Optimal Sensor Placement for the Efficient Contaminant Detection in Water Distribution Systems

This paper presents a novel risk-based approach to determine efficiently and effectively the optimal sensor locations to detect contaminants in a water distribution network (WDS). If the contamination is detected in time, it is prevented from spreading through the network and customer safety is ensured. The optimisation task formulated here has two objectives: (1) minimise the risk of contamination and (2) minimise the number of sensors used. The former expresses the likelihood that a set of sensors cannot detect a contaminant and the associated amount of polluted water that is consumed prior to detection, while the latter acts as a surrogate for the sampling design cost. The problem is solved using the NSGA-II algorithm (Deb et al., 2002) assuming that a single contaminant intrusion can occur anywhere in the system at any ten minute time interval during 24 hours. An importance-based contamination scenario selection is used to reduce the computational requirements associated with evaluating a potentially large number of samples and to accelerate the optimisation process while preserving accuracy. The methodology is tested on a benchmark network from the Battle of the Water Sensor Networks held in 2006 (Ostfeld et al., 2008). The results obtained demonstrate that only a small portion of a network needs to be monitored to provide a reasonable amount of safety.