融合小波与MS-TCN的锂电池热失控自适应预警

吴文进; 沙晗; 郑鹏贤; 詹文法; 吴晶

doi:10.19799/j.cnki.2095-4239.2026.0019

您当前的位置：

首页 >

文章列表页 >

融合小波与MS-TCN的锂电池热失控自适应预警

储能系统与工程 | 更新时间：2026-08-26

- 融合小波与MS-TCN的锂电池热失控自适应预警
- Adaptive thermal runaway early warning for Li-ion batteries via wavelet and MS-TCN fusion
- 储能科学与技术 2026年15卷第8期页码：3143-3152
- 作者机构：
  
  安庆师范大学电子工程与智能制造学院，安徽安庆 246133
- 作者简介：
  
  吴文进（1975—），男，博士，教授，研究方向为电力电子系统及其控制、电能储存与节电技术，E-mail：wuwenjinaq@163.com。
- 基金信息：
  
  国家自然科学基金重大研究计划项目资助(92573109);国家自然科学基金面上项目资助(62474002);安徽高校协同创新项目资助(GXXT-2021-025)
- DOI：10.19799/j.cnki.2095-4239.2026.0019
  中图分类号： TM 912
- 收稿：2026-01-08，
  
  修回：2026-02-01，
  
  网络首发：2026-08-26，
  
  纸质出版：2026-08-28
- 稿件说明：
移动端阅览
吴文进, 沙晗, 郑鹏贤, 等. 融合小波与MS-TCN的锂电池热失控自适应预警[J]. 储能科学与技术, 2026, 15(8): 3143-3152.

WU Wenjin, SHA Han, ZHENG Pengxian, et al. Adaptive thermal runaway early warning for Li-ion batteries via wavelet and MS-TCN fusion[J]. Energy Storage Science and Technology, 2026, 15(8): 3143-3152.
吴文进, 沙晗, 郑鹏贤, 等. 融合小波与MS-TCN的锂电池热失控自适应预警[J]. 储能科学与技术, 2026, 15(8): 3143-3152. DOI： 10.19799/j.cnki.2095-4239.2026.0019.

WU Wenjin, SHA Han, ZHENG Pengxian, et al. Adaptive thermal runaway early warning for Li-ion batteries via wavelet and MS-TCN fusion[J]. Energy Storage Science and Technology, 2026, 15(8): 3143-3152. DOI： 10.19799/j.cnki.2095-4239.2026.0019.

摘要

针对锂离子电池热失控早期微短路（ISC）信号极易被动态负载与环境噪声掩盖，以及传统固定阈值法在复杂工况下面临严重“报警疲劳”与漏报的难题，设计并验证了一种融合小波特征增强、多尺度时域卷积网络（MS-TCN）与自适应阈值的早期预警方法。首先，分析了微短路演化过程中的多维物理特征，针对初期电压跌落特征不明显的缺陷，构建了物理信息增强的特征工程；选用具有二阶消失矩与良好平滑性的db4小波基，对电压信号进行3层离散小波分解，通过提取高频细节能量并耦合表面温度趋势，显著增强了微弱故障信号在强噪声背景下的信噪比。其次，构建了基于指数增长膨胀因果卷积的MS-TCN架构，利用其线性计算复杂度与超大感受野优势，替代传统LSTM网络，实现了对电压秒级瞬态突变与温度分钟级热积累趋势的并行捕捉，建立了高精度的电池正常行为预测基准。进而，提出了基于指数加权移动平均（EWMA）的自适应决策策略，摒弃固定阈值，根据预测残差的实时统计特性动态调整报警边界（灵敏度系数

=3），实现了对异常的连续性判定。实验结果表明：基于欧盟JRC感应加热热失控数据集的验证显示，该方法在同构软包电池上的

1分数高达0.9784，平均提前预警时间约为29 s；与Transformer及CNN模型相比，

1分数分别提升了8.68%和9.15%，且计算效率更高。研究还发现，在跨封装类型（软包迁移至圆柱）的少样本场景下，本方法仍保持0.9554的

1分数与零误报率。结论指出，本方法摆脱了对电流传感器的依赖，有效解决了宽温域工况下误报与漏报难以平衡的问题，为电动汽车与储能电站的安全监控提供了高鲁棒性的解决方案。

Abstract

Early internal short circuit (ISC) signals in Li-ion batteries are easily masked by dynamic loads and environmental noise

which subjects traditional fixed-threshold methods to severe "alarm fatigue" and missed detections under complex conditions. To address

these challenges

this study designed and validated an early warning framework fusing wavelet feature enhancement

a multi-scale temporal convolutional network (MS-TCN)

and adaptive thresholding. First

the multidimensional physical characteristics during ISC evolution were analyzed. A feature engineering approach enhanced with physical information enhancement was constructed to address unobvious initial voltage drops. Specifically

the Daubechies 4 wavelet base

characterized by a second-order vanishing moment and good smoothness

was selected to perform a three-level discrete wavelet decomposition on the voltage signals. Extracting the high-frequency detail energy and coupling it with surface-temperature trends significantly improved the signal-to-noise ratio of weak fault features under intense background noise. Second

an MS-TCN architecture based on exponentially increasing dilated causal convolutions was developed. This model

leveraging its linear computational complexity and extensive receptive field

was employed instead of traditional long short-term memory networks to simultaneously capture transient voltage mutations (seconds scale) and thermal-accumulation trends (minutes scale)

establishing a high-precision baseline for normal battery behavior. Furthermore

an adaptive decision strategy based on exponentially weighted moving averages was proposed. By dynamically adjusting alarm boundaries based on the real-time statistical properties of prediction residuals (sensitivity coefficient = 3)

this strategy achieved continuous anomaly determination without relying on fixed thresholds. Experimental results on the European Commission Joint Research Centre dataset demonstrate that the proposed method achieved an

1-score of 0.9784 on pouch cells

providing an average warning lead time of approximately 29 s. Compared with transformer and convolutional neural network models

the

1-scores were improved by 8.68% and 9.15%

respectively

while exhibiting higher computational e

fficiency. Crucially

in cross-package transfer scenarios (from pouch to cylindrical cells) comprising limited samples

the method maintained an

1-score of 0.9554 and a zero-false-alarm rate. In conclusion

this method eliminated the dependency on current sensors and effectively resolved the trade-off between false alarms and missed detections across wide temperature ranges

providing a highly robust solution for safety monitoring in electric vehicles and energy storage stations.

关键词

Keywords

references

郭翠静, 褚永金, 刘庆. 锂离子电池表面形变影响因素分析及在热失控预警中的应用[J]. 储能科学与技术, 2025, 14(10): 3990-3995.

GUO C J, CHU Y J, LIU Q. Analysis of factors affecting surface deformation of lithium-ion battery and its application in thermal runaway early warning[J]. Energy Storage Science and Technology, 2025, 14(10): 3990-3995.

ZHANG X, CHEN S, ZHU J Z, et al. A critical review of thermal runaway prediction and early-warning methods for lithium-ion batteries[J]. Energy Material Advances, 2023, 4: 8. DOI:10.34133/energymatadv.0008.

李昌豪, 曹志成, 汪书苹, 等. 基于时序模式注意力机制和孤立森林的电池过充热失控预警方法研究[J]. 储能科学与技术, 2025, 14(10): 3742-3754.

LI C H, CAO Z C, WANG S P, et al. Warning method for battery-overcharge thermal runaway based on a temporal pattern attention mechanism and an isolated forest algorithm[J]. Energy Storage Science and Technology, 2025, 14(10): 3742-3754.

WANG Y B, ZHANG C P, HU J, et al. Research on internal short circuit detection method for lithium-ion batteries based on battery expansion characteristics[J]. Journal of Power Sources, 2023, 587: 233673. DOI:10.1016/j.jpowsour.2023.233673.

REN S, SUN J. Multi-fault diagnosis strategy based on a non-redundant interleaved measurement circuit and improved fuzzy entropy for the battery system[J]. Energy, 2024, 292: 130603. DOI:10.1016/j.energy.2024.130603.

WANG S H, WANG Z P, CHENG X M, et al. A double-layer fault diagnosis strategy for electric vehicle batteries based on Gaussian mixture model[J]. Energy, 2023, 281: 128318. DOI:10. 1016/j.energy.2023.128318.

CHANG C, WANG Q Y, JIANG J C, et al. Voltage fault diagnosis of a power battery based on wavelet time-frequency diagram[J]. Energy, 2023, 278: 127920. DOI:10.1016/j.energy.2023.127920.

ZHAO K, LIU Y, ZHOU Y, et al. Digital twin-supported battery state estimation based on TCN-LSTM neural networks and transfer learning[J]. CSEE Journal of Power and Energy Systems, 2025, 11(2): 567-579. DOI:10.17775/CSEEJPES. 2024.00900.

LI M R, DONG C Y, WANG R, et al. Synthetic thermal convolutional-memory network for the lithium-ion battery behaviour diagnosis against noise interruptions[J]. IET Energy Systems Integration, 2023, 5(1): 29-39. DOI:10.1049/esi2.12080.

GHAZALI A K, AZIZ N A A, HASSAN M K. Advanced algorithms in battery management systems for electric vehicles: A comprehensive review[J]. Symmetry, 2025, 17(3): 321. DOI:10. 3390/sym17030321.

张朝龙, 陈阳, 刘梦玲, 等. 一种基于ICA-T特征和CNN-LA-BiLSTM的锂离子电池健康状态估计方法[J]. 储能科学与技术, 2025, 14(3): 1258-1269.

ZHANG C L, CHEN Y, LIU M L, et al. A state of health estimation method for lithium-ion batteries using ICA-T features and CNN-LA-BiLSTM[J]. Energy Storage Science and Technology, 2025, 14(3): 1258-1269.

黄强, 陈一童, 王备, 等. 融合多尺度TCN与动态注意力的电池容量预测[J]. 电池, 2025, 55(5): 1034-1041. DOI:10.19535/j.1001-1579.2025.05.023.

HUANG Q, CHEN Y T, WANG B, et al. Prediction of battery capacity based on multi-scale TCN and dynamic attention[J]. Battery Bimonthly, 2025, 55(5): 1034-1041. DOI:10.19535/j.1001-1579.2025.05.023.

DONG Z K, GU S Y, ZHOU S Q, et al. Periodic segmentation transformer-based internal short circuit detection method for battery packs[J]. IEEE Transactions on Transportation Electrification, 2025, 11(1): 3655-3666. DOI:10.1109/TTE. 2024.3444453.

ANSARI S, AYOB A, HOSSAIN LIPU M S, et al. Data-driven remaining useful life prediction for lithium-ion batteries using multi-charging profile framework: A recurrent neural network approach[J]. Sustainability, 2021, 13(23): 13333. DOI:10.3390/su132313333.

LI S P, ZHANG H, MISUL D A, et al. Battery temperature anomaly early warning for electric vehicles under real driving conditions using a temporal convolutional network[J]. eTransportation, 2025, 25: 100445. DOI:10.1016/j.etran.2025. 100445.

CHENG Y M, GAO D X, ZHAO F M, et al. A thermal runaway early warning method for electric vehicles based on hybrid neural network model[J]. Journal of Electrical Engineering & Technology, 2024, 19(6): 3735-3748. DOI:10.1007/s42835-024-01825-8.

胡杰, 王浩杰, 魏敏, 等. 基于TCN-LSTM模型的实车电池健康状态评估方法[J]. 汽车工程, 2025, 47(6): 1060-1071.

HU J, WANG H J, WEI M, et al. A TCN-LSTM model-based approach for real vehicle battery health state evaluation[J]. Automotive Engineering, 2025, 47(6): 1060-1071.

周航, 程泽, 弓清瑞, 等. 基于TCN编码的锂离子电池SOH估计方法[J]. 湖南大学学报(自然科学版), 2023, 50(4): 185-192. DOI:10.16339/j.cnki.hdxbzkb.2023228.

ZHOU H, CHENG Z, GONG Q R, et al. SOH estimation method of lithium-ion battery based on TCN encoding[J]. Journal of Hunan University (Natural Sciences), 2023, 50(4): 185-192. DOI:10.16339/j.cnki.hdxbzkb.2023228.

叶震, 李琨, 李梦男, 等. 改进EWMA和多通道混合模型的锂电池容量预测[J]. 机械科学与技术, 2025, 44(3): 513-520. DOI:10.13433/j.cnki.1003-8728.20230185.

YE Z, LI K, LI M N, et al. Lithium battery capacity prediction using improved exponentially weighted moving average and multi-channel hybrid model[J]. Mechanical Science and Technology for Aerospace Engineering, 2025, 44(3): 513-520. DOI:10.13433/j.cnki.1003-8728.20230185.

ZHANG X Y, YANG W H, YAN L S, et al. Adaptive internal short-circuit fault detection for lithium-ion batteries of electric vehicles[J]. Journal of Energy Storage, 2024, 84: 110874. DOI:10.1016/j.est.2024.110874.

KRISTON A, KERSYS A, ANTONELLI A, et al. Initiation of thermal runaway in Lithium-ion cells by inductive heating[J]. Journal of Power Sources, 2020, 454: 227914. DOI:10.1016/j.jpowsour.2020.227914.

ZHAO J Y, FENG X N, WANG J B, et al. Battery fault diagnosis and failure prognosis for electric vehicles using spatio-temporal transformer networks[J]. Applied Energy, 2023, 352: 121949. DOI:10.1016/j.apenergy.2023.121949.

SHEN D X, LYU C, YANG D Z, et al. Concurrent multi-fault diagnosis of lithium-ion battery packs using random convolution kernel transformation and Gaussian process classifier[J]. Energy, 2024, 306: 132467. DOI:10.1016/j.energy.2024.132467.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

功率曲线分解方法研究综述

高温固体颗粒储热耦合燃煤机组深度灵活调峰技术研究进展

S掺杂/Li2SO4包覆富锂锰基正极材料的电化学性能研究

不同应用场景下超级电容储能经济性对比研究