In this study a multi-objective optimization model is developed for water sensor
network design in water distribution systems. In this model the three criteria used for
evaluating the performance of the water sensor placement designed are directly used
as the objectives of the optimization problem. These include minimizing the expected
water volume contaminated, minimizing the expected time of detection and
maximizing the detection likelihood. Due to the difficulty of determining sensor
placement locations within thousands of junction combinations in the system, the
sub-domain concept is introduced, which identifies a subset of junctions for candidate
sensor locations. The sub-domains are determined using the roulette wheel method
based on junction water demand values. The junctions with larger water demand have
higher probabilities to be selected to the candidate sensor subset. For solution of the
model an improved approach that is based on the non-dominated sorting genetic
algorithm (NSGA-II) is used. The approach works over the sub-domain and the final
Pareto optimal front is obtained through the sub-domain iteration process. The two
water distribution systems provided in BWSN 2006 are chosen as examples to
demonstrate the performance of the model and algorithm proposed. The impact of the
non-detected scenarios in calculating objectives on the Pareto optimal front is also
addressed in this study. The results show that the proposed model and the algorithm
are effective in solving this problem.