利用非负矩阵分解识别县域人群活动时空特征

Identification of Spatiotemporal Characteristics of Crowd Activity in County with Non-negative Matrix Factorization

对地区的人群活动时空特征进行识别,有利于认识地区空间结构特质,但现有研究对县域及以下层面的关注相对不足,且用于表征县域尺度人群活动的手机信令数据呈现一定的稀疏性,欠缺系统的数据处理方法。为此,以广东省新兴县为例,选取2020年工作日和周末两个特征日的手机信令数据构建时空矩阵,利用非负矩阵分解方法提取隐含的活动模式特征;并基于这些特征,通过k-means聚类算法得出各类功能区在县域的空间分布。研究表明,新兴县周末的夜间经济繁荣、人群活动模式丰富,中心城区的人群活动强度高,但同周边村镇的联系弱;县域内以居住功能为主导,不存在显著的功能分区。研究结果说明了利用非负矩阵分解方法可有效提取稀疏时空矩阵中的模式特征,可为县域国土空间规划编制提供科学的支持和帮助。

The identification of spatiotemporal characteristics of crowd activity is conducive to the understanding of regional spatial structure characteristics. However, the existing studies have not paid enough attention to the county level and below.Moreover, the mobile signaling data used to characterize the crowd activities at the county level are sparse to a certain extent and systematic data processing methods are lacking. We take Xinxing County in Guangdong Province as an example, select the mobile signaling data of two characteristic days of working day and weekend in 2020 to construct a spatiotemporal matrix, and use the non-negative matrix factorization method to extract the implied activity pattern features. Based on these features, the spatial distribution of various functional areas in this county is obtained by k-means clustering algorithm. The results show that Xinxing County has prosperous economy at night and abundant crowd activity mode on weekends. The intensity of crowd activity in the central city is high, but the connection with the surrounding villages and towns is weak. The county is dominated by residential function, and there is no significant functional partition. The results demonstrate that the non-negative matrix factorization method can effectively extract the pattern features in the sparse spatiotemporal matrix, which can provide scientific support and help for the county land spatial planning.