华南理工大学学报(自然科学版) ›› 2025, Vol. 53 ›› Issue (2): 58-67.doi: 10.12141/j.issn.1000-565X.240219

• 交通运输工程 • 上一篇    下一篇

基于LSTM-FC模型的充电站短期运行状态预测

毕军1,2, 王嘉宁1, 王永兴1   

  1. 1.北京交通大学 交通运输学院,北京 100044
    2.北京交通大学 综合交通运输大数据应用技术交通运输行业重点实验室,北京 100044
  • 收稿日期:2024-05-06 出版日期:2025-02-25 发布日期:2025-02-03
  • 通信作者: 王永兴(1990—),男,博士,讲师,主要从事电动汽车运行管理等研究。 E-mail:yx.wang@bjtu.edu.cn
  • 作者简介:毕军(1973—),男,博士,教授,主要从事大数据智能交通决策等研究。E-mail: jbi@bjtu.edu.cn
  • 基金资助:
    国家自然科学基金项目(72171019)

Short-Term Operation State Prediction of Charging Station Based on LSTM-FC Model

BI Jun1,2, WANG Jianing1, WANG Yongxing1   

  1. 1.School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044,China
    2.Key Laboratory of Transport Industry of Big Data Application Technologies for Comprehensive Transport,Beijing Jiaotong University,Beijing 100044,China
  • Received:2024-05-06 Online:2025-02-25 Published:2025-02-03
  • Contact: 王永兴(1990—),男,博士,讲师,主要从事电动汽车运行管理等研究。 E-mail:yx.wang@bjtu.edu.cn
  • About author:毕军(1973—),男,博士,教授,主要从事大数据智能交通决策等研究。E-mail: jbi@bjtu.edu.cn
  • Supported by:
    the National Natural Science Foundation of China(72171019)

摘要:

公共充电站可用充电桩数量预测对于制定智能充电推荐策略和减少用户的充电排队时间具有重要意义。现阶段充电站运行状态研究通常集中于充电负荷预测,对于站内充电桩占用情况的研究较少,同时缺乏实际数据支撑。为此,基于充电站实际运行数据,提出一种基于长短时记忆(LSTM)网络与全连接(FC)网络结合的充电站内可用充电桩预测模型,有效结合了历史充电状态序列和相关特征。首先,将兰州市某充电站的订单数据转化为可用充电桩数量,并进行数据预处理;其次,提出了基于LSTM-FC的充电站运行状态预测模型;最后,将输入步长、隐藏层神经元数量和输出步长3种参数进行单独测试。为验证LSTM-FC模型的预测效果,将该模型与原始LSTM网络、BP神经网络模型和支持向量回归(SVR)模型进行对比。结果表明:LSTM-FC模型的平均绝对百分比误差分别降低了0.247、1.161和2.204个百分点,具有较高的预测精度。

关键词: LSTM神经网络, 全连接网络, 电动汽车, 充电站运行状态

Abstract:

The prediction of the number of available charging piles in public charging stations is of great significance for the formulation of intelligent charging recommendation strategy and the reduction of users’ charging queue time. At present, the research on the operating state of charging stations typically focuses on charging load forecasting, with relatively little attention given to the utilization of charging piles within the stations. Additionally, there is a lack of support from real-world data. Therefore, based on the actual operation data of charging stations, this paper proposed a prediction model of available charging piles in charging stations based on the combination of long short-term memory network (LSTM) and fully connected network (FC), which effectively combines the historical charging state sequence and related features. Firstly, the order data from a specific charging station in Lanzhou was transformed into the number of available charging piles, followed by data preprocessing. Secondly, an LSTM-FC-based predictive model for the operational status of the charging station was proposed. Finally, three parameters—input step size, number of hidden layer neurons, and output step size—were individually tested. To validate the predictive performance of the LSTM-FC model, it was compared with the original LSTM network, BP neural network model, and support vector regression (SVR) model. The results show that the mean absolute percentage error of LSTM-FC model is reduced by 0.247%, 1.161% and 2.204% respectively, which shows high prediction accuracy.

Key words: long short-term memory neural network, fully connected network, electric vehicles, charging station operating status

中图分类号: