There are significant spatiotemporal correlations among the traffic flows of neighboring road sections in the road network. Correctly identifying such correlations makes an essential contribution for improving the accuracy of traffic flow prediction. Many efforts have been made by several researchers to solve this issue, but they assume that the spatiotemporal correlations among traffic flows are stationary in both time and space, i.e., the degrees to which traffic flows affect each other are fixed. In this study, we propose a clustering based traffic flow prediction method that considers the dynamic nature of spatiotemporal correlations. In order to express the short-term dependence between the target road section and neighboring ones, the spatiotemporal correlation matrices are introduced. The historical traffic data are divided into several clusters according to the similarity between spatiotemporal correlation matrices. The spatiotemporal correlation analysis and the predictor selection based on the mutual information are performed in each cluster, and the multiple prediction models are trained separately. A prediction model corresponding to the cluster to which the current traffic pattern belongs is selected to output the prediction result. Experimental results on real traffic data show that the proposed method achieves good prediction accuracy by distinguishing the heterogeneity of spatiotemporal correlations among the traffic flows.

路网中相邻路段的交通流量之间存在显着的时空相关性。正确识别此类相关性对于提高交通流预测的准确性做出了重要贡献。一些研究人员为解决这个问题做了很多努力，但他们假设交通流之间的时空相关性在时间和空间上都是固定的，即交通流相互影响的程度是固定的。在这项研究中，我们提出了一种基于聚类的交通流预测方法，该方法考虑了时空相关性的动态特性。为了表达目标路段与相邻路段之间的短期相关性，引入了时空相关矩阵。根据时空相关矩阵之间的相似性，将历史交通数据分成若干簇。在每个簇中进行时空相关分析和基于互信息的预测器选择，分别训练多个预测模型。选择当前交通模式所属的聚类对应的预测模型输出预测结果。在真实交通数据上的实验结果表明，该方法通过区分交通流之间时空相关性的异质性，实现了良好的预测精度。在每个簇中进行时空相关分析和基于互信息的预测器选择，分别训练多个预测模型。选择当前交通模式所属的聚类对应的预测模型输出预测结果。在真实交通数据上的实验结果表明，该方法通过区分交通流之间时空相关性的异质性，实现了良好的预测精度。在每个簇中进行时空相关分析和基于互信息的预测器选择，分别训练多个预测模型。选择当前交通模式所属的聚类对应的预测模型输出预测结果。在真实交通数据上的实验结果表明，该方法通过区分交通流之间时空相关性的异质性，实现了良好的预测精度。