Extreme impact contamination events sampling for real-sized water distribution systems

Contamination warning systems are being designed to protect water distribution systems against deliberate contamination intrusions. To design a contamination warning system, contamination intrusion events need to be selected. Because contamination intrusions are random, even for a medium-size network the theoretical number of possible injection events is huge, and thus the number of contamination events which can be considered in the design process is limited. To effectively cope with the threat of contamination events there is a need to identify those critical instances. A straightforward approach of enumerating all possible contamination intrusions from which critical events can be selected is limited to small systems. As critical events are rare the probability of revealing them using common Monte Carlo randomized simulations is very small or requires an extensive impractical computational amount of trials. In this study a methodology utilizing importance sampling and cross entropy based on a recent published work of the authors is further tested on real-sized water distribution systems of increasing complexity. The results demonstrate the robustness of the methodology in terms of improved run times, suggesting computational feasibility for problems in which size prevents full enumeration or application of direct Monte Carlo simulation techniques.