Data-driven optimal switching of switched systems

Switched systems are complicated due to the switching among the subsystems. When the subsystem models are unknown, control problems on switched systems turn to be more intractable. In this paper, the optimal switching problems are investigated for continuous-time switched autonomous systems with unknown dynamics and a finite-horizon cost function. Firstly, a novel data-driven optimal scheduling approach is proposed based on the estimated insertion gradients. Secondly, aiming at switched systems with a prescribed switching sequence, a data-driven optimal switching time approach is proposed based on the estimated derivatives of the cost with respect to the switching times. The two approaches take advantages of plenty state data containing necessary information instead of the system models. Furthermore, the errors of the approaches are analysed and bounded. Finally, simulation results of two examples are given to show the validity of the two approaches.