Prediction-evaluation framework for anomaly detection in electric vehicle lithium-ion battery

doi:10.11772/j.issn.1001-9081.2025050574

Abstract

Abstract:

To address the challenges of high complexity in multi-source heterogeneous time-series data， scarcity of anomaly samples， and strong inter-variable dependencies in lithium-ion battery fault detection for electric vehicles， a prediction-evaluation framework based on Dynamic Transformer Memory autoencoder for Anomaly Detection （DTMAD） was proposed to enhance fault identification accuracy and model generalization capability. Firstly， a joint feature encoder was designed by integrating Dynamical autoencoder for Anomaly Detection （DyAD） with Gated Recurrent Unit （GRU） to perform feature fusion and dimensionality reduction on multi-source time-series data， extracting deep cross-modal representations. Simultaneously， a pre-response encoder based on a self-attention mechanism was constructed to capture long-term dependencies in time-series data， enhancing the efficiency and accuracy of feature extraction. Furthermore， a memory parsing module was introduced， which fused predicted path with the actual response path via residual contrastive learning， improving the model's capability to detect anomalous patterns. Secondly， based on the distribution characteristics of reconstruction error， an evaluation model was designed through a collaborative anomaly detection algorithm. Finally， through the comprehensive prediction-evaluation framework， key response patterns were extracted from complex multi-source data， and latent anomalies were identified under unsupervised learning conditions. Experimental results on multi-group and multi-source electric-vehicle lithium-ion battery datasets demonstrate that the proposed framework significantly outperforms baseline methods such as AutoEncoder （AE）， Deep Support Vector Data Description （DeepSVDD）， and Graph Deviation Network （GDN） in terms of detection accuracy and model robustness. Notably， compared to the DyAD model， the DTMAD model achieves an Area Under the Receiver Operating Characteristic curve （AUROC） of 0.900 8， with result variability reduced from 0.029 to 0.026， indicating superior detection stability and generalization capability.

Key words: lithium-ion battery, fault detection, feature fusion, Variational AutoEncoder (VAE)

摘要：

针对电动汽车锂离子电池故障检测中多源异构时序数据复杂性高、异常样本稀缺及多变量关联性强的挑战，提出一种基于动态变换记忆自编码器的预测-评估故障检测框架（DTMAD），以提升故障识别准确性与模型泛化能力。首先，设计融合动态自编码器（DyAD）与门控循环单元（GRU）的联合特征编码器，对多源时序数据进行特征融合与降维处理，提取跨模态深层特征表示；同时，构建基于自注意力机制的预响应编码器，捕捉时序数据中的长期依赖关系，提升特征提取效率与精度；进一步地，引入记忆解析模块，通过残差对比学习机制融合预测路径与实际响应路径，增强模型对异常模式的检测能力。其次，基于重构误差的分布特性，通过协同异常检测算法设计评估模型。最后，通过综合的预测-评估框架，在无监督学习条件下从多源数据中提取关键响应模式并识别潜在异常。在多组多源电动汽车锂离子电池数据集上的实验结果表明，所提框架的故障检测准确率和模型稳定性均优于对比的编码器（AE）、深度支持向量数据描述（DeepSVDD）与图偏差网络（GDN）等。其中，相较于DyAD模型，DTMAD模型的接收者操作特征曲线下面积（AUROC）提升至0.900 8，且结果波动幅度由0.029降至0.026，展现出更高的检测稳定性与泛化能力。

关键词: 锂离子电池, 故障检测, 特征融合, 变分自编码器

CLC Number:

TP277

Xuechao LIAO, Rui CHEN. Prediction-evaluation framework for anomaly detection in electric vehicle lithium-ion battery[J]. Journal of Computer Applications, 2026, 46(5): 1614-1623.

廖雪超, 陈睿. 电动汽车锂离子电池预测-评估故障检测框架[J]. 《计算机应用》唯一官方网站, 2026, 46(5): 1614-1623.

