Real-Time Water Distribution System Calibration Using Genetic Algorithm

This paper aims to calibrate, in real-time, a hydraulic model by combining an implicit technique with a genetic algorithm (GA). The approach uses demand multipliers as the calibration parameter. The hydraulic model yields an optimal solution through minimizing a nonlinear objective function between observed and simulated values of nodal pressures and pipe flow rates at a five-minute time interval, subject to a set of explicit bound constraints for the demand multipliers and implicit nonlinear hydraulic constraints. The hydraulic modeling software EPANET is applied to solve hydraulic constraints, including the equations of mass conservation for each junction and energy conservation for each pipe, and to retrieve the simulated values required in the objective function after assigning demand multipliers to nodes in the hydraulic model. The simulated values are computed under the same boundary conditions, including tank levels and pump status, as the measurements are collected. The approach is applied to the L-Town hypothetical network as a case study, yielding not only simulated values in good agreement with measurements but also consistent variation trends over time between simulated values and measurements. The case study shows that demand multipliers expressed in real numbers lead to more accurate simulated values, closer to the measurements, than expressed in a string of bits. Further efforts will focus on leak detection using the well-calibrated model.