基于单步滑动窗口-长短期记忆网络的锂电池SOC估计算法  

SOC estimation algorithm of lithium-ion battery based on single step sliding window-long short-term memory network

作　　者：王志亮 吴勇 韩尚卿 范晓东 王猛 于承航 WANG Zhiliang;WU Yong;HAN Shangqing;FAN Xiaodong;WANG Meng;YU Chenghang(Beijing Institute of Mechanical and Electrical Engineering,Beijing 100071,China;CETC Lantian Technology Co.,Ltd.,Tianjin 300384,China)

机构地区：[1]北京机电工程研究所,北京100071 [2]中电科蓝天科技股份有限公司,天津300384

出　　处：《电源技术》2024年第2期306-311,共6页Chinese Journal of Power Sources

摘　　要：准确的荷电状态(SOC)预估是电池管理系统安全稳定运行的基础,对锂离子电池的推广应用具有重要意义。为提高荷电状态的估计精度,建立了一种长短期记忆网络(LSTM)与单步滑动窗口技术相结合的荷电状态估计模型。引入单步滑动窗口技术对输入数据进行预处理。构建单步预估LSTM模型,利用错时间步数据结构增强LSTM算法的鲁棒性,达到提高SOC估计精度的目的。分别在自定义的充电、放电与模拟真实飞行器充放电实验工况下对所提算法进行了验证。结果表明,算法能够在充电与放电工况下实现10 s内收敛,模拟真实飞行器充放电实验工况下0.01 s收敛至预估精度2%以下,收敛后3种工况下最大预估误差均不超过0.005。证明了所提算法具有较强的鲁棒性与快速性,为动力电池的SOC估计优化提供了理论指导。The accurate estimation of the state of charge(SOC)is the foundation for the secure and stable operation of battery management systems,and is of great significance for the widespread application of lithium-ion batteries.To enhance the estimation precision of SOC,the estimation model combining long short-term memory(LSTM)network and single step sliding window was proposed.The single step sliding window technique was introduced to preprocess the input data.A single step predictive LSTM model was constructed,and the time shifted data structure was utilized to improve the robustness of the LSTM algorithm,improving SOC estimation precision.The algorithm was validated under charging,discharging,and simulated real aircraft charging and discharging experimental conditions.The results indicate that the algorithm achieves convergence within 10 s under charging and discharging conditions.Within 0.01 s,the algorithm attains an estimated accuracy of less than 2%under simulated real aircraft charging and discharging conditions.After convergence,the maximum estimation error is less than 0.005 under the three conditions.The results demonstrate that the proposed algorithm possesses strong robustness and rapidity,providing theoretical guidance for optimizing SOC estimation of lithium-ion batteries.

关 键 词：锂离子电池 长短期记忆神经网络 荷电状态 滑动窗口 

分 类 号：TM912.9[电气工程—电力电子与电力传动]