中文摘要：

为了克服人为地对交通控制进行时段划分的随意性和K-means方法本身的缺陷,使用谱聚类算法得到最优的时段划分结果.选择道路交叉口各相位的流量作为聚类数据,以尽可能代表交叉口的状态,识别出动态交通中的不同交通模式.对谱聚类中的经典NJW（Ng-Jordan-Weiss）算法进行改进,得到初始时段划分结果,再进行离群点的修正后,得到给定聚类数目下的时段划分结果.通过Synchro软件为每个时段建立最佳信号配时方案,使用SimTraffic对不同聚类数目下的时段划分结果进行仿真评价,以选择最佳的聚类数目.与K-means方法仿真对比结果表明：提出的方法使得总延误减少了6.8%、停车次数降低了5.4%.

英文摘要：

In order to overcome the artificial arbitrariness on time-of-day breakpoints determination in traffic control and the defects of K-means method,spectral clustering algorithm was used to obtain the optimal time-of-day（TOD）breakpoints.Each phase＇s volume was selected as clustering data to reflect the system status as closely as possible and identify different traffic patterns in the dynamic traffic movements.The classic Ng-Jordan-Weiss（NJW）spectral clustering algorithm was improved to obtain an initial breakpoints,and then the final breakpoints at a given number of clusters were obtained after dealing with the outliers.After the best timing plans were obtained by Synchro software for each period,the best numbers of clusters was selected by simulation evaluation under different numbers of clusters using SimTraffic.And comparing with K-means method on simulation results,the proposed method contributed to a reduction of 6.8%in total delay and 5.4%in stops.