主办：陕西省汽车工程学会
ISSN 1671-7988  CN 61-1394/TH
创刊：1976年
汽车实用技术

汽车实用技术 ›› 2025, Vol. 50 ›› Issue (11): 1-7.DOI: 10.16638/j.cnki.1671-7988.2025.011.001

• 新能源汽车 •    

基于 CNN-Attention-LSTM 的新能源汽车 充电中 SOC 预测

刘丰 1，高有山 1，黄敏 2，孙浩然 1，王爱红 1，任鸿 1   

  1. 1.太原科技大学 机械工程学院； 2.岚图汽车科技有限公司 新能源技术事业部
  • 发布日期:2025-06-06
  • 通讯作者: 刘丰
  • 作者简介:刘丰（2000－），男，硕士研究生，研究方向为车辆工程新能源技术
  • 基金资助:
    山西省重型机械电液控制及健康管理技术创新中心；山西省留学人员科技活动择优资助项目（20240021）； 山西省研究生实践创新项目（2023SJ257）

Prediction of SOC in New Energy Vehicle Charging Based on CNN-Attention-LSTM

LIU Feng1 , GAO Youshan1 , HUANG Min2 , SUN Haoran1 , WANG Aihong1 , REN Hong1   

  1. 1.School of Mechanical Engineering, Taiyuan University of Science and Technology; 2.New Energy Technology Division, VOYAH Automobile Technology Company Limited
  • Published:2025-06-06
  • Contact: LIU Feng

PDF

摘要： 准确预测新能源汽车充电过程中的荷电状态（SOC）可以提升充电效率与安全性、延 长电池使用寿命以及提高驾驶体验与智能化水平，具有重要研究价值。文章基于卷积神经网 络（CNN）与注意力机制和长短时记忆（LSTM）网络结合的方法，首先获取预测所需信息， 划分数据集并且进行数据预处理，提高模型的训练效果和泛化能力，然后输入 CNN-AttentionLSTM 模型进行训练，最后依据评价指标验证该模型的有效性。研究结果表明，该模型能够 从大量数据中提取到重要特征，理论上只通过实时获取充电过程中的数据，可以实现充电过 程中 SOC 的准确预测。通过设计不同充电实验，对比充电过程中的实际 SOC 与预测 SOC， 得出 CNN-Attention-LSTM 模型具有可行性与良好的泛化性并且比 CNN-LSTM 模型性能更 佳，在未来将具有广泛应用价值。

关键词: 荷电状态；卷积神经网络；注意力机制；长短时记忆网络

Abstract: Accurately predicting the state of charge (SOC) in the charging process of new energy vehicles can improve charging efficiency and safety, extend battery life, and improve driving experience and intelligence level, which has important research value. This paper is based on the method of combining convolution neural network (CNN) with attention mechanism and long shortterm memory (LSTM) network, information needed for prediction is obtained first. The data set is divided and the data is preprocessed to improve the training effect and generalization ability of the model. Then the CNN-Attention-LSTM model is input for training, and the effectiveness of the model is verified according to the evaluation index. The research results show that the model can extract important features from a large number of data, and in theory, the SOC can be accurately predicted during the charging process only by acquiring the data during the charging process in real time. By designing different charging experiments and comparing the actual SOC and the predicted SOC in the charging process, it is concluded that the CNN-Attention-LSTM model is feasible and has good generalization and better performance than the CNN-LSTM model, which will have wide application value in the future.

Key words: state of charge; convolution neural network; attention mechanism; long short-term memory network