Online Detecting Spreading Events with the Spatio-temporal Relationship in Water Distribution Networks.

In a water distribution network, massive streams come from multiple sensors concurrently. In this paper, we focus on detecting abnormal events spreading among streams in real time. The event is defined as a combination of multiple outliers caused by one same mechanism and once it breaks out, it will spread out in networks. Detecting these spreading events timely is an important and urgent problem both in research community and for public health. To the best of our knowledge, few methods for discovering abnormal spreading events in networks are proposed. In this paper, we propose an online method based on the spatial and temporal relationship among the streams. Firstly we utilize Bayesian Network to model the spatial relationship among the streams, and a succinct data structure to model the temporal relationship within a stream. Then we select some nodes as seeds to monitor and avoid monitoring all sensor streams, thus improving the response speed during detection. The effectiveness and strength of our method is validated by experiments on a real water distribution network. © Springer-Verlag 2013.