变模态分解下SSA-LSTM组合的锂离子电池剩余使用寿命预测方法  

Prediction method for remaining service life of lithium batteries using SSA-LSTM combination under variable mode decomposition

作　　者：李嘉波 王志璇 田迪 孙中麟 LI Jiabo;WANG Zhixuan;TIAN Di;SUN Zhongin(School of Mechanical Engineering,Xi'an University of Petroleum,Xi'an 710065,Shaanxi,China;Engineering Research Center for Highway Construction and Maintenance Equipment and Technology of Chang'an University,Ministry of Education,Xi'an 710064,Shaanxi,China)

机构地区：[1]西安石油大学机械工程学院,陕西西安710065 [2]长安大学,高速公路筑养装备与技术教育部工程研究中心,陕西西安710064

出　　处：《储能科学与技术》2025年第2期659-670,共12页Energy Storage Science and Technology

基　　金：陕西省教育厅科研计划项目(23JK0599);陕西省自然科学基础研究计划资助项目(2023-JC-QN-0658);高速公路筑养装备与技术教育部工程研究中心(300102253513);“咸阳市二〇二三年重点研发计划”项目动力电池过充电故障诊断研究(L2023-ZDYF-0YCX-032);陕西省教育厅科研计划项目(23JK0599)。

摘　　要：锂离子电池在电动汽车、可再生能源等领域广泛应用,对其剩余使用寿命(remaining useful life,RUL)进行精确预测,能够实时把握电池的内在性能退化状态,降低电池使用风险。本工作提出了一种基于变模态分解(variational mode decomposition,VMD)、麻雀优化算法(sparrow search algorithm,SSA)和长短期记忆网络(long short-term memory,LSTM)的组合预测算法对锂离子电池剩余寿命进行预测。首先,基于锂离子电池电流、电压以及温度曲线,提取等压差充电时间、等压差充电能量、放电温度峰值和恒流充电时间作为预测RUL的间接健康因子。其次,采用变模态分解法分解容量以避免容量回升的局部波动和测试噪声对RUL预测结果造成干扰。针对传统LSTM模型超参数设置易受到经验和随机性的影响,提出了麻雀优化算法对LSTM模型参数进行优化,以提升模型的预测能力。最后,应用NASA和CALCE数据集,将所提模型与其他模型进行对比。实验结果表明,锂离子电池RUL预测均方根误差控制在2%以内,所提方法具有较高的预测性能。The widespread application of lithium-ion batteries in electric vehicles,renewable energy,and other fields necessitates accurate prediction of their remaining useful life(RUL).Such predictions enable real-time monitoring of the battery's internal performance degradation,thereby reducing the risk associated with battery usage.We propose a combined prediction algorithm utilizing variational mode decomposition,sparrow search algorithm(SSA),and long short-term memory(LSTM)for predicting the remaining life of lithium-ion batteries.Initially,indirect health indicators for predicting RUL were extracted from the current,voltage,and temperature curves of the batteries.These indicators included isobaric charging time,isobaric charging energy,peak discharge temperature,and constant-current charging time.Subsequently,the VMD method was employed to decompose the capacity,aiming to avoid local fluctuations in capacity recovery and interference from test noise that could affect RUL prediction results.To address the susceptibility of traditional LSTM model hyperparameter settings toward experience and randomness,an SSA was proposed to optimize the parameters of the LSTM model,thereby enhancing the model's predictive capabilities.Ultimately,by utilizing NASA and CALCE datasets,a comparison was conducted between the proposed model and other models.Experimental results demonstrate that the proposed method achieves high predictive performance,with the root mean square error for RUL prediction of lithium-ion batteries consistently maintained within 2%.

关 键 词：锂离子电池 剩余使用寿命 变模态分解 麻雀优化算法 长短期记忆网络 

分 类 号：TM912[电气工程—电力电子与电力传动]