Figures/Tables 18

Fig. 1 Prediction-evaluation framework

Tab. 1 Three types of loss functions

训练函数	定义	计算公式
MseLoss	均方误差	$1 N ∑ i = 1 N (y i - y ̂ i) 2$
MaeLoss	平均绝对误差	$1 N ∑ i = 1 n y i - y ̂$
SmoothLoss	光滑损失	$0.5 x 2, x < 1 x - 0.5, 其他$

Tab. 1 Three types of loss functions

训练函数	定义	计算公式
MseLoss	均方误差	$1 N ∑ i = 1 N (y i - y ̂ i) 2$
MaeLoss	平均绝对误差	$1 N ∑ i = 1 n y i - y ̂$
SmoothLoss	光滑损失	$0.5 x 2, x < 1 x - 0.5, 其他$

Tab. 2 Feature mapping

特征	英文	中文	特征
X1	SoC	剩余电池电量	输入特征
X2	current	电流	输入特征
Y1	Min_temp	最小温度	响应特征
Y2	Max_single_volt	最大单体电压	响应特征
Y3	Max_temp	最大温度	响应特征
Y4	Min_single_volt	最小单体电压	响应特征
Y5	volt	电压	响应特征
Y6	mileage	里程	弱监督特征

Fig. 2 Random sampling results of temperature distribution

Fig. 3 Sampling results of charging time slices

Fig. 4 Feature correlation coefficient matrix

Tab. 3 Model parameter settings

模型设置	参数设置	模型设置	参数设置
优化器	AdamW	隐藏层大小	128
优化器参数	0.1	VAE特征维度	16
学习率	0.01	噪声比例	0.01
训练轮数	20	dropout	0.1
w₃	0.1	多头头数	5
批次大小	128	前馈网络层数	1 024
RNN网络类型	双向GRU	潜在空间层数	16
w₁	10	w₂	0.001

Tab. 4 Average training losses of different loss functions

模型	MseLoss	MaeLoss	SmoothLoss
AE	0.134 47	0.274 87	0.066 63
DeepSVDD	1.958 68	2.058 92	1.985 98
DyAD	0.075 50	0.426 73	0.034 20
DTAD	0.092 13	0.470 81	0.033 44
DMAD	0.073 71	0.389 50	0.037 10
DTMAD	0.069 96	0.499 19	0.029 25

Fig. 5 Training process of multi-dimensional losses

Tab. 5 Total loss comparison of each model

模型	主损失 $l o s s S M$	正则化损失 $l o s s K L$	里程损失 $l o s s M I L E$	综合损失 $l o s s T O T A L$
DyAD	0.005 31	0.168 99	9.384 19	0.079 45
DTAD	0.005 56	0.155 06	9.358 67	0.080 48
DMAD	0.005 46	0.146 05	9.355 63	0.078 61
DTMAD	0.004 92	0.154 87	9.322 03	0.074 02

Tab. 5 Total loss comparison of each model

模型	主损失 $l o s s S M$	正则化损失 $l o s s K L$	里程损失 $l o s s M I L E$	综合损失 $l o s s T O T A L$
DyAD	0.005 31	0.168 99	9.384 19	0.079 45
DTAD	0.005 56	0.155 06	9.358 67	0.080 48
DMAD	0.005 46	0.146 05	9.355 63	0.078 61
DTMAD	0.004 92	0.154 87	9.322 03	0.074 02

Tab. 6 Model performance under different hyperparameters

隐藏层大小	综合损失	AUROC	批次大小	综合损失	AUROC
32	0.100 6	0.847	32	0.114 8	0.826
64	0.096 1	0.841	64	0.107 9	0.861
128	0.075 6	0.913	128	0.075 6	0.913
256	0.094 1	0.909	256	0.091 9	0.888

