A distributed hierarchical algorithm for multi-cluster constrained optimization

In this paper, we consider a constrained optimization problem for a large-scale multi-cluster agent system, in which a number of clusters already exist as a priori. The aim is to minimize a global objective function being the sum of multi-cluster local agents’ cost functions subject to certain global constraints. To solve this problem, a novel distributed hierarchical algorithm based on projected gradient method is proposed by using synchronous and sequential communication strategies. We firstly assign one agent as leader agent in each cluster, which can communicate with the leaders of its neighboring clusters. The agents in the same cluster conduct local optimization and communicate with their neighboring agents synchronously while the leader agents of different clusters exchange information in a sequential way. Then a scheme is proposed for each agent to iteratively estimate a solution of the optimization problem in a distributed manner. It is theoretically proved that the estimated solutions of all the agents reach consensus of the optimal solution asymptomatically when the chosen stepsizes are diminishing. Numerical examples are provided to validate the proposed method.