基于大数据的城市级别人流预测和影响评估

The rapid progress of urbanization leads to massive people streaming into cities, generating not only serious traffic problems but also catastrophic accidents (e.g. the bloody stampede occurred in Shanghai on the first day of 2015). If being able to predict the flow of crowds throughout an entire city, we can launch emergency mechanisms, such as sending security alerts and conducting traffic controls, in a location before it would be overloaded by people. In this research project, first, we aim to predict the flow of crowds in each and every location of a city (i.e. city-scale) over the next time interval, by fusing a diversity of datasets, such as traffic flow, road networks, points of interests, meteorology, and social media. The location can be a road segment, or a region bounded by major roads. The flow of crowds can be denoted by the number of people traveling by different transportation modes, including driving, subway, bus, bike, and on foot. Thus, the flow of crowds in a location can be the volume of vehicles traversing a road segment, or the number of people entering (or departing from) a region. Second, we would like to detect anomalous events in a city according to the flow of crowds, and further diagnose the root course of those anomalies by using big data. Third, we would like to estimate the impact of a significant flow increase in a location, forecasting its damaging level, impacting area and cascading path. We will build a prototype system to demonstrate the aforementioned three functions. The research can inform governmental officials’ decision making on traffic management and public security, protecting people and our cities from the damage of catastrophic accidents.

人口密度渐增与人群流动的动态性和不可控性极易造成拥堵或异常事件发生，给城市带来不可估量的安全隐患与经济损失，已成为城市发展瓶颈。人流预测是实现流动需求与资源相匹配，缓堵降异的核心，迫切性不言而喻。本项目以解决城市面临的实际挑战为出发点，以城市数据为基础，数据驱动城市发展为主线，提供一体化的城市管理辅助解决方案。主要研究点：（1）建立有效的人流预测模型，融合多源异构数据实现人群流动方向的准确而实时的预测，挖掘人流大量聚集的热点区域；（2）研究异常事件检测模型，通过融合多源异构数据来实时捕捉人群流动过程出现的异常，分析和解释异常出现的原因；（3）建立影响力传播模型，推演人流聚集、异常事件发生后可能导致的后果或影响的传播模式；（4）搭建可预测未来人群流动、异常事件检测与后果预知的一体化城市管理辅助系统，为城市管理者制定更加精准而有效的应急预案提供辅助，有效保障城市安全，降低经济损失。

本项目以解决城市面临的实际挑战为出发点，以城市数据为基础，数据驱动城市发展为主线，提供一体化的城市管理辅助解决方案。本项目按照预定计划对基于大数据的城市级别人流预测和影响评估进行了深入的探索与研究，并按时、按量甚至大幅度超额完成研究计划中的预期研究成果。在该项目受资助期间，共发表国际顶级会议或期刊论文39篇，大部分发表于CCF推荐A类会议或A类期刊上，相关研究成果获得了领域内的广泛认可，提出的ST-ResNet模型被引用次数达到700+。项目的主要研究成果包括：首先，针对海量轨迹数据的管理问题，我们开发了一个高效的轨迹数据管理系统，弥合了轨迹数据与城市应用之间的鸿沟，作为后续研究工作的数据基石。其次，针对城市级别人流预测和异常检测等问题，我们提出了一系列有效的算法模型，融合多源异构数据实现人群流动方向的准确而实时的预测，挖掘人流大量聚集的热点区域。再进一步，我们利用收集管理的轨迹数据对其他若干城市应用问题进行了研究，包括空气质量预测、共享单车车道规划以及资源的动态分配等问题。