Abstract

The resurgence of data-driven dynamic models offers the tantalising prospect of being able to implement feedback controllers directly from measurements of the trajectories of the system to be controlled. Data-enabled predictive control (DeePC), data-driven predictive control (DDPC), and similar variants circumvent the traditional approach of identifying a dynamic model as an intermediate step in the control design process. Such approaches require regularisation to trade off between the estimation and control objectives. Another weakness is the inability to effectively handle unmeasured disturbances. We take a somewhat different view here that the data matrices used for data-driven control are themselves models (signal matrix models) that use the system trajectories as the representation. We will use this approach to construct Kalman filters and predictive controllers. Regularisation is no longer necessary and unmeasured disturbances can be effectively controlled.

The seminar will be held in JDB Seminar Room, Department of Engineering