基于改进递归小波变换的交通流异常点与变点检测算法  

An Algorithm for Detecting Outlier and Change Point of Traffic Flow Based on Improved Recursive Wavelet Transform

作　　者：商明菊 胡尧 周江娥 SHANG Ming-ju;HU Yao;ZHOU Jiang-e(School of Mathematics and Statistics,Guizhou University,Guiyang Guizhou 550025,China;Guizhou Provincial Key Laboratory of Public Big Data,Guiyang Guizhou 550025,China)

机构地区：[1]贵州大学数学与统计学院,贵州贵阳550025 [2]贵州省公共大数据重点实验室,贵州贵阳550025

出　　处：《公路交通科技》2019年第8期133-143,共11页Journal of Highway and Transportation Research and Development

基　　金：国家自然科学基金项目(11661018);贵州省科技计划项目(黔科合平台人才[2017]5788号)

摘　　要：为准确界定交通流状态,辅助交通管理者对交通异常事件进行及时处理,提出了一种基于改进递归小波变换的异常点与变点快速检测算法,并将其应用于交通流实时监控与预警。首先,对历史交通流序列建立自回归模型,将残差序列的标准化有效分数向量作为统计量,利用3-Sigma原则,提出为统计量差分时序设定监控阈值的方法,实现了交通流状态的实时预警。其次,利用改进递归小波变换统计量,结合小波复合信息并综合考虑真实变点与估计变点之间的差异,选取小波变换特征频率与最优搜索长度,快速检测并估计交通流异常点与变点,实现了交通流状态的在线监控。最后,仿真试验和实例分析验证了算法的合理性与可行性。研究结果表明:设定的阈值对交通流变化趋势掌控明显,能够对交通异常状态进行及时预警;结合特征频率的复小波变换信息,能够有效检测并区分交通流异常点与变点;与基于有效分数向量的传统变点检测算法相比,算法的检测性能在延迟与收敛性两方面均有明显改善。该算法能够对交通流状态进行在线监控,这将为断面车流实时预警提供支持。In order to define the traffic flow state accurately and assist traffic managers to deal with traffic abnormal events in time, an algorithm for fast detecting outlier and change point based on improved recursive wavelet transform is proposed, and the algorithm is applied to real-time monitoring and forewarning of traffic flow. First, an autoregressive model for the historical traffic flow sequence is established. Taking the normalized efficient score vectors of residuals as the statistics, a method of setting the monitoring threshold for differential time series of the statistics is proposed by using 3-Sigma principle to real-time forewarn the traffic flow state. Second, the statistics are transformed by using improved recursive wavelet. By combining the composite information of wavelet transform and considering the difference between the real change point and the estimated change point, the characteristic frequency of wavelet transform and the optimal search length are selected. As a result, the outliers and change points of traffic flow are detected and estimated rapidly, and the online monitoring of traffic flow state is realized. Finally, the rationality and feasibility of the algorithm are verified by simulation experiment and case analysis. The result shows that (1) the trend of traffic flow can be effectively monitored with the set threshold, and the abnormal traffic state can be warned in time;(2) by combining the complex wavelet transform information of the characteristic frequency, the outliers and change points of traffic flow can be effectively detected and distinguished;(3) compared with the traditional change point detection algorithm based on the efficient score vector, the detection performance of the proposed algorithm can be significantly improved in terms of delay time and convergence. The algorithm is effective in the online monitoring of traffic flow state, which will provide a support for real-time forewarning of section traffic flow.

关 键 词：城市交通 变点 改进递归小波变换 交通流 异常点 

分 类 号：U491.112[交通运输工程—交通运输规划与管理]