计算机工程与应用 ›› 2024, Vol. 60 ›› Issue (9): 65-78.DOI: 10.3778/j.issn.1002-8331.2308-0127

• 热点与综述 • 上一篇    下一篇

基于深度学习的公交行驶轨迹预测研究综述

杨晨曦,庄旭菲,陈俊楠,李衡   

  1. 内蒙古工业大学 信息工程学院,呼和浩特 010080
  • 出版日期:2024-05-01 发布日期:2024-04-29

Review of Research on Bus Travel Trajectory Prediction Based on Deep Learning

YANG Chenxi, ZHUANG Xufei, CHEN Junnan, LI Heng   

  1. College of Information Engineering, Inner Mongolia University of Technology, Hohhot 010080, China
  • Online:2024-05-01 Published:2024-04-29

摘要: 公交行驶轨迹预测是对公交车到达线路上的重要轨迹点,如站点和道路交叉口等,进行到达时间预测。准确预测公交车到达站点和道路交叉口的时间,可以提高城市公交系统的运行效率和服务质量,对于城市公共交通规划和公交调度至关重要。从公交行驶轨迹预测方法的发展现状入手,分析了影响公交运行的相关因素,归纳并探讨了不同类型的相关数据集以及数据预处理方法。依照其发展脉络将公交行驶轨迹预测方法分为基于历史数据的模型、以时间序列模型为代表的参数模型以及包括机器学习和深度学习方法的非参数模型三大类,并总结分析了不同方法的优势和局限性。由于基于深度学习的相关模型在时间序列预测任务中表现出了优越性能,因此越来越多的学者开始采用基于深度学习的模型来解决公交行驶轨迹预测问题,同时考虑将城市道路所展现的空间特征与时间特征相结合以进一步提高预测精度。最后,阐述了公交行驶轨迹预测研究领域中面临的挑战,并对该领域未来的发展进行总结分析与趋势展望。

关键词: 公交行驶轨迹预测, 深度学习, 时空特征, 时间序列预测, 智能交通

Abstract: Bus travel trajectory prediction predicts when the bus arrives at important track points on its route, such as stops and road intersections. Accurate bus arrival time prediction at road intersections and stops can improve the efficiency and service quality of urban public transport system, which is crucial for urban public transport planning and bus dispatch. From the perspective of the development of bus travel trajectory prediction methods, this paper analyzes the factors that affect bus operation, explores the types of datasets, and summarizes the data preprocessing methods. According to their development venation, bus travel trajectory prediction methods are divided into three categories:historical methods, parametric models represented by time series models, and non-parametric models including machine learning and deep learning methods. The advantages and limitations of different methods are summarized. Due to the superior performance of deep learning models in time series prediction tasks, more and more scholars begin to adopt deep learning based models to solve the problem of bus travel trajectory prediction, and consider combining the spatial and temporal features exhibited by urban roads to improve prediction accuracy further. Finally, the challenges faced in bus travel trajectory prediction field are analyzed, and future development and research directions in this field are prospected.

Key words: bus travel trajectory prediction, deep learning, spatial-temporal features, time series forecasting, intelligent transportation