Quadratic optimal control of switched linear stochastic systems

This paper studies a quadratic optimal control problem for discrete-time switched linear stochastic systems with nonautonomous subsystems perturbed by Gaussian random noises. The goal is to jointly design a deterministic switching sequence and a continuous feedback law to minimize the expectation of a finite-horizon quadratic cost function. Both the value function and the optimal control strategy are characterized analytically. A numerical relaxation framework is developed to efficiently compute a control strategy with a guaranteed performance upper bound. It is also proved that by choosing the relaxation parameter sufficiently small, the performance of the resulting control strategy can be made arbitrarily close to the optimal one.

Research highlights
► Studied the optimal determinist scheduling problem for LQG controllers.
► Value functions and the optimal control and scheduling strategies are characterized analytically in terms of the Switched Riccati Mapping.
► Efficient algorithm is developed to solve the problem with guaranteed suboptimal solution.
► Promising applications in networked control and estimation problems.