Fig. 6 Comparison of predicted trends for different target features

Fig. 7 Comparison of reconstruction error

Tab. 7 Comparison of total time consumption

模型	第1次	第2次	第3次	第4次	平均耗时
DyAD	708.7	709.4	716.4	680.9	703.875
DTAD	687.5	662.7	770.5	746.9	716.900
DMAD	781.6	752.1	779.3	770.8	770.925
DTMAD	772.2	802.9	750.4	794.6	780.025

Tab. 8 Confusion matrix

真实值	预测为正	预测为负
实际正例	TP	FN
实际负例	FP	TN

Tab. 9 AUROC comparison under different combinations of response features

特征	AUROC	特征	AUROC	特征	AUROC
Y1	0.865	Y12	0.869	Y123	0.872
Y2	0.901	Y13	0.871	Y124	0.914
Y3	0.871	Y14	0.924	Y125	0.895
Y4	0.922	Y15	0.872	Y134	0.890
Y5	0.880	Y23	0.869	Y135	0.874
Y2345	0.923	Y24	0.884	Y145	0.884
Y1345	0.905	Y25	0.914	Y234	0.907
Y1245	0.906	Y34	0.897	Y235	0.865
Y1235	0.909	Y35	0.869	Y245	0.884
Y1234	0.895	Y45	0.881	Y345	0.885

Tab. 10 AUROC comparison of models

模型	AUROC	模型	AUROC
AE	0.668 6	DyAD	0.869 1±0.029 1
DeepSVDD	0.684 1	DTAD	0.887 3±0.031 4
GDN	0.801 8	DMAD	0.885 9±0.036 7
GP	0.706 3	DTMAD	0.900 8±0.026 1

