采用递归有序聚类的信号控制时段划分方法

为了弥补传统聚类思想下的信号控制时段划分算法忽略了交通流量序列的时间特性的缺点，引入有序聚类建立智能化的交通控制时段划分方法.针对特定分割数目下的任意一种可能划分方案，用类表示特定时段内部的数据序列集合，以直径为参数测算类内样本差异性，以所有类内直径总和作为指标衡量划分结果损失值及方案优劣性.为了降低传统有序聚类时间复杂度，引入动态递归策略，建立特定分割数目下最佳方案的快速求解方法，通过识别不同分割个数下最小损失值突变点，获取最佳分割数和最优方案.基于该方法得到的最优划分在实际交通规划中对比常用方法，交通运行效率得到了显著提升.
Abstract: An intelligent partition method for the traffic flow time series data was proposed based on order clustering to compensate for the technical defects of traditional methods which neglect time characteristic of traffic flow for traffic time-of-day (TOD) breakpoints optimization. The parameter of diameter was selected with fixed number of each cluster to measure the difference between any two samples within one cluster. The sum of the diameters was the loss value for this cluster. A fast solution method for seeking the optimal plan among all possible scenarios with known number of cluster was advanced based on dynamic recurrence algorithm in order to reduce the time complexity of the original method. The optimal number of clusters and the TOD plan was determined by identifying the elbow point in the change pattern of the minimum loss values with different numbers of clusters. The optimal partition used in the actual traffic planning can significantly improve the efficiency of traffic operation.