基于狄利克雷过程混合模型的城市活动聚类方法研究

手机信令数据不仅记录个体出行轨迹,也为分析城市活动空间分布特征提供了基础.本文提出一种基于狄利克雷混合模型的城市活动特征聚类方法,以手机信令提取居民出行OD为基础,将每个基站的到发出行量作为表征该基站所处空间位置的活动特征,研究特征的聚类方法.引入狄利克雷分布作为先验分布,由中餐馆模型推定特征聚类数量.与其他聚类方法相比,该方法最大的优点在于无需事先指定聚类数量,避免了传统聚类方法的缺陷.将本文方法应用到三亚市城市活动特征聚类当中,结果能够有效地反应不同城市功能组团的活动特征.

Urban Activity Clustering Method Based on Dirichlet Process Mixture Model

The mobile phone data not only provides the track record of individual travel, but also provides a basis for analyzing the spatial distribution characteristics of urban activities. This paper proposes a city activity feature clustering method based on the Dirichlet Process Mixture Model. Based on the residents' OD matrix extracted from mobile phone signaling, the number of arrivals and departures of each base station is used as a representation of the spatial location of the base station. The activity characteristics of the research feature clustering method. This method introduces Dirichlet distribution as the prior distribution, and estimates the number of feature clusters from the Chinese restaurant model. Compared with other clustering methods, the major advantage of this method is that there is no pre- specified number of clusters in the method, which avoids the defect of pre- specifying the cluster number in traditional clustering methods. This method is applied to the clustering of urban activity characteristics in Sanya, which can effectively reflect the activity characteristics of different urban functional groups.