城市群体性活动人群移动模式挖掘与预测

With the development of urbanization, more and more group activities such as tourism, shopping, football game, entertainment, have enriched our lives. This situation of crowd swarming in group activities poses grave challenges for public safety risk and urban planning. Mining human mobility pattern and understanding human behavior belongs to the foundational research in the field of Geographic Information System. Traditional questionnaires and surveys have been unable to uncover the long-term laws of human movements due to limited samples, especially in the fine scale. Fortunately, in recent years, with the development of positioning techniques, such as GPS, WiFi, Cell-ID, and with the widely used of mobile devices, location-based services motive the explosion of spatial-temporal trajectories. It provides critical data sources for the research of human mobility. . In this study, we utilize multi-sources spatial-temporal data including mobile phone data, floating car data and the basic geographic data to reveal the underlying mechanism of people’s movement. First, we need to identify group activities and extract their semantic information. We propose a method of extracting their semantic information of group activities based on Spatial Profile. Second, Spatial-temporal patterns of human convergence and dispersion in citywide group activities are needed to explore in this research. The temporal-spatial distribution model of main transport hubs, which include parking lots, bus stations and the subway exit and entrance near the sites of group activities, can be built to analyze and find the laws of human movements. We bring in a new indicator to measure the people’s moving characteristics by the time and space distance entropy. Then, the gravity model is used to discover these patterns and analyze the influencing factors. Finally, understanding the spatial-temporal patterns of human convergence and dispersion could provide good knowledge of human behavior in group activates. Based on these knowledge, we present a method of predicting the flow of crowds using the method of convolution neural network. The recognition of human mobility pattern could provide strong evidence for urban management and decision-making. This research has important practical significance to reduce public safety risk and improve the operational efficiency of the city.

城市群体性活动因短时局部空间内汇集大量人流、车流等，给城市规划、公共安全风险管控带来巨大挑战，理解与揭示城市群体性活动人群移动模式是人类移动性研究的基础问题之一。传统问卷调查方式难以大规模、长时间地观测和记录人群的空间移动行为，近年，移动定位技术迅猛发展，位置服务应用不断增多，获取时空精细度更高的海量移动对象时空轨迹成为可能，这为城市群体性活动研究提供了重要数据支撑。本课题针对城市群体性活动人群高度聚集所带来公共安全风险问题，立足于手机信令数据、浮动车轨迹、基础地理信息等多源时空数据，重点研究城市群体性活动提取识别，挖掘城市大型活动中人群聚集消散模式，分析城市居民的出行行为规律，基于深度学习开展城市群体性活动人流预测，实现对城市群体性活动人群移动特征分析与挖掘，揭示城市人群空间移动行为时空特征，为城市居民公众出行规划，提高城市运行效率，提升城市管理部门对群体性活动管控预警能力提供决策支持。

城市群体性活动会在较短时间与局部空间内汇集大量的人流、车流等，给城市规划、公共安全风险管理带来巨大挑战，理解与解释城市群体性活动人群移动模式是人类移动性研究的基础问题之一。本项目立足于手机信令数据、浮动车轨迹等多源时空数据，重点研究城市群体性活动提取识别技术，挖掘城市大型活动中人群聚集消散模式，分析城市居民的出行行为规律，本项目取得在主要成果在于：.（1）本研究提出了一种融合多源时空信息的GBDT模型轨迹交通模式识别算法。本方法除了添加基本的特征向量外，还将地理信息添加到特征向量中，实验结果表明，地理信息特征向量在模式识别中的贡献量为12%，交通模式识别精度达到84%，比不添加地理信息特征向量的识别精度提高了2%。.（2）本研究提出了新的项集关联度的概念，在其计算过程中加入了网络嵌入算法 node2vec，使得关联度能综合考虑室内空间实体之间的空间临近和语义倾向等潜在的关联信息，更好地反映关联规则中室内空间实体集合之间的关联强度。接着，本项目提出新的基于支持度-置信度-关联度约束框架的R-FP-growth关联规则挖掘算法，试验结果表明，算法的准确率提高了19%，大大提高了关联规则挖掘结果的质量和在室内位置服务中的应用价值。.（3）本研究提出一种基于众包轨迹的导航网络提取算法，通过轨迹停留点检测与简化，KNN迹自适应栅格化算法以及优化的CFSFDP算法实现了三维导航路网的提取与构建，试验结果表明，相比传统的KDE法分别提高1.83%，拓扑正确率提高13.7%。.本项目经过三年时间研发，总体来说研发过程顺利，共发表论文7篇，其中2篇SCI，登记软件著作权3项，获得科研奖励3项，已完成规定的预期成果考核指标。