and improve the method’s detection ability by crossover between night subsequences. The method was applied to a network’s hydraulic model and three real-life district metering areas. Results show that the method’s detection performance is only slightly affected by seasonal changes of data and is insensitive to data sets from different networks.