Fast Clustering for Multi-agent Model Predictive Control

Abstract

In <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">coalitional</i> model predictive control, the overall system is controlled by a set of networked agents that are dynamically arranged into <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">clusters</i> of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">connected</i> agents that coordinate their actions, also called coalitions. In this way, the overall coordination burden and the need for sharing information are reduced. In this article, the clustering problem is formulated as a binary quadratic program (BQP), where each variable represents one agent-to-agent connection. A supervisory layer decides periodically the number and composition of the coalitions by solving the BQP while, at a bottom layer, each cluster computes the control inputs. The performance of this method is illustrated through numerical examples.