Traditional control and estimation methods for large-scale energy and process systems can roughly be divided into centralized and decentralized approaches. While centralized methods can achieve the best performance, decentralized methods are often more flexible and easier to implement. Recently, distributed methods have been developed with the aim to combine all of these features in a single approach.
In this talk, we present a distributed, or, more precisely, a partition-based moving horizon estimator for optimal state estimation in large-scale systems that are composed of geographically separated, yet interacting subsystems. Each subsystem is connected to a dedicated estimator, which estimates only the state of its associated subsystem, and communicates these estimates with the other subsystem estimators. The resulting estimation scheme is iterative, and its unique feature is its ability to asymptotically approach the optimal state estimates of a centralized moving horizon estimator.
We discuss convergence and stability of the proposed algorithm, as well as possibilities to improve its performance by incorporating additional inequality constraints on the estimated states. Finally, these properties will be demonstrated with the help of numerical examples and a case study.
René Schneider studied Engineering Cybernetics at the Otto-von-Guericke University Magdeburg in Germany from 2004 to 2009. From 2008 to 2009 he was a visiting student at the Norwegian University of Science and Technology in Trondheim, an intern at the Statoil Research Centre in Trondheim, as well as a visiting student at the Technical University Berlin. In 2009 he joined the Process Systems Engineering group of Prof. Wolfgang Marquardt at RWTH Aachen University, Germany, where he works as a research associate. His research interest is the application of optimization methods for control and state estimation, with a current focus on partition-based moving horizon estimators.