Distributed gradient descent method with edge-based event-driven communication for non-convex optimization
Distributed gradient descent method with edge-based event-driven communication for non-convex optimization
复制标题
DOI：
10.1049/cth2.12127
复制
发表时间：
2021-05-22
影响因子：
2.6
通讯作者：
Takai, S.
中科院分区：
文献类型：
--
作者：
Adachi, T.;Hayashi, N.;Takai, S.
考虑事件驱动的多智能体系统的分布式非凸优化算法，其中每个智能体都有一个非凸代价函数。多智能体系统的目标是以一种分布式的方式最小化全局目标函数，即这些局部代价函数的总和。为此，每个智能体通过基于共识的梯度下降算法来更新自己的状态。邻居代理之间的本地信息交换采用事件触发机制，以较少的代理间通信来达成共识。数值算例表明，该算法收敛于目标函数的一个临界点，并证明了算法的有效性。
This paper considers an event-driven distributed non-convex optimization algorithm for a multi-agent system, where each agent has a non-convex cost function. The goal of the multi-agent system is to minimize the global objective function, which is the sum of these local cost functions, in a distributed manner. To this end, each agent updates the own state by a consensus-based gradient descent algorithm. The local information exchange among neighbor agents is carried out with an event-triggered scheme to achieve consensus with less inter-agent communication. Convergence to a critical point of the objective function and the validity of the proposed algorithm in numerical examples are shown.