Maben Rabi, KTH, School of Electrical Engineering
This talk concerns some switching control problems - where the control signal is switched between different constant levels at times when the state crosses some levels. Such event-driven control generation is more efficient than traditional deterministic policies such as periodic sample-and-hold control; the event-driven polices require lower switching rates for the same performance. This becomes important in Networked Control where the switching rate directly translates into either a communication cost or an attention index.
The search for optimal event-driven control policies leads to problems that combine both optimal control and optimal stopping.
First, I will present a simple problem in finite horizon and present its solution. Then, I will discuss an infinite horizon, average cost control problem. I will describe some plausible lines of attack mainly focussing on strategies that yield approximate solutions. I will characterize the performance of these sub-optimal strategies.