Optimality Loss Minimization in Distributed Control: A Multi-Agent Partitioning Approach

This paper presents a method for finding the partition that minimizes the total cost at the Nash Equilibrium in a communication-based control of large-scale systems via an evolving tree search. By directly considering the control objectives, including the cost function and constraints, the proposed partitioning method finds the distributed control structure that minimizes the cost increase introduced when converting from a centralized control problem to a distributed one. The performance of the proposed partitioning method is compared to a state-coupling-based method for a set of 30 constrained linear quadratic regulator control problems, where it finds feasible solutions 23% more often and reduces the average total cost by 14.9%.