This project is concerned with the controller design for networks of switching servers with setup times, e.g. manufacturing systems or urban road networks (traffic light control). Control of these networks is difficult, since using controllers that are stable for a server in isolation might render the network unstable. Sofar, in literature, most people first propose a policy, and then study the resulting behavior of the network under this policy.
In this project we propose an entirely different way of looking at the problem of controlling a network of switching servers with setup times. Instead of starting from a policy and then analyzing the proposed policy, we start from a priori specified desired network behavior. Using this desired behavior for the network under consideration as a starting point, we derive a policy which guarantees convergence of the system towards this desired behavior. Though the policy is tailor made for both the network under consideration and its desired behavior, we aim for a general methodology that is applicable to arbitrary networks and arbitrary feasible network behavior.
First studies show that this reverse approach by using Lyapunov's direct method is indeed promising for a large class of networks and arbitrary desired network behavior. However, the controllers resulting from this approach are global controllers, or central controllers.
To derive distributed controllers the control theoretical concept of observers is used. As a first study from the applicant shows, this idea not only results in tailor made distributed policies that achieve feasible desired network behavior, but the resulting policies are also different from policies as commonly considered in literature.
The outcome from this project are control strategies and rules for deriving network controllers. These algorithms will be implemented at NXP Semiconductors and the city of Eindhoven.
Flow networks, switched servers, setup times, controller/observer design, deterministic multi-class fluid queues.
See the following link: Research proposal