| 引用本文: | 王浩聪,王栎阳,付主木,陈启宏,陶发展.燃料电池混合动力汽车深度强化学习能量管理优化[J].控制理论与应用,2024,41(10):1831~1841.[点击复制] |
| WANG Hao-cong,WANG Yue-yang,FU Zhu-mu,CHEN Qi-hong,TAO Fa-zhan.Energy management optimization of fuel cell hybrid electric vehicle based on deep reinforcement learning[J].Control Theory and Technology,2024,41(10):1831~1841.[点击复制] |
|
| 燃料电池混合动力汽车深度强化学习能量管理优化 |
| Energy management optimization of fuel cell hybrid electric vehicle based on deep reinforcement learning |
| 摘要点击 3451 全文点击 48 投稿时间：2022-05-30 修订日期：2024-05-12 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2023.20468 |
| 2024,41(10):1831-1841 |
| 中文关键词 燃料电池混合动力汽车 深度强化学习 能量源退化 等效消耗最小 数据驱动 能量管理策略 |
| 英文关键词 fuel cell hybrid electric vehicle deep reinforcement learning energy sources degradation equivalent consumption minimization date driven energy management strategy |
| 基金项目 国家自然科学基金项目(62201200), 河南省高校科技创新人才计划项目(23HASTIT021), 河南省重点研发与推广专项科技攻关项目(21210221 0153, 222102240009), 河南省博士后科研项目(202003077), 河南省科技研发计划联合基金项目(222103810036), 内蒙古机电控制重点实验室开放 基金项目(IMMEC2022001, IMMEC2022002)资助. |
|
| 中文摘要 |
| 针对配备有锂电池与超级电容的燃料电池混合动力汽车, 为降低车辆总体运行成本, 延长能量源寿命, 本文提出一种基于深度强化学习的能量管理策略. 首先, 依据超级电容高功率密度特性, 建立基于模糊自适应滤波的功率分层结构, 并依据燃料电池与锂电池的经验退化模型, 建立能量源退化的成本函数, 采用等效消耗最小策略平衡氢耗成本与能量源退化成本, 以最小化总体运行成本为目标来优化能量源功率分配; 然后, 引入优先经验回放与软更新以提高深度强化学习的离线训练效率; 最后, 在多种工况下进行仿真, 结果表明, 与未考虑退化的策略相比,本文所提出策略在全球统一轻型车辆测试循环下可使氢耗量降低11.8%, 并可有效减缓燃料电池与锂电池的退化速率, 降低燃料电池混合动力汽车的总体运行成本. |
| 英文摘要 |
| For the fuel cell hybrid electric vehicle equipped with lithium battery and ultracapacitor, to reduce the overall operation cost and prolong the lifespan of energy sources, an energy management strategy based on deep reinforcement learning is proposed in this paper. Firstly, according to the high power density characteristics of ultracapacitor, a power hierarchical structure based on a fuzzy adaptive filter is established, and based on the empirical degradation model of fuel cell and lithium battery, the cost function of energy source degradation is established. The equivalent consumption minimum strategy is used to balance the hydrogen consumption cost and energy source degradation cost, and the power allocation of energy sources is optimized to minimize the overall operation cost. Then, prioritized experience replay and soft update are introduced to improve the off-line training efficiency of deep reinforcement learning. Finally, the simulation is carried out under various driving cycles. The results show that compared with the strategy without considering degradation, the proposed strategy reduces hydrogen consumption by 11.8% under the world light vehicle test cycle, and can effectively slow down the degradation rate of fuel cell and lithium battery and reduce the overall operating cost of fuel cell hybrid electric vehicle. |
|
|
|
|
|