一种支持轨迹大数据潜在语义相关性挖掘的谱聚类方法

A Spectral Clustering Method for Big Trajectory Data Mining with Latent Semantic Correlation

针对交通管理优化和轨迹大数据挖掘的实际应用需求,本文提出了一种支持交通轨迹大数据潜在语义相关性挖掘的交通路网谱聚类方法(TSSC).首先研究了交通轨迹数据的向量空间建模方法,其次通过随机投影法快速提取大规模轨迹数据矩阵的特征信息并构建其低维语义子空间,然后基于语义子空间挖掘轨迹数据的潜在语义相关特性,在此基础上通过谱聚类方法实现了交通路网的快速聚类.通过本文提出的方法对总里程1400多万公里的实际交通轨迹数据进行实验分析表明,本方法可根据交通轨迹大数据的潜在语义相关性对交通路网进行快速的谱聚类处理,从而在复杂的交通路网间快速挖掘其潜在特性,为交通规划及其管理优化提供决策支持信息,同时也为时空大数据的聚类挖掘提供了一种新的解决方案.

To facilitate traffic understanding, planning and management optimization, we present a new spectral clustering method（TSSC） for big trajectory data mining based on latent semantic correlation. First, a matrix model is proposed to represent ve- hicle trajectories and the underlying road -twork with a grid-vehicle matrix,which is then transfotmed to a low-dimensional seman- tic subspace with random projection. Second, through matrix decomposition we extract hidden characteristics of the mass trajectory data and construct a similarity matrix for road network cells. Third, we adopt and implement a fast spectral clustering method to discover road network clusters based on the similarity matrix in the semantic space.Finally, we evaluate our approach with a large tra- jectory data set collected by the Fujian Communications Department, which has 19,719 vehicles and a total mileage of more than 14 million kilometers. Experiment results show that the approach can efficiently cluster the road network with traffic context semantic information derived from massive trajectory data. The approach is capable to discover inherent characteristics of complex road net- works and provide insights for traffic planning and management optimization.