Coalitional control: a bottom-up approach

TL;DR

This paper introduces a bottom-up coalitional control approach for multi-agent systems, where controllers adapt to changing interactions to form coalitions that improve overall system performance.

Contribution

It presents a novel method for dynamic coalition formation among control agents based on adaptive model predictive controllers in distributed systems.

Findings

Coalitions form dynamically based on interaction strength.

System performance improves with adaptive coalition strategies.

Controllers adapt to time-varying coupling conditions.

Abstract

The recent major developments in information technologies have opened interesting possibilities for the effective management of multi-agent systems. In many cases, the important role of central control nodes can now be undertaken by several controllers in a distributed topology that suits better the structure of the system. This opens as well the possibility to promote cooperation between control agents in competitive environments, establishing links between controllers in order to adapt the exchange of critical information to the degree of subsystems' interactions. In this paper a bottom-up approach to coalitional control is presented, where the structure of each agent's model predictive controller is adapted to the time-variant coupling conditions, promoting the formation of coalitions - clusters of control agents where communication is essential to ensure the cooperation - whenever it can bring benefit to the overall system performance.