居民出行异质性与城市活动结构

Residents’ travel heterogeneity and urban mobility structure

城市活动结构在居民出行过程中会产生动态变化,发现城市活动结构已经成为城市地理学、交通地理学研究的热点问题,对于城市规划与管理、公交线路调配、流量估计等具有重要意义。传统的城市结构发现方法是从计算机技术发展而来的,通常只考虑区域之间的交互,忽视了区域的属性信息,并且在顾及居民出行异质性方面研究不足。本文基于自编码模型,提出一种顾及居民出行异质性探测城市活动结构的表示学习方法,实现静态属性信息与动态出行信息的融合,并引入多元高斯分布模型,旨在完成从节点表示到社区表示的完整过程,实现更为准确的表示结果;借助地理探测器思想,顾及不同社区之间的居民出行异质性特征。试验发现,本文方法能够较为准确地发现城市活动结构以及居民的出行规律,通过与传统方法的比较,证明该方法在居民出行行为的挖掘上具有独特优势,便于从居民出行的角度重新认识城市活动结构。

The urban mobility structure could produce dynamic changes in the process of residents ’ travel. Discovering urban mobility structure has become a hot issue in urban geography and traffic geography, which is of great significance for urban planning and management, line allocation, traffic estimation, etc. Traditional urban structure discovery methods usually consider a single factor, and lack of research in considering the heterogeneity of residents ’ travel. Based on the auto-encoder model, this paper proposes a representational learning method considering the heterogeneity of residents ’ travel to detect the urban mobility structure. Our method realizes the fusion of static attribute information and dynamic travel information and introduces the multivariate Gaussian distribution model to complete the whole process from node expression to community expression considering the residents ’ travel heterogeneity between different communities, finally can achieve more accurate representational results. The experimental results show that the proposed method can more accurately discover the urban mobility structure and the residents ’ travel patterns. Compared with the traditional methods, it is proved that this method has unique advantages in the mining of residents ’ travel behavior.