Lessons Learned In Solving The Contaminant Source Identification Problem In An Online Context
Leçons apprises en résolvant le problème de l'identification de sources de contaminant dans un contexte temps réel
Résumé
Protection of Water Distribution Networks (WDNs) against contamination events is of paramount importance. Either deliberate or accidental contamination of this infrastructure has strong negative consequences from both social and economical points of view. The project SMaRT-OnlineWDN aims to develop methods and software solutions to 1) detect contamination from non-specific sensors, 2) maintain an online water quantity and water quality model that is reliable and 3) use the past model predictions to backtrack the potential sources of contamination. The problem of source identification consists of determining the location and duration of a contamination taking into account sensor responses. Our solution is a two-step enumeration/exploration method. Firstly, we solve the transport equation in reverse time for enumeration of the potential solutions. This is made independent of the reaction kinetics of particular substances. The known boundary conditions are the responses of sensors that count successive contaminant fronts arriving at each sensor. In the second exploration step a probability calculation for ranking of the candidate solutions is proposed with two general stochastic methods (minimum relative entropy or least squares methods). An extensive use of simplification methods is carried out, both temporally and spatially on the dynamic graph. A sensitivity analysis is made with regards to the demand uncertainty. Results on real networks in France and Germany are presented.