电工技术学报  2021, Vol. 36 Issue (19): 4170-4182    DOI: 10.19595/j.cnki.1000-6753.tces.L90124
电能存储与应用 |
有轨电车基于工况识别的强化学习能量管理策略
莫浩楠, 杨中平, 林飞, 王玙, 安星锟
北京交通大学电气工程学院 北京 100044
Reinforcement Learning Energy Management Strategy of Tram Based on Condition Identification
Mo Haonan, Yang Zhongping, Lin Fei, Wang Yu, An Xingkun
School of Electrical Engineering Beijing Jiaotong University Beijing 100044 China
全文: PDF (8047 KB)   HTML
输出: BibTeX | EndNote (RIS)      
摘要 储能式混合动力有轨电车以储能系统作为唯一动力源,对能量管理策略进行优化设计,可以提高有轨电车的运行性能及经济效益。将有轨电车的需求功率看做马尔科夫过程,且为避免驾驶工况变化较大时对能量管理策略的影响,提出基于工况识别的强化学习能量管理策略。首先,通过历史行驶数据构建有轨电车驾驶工况并得到不同工况下的马尔科夫功率状态转移矩阵;然后,以混合储能系统能耗最小为目标,通过强化学习算法得到不同工况下的功率分配策略;最后,以改进的学习向量化(LVQ)神经网络对当前的驾驶工况进行实时识别,控制系统通过当前识别的工况以及列车状态做出实时决策。以实车数据进行仿真验证,优化后的策略可以降低储能系统损耗,并且可应用于不同的驾驶工况中。以90kW混合储能平台进行实验验证,验证了该策略在实际工程应用中的可行性。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
莫浩楠
杨中平
林飞
王玙
安星锟
关键词 有轨电车混合储能系统工况识别强化学习能量管理策略    
Abstract:Energy storage hybrid trams use the energy storage system as the only power source, and optimize the design of energy management strategy, which can improve the running performance and economic benefits of the tram. Regarding the demand power of the tram as a Markov process, and in order to avoid the impact on the energy management strategy when the driving conditions change greatly, a reinforcement learning energy management strategy based on the recognition of the operating conditions is proposed. Based on historical driving data, the tram driving conditions are constructed and the Markov power state transition matrix under different operating conditions is obtained. Then, with the goal of minimizing the energy consumption of the hybrid energy storage system, the power allocation strategy under different working conditions is obtained through the reinforcement learning algorithm. Finally, the improved learning vector quantization (LVQ) neural network is used to recognize the current driving conditions in real time, and the control system makes real-time decisions based on the current recognized conditions and train status. Real vehicle data is used for simulation verification. The optimized strategy can reduce the energy storage system loss and can be applied to different driving conditions. Experimental verification with a 90kW hybrid energy storage platform verifies the feasibility of the strategy in practical engineering applications.
Key wordsTram    hybrid energy storage system    operating condition recognition    reinforcement learning    energy management strategy   
收稿日期: 2020-06-30     
PACS: TM921  
基金资助:国家高速列车技术创新中心资助项目
通讯作者: 莫浩楠 男,1997年生,硕士研究生,研究方向为车载储能系统容量配置与能量管理策略。E-mail:19126139@bjtu.edu.cn   
作者简介: 杨中平 男,1970年生,教授,博士生导师,研究方向为轨道交通电力牵引传动、节能、高速列车系统优化设计等。E-mail:zhpyang@bjtu.edu.cn
引用本文:   
莫浩楠, 杨中平, 林飞, 王玙, 安星锟. 有轨电车基于工况识别的强化学习能量管理策略[J]. 电工技术学报, 2021, 36(19): 4170-4182. Mo Haonan, Yang Zhongping, Lin Fei, Wang Yu, An Xingkun. Reinforcement Learning Energy Management Strategy of Tram Based on Condition Identification. Transactions of China Electrotechnical Society, 2021, 36(19): 4170-4182.
链接本文:  
https://dgjsxb.ces-transaction.com/CN/10.19595/j.cnki.1000-6753.tces.L90124          https://dgjsxb.ces-transaction.com/CN/Y2021/V36/I19/4170