An optimal coordinated proton exchange membrane fuel cell heat management method based on large-scale multi-agent deep reinforcement learning

To improve the operating efficiency of proton exchange membrane fuel cells (PEMFCs), an optimal coordinated control strategy for addressing the poor coordination problem between the water pump and radiator in a PEMFC stack heat management system is proposed in this paper. To this end, a cooperative...

Full description

Saved in:
Bibliographic Details
Published in:Energy reports Vol. 7; pp. 6054 - 6068
Main Authors: Li, Jiawen, Li, Yaping, Yu, Tao
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.11.2021
Elsevier
Subjects:
Cooperative exploration strategy large-scale multi-agent twin-delay deep policy gradient (CESL-MATD3), proton exchange membrane fuel cell (PEMFC), coordinated control of stack temperature
Distributed deep reinforcement learning
Stack heat management system
Distributed deep reinforcement learning
Cooperative exploration strategy large-scale multi-agent twin-delay deep policy gradient (CESL-MATD3), proton exchange membrane fuel cell (PEMFC), coordinated control of stack temperature
Stack heat management system
ISSN:2352-4847, 2352-4847
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:To improve the operating efficiency of proton exchange membrane fuel cells (PEMFCs), an optimal coordinated control strategy for addressing the poor coordination problem between the water pump and radiator in a PEMFC stack heat management system is proposed in this paper. To this end, a cooperative exploration strategy large-scale multiagent twin-delay deep policy gradient (CESL-MATD3) algorithm has been developed for this control strategy. In this algorithm, both the water pump and radiator are treated as individual agents, and the strategies of centralized training and decentralized execution are applied; thus, coordinated control over the two agents is realized. Moreover, the concepts of curriculum learning, imitation learning, and various novel parallel computing techniques are incorporated into the design of this algorithm, resulting in enhanced training efficiency; thus, a coordinated control strategy with better robustness is obtained. According to the experimental results, compared with other advanced control algorithms, this coordinated control strategy-based algorithm achieves better performance and robustness for PEMFC stack temperature management. The proposed method can effectively improve the response speed of the controllers, reduce the fluctuation and oscillation of the stack temperature and the temperature difference between the stack outlet and inlet (stack temperature difference) during heat management, and reduce the maximum overshoot of the stack temperature by 99.12% and of the stack temperature difference by 97.97%. •A novel 9-order dynamic PEMFC stack heat management system model is proposed.•A novel PEMFC stack temperature coordinated control strategy is proposed.•A new large-scale deep reinforcement learning algorithm is proposed for the strategy.•The strategy coordinates of pump and radiator and improve the efficiency of PEMFC.•The proposed algorithm has better robustness compared with conventional algorithms.
ISSN:2352-4847
2352-4847
DOI:10.1016/j.egyr.2021.09.015