Continuous Summarization Of Co-Evolving Data In Large Water Distribution Network

While traditional water data stream analysis focuses mainly on single sensor node or monitoring station, having an accurate picture of the overall data patterns is more meaningful in understanding large water distribution network's behavior and characteristics, tracking important trends, and also making informed judgments about measurement or utilization operations. In this paper, we propose a continuous summarization scheme that aims to continuously provide Representative Patterns of the complete data in large water distribution network. Our core contributions are to propose to summarize Representative Pattern for describing the spatial-temporal pattern in water distribution network and employ a parameter-free algorithm based on the Minimum Description Length (MDL) Principle to automatically split data streams into episodes for generating the series of representative patterns. Moreover, we evaluate our approaches on a real water distribution network from the Battle of the Water Sensor Network (BWSN). Experiment results show that our online summarization methods are effective, scalable and interpretable; What's more, we discover interesting periodic time-evolving patterns on the chlorine data.