基于优化VMD-SSA-LSTM算法的锂离子电池RUL预测
RUL Prediction for Li-ion Batteries Based on Optimized VMD-SSA-LSTM Algorithm
朱宗玖 1顾发慧1
作者信息
- 1. 安徽理工大学电气与信息工程学院,安徽 淮南 232001
- 折叠
摘要
目的 为了避免锂电池在使用的过程中可能会出现容量虚假回升现象,从而导致电池在超出退化标准后继续使用造成风险.方法 提出基于鲸鱼优化算法(WOA)、分模态分解(VMD)、麻雀搜索算法(SSA)和长短时记忆神经网络(LSTM)的组合预测算法对锂离子电池剩余寿命(RUL)进行预测.首先对于变分模态分解模态数K和惩罚因子a以往需要凭经验确定的问题,提出使用WOA对VMD的两个参数进行寻优.其次将原始容量退化数据根据上一步确定的参数进行模态分解,得到有限个模态分量.由于经过分解过后得到的残差分量的起伏性较大,因此将其作为其中的一个分量.最后,使用SSA优化LSTM的超参数,并对得到的模态分量和残差分量进行预测,并将预测的各个分量重构得到预测结果.结果 采用NASA PCoE实验室公开的锂电池失效数据集进行实验,验证了所提出的WOA-VMD-SSA-LSTM优化算法相较于其他2种优化算法,在平均绝对误差(MAE)、均方根误差(RMSE)和平均相对百分误差(MAPE)3项评价标准中都是最低,且MAPE小于1％.结论 该优化算法对于锂电池RUL预测具有不错的精度和稳定性,为锂电池RUL预测提供了一种新的预测模型的同时,也为VMD超参数的选择和确定提供了 一种新方法.
Abstract
Objective In order to avoid the phenomenon of false regain capacity that may occur during the use of lithium batteries,which may lead to the risk of continued use of the battery beyond the degradation criteria.Meth-ods The prediction algorithm based on the Whale Optimization Algorithm(WOA),Variational Mode Decomposi-tion(VMD),Sparrow Search algorithm(SSA)and Long Short-Term Memory(LSTM)coupling was proposed to predict the Remaining Useful Life(RUL)of Li-ion batteries.Firstly,the use of WOA was proposed to optimise the modal number K and penalty factor a in the variable modal decomposition(VMD),which used to be deter-mined empirically.Secondly,the original capacity degradation data were modally decomposed according to the pa-rameters determined in the previous step to obtain a finite number of modal components.The residual component obtained after the decomposition was used as one of the components due to its large undulation.Finally,the hy-perparameters of the LSTM are optimised by using SSA and the obtained modal and residual components were predicted and the predictions were obtained by reconstructing each of the predicted components.Results Experi-ments were conducted by using the publicly available lithium battery failure dataset from NASA PCoE Laborato-ry,and it was verified that the proposed WOA-VMD-SSA-LSTM optimisation algorithm had the lowest errors in the three evaluation criteria of Mean Absolute Error(MAE),Root Mean Square Error(RMSE),and Mean Rela-tive Percentage Error(MAPE),with the MAPE less than 1％compared with the other 2 optimisation algorithms.Conclusion The optimisation algorithm has good accuracy and stability for lithium battery RUL prediction,and provides a new prediction model for lithium battery RUL prediction,at the same time,it also providing a new method for the selection and determination of VMD hyperparameters.
关键词
RUL预测/VMD/锂离子电池/LSTM/SSAKey words
RUL forecast/VMD/lithium-ion batteries/LSTM/SSA引用本文复制引用
基金项目
安徽省自然科学基金资助项目(1808085MF169)
安徽高校自然科学研究项目(KJ2018A0086)
出版年
2024
NETL
NSTL
万方数据