New ways to find and estimate leak size: moving horizon estimation for pipeline leak detection and localization

Speaker: Stevan Dubljevic

Venue: Room 211, College of Control Science and Engineering, Yuquan Campus

Abstract: Effective pipeline leak detection and localization are essential for mitigating greenhouse gas emissions in hydrocarbon transportation systems. However, the complex spatiotemporal dynamics, limited sensor coverage, and measurement disturbances pose significant challenges. This talk presents advanced estimation and control strategies designed for pipeline networks modeled by infinite-dimensional systems governed by partial differential equations (PDEs). Specifically, a novel moving horizon estimation (MHE) framework is introduced for constrained estimation of leak size and location, using a discrete-time pipeline hydraulic model derived via the structure-preserving Cayley-Tustin discretization. By leveraging coordinate transformation, the estimation problem is decoupled to improve leak localization accuracy. On the other hand, a discrete Luenberger observer is designed for state reconstruction under limited measurements, and support vector machines (SVM) are employed for data-driven leak classification and localization. The MHE framework is further extended by integrating state and parameter estimation with model predictive control (MPC) for set-point tracking in PDEs-governed pipeline network systems. Finally, an industrial application involving a pipeline system in Alberta will be discussed.