State estimation based on enhanced Bayesian approach: Application in water distribution systems

Abstract

Hydraulic models have been regarded as an essential tool for controlling and operating water distribution systems (WDSs). With the rapid development of communication networks, real-time control is becoming a hot issue. The nodal water demand as the state variable should be estimated for the control and management of WDS. Due to the lack of sufficient observed data to determine the nodal water demand, the hydraulic model may result in unreasonable model outputs, which are undesirable in actual engineering. We develop a Bayesian inference method to fuse nodal pressure prior information into the state estimation process to address this problem. The developed approach can constrain the nodal water demand estimation process and keep the model outputs (e.g., nodal water pressure) in the feasible domain. This approach is applied to a simple network and a realistic large-scale network. Results show that prior nodal pressure information can significantly increase estimation accuracy and decrease the uncertainty, and the estimated nodal pressure can be confined to a feasible domain.