摘要
基于GPS轨迹数据的交通调查技术能够有效弥补传统居民出行调查方式的不足,该技术在高峰拥堵时段的交通方式识别效果有待进一步研究.针对公交车和小汽车识别精度较低的问题,本文提出基于支持向量机(SVM)的流程优化方法,加入基于短时傅里叶变换(STFT)的频域属性,利用遗传算法(GA)对SVM的惩罚系数和核参数进行联合优化,评估不同交通状态下交通方式和方式转换点的识别效果.结果表明:频域属性的加入能够有效提升交通方式识别精度,在道路畅通状态和一般拥堵状态下,交通方式和方式转换点的识别效果均较为理想;在严重拥堵状态下,机动化方式易与非机动化方式相混淆,方式转换点最大识别误差在13 min以内,相比于基于主观回忆的人工问卷调查方式仍具有参考性.
This study focuses on the transportation mode recognition for the Global Positioning System(GPS)-based travel survey technology.The study proposed a procedure optimization method that is based on the Support Vector Machine(SVM)to improve the recognition accuracy of buses and cars.The proposed model included the new frequency domain features generated from Short-time Fourier Transform(STFT).The Genetic Algorithm(GA)was used to optimize the penalty parameter and the nuclear parameter of SVM.The recognition results of the transportation modes and mode transfer time under different traffic conditions were evaluated,and the result showed the newly added frequency domain features effectively improved the recognition accuracy of the transportation modes.In the free-flow and slightly congested traffic conditions,the transportation mode recognition and mode transfer time both obtained satisfied results.In severe congestions,the motorized modes are relatively easy to be mixed with the non-motorized modes.The maximum error of mode transfer time is within 13 minutes,which might still be informative compared with traditional manual questionnaire surveys.
作者
杨飞
姜海航
刘好德
姚振兴
霍娅敏
周子一
YANG Fei;JIANG Hai-hang;LIU Hao-de;YAO Zhen-xing;HUO Ya-min;ZHOU Zi-yi(School of Transportation and Logistics,Southwest Jiaotong University,Chengdu 611756,China;Urban Transportation Center,China Academy of Transportation Science,Beijing 100029,China;School of Highway,Chang'an University,Xi'an 710054,China;School of Transportation,Southeast University,Nanjing 211189,China)
出处
《交通运输系统工程与信息》
EI
CSCD
北大核心
2020年第4期83-89,105,共8页
Journal of Transportation Systems Engineering and Information Technology
基金
国家重点研发计划(2018YFB1601300)
中央高校基本科研业务费专项资金(300102219301)
国家自然科学基金(51678505).
关键词
智能交通
交通方式识别
支持向量机
GPS轨迹数据
遗传算法
频域属性
intelligent transportation
transportation mode recognition
support vector machine
GPS trajectory data
genetic algorithm
frequency domain feature
