中国邮电高校学报(英文)

中国邮电高校学报(英文) ›› 2021, Vol. 28 ›› Issue (3): 63-75.doi: 10.19682/j.cnki.1005-8885.2021.0010

• Artificial Intelligence • 上一篇    下一篇

TCL: a taxi trajectory prediction model combining time and space features

焦继超,陈新平,管孟,赵亚鑫   

  1. 北京邮电大学
  • 收稿日期:2020-07-19 修回日期:2020-11-24 出版日期:2021-06-30 发布日期:2021-06-22
  • 通讯作者: 陈新平 E-mail:xinping.555@bupt.edu.cn
  • 基金资助:
    国家重点研究发展计划;中央大学基础研究基金

TCL: a taxi trajectory prediction model combining time and space features

  • Received:2020-07-19 Revised:2020-11-24 Online:2021-06-30 Published:2021-06-22
  • Contact: CHEN Xinping E-mail:xinping.555@bupt.edu.cn

PDF

161

摘要：

Vehicle trajectory modeling is an important foundation for urban intelligent services. Trajectory prediction of cars is a hot topic. A model including convolutional neural network (CNN) and long short-term memory (LSTM) was proposed, which is named trajectory-CNN-LSTM (TCL). CNN can extract the spatial features of the trajectory in the input image. Besides, LSTM can extract the time-series features of the input trajectory. After that, the model uses fully connected layers to merge the two features for the final predicting. The experiments on the Porto dataset of The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML-PKDD) show that the average prediction error of TCL is reduced by 0.15 km, 0.42 km, and 0.39 km compared to the trajectory-convolution (T-CONV), multi-layer perceptron (MLP), and recurrent neural network (RNN) model, respectively.

关键词: GNSS轨迹数据|轨迹预测|空间特征|时间特征。

Abstract:

Vehicle trajectory modeling is an important foundation for urban intelligent services. Trajectory prediction of cars is a hot topic. A model including convolutional neural network (CNN) and long short-term memory (LSTM) was proposed, which is named trajectory-CNN-LSTM (TCL). CNN can extract the spatial features of the trajectory in the input image. Besides, LSTM can extract the time-series features of the input trajectory. After that, the model uses fully connected layers to merge the two features for the final predicting. The experiments on the Porto dataset of The European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML-PKDD) show that the average prediction error of TCL is reduced by 0.15 km, 0.42 km, and 0.39 km compared to the trajectory-convolution (T-CONV), multi-layer perceptron (MLP), and recurrent neural network (RNN) model, respectively.

Key words: GNSS trajectory data| trajectory prediction| spatial features| time features.

中图分类号: 

版权所有 © 2020 中国邮电高校学报(英文)编辑部　
　 地址：北京邮电大学图书馆1层119 室　邮政编码：100081
Tel：86-010-62282493　Fax： 86-010-62283461　E-mail: jchupt@bupt.edu.cn
本系统由北京玛格泰克科技发展有限公司设计开发 技术支持：support@magtech.com.cn