基于频繁序列模式挖掘的卡口短时交通量预测  

Short-Term Traffic Flow Forecasting at Intersections Based on Frequent Sequence Pattern Mining

作　　者：刘冉 李岩 毛海虓[1] 钱剑培 王继峰[1] 马悦 LIU Ran;LI Yan;MAO Haixiao;QIAN Jianpei;WANG Jifeng;MAYue(China Academy of Urban Planning&Design,Beijing 100037,China;Research Institute of Standards and Norms Ministry of Housing and Urban-Rural Development,Beijing 100835,China)

机构地区：[1]中国城市规划设计研究院,北京100037 [2]住房和城乡建设部标准定额研究所,北京100835

出　　处：《城市交通》2023年第4期87-98,共12页Urban Transport of China

基　　金：国家重点研发计划资助项目“基于城市高强度出行的道路空间组织关键技术”(2020YFB1600500)。

摘　　要：基于数据的城市交通管理和控制方法是广大学者和交通管理部门的关注重点。以频繁序列模式挖掘算法为基础,对卡口车辆轨迹序列进行时空特征分析。选用7种典型的机器学习算法进行预测,并分析了卡口空间区位、交通量以及连接道路等级对预测精度的影响。研究结果表明,集成学习算法特别是RF的预测性能最好,误差较小且训练速度快;SVR和神经网络算法(MLP、LSTM)在预测误差表现上相近,但是基于神经网络算法的预测模型耗时较长。此外,不同模型的预测误差在空间上的分布具有相似性,在卡口密布的区域预测精度更高,在外围边缘区域误差较大;卡口交通量越大、连接的道路等级越高,预测精度越高。随着城市交通电子卡口设备在路网中的完善,该预测方法的准确性可以进一步提高。Data-based urban traffic management and control strategies are major focus of scholars and traffic management departments.This paper conducts a spatiotemporal features analysis on the trajectory sequences based on frequent sequence pattern mining algorithms.Seven typical machine learning algorithms are employed for short-term traffic flow prediction,and the impact of spatial location,traffic volume,and grade of intersecting roads on prediction accuracy are analyzed.The results reveal that ensemble learning algorithms,particularly the RF model,demonstrate superior predictive performance with smaller errors and faster training speeds.SVR and neural network algorithms(MLP,LSTM)show comparable predictive error performance,but neural network-based models are more time-consuming.Besides,the prediction errors of the model were similar in space.The prediction accuracy is higher in the area where the checkpoints are densely distributed,and lower in the periphery area.The prediction error is lower where traffic volume is larger and the connected roads with higher grades.With the improvements of electronic devices at checkpoint in urban road network,the forecasting accuracy can be further enhanced.

关 键 词：短时交通流量预测 频繁序列模式挖掘 机器学习 

分 类 号：U491.14[交通运输工程—交通运输规划与管理]