Abstract
The performance of model-based leak detection and localization techniques heavily depends on the configuration of a limited number of sensors. This paper presents a sensor placement optimization strategy that guarantees sufficient diagnosability while satisfying the budget constraint. Based on the theory of compressed sensing, the leak localization problem could be transformed into acquiring the sparse leak-induced demands from the available measurements, and the average mutual coherence is devised as a diagnosability criterion for evaluating whether the measurements contain enough information for identifying the potential leaks. The optimal sensor placement problem is then reformulated as a {0, 1} quadratic knapsack problem, seeking an optimal sensor placement scheme by minimizing the average mutual coherence to maximize the degree of diagnosability. To effectively handle the complicated real-life water distribution networks, a validated binary version of artificial bee colony algorithm enhanced by genetic operators, including crossover and swap, is introduced to solve the binary knapsack problem. The proposed strategy is illustrated and validated through a real-life water distribution network with synthetically generated field data.
Optimal sensor placement for leak location in water distribution networks using genetic algorithms
