You are here

Direct data-driven design of explicit predictive controls

15 June 2022
11:00 am
San Ponziano Complex - Conference Room

In this talk, we deal with constrained predictive control of linear time-invariant systems. Specifically, byemploying the Willems' lemma, we show how explicit predictive laws can be learnt directly from data, without theneedto identify the plant to control. Moreover, by leveraging on the knowledge that the optimal controller turns out tobe a piecewise affine law, we derive a purely data-driven stability check that can be safely performed prior to thecontroller deployment. Finally,as one critical point of explicit predictive control is its computational load when appliedto large scale systems, we discuss a possible warm-start procedure to speed up the detection of the optimal controlaction when the size of the state vector is large.The effectiveness of the proposed data-driven design approach isillustrated via two benchmark simulation case studies.


Join at

Simone Formentin - Politecnico di Milano