基于深度特征融合模型的电动汽车不同模式充电时长自适应预测

doi:10.19799/j.cnki.2095-4239.2025.0490

摘要/Abstract

摘要：

精准预测锂电池充电时长能够提升充电效率、优化资源分配，对电动汽车的发展具有重要意义。本工作提出了一种基于深度特征融合模型的电动汽车不同模式下充电时长自适应预测方法。首先，对新能源汽车监控平台采集的车辆运行数据进行清洗和分割，结合充电电压、电流及平均功率划分充电模式，形成快慢充数据集；接着，基于充电数据集采用主成分分析(PCA)提取模型输入特征；其次，融合Attention注意力机制构建多层感知器(MLP)模型，对输入特征进行非线性映射获取中间特征；考虑到从原始数据中直接提取的特征无法全面反映与充电时长间的复杂关系，故引入随机森林(RF)模型，根据RF内部分裂原理构造叶节点规则特征，发掘隐含的特征信息；然后，在MLP建立“规则层”处理融合的中间特征与规则特征，实现两模型的结构性融合。最后，对Attention MLP-RF融合模型预测结果进行验证，结果显示，融合模型在快、慢充模式下预测平均绝对误差分别为4.25分钟和6.68分钟，平均绝对百分比误差分别仅有4.33%和3.86%，实现了不同电动汽车充电时长的精准预测。同时，该方法对于电池老化和短时充电情况下的充电时长预测仍具有很高的精度，平均预测误差不超过2分钟，融合模型整体具有很强的预测性能及泛化能力。

关键词: 锂离子电池, 数据驱动, 充电模式, 充电时长, 特征融合

Abstract:

Accurately predicting the charging time of lithium batteries can improve charging efficiency and optimize resource allocation-an important factor for developing electric vehicles. This study proposes an adaptive prediction method for electric vehicle charging duration under different modes based on a deep feature fusion model. First, vehicle operation data collected by a new energy vehicle monitoring platform are cleaned and segmented, and charging modes are classified based on voltage, current, and average power to form fast- and slow-charging datasets. Next, based on the charging dataset, principal component analysis is applied to extract model input features. Then, a multilayer perceptron (MLP) model is constructed by integrating the attention mechanism to obtain intermediate features through a nonlinear mapping of input features. As features directly extracted from raw data cannot fully capture the complex relationship with charging duration, a random forest (RF) model is introduced to construct leaf-node rule features based on the internal splitting principle of RF, exploring implicit feature information. A "rule layer" is subsequently established in the MLP to fuse intermediate and rule features, achieving structural fusion of the two models. Finally, the prediction results of the attention MLP-RF fusion model are validated, demonstrating an average absolute error of 4.25 and 6.68 minutes for fast-and slow-charging modes, respectively, with an average absolute percentage error of 4.33% and 3.86%, indicating accurate prediction of different electric vehicle charging durations. Moreover, this method maintains high accuracy in predicting charging duration under battery aging and short-term charging conditions, with an average prediction error of less than 2 min. Overall, the fusion model demonstrates strong predictive performance and generalization capabilities.

Key words: lithium-ion battery, data driven, charge mode, charging duration, feature fusion

中图分类号:

TM 912

韦介洋, 申江卫, 陈峥, 魏福星, 夏雪磊, 刘永刚. 基于深度特征融合模型的电动汽车不同模式充电时长自适应预测[J]. 储能科学与技术, 2025, 14(11): 4330-4345.

Jieyang WEI, Jiangwei SHEN, Zheng CHEN, Fuxing WEI, Xuelei XIA, Yonggang LIU. Adaptive prediction of charging duration for different modes of electric vehicles based on a deep feature fusion model[J]. Energy Storage Science and Technology, 2025, 14(11): 4330-4345.

图/表 20

图1

图2

图3

图4

图5

表1

图6

表2

图7

表3

表4

图8

图9

图10

图11

表5

表6

图12

图13

图14

参考文献 34

[1]	GIL-GARCÍA I C, GARCÍA-CASCALES M S, DAGHER H, et al. Electric vehicle and renewable energy sources: Motor fusion in the energy transition from a multi-indicator perspective[J]. Sustainability, 2021, 13(6): 3430. DOI: 10.3390/su13063430.
[2]	SHEN J W, ZHANG Z, SHEN S Q, et al. Accurate state of temperature estimation for lithium-Ion batteries based on square root cubature Kalman filter[J]. Applied Thermal Engineering, 2024, 242: 122452. DOI: 10.1016/j.applthermaleng.2024.122452.
[3]	TAO Y C, QIU J, LAI S Y, et al. Adaptive integrated planning of electricity networks and fast charging stations under electric vehicle diffusion[J]. IEEE Transactions on Power Systems, 2023, 38(1): 499-513. DOI: 10.1109/TPWRS.2022.3167666.
[4]	EDDINE M D, SHEN Y M. A deep learning based approach for predicting the demand of electric vehicle charge[J]. The Journal of Supercomputing, 2022, 78(12): 14072-14095. DOI: 10.1007/s11227-022-04428-0.
[5]	YAGHOUBI E, YAGHOUBI E, KHAMEES A, et al. A systematic review and meta-analysis of machine learning, deep learning, and ensemble learning approaches in predicting EV charging behavior[J]. Engineering Applications of Artificial Intelligence, 2024, 135: 108789. DOI: 10.1016/j.engappai.2024.108789.
[6]	申江卫, 刘伟强, 高承志, 等. 宽温度环境下基于迁移模型的锂电池组SOC估计[J]. 中国公路学报, 2024, 37(5): 383-396. DOI: 10. 19721/j.cnki.1001-7372.2024.05.025.
	SHEN J W, LIU W Q, GAO C Z, et al. State of charge estimation of lithium battery packs in wide temperature environments based on migration model[J]. China Journal of Highway and Transport, 2024, 37(5): 383-396. DOI: 10.19721/j.cnki.1001-7372.2024.05.025.
[7]	SUBASHINI M, SUMATHI V. Smart algorithms for power prediction in smart EV charging stations[J]. Journal of Engineering Research, 2024, 12(2): 124-134. DOI: 10.1016/j.jer.2023.11.028.
[8]	ADAIKKAPPAN M, SATHIYAMOORTHY N. Modeling, state of charge estimation, and charging of lithium-ion battery in electric vehicle: A review[J]. International Journal of Energy Research, 2022, 46(3): 2141-2165. DOI: 10.1002/er.7339.
[9]	RAGONE M, YURKIV V, RAMASUBRAMANIAN A, et al. Data driven estimation of electric vehicle battery state-of-charge informed by automotive simulations and multi-physics modeling[J]. Journal of Power Sources, 2021, 483: 229108. DOI: 10.1016/j.jpowsour.2020.229108.
[10]	SHI J Z, TIAN M, HAN S, et al. Electric vehicle battery remaining charging time estimation considering charging accuracy and charging profile prediction[J]. Journal of Energy Storage, 2022, 49: 104132. DOI: 10.1016/j.est.2022.104132.
[11]	ZHOU B R, FAN G D, WANG Y S, et al. Life-extending optimal charging for lithium-ion batteries based on a multi-physics model and model predictive control[J]. Applied Energy, 2024, 361: 122918. DOI: 10.1016/j.apenergy.2024.122918.
[12]	OEHLER F F, NÜRNBERGER K, STURM J, et al. Embedded real-time state observer implementation for lithium-ion cells using an electrochemical model and extended Kalman filter[J]. Journal of Power Sources, 2022, 525: 231018. DOI: 10.1016/j.jpowsour. 2022.231018.
[13]	POWELL S, VIANNA CEZAR G, APOSTOLAKI-IOSIFIDOU E, et al. Large-scale scenarios of electric vehicle charging with a data-driven model of control[J]. Energy, 2022, 248: 123592. DOI: 10.1016/j.energy.2022.123592.
[14]	ZHAO Y, WANG Z P, SHEN Z M, et al. Data-driven framework for large-scale prediction of charging energy in electric vehicles[J]. Applied Energy, 2021, 282: 116175. DOI: 10.1016/j.apenergy. 2020.116175.
[15]	ALMAGHREBI A, ALJUHESHI F, RAFAIE M, et al. Data-driven charging demand prediction at public charging stations using supervised machine learning regression methods[J]. Energies, 2020, 13(16): 4231. DOI: 10.3390/en13164231.
[16]	ULLAH I, LIU K, YAMAMOTO T, et al. Grey wolf optimizer-based machine learning algorithm to predict electric vehicle charging duration time[J]. Transportation Letters, 2023, 15(8): 889-906. DOI: 10.1080/19427867.2022.2111902.
[17]	SHAHRIAR S, AL-ALI A R, OSMAN A H, et al. Prediction of EV charging behavior using machine learning[J]. IEEE Access, 2021, 9: 111576-111586.
[18]	YI Z Y, LIU X C, WEI R, et al. Electric vehicle charging demand forecasting using deep learning model[J]. Journal of Intelligent Transportation Systems, 2022, 26(6): 690-703. DOI: 10.1080/15472450.2021.1966627.
[19]	王毅, 谷亿, 丁壮, 等. 基于模糊熵和集成学习的电动汽车充电需求预测[J]. 电力系统自动化, 2020, 44(3): 114-121.
	WANG Y, GU Y, DING Z, et al. Charging demand forecasting of electric vehicle based on empirical mode decomposition-fuzzy entropy and ensemble learning[J]. Automation of Electric Power Systems, 2020, 44(3): 114-121.
[20]	ULLAH I, LIU K, YAMAMOTO T, et al. Prediction of electric vehicle charging duration time using ensemble machine learning algorithm and Shapley additive explanations[J]. International Journal of Energy Research, 2022, 46(11): 15211-15230. DOI: 10.1002/er.8219.
[21]	胡杰, 陈琳, 王志红, 等. 基于Transformer的纯电动汽车充电时间预测[J]. 汽车工程, 2024, 46(11): 2059-2067. DOI: 10.19562/j.chinasae. qcgc.2024.11.012.
	HU J, CHEN L, WANG Z H, et al. Transformer-based prediction of charging time for pure electric vehicles[J]. Automotive Engineering, 2024, 46(11): 2059-2067. DOI: 10.19562/j.chinasae.qcgc.2024. 11.012.
[22]	陈媛, 章思源, 蔡宇晶, 等. 融合多项式特征扩展与CNN-Transformer模型的锂电池SOH估计[J]. 储能科学与技术, 2024, 13(9): 2995-3005. DOI: 10.19799/j.cnki.2095-4239.2024.0465.
	CHEN Y, ZHANG S Y, CAI Y J, et al. State-of-health estimation of lithium batteries based on polynomial feature extension of the CNN-transformer model[J]. Energy Storage Science and Technology, 2024, 13(9): 2995-3005. DOI: 10.19799/j.cnki.2095-4239.2024.0465.
[23]	PIAS T S, EISENBERG D, FRESNEDA FERNANDEZ J. Accuracy improvement of vehicle recognition by using smart device sensors[J]. Sensors, 2022, 22(12): 4397. DOI: 10.3390/s22124397.
[24]	时珊珊, 王凯, 张宇, 等. 城市轨道交通混合储能系统无源性协同控制方法[J]. 储能科学与技术, 2024, 13(11): 4040-4052. DOI: 10. 19799/j.cnki.2095-4239.2024.0516.
	SHI S S, WANG K, ZHANG Y, et al. Collaborative passivity-based control method for hybrid energy storage systems in urban rail transit[J]. Energy Storage Science and Technology, 2024, 13(11): 4040-4052. DOI: 10.19799/j.cnki.2095-4239.2024.0516.
[25]	宾涛. 电动汽车混合充换电站选址定容建模及求解[D]. 重庆: 重庆交通大学, 2024. DOI: 10.27671/d.cnki.gcjtc.2024.000200.
	BIN T. Modeling and solution of site selection and constant volume of hybrid charging and replacing power station for electric vehicle[D]. Chongqing: Chongqing Jiaotong University, 2024. DOI: 10.27671/d.cnki.gcjtc.2024.000200.
[26]	李兴建. 电动汽车充电桩运行管理系统设计与实现[D]. 成都: 电子科技大学, 2020. DOI: 10.27005/d.cnki.gdzku.2020.004222.
	LI X J. Design and implementation of electric vehicle charging station operation management system[D]. Chengdu: University of Electronic Science and Technology of China, 2020. DOI: 10. 27005/d.cnki.gdzku.2020.004222.
[27]	王明深, 袁晓冬, 曾飞, 等. 电动汽车充电设施规划运行关键技术研究综述[J]. 电力自动化设备, 2025, 45(5): 65-76. DOI: 10.16081/j.epae.202502002.
	WANG M S, YUAN X D, ZENG F, et al. Review of key technologies research for electric vehicle charging facilities planning and scheduling[J]. Electric Power Automation Equipment, 2025, 45(5): 65-76. DOI: 10.16081/j.epae.202502002.
[28]	NASKATH J, SIVAKAMASUNDARI G, BEGUM A A S. A study on different deep learning algorithms used in deep neural nets: MLP SOM and DBN[J]. Wireless Personal Communications, 2023, 128(4): 2913-2936. DOI: 10.1007/s11277-022-10079-4.
[29]	AZAD M A K, MALLICK J, ISLAM A R M T, et al. Estimation of solar radiation in data-scarce subtropical region using ensemble learning models based on a novel CART-based feature selection[J]. Theoretical and Applied Climatology, 2024, 155(1): 349-369. DOI: 10.1007/s00704-023-04638-3.
[30]	李旭东, 张向文. 基于主成分分析与WOA-Elman的锂离子电池SOH估计[J]. 储能科学与技术, 2022, 11(12): 4010-4021. DOI: 10. 19799/j.cnki.2095-4239.2022.0384.
	LI X D, ZHANG X W. State of health estimation method for lithium-ion batteries based on principal component analysis and whale optimization algorithm-Elman model[J]. Energy Storage Science and Technology, 2022, 11(12): 4010-4021. DOI: 10.19799/j.cnki.2095-4239.2022.0384.
[31]	ZHANG Y, SHI X H, ZHANG H X, et al. Review on deep learning applications in frequency analysis and control of modern power system[J]. International Journal of Electrical Power & Energy Systems, 2022, 136: 107744. DOI: 10.1016/j.ijepes.2021.107744.
[32]	刘素贞, 袁路航, 张闯, 等. 基于超声时域特征及随机森林的磷酸铁锂电池荷电状态估计[J]. 电工技术学报, 2022, 37(22): 5872-5885. DOI: 10.19595/j.cnki.1000-6753.tces.211585.
	LIU S Z, YUAN L H, ZHANG C, et al. State of charge estimation of LiFeO₄ batteries based on time domain features of ultrasonic waves and random forest[J]. Transactions of China Electrotechnical Society, 2022, 37(22): 5872-5885. DOI: 10.19595/j.cnki.1000-6753.tces. 211585.
[33]	THAJEEL I K, SAMSUDIN K, HASHIM S J, et al. Machine and deep learning-based XSS detection approaches: A systematic literature review[J]. Journal of King Saud University - Computer and Information Sciences, 2023, 35(7): 101628. DOI: 10.1016/j.jksuci.2023.101628.
[34]	韩毅, 侯震梅. 基于RF-RFECV特征选择的BO-CNN-BiLSTM-attention中国资源型城市碳排放影响因素分析与达峰情景模拟[J/OL]. 环境科学, 2025: 1-26. (2025-05-14). https://link.cnki.net/doi/10.13227/j.hjkx.202501278.
	HAN Y, HOU Z M. BO-CNN-BiLSTM-attention based on RF-RFECV feature selection analysis and peak simulation of carbon emissions in resource-based cities of China[J/OL]. Environmental Science, 2025: 1-26. (2025-05-14). https://link.cnki.net/doi/10. 13227/j.hjkx.202501278.

初始SOC	结束SOC	初始总电压/V	初始最低温度/℃	单体最高电压/V	单体最低电压/V	充电时长/s
59	100	391.1	0	3.77	3.74	14894
70	90	400	6	3.85	3.83	6820
39	100	377.5	3	3.63	3.62	20940
23	99	370.4	4	3.57	3.55	26285
14	97	364.6	4	3.51	3.49	31240
20	100	363.4	5	3.62	3.60	28152

“规则层”使用激活函数	MAE/min	MAPE/%
ReLU	4.25	4.33
LeakyReLU	6.73	6.98
Swish	7.18	7.24

注意力头数	MLP每层神经元数	树最大深度	MAE/min	MAPE/%
4	32	10	5.82	6.03
8	32	10	5.37	5.48
12	32	10	7.18	7.24
8	64	10	4.25	4.33
8	128	10	8.74	8.92
8	64	5	6.51	6.83
8	64	15	10.06	9.55

模型结构	是否含“规则层”	是否含Attention	MAE/min	MAPE/%
Attention MLP-RF	是	是	4.25	4.33
Attention MLP-RF	否	是	6.62	6.87
MLP-RF	是	否	9.13	9.51
MLP	否	否	13.52	14.02
RF	否	否	12.14	12.67

模型名称	MAE/min	RMSE/min	MAPE/%
Attention MLP-RF	4.25	5.16	4.33
BiLSTM-RF	7.51	8.81	6.52
RF	9.01	10.68	8.71
Transformer	4.92	6.07	5.14
XGBoost	6.08	7.25	5.85