Sampling design for network model calibration using genetic algorithms

Today, most municipal water utilities use computerized numerical models of their distribution networks. These models must be calibrated to the physical system if they are to provide accurate results. Calibration entails adjusting certain model parameters, usually the pipe roughness coefficients, until the model accurately predicts the results of a series of flow tests. The choice of flow test locations, the sampling design, is currently done on an ad hoc basis. If the flow test locations are less than optimal, the data collected may yield insufficient information for an accurate calibration, leaving the modeler a choice between collecting more data, at additional expense, or using a less accurate model. This paper describes the use of a genetic algorithm to optimize the sampling design. The genetic algorithm was applied to a network model for a small town in Ohio and shown to perform extremely well, matching the optimal solutions produced by complete enumeration in a series of validation tests.