周期数据时段划分的双截断高斯混合模型及其EM算法

有限混合模型是多模态数据拟合和聚类的有力工具,本文针对具有多模态的周期数据提出了双截断高斯混合糢型,并推导出相应的EM算法,再通过BIC准則确定混合成分个数,该方法的优点是可以将相邻周期上距离较近的数据聚为一类.模拟研究显示,在具体参数设置下,EM算法和BIC准则是相合的。最后,该方法应用于车流量数据的时段划分,将一天划分为具有显著特征的6个时段,有助于交通部门采取相应策略,为优化交通灯信号配时提供参考依据.

Double Truncated Gaussian Mixture Model and Its EM Algorithm for Time Segmentation of Periodic Data

Finite mixture model is a powerful tool for fitting and clustering of multimodal data.In this paper,a double truncated Gaussian mixture model is proposed for periodic data with multiple modes,and the corresponding EM algorithm is derived.Then the number of components is determined by BIC criterion.The advantage of this method is that the data points closed to each other in the adjacent periods can be clustered into one group.The simulation results show that the EM algorithm and the BIC criterion are consistent under the specific parameter setting.Finally,the method is applied to the time segmentation of traffic flow data,dividing one day into six time periods with significant characteristics,which is helpful for traffic departments to adopt corresponding strategies and provide reference for the optimal allocation of traffic lights.