基于异质数据的城市公交到站时间预测研究

With the acceleration of urbanization process in our country, predicating public bus arriving time in a metropolis is the key step for information services and public buses surveillance in the transportation management section. Predicating bus arriving time is not only the key problem in the intelligent transportation community, but also the challenging practice problem in the public transportation service. It is possible to efficiently predict bus arriving time, only if public bus transportation is recognized to be affected by the heterogeneous factors, i.e., dynamic road traffic states, the spatio-temporal traveling pattern of people in a city, and data generated from the traffic community. Based on the corporation of the multiple types of heterogeneous data, this proposal will study the predication of public bus arriving time. The main contents include: the mining of the people traveling patterns from social media, the extracting of the road traffic states from surveillance videos for traffics, and the models of predicting bus arriving time from multiple types of data. In brief, the expected results of this proposal will contribute to mine the collaborative evolution of urban road traffic states and traffic demands, expanding the applications of big data technology in the field of intelligent transportation, and improving the urban transportation planning and managements as well as the scientific level of decision-making.

随着我国城市化进程的加快，城市公交到站时间预测，是交通管理部门提供信息服务和公交管控的重要依据。它不仅是智能交通领域的关键科学问题之一，也是城市公共交通亟需解决的实际难点之一。只有充分考虑到城市公交受到多种异质因素影响，既动态道路交通状况、城市人群出行模式的时空分布，并结合行业内的交通数据，才能有效的对公共到站时间进行预测。基于异质数据协同的思想，本项目研究城市公交到站时间预测问题，主要研究内容包括：基于社交媒体的人群出行模式挖掘、基于交通摄像头的道交通参数提取、和基于多种数据的公交到站时间预测模型。预期研究成果对于发掘城市道路交通状态与交通需求的演化规律，拓展大数据技术在智能交通领域的应用，提升城市交通规划、管理决策的科学水平

针对城市公交到站时间预测，本项目面向智能交通的实际需求出发，从多种异质数据中去感知交通状况，从而实现公交到站时间预测。主要研究内容按照项目计划书进行，包括出行热点建模和发现、社交媒体感兴趣内容的挖掘与评价、基于再生神经网络的公交到站研究、基于强化学习的车辆重现检测等方面研究工作，取得了一系列的阶段性研究成果。共发表学术论文9篇，包括TMM、TNNLS、TITS本领域权威国际期刊5篇、Neurocomputing、INformation Sciences等本领域内知名国际期刊2篇、MMM国际会议论文1篇；申请国内发明专利4项，发布了目前世界上最大规模的公交到站时间预测数据集。项目组成员参加了MMM国际会议。. 研究将有助于从理论上系统深入地探讨城市公共交通的复杂影响因素，把握道路交通状态，建立高效和可靠的公交到站时间预测。通过本项目实施，一方面，可以科学的为城市公交运行监控分析和运营管理提供基础理论和方法支持；另一方面，将对提升我国在海量异质数据挖掘、多媒体技术等方面的创新研究。