全文下载:
202605015.pdf
文章编号:1672-6987(2026)05-0124-08;DOI:10. 16351/j. 1672-6987. 2026. 05. 015
杜玉蓉,高德欣* (青岛科技大学 自动化与电子工程学院,山东 青岛 266061)
摘 要:针对电动汽车充电过程受多特征输入因素影响导致温度预测精度低的问题,为充分 挖掘深层特征信息影响,提出了一种基于一维卷积神经网络(1D CNN)、双向嵌套式长短期记 忆网络(BiNLSTM)和注意力机制(AM)的深度学习预测模型,并通过残差分析方法确定动态 预警阈值,保障充电安全。该方法通过 1D CNN 进行层级特征提取,使用 BiNLSTM 模块进行 时间信息融合,使用 AM 模块进行变量加权,更好地捕捉充电过程中的复杂关系。选取春/秋、 夏、冬三组充电历史数据进行模型训练,通过独立充电热失控实例进行验证。结果表明, CNN-BiNLSTM-AM 模型较于目前主流预测模型预测精度更高,且利用温度异常特性可提 前判别充电热失控事故。
关键词:电动汽车热失控;温度预测;一维卷积神经网络;双向嵌套式长短期记忆网络;注意 力机制
中图分类号:U 491. 8; TM 910. 6 文献标志码:A
引用格式:杜玉蓉,高德欣 . 基于深度学习的电动汽车充电热失控安全预警方法[J]. 青岛科 技大学学报(自然科学版),2026,47(5):124-131.
DU Yurong, GAO Dexin. A safety pre-warning method for electric vehicle charging thermal runaway based on deep learning[J]. Journal of Qingdao University of Science and Technology (Natural Science Edition),2026,47(5):124-131.
A Safety Pre-Warning Method for Electric Vehicle Charging Thermal Runaway Based on Deep Learning
DU Yurong,GAO Dexin (College of Automation and Electronic Engineering, Qingdao University of Science and Technology, Qingdao 266061, China)
Abstract:Aiming at the problem of low-temperature prediction accuracy due to the influence of multi-feature input factors in the charging process of electric vehicles, in order to fully explore the influence of deep feature information, a deep learning prediction model based on 1D Convolutional Neural Network (1D CNN), Bi-directional Nested Long Short-term Memory (BiNLSTM), and Attention Mechanism (AM) is proposed, and the residual analysis method is used to determine the dynamic warning thresholds for charging safety. The method performs hierarchical feature extraction via 1D CNN, temporal information fusion using the BiNLSTM module, and variable weighting using the AM module to better capture the complex relation⁃ships in the charging process. Three sets of charging history data, Spring/Autumn, Summer, and Winter, are selected for model training and validated by independent charging thermal run⁃ away examples. The results show that the CNN-BiNLSTM-AM model has higher prediction accuracy than the current mainstream prediction model, and the charging thermal runaway acci⁃ dent can be identified in advance by using the temperature anomaly characteristic.
Key words:electric vehicle thermal runaway; temperature prediction;1D convolutional neural network; bi-directional nested long short-term memory; attention mechanism
收稿日期:2026-03-15
基金项目:山东省重点研发计划项目(公益类)(2019GGX101012);山东省自然科学基金项目(ZR2026MS1081).
作者简介:杜玉蓉(2000—),女,硕士研究生 . * 通信联系人 .