大数据下锂电池健康状态运维挖掘模型仿真
Simulation of Lithium Battery Health State Operation and Maintenance Mining Model under Big Data
王红林 1张伽浩1
作者信息
- 1. 南京信息工程大学,江苏 南京 210044
- 折叠
摘要
对锂电池健康状态的运维挖掘,可以实时掌控锂电池的当前健康状态,保障用电设备正常使用.提出大数据下锂电池健康状态运维挖掘模型,分析模型整体架构,采用CNN-LSTM神经网络模型进行锂电池健康状态估算.首先选取十个备选锂电池健康特征因子,然后进行灰色关联分析法和主成分分析,提取出五个健康特征因子,分别为循环次数、放电阶段平均温度、充电阶段平均温度、放电总时间、和恒定电流放电时间;然后利用卷积神经网络提取健康特征因子的局部特征,长短期神经网络挖掘时间序列特征因子,构建CNN-LSTM组合神经网络模型;最后利用NASA锂电池数据集进行仿真,预测锂电池健康状态.实验结果表明,所提的组合网络模型具有较高的预测精度,相较于LSTM和BP 单一网络模型,平均绝对误差降低 37%和 62%,均方根误差降低了17%和39%.
Abstract
The operation and maintenance mining of lithium battery health status can control the current health status of lithium battery in real time and ensure the normal use of power-using equipment.We proposed an O&M mining model for lithium battery health status under big data,analyzed the overall architecture of the model,and used CNN-LSTM neural network model for lithium battery health status estimation.Firstly,ten alternative lithium battery health feature factors were selected,and then gray correlation analysis and principal component analysis were per-formed to extract five health feature factors,which are cycle number,average temperature in discharge phase,average temperature in charge phase,total discharge time,and constant current discharge time;Then convolutional neural net-work was used to extract local features of health feature factors,and long and short-term neural network was used to mine time series feature The CNN-LSTM combined neural network model was constructed by using convolutional neu-ral network to extract local features of health feature factors and long and short term neural network to mine time se-ries feature factors.The experimental results show that the proposed combined network model has high prediction ac-curacy,and the average absolute error is reduced by 37%and 62%,and the root mean square error is reduced by 17%and 39%compared with the LSTM and BP single network models.
关键词
锂电池/神经网络/大数据Key words
Lithium battery/CNN-lstm/Bigdata引用本文复制引用
基金项目
国家自然科学基金委员会青年项目(62101274)
出版年
2024
NETL
NSTL
万方数据