A novel data-driven model-free synchronization protocol for discrete-time multi-agent systems via TD3 based algorithm

System modeling is a complex and time-consuming task in engineering, and traditional model-based or observer-based methods are unable to deal situations when the system models are unknown or affected by elements such as temperature. As a result, it is critical to build model-free methods capable of...

Full description

Saved in:
Bibliographic Details
Published in:Knowledge-based systems Vol. 287
Main Authors: Liu, Zhongxin, Li, Ye, Lan, Ge, Chen, Zengqiang
Format: Journal Article
Language:English
Published: Elsevier B.V 05.03.2024
Subjects:
Multi-agent consensus
Optimal control
Reinforcement learning
Twin Delayed Deep Deterministic Policy Gradient Algorithm (TD3)
Multi-agent consensus
Twin Delayed Deep Deterministic Policy Gradient Algorithm (TD3)
Optimal control
Reinforcement learning
ISSN:0950-7051, 1872-7409
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:System modeling is a complex and time-consuming task in engineering, and traditional model-based or observer-based methods are unable to deal situations when the system models are unknown or affected by elements such as temperature. As a result, it is critical to build model-free methods capable of iterative learning and real-time updating. In this paper, a data-driven, model-free TD3-based algorithm is proposed. The neural network is used to address the issue of dimensional explosion in the state and action space. Besides, each agent can set a variable number of neighbors via virtual neighbor technology, giving the connection topology more flexibility. What is more, the proposed controller is only constructed based on the consensus error, which can quickly realize synchronization while consuming the least amount of energy. Finally, proofs and some simulation examples are given to verify the efficiency of our algorithm.
ISSN:0950-7051
1872-7409
DOI:10.1016/j.knosys.2024.111430