Journal of Transportation Systems Engineering and Information Technology ›› 2018, Vol. 18 ›› Issue (5): 129-135.

• Systems Engineering Theory and Methods • Previous Articles     Next Articles

An Online Estimation Method for Passenger Flow OD of Urban Rail Transit Network by Using AFC Data

JIANG Xi, JIA Fei-fan, FENG Jia-ping   

  1. State Key Lab of Rail Traffic Control and Safety, Beijing Jiaotong University, Beijing 100044, China
  • Received:2018-05-22 Revised:2018-07-10 Online:2018-10-25 Published:2018-10-26

基于AFC数据的城轨路网客流OD在线动态估计

蒋熙*,贾飞凡,冯佳平   

  1. 北京交通大学 轨道交通控制与安全国家重点实验室,北京 100044
  • 作者简介:蒋熙(1971-),女,四川江油人,副教授,博士.
  • 基金资助:

    国家重点研发计划/ National Key Basic Research Program of China (2016YFB1200402);国家重点实验室自主课题/ State Key Lab Self-topic Program (RCS2018ZT005).

Abstract:

The real-time passenger flow of the urban rail transit network is the main basis for the scientific decision-making of the daily operation organization. The timely and accurate online estimation of the passenger flow OD is a prerequisite. The problem of passenger flow OD online dynamic estimation by using real-time AFC data is analyzed. A dynamic estimation method for passenger flow OD combined with machine learning and recursive Bayes is proposed; the LSTM based OD state transfer model and the passenger flow OD recursive Bayesian estimation model embedded with the LSTM model are constructed. Considering the nonlinearity and uncertainty of the OD dynamic state, a particle filter based method is proposed to solve the OD recursive Bayesian estimation problem. Oriented to the third-order Markov process of state transition formed by the LSTM model embedding, the general particle filter algorithm is extended in high order and the implementation of the algorithm is studied. Finally, an example is used to verify the method proposed in this paper.

Key words: traffic engineering, passenger flow OD online estimation, particle filter, urban rail transit, LSTM mode

摘要:

路网实时客流状态是城市轨道交通日常运营组织科学决策的主要依据,而精准地在线估计客流OD是前提条件.本文分析了准实时AFC数据接入条件下客流OD在线动态估计问题及其特点,提出了将机器学习与递归贝叶斯相结合的客流OD动态估计方法;构建了基于LSTM的客流OD状态转移模型,以及LSTM模型嵌入下的客流OD递归贝叶斯估计模型;针对客流OD状态变化的非线性、不确定性特点,提出基于粒子滤波算法求解客流OD递归贝叶斯估计问题.面向LSTM模型嵌入所形成的客流OD状态转移三阶马尔科夫过程,对一般的粒子滤波算法进行高阶扩展,研究了算法的实现.最后用实例对本文提出的方法进行了验证.

关键词: 交通工程, 客流OD在线动态估计, 粒子滤波, 城市轨道交通, LSTM模型

CLC Number: