关键词: 城市交通, 交通信号时段划分, 流形学习算法, K均值聚类, 信号控制

Abstract: In the time of day control schemes，the basic problem is to program the traffic intervals rationally.But，as the traditional main means to fix the lengths of intervals，the artificial method has subjectivity and one-sidedness.To avoid the shortcomings，this paper puts forward a new method to fix the lengths of intervals.Based on the concepts of manifold and manifold learning algorithm，it raises an estimation method which can discover the intrinsic dimensions of the flow data in isolated intersection.Firstly，the paper supposes the measured data that has its lower-dimensional manifold embedded in the high-dimension manifold；with Isomap algorithm，it finds out the intrinsic dimensions；finally using the reduced sample data，it clusters them with K-means algorithm and gets corresponding traffic intervals.The results indicate that the Isomap and K-means algorithm based clustering method outperforms the traditional artificial method as well as the other methods in the traffic signal periods division of day.

Key words: urban traffic, traffic intervals programming, manifold learning algorithm, K-means clustering algorithm, traffic signal control