Distributed extremum-seeking control over networks of unstable dynamic agents

Abstract

In this paper, a distributed extremum seeking control technique is proposed to solve a class of real-time optimization problems over a network of dynamic agents with unknown unstable dynamics. Each dynamic agent measures a cost that is shared over a network. A dynamic average consensus approach is used to provide each agent with an estimate of the total network cost. The extremum seeking controller uses the local estimate of the total cost to adjust the value of the local decision variables. The contribution of the proposed technique is the simultaneous stabilization of the network dynamics and the distributed optimization of the total network cost. A dynamic network simulation example is presented to demonstrate the effectiveness of the technique.