Abstract:
The development of water quality sensors and sensor technologies makes it feasible to establish water quality monitoring networks in drinking water distribution systems. The collected water quality data can be used to determine the location and time of the source contamination in the case of terrorist attack or accidental contamination intrusion. If the contaminant reaction can be modeled reliably within the pipe network and the sensors can measure water quality quantitatively, minimization of the difference between modeled and measured water quality is one approach to solution of the contaminant source determination problem. The underling mathematical problem is, however, inherently ill-posed, due to the shortage of measurements compared to source parameters, and regularization methods are required to force identification of a unique solution. Further, it is usually the case that the contaminant reaction dynamics are unknown, and/or the sensor can only detect the presence or absence of the contaminant and not the quantitative concentration. Even if the source determination problem can be formulated mathematically and optimization algorithms can be applied to solve the problem, the decision variable dimension can be huge since contamination can happen anywhere and anytime in the network. An alternative practical method is developed in this paper to identify all possible locations and times that explain contamination incidents detected by the water quality sensors. It is assumed that contaminant injections occur at network junctions and over discrete time intervals. The method only requires the positive/negative sensor status over time, and knowledge of network hydraulics. A particle backtracking algorithm is used to identify the water flow paths leading to each sensor measurement and the travel time from the junctions along the flow paths to the measurement. Those locations and times that are connected to positive sensor measurements – and are not connected to negative measurements – are the possible sources, assuming no false positive/negative readings and an accurate hydraulic model. This method can also be used as a pruning step for solving the source identification problem using optimization algorithms, as it reduces the number of decision variables by eliminating locations and times that are inconsistent with the sensor responses. The method also forms the basis for incorporating important concerns about hydraulic and sensor uncertainty, which are likely to enlarge the set of possible sources.
