Contamination Event Detection Method Based on the Longest Common Subsequence Analysis Using Multiple Water Quality Parameters

As the irreplaceable natural resource, water plays an important part in human life. It’s necessary to establish of a set of water quality early warning system which can quickly and accurately identify contamination. The existing anomaly detection methods are either simple fusion of abnormal judgment results from single index, or correlation analysis of multiple parameters at the cost of high time complexity. Aimed to solve these problems, this paper describes a multivariate correlation analysis method based on the longest common subsequence (LCSS) analysis combined with Dempster-Shafer (D-S) evidence theory. The proposed method utilizes adaptive sliding window scale to acquire water quality parameter time sequence, LCSS to describe similarity between different water quality series fluctuation, and D-S evidence theory to calculate abnormal probabilities. The ROC curve is introduced to verify the algorithm performance. Tested by data from the laboratory contaminant injection experiment, this approach shows nice ability in dealing with time series deformation problem rapidly, which to a certain extent can improve the reliability of multi-index correlation degree analysis, and ultimately reduce the abnormal water contamination event detection false alarm rate.