stochastic sampling design for water distribution model calibration

a novel approach to determine optimal sampling locations under parameter uncertainty in a water distribution system (wds) for the purpose of its hydraulic model calibration is presented. the problem is formulated as a multi-objective optimisation problem under calibration parameter uncertainty. the objectives are to maximise the calibrated model accuracy and to minimise the number of sampling devices as a surrogate of sampling design cost. model accuracy is defined as the average of normalised traces of model prediction covariance matrices, each of which is constructed from a randomly generated sample of calibration parameter values. to resolve the computational time issue, the optimisation problem is solved using a multi-objective genetic algorithm and adaptive neural networks (moga-ann). the verification of results is done by comparison of the optimal sampling locations obtained using the moga-ann model to the ones obtained using the monte carlo simulation (mcs) method. in the mcs method, an equivalent deterministic sampling design optimisation problem is solved for a number of randomly generated calibration model parameter samples.the results show that significant computational savings can be achieved by using moga-ann compared to the mcs model or the ga model based on all full fitness evaluations without significant decrease in the final solution accuracy.