Machine Learning Approach for Contamination Source Identification in Water Distribution Systems

Contamination source identification involves the characterization of the contamination event attributes using threat observations such as sensor network measurements. The defining attributes of a contamination event may include contaminant type, site(s) of contaminant intrusion, contaminant amount, the time of day the contamination event is initiated, and the intrusion duration. Accurate and prompt determination of these attributes is central to validity of impact assessments conducted and effectiveness of response strategies taken. Focusing on high accuracy, past efforts have successfully applied optimization and back-tracking approaches to deal with this critical task. However, these techniques are typically computationally burdensome and thus not acceptably fast, specifically when they are applied to realistically large water distribution networks. This ongoing study accordingly investigates performance of machine learning tools for real-time source identification. In contrast to traditional approaches that perform whole analysis after a contamination event occurs, machine learning methods gain system knowledge in advance and use this extracted information to identify contamination attributes after an incident occurs. 2012 ASCE.

Machine Learning Approach for Contamination Source Identification in Water Distribution Systems Conference Paper