交通运输系统工程与信息 ›› 2023, Vol. 23 ›› Issue (4): 80-87.DOI: 10.16097/j.cnki.1009-6744.2023.04.009

• 智能交通系统与信息技术 • 上一篇    下一篇

混合示教长短时记忆网络的车辆轨迹预测研究

方华珍,刘立,肖小凤,顾青*,孟宇   

  1. 北京科技大学,机械工程学院,北京 100083
  • 收稿日期:2023-03-05 修回日期:2023-05-28 接受日期:2023-05-29 出版日期:2023-08-25 发布日期:2023-08-21
  • 作者简介:方华珍(1996- ),男,江西鄱阳人,博士生
  • 基金资助:
    国家自然科学基金青年科学基金(52202505);国家重点研发计划(2019YFC0605300)

Vehicle Trajectory Prediction Based on Mixed Teaching Force Long Short-term Memory

FANG Hua-zhen, LIU Li, XIAO Xiao-feng, GU Qing*, MENG Yu   

  1. School of Mechanical Engineering, University of Science and Technology Beijing, Beijing 100083, China
  • Received:2023-03-05 Revised:2023-05-28 Accepted:2023-05-29 Online:2023-08-25 Published:2023-08-21
  • Supported by:
    Young Scientists Fund of the National Natural Science Foundation of China (52202505);National Key Research and Development Program of China (2019YFC0605300)

摘要: 为实现智能网联车对周围车辆运行轨迹准确地长时预测,本文提出一种混合示教解码的长短时记忆网络的车辆轨迹预测方法。首先,通过特征筛选和历史轨迹序列标注建立轨迹预测数据集;其次,构建长短时记忆网络的编码器-解码器模型,编码器将自车和周围车辆历史轨迹及道路环境信息编码为上下文向量,解码器采用混合示教的模式将上下文向量解码动态解码为未来轨迹;最后,采用真实道路数据集NGSIM US101和I-80路段验证模型的可行性。多组对比分析实验结果表明:本文所提方法在长时域预测的终点位移误差指标上的有效性和优越性,5 s的终点位移误差在2.7 m以内;并且模型在稀疏采样后的数据集上达到更高的预测准确率,5 s的位移误差在1.3 m以内。

关键词: 智能交通, 混合示教, 长短时记忆网络, 人工驾驶车辆, 智能网联车, 车辆轨迹预测

Abstract: To improve the long-term trajectory prediction of the intelligent connected vehicle to the surrounding vehicles, this paper proposes an interaction-aware network framework based on mixed teacher forcing Long Short�Term Memory (LSTM) encoder-decoder. First, a trajectory prediction dataset is established through feature selection and trajectory sequence labeling. Then, the LSTM encoder-decoder model is developed. The encoder encodes the historical trajectory of the target vehicle, the information of surrounding vehicles, and the road environment into the context vector. The decoder adopts the mixed teaching mode to decode the context vector dynamically into the future trajectory. At last, the model is verified on the real road datasets NGSIM US101 and I-80 and compared with the traditional models. The experimental results show that the proposed model performs better than the traditional methods in long-term prediction. The 5 seconds final displacement error is 2.7 meters. The accuracy of the model after sparse sampling has been significantly improved compared with other methods, the 5 seconds final displacement error is 1.3 meters.

Key words: intelligent transportation, mixed teaching force, long short-term memory(LSTM), human driven vehicles, intelligent connected vehicle, vehicle trajectory prediction

中图分类号: