基于Gamma分布的交通流时间序列分割模型

Gamma distribution based traffic flow time series segmenting model

准确获取交通流量变化点,对后续的交通流预测、分类及多时段控制具有重要意义.鉴于交通流时间序列的非负性及异方差性,采用Gamma分布拟合交通流时间序列,并对其进行有效分割.针对多元交通流时间序列,首先利用非负主成分分析方法实现降维并提取特征序列,之后利用最大似然估计得到Gamma分布参数,通过不同参数的Gamma分布拟合特征序列的不同片段,并由赤池信息准则(AIC)确定最优分割边界及分割阶数.实验结果表明,所建立的分割模型能够反映不同时段的交通流变化,与现有分割方法相比,取得了更好的分割结果.

It is significant to accurately obtain the change points of traffic flow for the subsequent traffic flow prediction,classification and multi-time control.Considering the nonnegative and heteroscedasticity,the traffic flow time series are fitted by Gamma distribution and segmented effectively.For multiple traffic flow time series,dimension reduction is carried out by the nonnegative principal component analysis(NPCA)for feature extraction.Then,the likelihood of the principal component is constructed to obtain the parameters of the Gamma distribution.Consequently,the change points are determined from degree of fitting using the different parameters of the Gamma distribution by maximizing the likelihood.The Akaike information criterion(AIC)is used to select the optimal segmentation order and boundary.The experimental results indicate that the proposed segmenting model can reflect the change of traffic flow at different time and has better segmentation results than other existing methods.