Optimal meter placement for water distribution system state estimation

Real-time state estimates (SEs) of nodal demands in a water distribution system (WDS) can be developed using data from a supervisory control and data acquisition (SCADA) system. These estimates provide information for improved operations and customer service in terms of energy consumption and water quality. The SE results in a WDS are significantly affected by measurement characteristics, i.e., meter types, numbers, and topological distributions. The number and type of meters are generally selected prior to a SCADA layout. Thus, selecting measurement locations is critical. The aim of this study is to develop a methodology that optimally locates field measurement sites and leads to more reliable SEs. An optimal meter placement (OMP) problem is posed as a multiobjective optimization form. Three distinctive objectives are formulated: (1) minimization of nodal demand estimation uncertainty; (2) minimization of nodal pressure prediction uncertainty; and (3) minimization of absolute error between demand estimates and their expected values. Objectives (1) and (2) represent the model precisions while Objective (3) describes the model accuracy. The OMP is solved using a multiobjective genetic algorithm (MOGA) based on Pareto-optimal solutions. The trade-off between model precision and accuracy is clearly observed in two case studies and it is recommended to use both criteria as objectives. It is also concluded that the proposed objectives are more appropriate for OMP purposes compared to calibration sampling design studies in which minimization of metering costs (i.e., number of meters) is used as one of the multiple objectives. The MOGA saves computational effort while providing optimal Pareto solutions compared to full enumeration for a small hypothetical network. For real networks, GA solutions, although not guaranteed to be globally optimal, are improvements over those obtained using less robust methods or designers' experienced judgment.