Duality-Based Distributed Optimization With Communication Delays in Multi-Cluster Networks

TL;DR

This paper introduces an asynchronous distributed dual proximal gradient algorithm for multi-cluster network optimization with communication delays, enabling agents to reach consensus on complex, coupled decision problems.

Contribution

It proposes a novel asynchronous dual optimization method tailored for multi-cluster networks with delays, incorporating a cluster-based consensus protocol.

Findings

Algorithm converges ergodically under communication delays.

Effective in solving social welfare optimization problems.

Communication delays do not hinder convergence.

Abstract

In this work, we consider solving a distributed optimization problem (DOP) in a multi-agent network with multiple agent clusters. In each cluster, the agents manage separable cost functions composed of possibly non-smooth components and aim to achieve an agreement on a common decision of the cluster. The global cost function is considered as the sum of the individual cost functions associated with affine coupling constraints on the clusters' decisions. To solve this problem, the dual problem is formulated by the concept of Fenchel conjugate. Then an asynchronous distributed dual proximal gradient (Asyn-DDPG) algorithm is proposed based on a cluster-based partial and mixed consensus protocol, by which the agents are only required to communicate with their neighbors with communication delays. An ergodic convergence result is provided, and the feasibility of the proposed algorithm is verified by solving a social welfare optimization problem in the simulation.