Fig. 8 Comparison of AUROC

References 27

[1]	WASEEM M， AHMAD M， PARVEEN A， et al. Battery technologies and functionality of battery management system for EVs： current status， key challenges， and future prospectives［J］. Journal of Power Sources， 2023， 580： No.233349.
[2]	HUANG L， LIU L， LU L， et al. A review of the internal short circuit mechanism in lithium-ion batteries： inducement， detection and prevention［J］. International Journal of Energy Research， 2021， 45（11）： 15797-15831.
[3]	KONG D， LV H， PING P， et al. A review of early warning methods of thermal runaway of lithium ion batteries［J］. Journal of Energy Storage， 2023， 64： No.107073.
[4]	TANG T， YANG X， LI M， et al. Deep learning model-based real-time state-of-health estimation of lithium-ion batteries under dynamic operating conditions［J］. Energy， 2025， 317： No.134697.
[5]	YIN Y， XU G， XIE Y， et al. Utilizing deep learning for crystal system classification in lithium-ion batteries［J］. Journal of Theory and Practice of Engineering Science， 2024， 4（3）： 199-206.
[6]	NIU D， SONG D. Model-based robust fault diagnosis of incipient ITSC for PMSM in elevator traction system［J］. IEEE Transactions on Instrumentation and Measurement， 2023， 72： No.3533512.
[7]	MASOUMI Z， MOAVENI B， GAZAFRUDI S M M， et al. Signal-model-based fault diagnosis in windings of synchronous generator［J］. IEEE Transactions on Industrial Informatics， 2023， 19（3）： 2942-2951.
[8]	PURBOWASKITO W， LAN C Y， FUH K. The potentiality of integrating model-based residuals and machine-learning classifiers： an induction motor fault diagnosis case［J］. IEEE Transactions on Industrial Informatics， 2024， 20（2）： 2822-2832.
[9]	LIU Y， CHEN Z， WEI L， et al. Braking sensor and actuator fault diagnosis with combined model-based and data-driven pressure estimation methods［J］. IEEE Transactions on Industrial Electronics， 2023， 70（11）： 11639-11648.
[10]	CEN J， YANG Z， LIU X， et al. A review of data-driven machinery fault diagnosis using machine learning algorithms［J］. Journal of Vibration Engineering and Technologies， 2022， 10（7）： 2481-2507.
[11]	GUNERKAR R S， JALAN A K， BELGAMWAR S U. Fault diagnosis of rolling element bearing based on artificial neural network［J］. Journal of Mechanical Science and Technology， 2019， 33（2）： 505-511.
[12]	陈鑫，肖明清，文斌成，等.基于变分模态分解和混沌麻雀搜索算法优化支持向量机的滚动轴承故障诊断［J］.计算机应用，2021，41（S2）：118-123.
	CHEN X， XIAO M Q， WEN B C， et al. Rolling bearing fault diagnosis based on variational mode decomposition， chaotic sparrow search algorithm and support vector machine［J］. Journal of Computer Applications， 2021， 41（S2）： 118-123.
[13]	CHEN Z， ZHOU D， ZIO E， et al. Adaptive transfer learning for multimode process monitoring and unsupervised anomaly detection in steam turbines［J］. Reliability Engineering and System Safety， 2023， 234： No.109162.
[14]	MACULOTTI G， GENTA G， QUAGLIOTTI D， et al. Gaussian process regression‐based detection and correction of disturbances in surface topography measurements［J］. Quality and Reliability Engineering International， 2022， 38（3）： 1501-1518.
[15]	ZHANG Z， DENG X. Anomaly detection using improved deep SVDD model with data structure preservation［J］. Pattern Recognition Letters， 2021， 148： 1-6.
[16]	DENG A， HOOI B. Graph neural network-based anomaly detection in multivariate time series［C］// Proceedings of the 35th AAAI Conference on Artificial Intelligence. Palo Alto： AAAI Press， 2021： 4027-4035.
[17]	TANG C， XU L， YANG B， et al. GRU-based interpretable multivariate time series anomaly detection in industrial control system［J］. Computers and Security， 2023， 127： No.103094.
[18]	ZHANG X， LIU P， LIN N， et al. A novel battery abnormality detection method using interpretable Autoencoder［J］. Applied Energy， 2023， 330（Pt B）： No.120312.
[19]	ZHANG Y， WANG J， CHEN Y， et al. Adaptive memory networks with self-supervised learning for unsupervised anomaly detection［J］. IEEE Transactions on Knowledge and Data Engineering， 2023， 35（12）： 12068-12080.
[20]	SUN C， HE Z， LIN H， et al. Anomaly detection of power battery pack using gated recurrent units based variational autoencoder［J］. Applied Soft Computing， 2023， 132： No.109903.
[21]	ZHANG J， WANG Y， JIANG B， et al. Realistic fault detection of li-ion battery via dynamical deep learning［J］. Nature Communications， 2023， 14： No.5940.
[22]	LI Y， CHEN W， CHEN B， et al. Prototype-oriented unsupervised anomaly detection for multivariate time series［C］// Proceedings of the 40th International Conference on Machine Learning. New York： JMLR.org， 2023： 19407-19424.
[23]	XU J， WU H， WANG J， et al. Anomaly Transformer： time series anomaly detection with association discrepancy［EB/OL］. ［2024-03-22］..
[24]	郭秋亚，张兆功，胡本然，等. 数据融合在能源互联网故障诊断中的应用［J］. 计算机应用， 2024， 44（S2）： 309-315.
	GUO Q Y， ZHANG Z G， HU B R， et al. Application of data fusion in fault diagnosis of energy Internet［J］. Journal of Computer Applications， 2024， 44（S2）： 309-315.
[25]	TULI S， CASALE G， JENNINGS N R. TranAD： deep Transformer networks for anomaly detection in multivariate time series data［EB/OL］. ［2024-03-22］..
[26]	YU X， ZHANG K， LIU Y， et al. Adversarial Transformer-based anomaly detection for multivariate time series［J］. IEEE Transactions on Industrial Informatics， 2025， 21（3）： 2471-2480.
[27]	YANG Y， ZHANG C， ZHOU T， et al. DCdetector： dual attention contrastive representation learning for time series anomaly detection［C］// Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. New York： ACM， 2023： 3033-3045.