多尺度特征融合的锂离子电池循环寿命及拐点预测  

Prediction of cycle life and knee points in lithium-ion batteries through multi-scale feature fusion

作　　者：史永胜[1] 胡玙珺 翟欣然 SHI Yongsheng;HU Yujun;ZHAI Xinran(School of Electrical and Control Engineering,Shaanxi University of Science and Technology,Xi'an Shaanxi,710021,China)

机构地区：[1]陕西科技大学电气与控制工程学院,陕西西安710021

出　　处：《电源技术》2023年第12期1603-1608,共6页Chinese Journal of Power Sources

基　　金：国家自然科学基金项目(22279076);陕西省科技厅工业科技攻关计划项目(2019GY-175)。

摘　　要：锂离子电池在反复充放电过程中容量逐渐衰减,到达拐点后容量急剧衰减至寿命结束。拐点的出现可能导致电池故障或寿命缩短等问题,为了确保系统的安全使用,提出了一种基于多尺度特征融合模型预测循环寿命及拐点。首先,使用多个扩张率的卷积神经网络(CNN)从电压、电流和温度等数据中提取不同时间尺度的健康特征,并通过长短期记忆神经网络(LSTM)挖掘特征的时间序列依赖关系。其次,对电池早期和后期的衰退轨迹拟合以识别拐点。最后,利用Stanford-MIT数据集验证模型的有效性和准确性,结果显示该模型使用前80个循环数据能准确预测电池循环寿命和拐点,误差RMSE分别低于30和60次,MAPE分别低于2%和5%。准确预测拐点有利于电池性能和寿命的改进,对电池健康管理至关重要。Lithium-ion batteries exhibit capacity decay during charge-discharge cycles,followed by a sharp decline near end-of-life.Knee points can cause malfunctions and premature aging,necessitating the development of a multiscale feature fusion approach for the prediction of cycle life and knee points in lithium-ion batteries,ensuring the safe utilization of such systems.Primarily,a multi-scale convolutional neural network(CNN)was employed to extract health-related features from raw data,including discharge voltage,current,and temperature,spanning various temporal scales.Subsequently,the long short-term memory(LSTM)neural network was employed to capture the long-term dependencies of these features.Furthermore,fitting of degradation trajectories in the early and late stages of battery operation enabled the identification of knee points.Lastly,the effectiveness and accuracy of the proposed model were validated using the Stanford-MIT dataset.The results demonstrate that the model,utilizing the first 80 cycles of data,accurately predicts both the cycle life and turning points of the battery.The RMSE for cycle life prediction is below 30 cycles,while for knee points,it is below 60 cycles.Additionally,the MAPE is below 2%for cycle life prediction and below 5%for knee points.Accurate prediction of knee points proves advantageous for continuous improvement of battery performance and lifespan,making it vital for effective battery health management.

关 键 词：锂离子电池 寿命预测 拐点 卷积神经网络 长短期记忆神经网络 

分 类 号：TM912.9[电气工程—电力电子与电力传动]