An analog "neural net" based suboptimal controller for constrained discrete-time linear systems

Abstract A large class of problems frequently encountered in practice involves the control of linear time invariant systems with states and controls restricted to closed convex regions of their respective spaces. In spite of the significance of this problem, to date it has not been solved satisfactorily except in some restricted cases. In this paper we propose a suboptimal feedback control algorithm based upon on-line optimization during the sampling interval. Theoretical results are presented showing that our approach yields asymptotically stable systems. Finally an implementation of the control algorithm using an analog circuit is discussed. This implementation provides an alternative to the use of digital computers in the feedback loop that offers advantages in terms of cost and reliability. We believe that it may prove to be specially valuable when the time available for computations is limited. A 5th-order model of a F-100 jet engine is used as an example application of the controller.