| 标 题： |
A Deep Reinforcement Learning Framework for Efficient and Resilient Power Management system in Medium Voltage DC Ship Power Systems
用于中压直流船舶电力系统高效且鲁棒电源管理系统的一种深度强化学习框架 |
| 作 者： |
Klaudia Bis
Laxman Timilsina
Christopher S. Edrington
Behnaz Papari
Asif Ahmed Khan
Gokhan Ozkan |
| 关键词： |
Power management [电源管理]
Reinforcement learning [ 强化学习]
DDPG agent [ DDPG智能体]
Neural network [ 神经网络] |
| 出版年： |
2025 |
| 文集名： |
2025 IEEE Electric Ship Technologies Symposium (ESTS 2025)
2025年IEEE船舶电气技术研讨会（ESTS 2025） |
| 会议名： |
2025 IEEE Electric Ship Technologies Symposium (ESTS 2025)
2025年IEEE船舶电气技术研讨会（ESTS 2025） |
| 届 次： |
|
| 摘 要： |
Next-generation naval vessels require highly efficient and robust power management systems to accommodate the dynamic and mission-critical demands of on-board loads. Traditional deterministic control methodologies often struggle to adapt to rapidly changing conditions, leading to suboptimal resource allocation and increased susceptibility to faults. To address these challenges, this paper presents a novel power management framework for a two-zone electric ship power system using a Deep Deterministic Policy Gradient (DDPG) agent implemented in MATLAB's Reinforcement Learning Toolbox. The simplified two-zone Simulink model captures the essential interactions between power generation and load distribution in an electric ship environment, facilitating both agent training and performance evaluation. Preliminary results demonstrate that the proposed DDPG-based power manager achieves enhanced power allocation, fast adaptation to load variations, and improved resilience compared to conventional control strategies.
新一代舰艇需要高效且可靠的电源管理系统，以满足船上负载动态变化和关键任务的需求。传统的确定性控制方法往往难以适应快速变化的环境，导致资源分配效率低下，并增加系统故障的风险。为应对这些挑战，本文提出了一种基于深度确定性策略梯度（DDPG）的新型电源管理框架，用于两区域电动船舶电力系统，该框架采用MATLAB强化学习工具箱中的DDPG智能体实现。简化后的两区域Simulink模型能够准确捕捉电动船舶环境中发电与负荷分配之间的核心交互关系，便于智能体训练和性能评估。初步结果表明，所提出的DDPG型电源管理器在功率分配、快速响应负荷变化以及提升系统鲁棒性方面，均优于传统控制策略。 |
| 中文内容为网络AI翻译，专业术语未必准确，仅供参考。 |
京公网安备 11010202009788