摘要
【目的】锂电池健康状态(state of health, SOH)的精确预测评估可以提高电池设备的安全性,降低故障的发生率。针对数据驱动方法在模型训练过程中需要大量标签样本数据的问题,提出了一种新的基于扩散模型和双向长短期记忆网络的锂电池SOH估计方法。【方法】首先,建立电池充电时间、电压和温度三者间的长期依赖关系云图;其次,设计一个时空信息捕捉模块,将该模块捕获的长期依赖信息作为扩散模型的生成条件,赋予扩散模型电池SOH数据生成能力;最后,利用双向长短期记忆网络(Bi-LSTM)对部分由原始数据和生成数据混合而成的电池数据集进行训练,并利用剩余的原始数据作为测试集对所提方法进行验证。【结果】验证结果表明,该方法不仅可以减少收集电池数据类型的周期和成本,而且能够有效预测电池SOH。【结论】该方法在电池SOH估计上具备良好的精度,可进一步探索其他电池数据集组合,优化模型结构,提高电池管理系统。
[Purposes]Accurate predictive assessment of the state of health(SOH)of lithium batteries can improve the safety of battery devices and reduce the risk of failure.To solve the problem that the datadriven method requires a large amount of label sample data in the process of model training,a new bat⁃tery SOH estimation method is proposed.[Methods]Firstly,the long-term dependence of battery charg⁃ing time,voltage and temperature was established.Then,a spatiotemporal perception module is de⁃signed,and the long-term dependent information captured by the module is used as the generation condi⁃tion of diffusion model,and the SOH data generation capability is given to the battery of diffusion model.Lastly,bidirectional long short-term memory(Bi-LSTM)network is used to train part of the original and generated hybrid battery data set,and the remaining raw data is used as a test set to verify the method.[Findings]The verification results show that this method can effectively predict SOH while reducing the cycle and cost of collecting battery data types.[Conclusions]This method has a good accuracy in SOH estimation,and can further explore other battery data set combinations,optimize the model structure,and improve the battery management system.
作者
柯欢
KE Huan(School of Statistics,University of International Business and Economics,Beijing 100029,China)
出处
《河南科技》
2024年第19期5-11,共7页
Henan Science and Technology
