动态复杂网络视角下的人类移动时空流异常模式挖掘方法

人类移动形成的时空流中蕴含着一种显著偏离移动过程时空演化规律的异常模式，对这种时空流异常模式的挖掘已成为地理空间大数据分析的一项重要研究内容。当前相关研究大多停留在数据表象层面的异常模式探测识别，缺乏对人类移动过程时空演变机制的深层挖掘，属于浅层计算，难以发现隐藏在人类移动复杂演化过程背后的深层异常模式。为此，本项目从基于动态复杂网络建模深层认知时空演化机制的视角出发，首先协同多源异构城市地理空间大数据，通过构建动态复杂网络定量描述人类移动时空流之间的拓扑结构特征；进而，综合考虑复杂网络多尺度结构特征约束，联合时空统计计算与图嵌入特征学习，揭示时空流演化过程蕴含的深层规律；最后，基于多层偏离度的统计建模，融合对内-外要素协同作用机制的挖掘，精准诊断偏离演化规律的时空流异常模式。本项目研究有望支撑突发交通拥堵、区域人群异常聚集等城市公共安全隐患的预测预警，辅助提升城市智能化管理与决策水平。

In the spatio-temporal flow produced by human mobility, there contains a kind of anomaly patterns with significant deviations from the spatio-temporal evolution regularities of human mobility. The mining of such spatio-temporal flow anomaly patterns has become an essential research contents in geospatial big data analysis. Current related researches mostly aim at identifying anomaly patterns based on the surface level of data computing. As being lack of mining the spatio-temporal evolution mechanism of human mobility, it is difficult for existing methods to deeply capture spatio-temporal flow anomaly patterns hidden behind the sophisticated evolution process of human mobility. In light of this, this project plans to conduct spatio-temporal flow anomaly pattern mining from the perspective of dynamic complex network modeling based evolution mechanism recognition. Firstly, through integrating multi-source heterogeneous urban sensing big data, the dynamic complex network is constructed to quantitative represent the topological structure characteristics among spatio-temporal flows that characterize the process of human mobility. On this basis, with the combination of spatio-temporal statistic computation and graph embedding based feature learning, we reveal the deep regularities implied in the spatio-temporal evolution process of human mobility by considering the constraints of multi-scale structure characteristics of complex networks. Finally, the statistical modeling of multi-level deviation degree and the coupling of quantitative effects of multiple factors on human mobility are carried out for the accurate detection of those spatio-temporal flow anomaly patterns. Through the research of this project, it will be valuable in supporting the forecasting and early warning of urban public security risks (i.e., unexpected traffic congestion, regional anomalous aggregation of populations, etc.) and promoting the urban intelligent management and decision level.