358 - State Estimation of Water Distribution Networks Using Virtual Measurements from Reduced Models

To ensure the smooth operation of water distribution networks (WDNs), practitioners require access to continuous and near-real time information on system states, i.e., pressures and flows in the WDN. State estimation (SE) is a common approach for inferring WDN states by fusing the information from hydraulic models with measurements from sensors placed in networks. However, the application of many advanced SE techniques in the context of WDNs is limited by (i) sparse sensor placement in WDNs leading to limited available sensor data and (ii) substantial computation time and effort required to solve the SE problem and system hydraulics for large networks. In order to address these challenges, this work proposes CAscading STAte estimation of water distribution NETworkS (CASTANETS), a framework that combines the Extended Kalman Filter (EKF) with reduced WDN models, i.e., models that are smaller in size but preserve original model pressures, to derive an accurate view of system states for large WDNs. Application of the EKF, which typically requires state space representation of the system, is achieved by formulating WDN hydraulics in terms of conservation of mass and momentum and modeling pressurized flow using the Preissmann slot. The SE problem is first solved to obtain estimates of nodal pressures for reduced models of WDNs, which then serve as “virtual measurements” to augment SE of larger sized models. We conduct sensitivity analyses to investigate the effects of reduced model structure and estimation uncertainty on SE accuracy and running times. The CASTANETS framework is validated using benchmark networks from the literature and a set of real-life WDNs. The methodology developed in this work aims to enhance overall awareness of system states, thus enabling real-time observation and control of
realistically-sized WDNs.