Contaminant Source Identification in Water Distribution Networks Under Conditions of Demand Uncertainty

Water distribution systems are susceptible to accidental and intentional chemical or biological contamination that could result in adverse health impact to the consumers. This study focuses on a water distribution forensics problem, contaminant source identification, subject to water demand uncertainty. Due to inherent variability in water consumption levels, demands at consumer nodes remain one of the major sources of uncertainty. In this research, the nodal demands are considered to be stochastic in nature and are varied using Gaussian and Autoregressive models. A hypothetical source identification problem is constructed by simulating observations at the sensor nodes from an arbitrary contaminant source. A simulation-optimization approach is used to solve the source identification problem with EPANET tool as the simulator and Genetic Algorithm (GA) as the optimizer. The goal is to find the source location and concentration by minimizing the difference between the simulated and observed concentrations at the sensor nodes. Two variations of GA, stochastic GA and noisy GA are applied to the same problem for comparison. Results show that noisy GA is more robust and is less computationally expensive than stochastic GA in solving the source identification problem. Moreover, the autoregressive demand uncertainty model better represents the uncertainty in the source identification process than the Gaussian model.