Dynamic Dual Decomposition for Distributed Control

Författare

Anders Rantzer

Summary, in English

We show how dynamic price mechanisms can be used for decomposition and distributed optimization of control systems.

A classical method to deal with optimization constraints is Lagrange relaxation, where dual variables are introduced in the optimization objective. When variables of different subproblems are connected by such constraints, the dual variables can be interpreted as prices in a market mechanism serving to achieve mutual agreement between the subproblems. In this paper, the same idea is used for decomposition of optimal control problems, with dynamics in both decision variables and prices. We show how the prices can be used for decentralized verification that a control law or trajectory stays within a prespecified distance from optimality. For example, approximately optimal decentralized controllers can be obtained by using simplified models for decomposition and more accurate local models for control.

Avdelning/ar