城市交通智能治理大数据计算平台及应用示范研究  

In Press Research Progress on a Big-data Based Smart Urban Transportation Management System and Its Application

作　　者：黎旭成 陈振武 张晓春 李熙莹 赵一新[3] 李健[4] 谭墍元[5] 刘维怡 LI Xucheng;CHEN Zhenwu;ZHANG Xiaochun;LI Xiying;ZHAO Yixin;LI Jian;TAN Jiyuan;LIU Weiyi(Shenzhen Urban Transport Planning Center Co.,Ltd.,Shenzhen 518021;School of Intelligent System Engineering,Sun Yatsen University,Guangzhou 510006;China Academy of Urban Planning and Development,Beijing 100044;College of Transportation Engineering,Tongji University,Shanghai 201804;School of Electrical and Control Engineering,North China University of Technology,Beijing 100144)

机构地区：[1]深圳市城市交通规划设计研究中心股份有限公司,深圳518021 [2]中山大学智能工程学院,广州510006 [3]中国城市规划设计研究院,北京100044 [4]同济大学交通运输工程学院,上海201804 [5]北方工业大学电气与控制工程学院,北京100144

出　　处：《中国基础科学》2021年第2期12-23,共12页China Basic Science

基　　金：国家重点研发计划项目(2018YFB1601100)。

摘　　要：我国有8亿城市化人口,每天出行约15亿人次,主要大城市每天出行人次超千万次。虽然各城市都广泛采用基础设施建设和政策调控等措施,但交通拥堵情况依然严重。本文主要介绍以"感-知-判-算-治"为主线,围绕提升交通全面感知能力、高可靠实时研判和态势推演能力、大规模高效计算能力,构建高效智能的城市交通智能治理大数据计算平台,进而缓解交通拥堵问题上所取得的研究进展。在"感"方面,建立了万路级视频分析引擎,实现了人、车特征识别率> 80%,跨摄像头重识别率> 60%;在"知"方面,研究了城市交通知识图谱的标准化设计方法,构建了包括2.87亿个实体和12亿条关系的深圳市交通知识图谱;在"判"方面,利用深圳市静态测试数据构建了大规模复杂系统推演算法,实现15 min交通状态的预测时间2.2 s,预测精度达94.09%;在"算"方面,研发了"云-边-端"协同的交通大数据智能平台软硬件,支持大规模智能治理大数据平台的集成与部署;在"治"方面,依托研发建成的城市交通智能治理Trans PaaS平台,在深圳和宣城实现了重大活动应急疏散、异常事件快速响应、出行服务、路段交通溯源、交叉口群信号协同五大示范应用。With over 1.5 billion daily trips travelled by about 800 million urban population in China,there are over 10 million trips per day in major cities.Although measures such as infrastructure constructions and new transport policies have been applied in various cities,traffic congestion is still a critical issue.This paper sets up the research pipeline as "Recognizing,Understanding,Judging,Computing,and Managing",aiming at building a highly efficient and smart urban transportation management system with comprehensive state perception ability,highly reliable real-time deduction and decision-making ability,and large-scale computation power,to support the mitigation of urban traffic congestion.For the research outcomes,the"Recognizing"ability is improved by building a millionchannel video structuring engine with the accuracy over 80% for vehicle and pedestrian recognition,and over 60% for the re-identification across multiple cameras;in the aspect of "Understanding" ability,the abstraction method of transportation knowledge graph is standardized,and a Shenzhen transportation knowledge graph with 287 million entities and 1.2 billion relationships is built;in terms of“Judging”ability,a large-scale complex traffic system simulation of Shenzhen is constructed and achieves a prediction time of 2.2 seconds and prediction accuracy of 94.09% for the traffic state of next 15 minutes;for the "Computing" ability,a serial of edge-cloud computing cooperation software and hardware are developed to support the integration and deployment of comprehensive intelligent analytical algorithms in the system;all technical developments are then integrated to form the"Managing"system,and five demo applications,including emergency evacuation,event management,smart mobility services,smart traffic estimation and guidance,and traffic signal group coordination,have been implemented in Shenzhen and Xuancheng.

关 键 词：多源数据融合 知识图谱 交通仿真 大数据计算 城市交通治理平台 

分 类 号：U495[交通运输工程—交通运输规划与管理]