Real-time inversion in large-scale water networks using discrete measurements

There are significant challenges related to contamination detection and characterization within large water distribution systems. Given current sensing technology and resources, source inversion algorithms will need to rely on manual grab samples providing only a discrete positive/negative indication of the presence of contaminant. We propose an integrated real-time strategy that identifies the set of likely locations and performs additional sampling cycles to accurately identify the contamination source. We present an MILP formulation that solves the source inversion problem using discrete (positive/negative) measurements from sparse manual grab samples at limited points in time and space. The water quality model is formulated using the origin-tracking approach and is then exactly and efficiently reduced prior to the formulation of the MILP, giving a much smaller problem that is solvable in real-time settings. The formulation is tested on a water network model comprised of over 10,000 nodes and more than 150 timesteps.

► The contamination source inversion problem is solved using an MILP formulation. ► We propose an iterative sampling strategy to gather discrete sparse measurement data. ► We present an exact and efficient reduced water quality model. ► The approach provides fast inversion in the presence of measurement error.