北京航空航天大学学报 ›› 2022, Vol. 48 ›› Issue (5): 890-897.doi: 10.13700/j.bh.1001-5965.2020.0662

• 论文 • 上一篇    下一篇

基于神经网络的车辆强制换道预测模型

崔洁茗1, 余贵珍1, 周彬1, 李存金2, 马继伟2, 徐国艳1   

  1. 1. 北京航空航天大学 交通科学与工程学院 特种车辆无人运输技术工业和信息化部重点实验室, 北京 100083;
    2. 内蒙古霍林河露天煤业股份有限公司, 通辽 028001
  • 收稿日期:2020-11-26 发布日期:2022-05-30
  • 通讯作者: 周彬 E-mail:binzhou@buaa.edu.cn
  • 基金资助:
    国家自然科学基金(51775016);北京市自然科学基金(L191002)

Mandatory lane change decision-making model based on neural network

CUI Jieming1, YU Guizhen1, ZHOU Bin1, LI Cunjin2, MA Jiwei2, XU Guoyan1   

  1. 1. Key Laboratory of Autonomous Transportation Technology for Special Vehicles, Ministry of Industry and Information Technology, School of Transportation Science and Engineering, Beihang University, Beijing 100083, China;
    2. Inner Mongolia Huolinhe Surface Coal Industry Co., Ltd., Tongliao 028001, China
  • Received:2020-11-26 Published:2022-05-30
  • Supported by:
    National Natural Science Foundation of China (51775016);Beijing Municipal Natural Science Foundation (L191002)

摘要: 针对高速公路上车辆行驶速度快,换道行为危险程度高的问题,聚焦于不可避免、发生频繁、一旦发生事故后果严重的强制换道行为,改进基于门控循环单元(GRU)的换道模型,对强制换道行为进行分析与预测。为保证模型的有效性,选取下一代仿真技术(NGSIM)数据作为模型的训练集与检测集,使用侧向加速度将车辆侧向摆动数据有效删除并得到强制换道的最迟换道点,进而实现车辆位置与换道决策的预测。实验结果证明,所提模型能够以96.01%的准确率判定车辆在最迟换道点的强制换道行为,相较于LSTM模型准确率提升了3.67%,同时相较于朴素贝叶斯网络准确率提高了7.31%。

关键词: 强制换道行为, 神经网络, 换道决策, 侧向加速度, 门控循环单元(GRU)

Abstract: Aiming at the problem of fast-speed and high risk of lane changing behavior on expressway, we focus on the ineviteable, freguent and serve mandatory lane-changing behaviors to improve the lane-changing model based on gated recurrent unit (GRU), and predict the decision-making behaviors of mandatony lane-changing. To verify the effectiveness of the model, adopt the next generation simulation (NGSIM) data as the training set and test set of the model. From this data, the lateral acceleration threshold is obtained to screen out the phenomenon of lateral swing of vehicles. The experimental results indicate that the optimized model could determine the location of mandatory lane change with an accuracy of 96.01%. The accuracy of the model is improved by 3.67% compared with the LSTM model, and is improved by 7.31% compared with the naive Bayes network.

Key words: mandatory lane change behavior, neural networks, lane change decision making, lateral acceleration, gated recurrent unit (GRU)

中图分类号: 


版权所有 © 《北京航空航天大学学报》编辑部
通讯地址:北京市海淀区学院路37号 北京航空航天大学学报编辑部 邮编:100191 E-mail:jbuaa@buaa.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发