基于VAR-LRTC-TNN的交通流量数据补全框架模型  

Traffic Flow Data Completion Framework Model Based on VAR-LRTC-TNN

作　　者：孙秋霞 王淇 李勍 孙璐 贾秀燕 SUN Qiuxia;WANG Qi;LI Qing;SUN Lu;JIA Xiuyan(College of Mathematics and Systems Science,Shandong University of Science and Technology,Qingdao 266590,Shandong,China;Business School,Qingdao University of Technology,Qingdao 266520,Shandong,China)

机构地区：[1]山东科技大学数学与系统科学学院,山东青岛266590 [2]青岛理工大学商学院,山东青岛266520

出　　处：《重庆交通大学学报（自然科学版）》2024年第6期47-53,86,共8页Journal of Chongqing Jiaotong University(Natural Science)

基　　金：山东省自然科学基金项目(ZR2021MF113);青岛市惠民基金项目(22-3-7-smjk-16-nsh)。

摘　　要：从各类传感系统收集到的交通流数据往往会因探测器或通信故障等缘故出现数据连续性的缺失,故准确补全缺失的交通流数据对制定合理的交通管理策略至关重要。鉴于交通流数据具有低秩的特性,通过低秩张量补全模型可较好地刻画出交通流数据的全局一致性,但却无法很好地捕捉数据的局部变化趋势,一定程度上影响了效果。基于此,提出了将VAR模型和基于残差序列的LRTC-TNN模型相结合的交通流补全框架模型;采用VAR模型对缺失数据进行粗略估计,移除平均趋势,利用LRTC-TNN模型对残差时间序列进行补全,再将平均趋势还原,从而完成对交通流量数据的高精度补全;该方法不仅保留了交通流数据的全局结构,还考虑了数据局部变化的特征。研究结果表明:与基于原始交通流量数据的填充方法相比,该模型框架对单传感器和多传感器数据的连续性缺失均具有更高的补全精度。The traffic flow data collected from various sensing systems frequently occurs continuously data missing due to detector and communication failure.Therefore,it is crucial to accurately complete the missing traffic flow data to develop a reasonable traffic management strategy.Given the low-rank nature of traffic flow data,the low-rank tensor completion model could effectively characterize the global consistency of traffic flow data,but it failed to well capture the local change trend of the data,which to some extent affected the effectiveness.In view of the above problems,a traffic flow completion framework model by combining VAR model with LRTC-TNN model based on residual sequence was proposed.The VAR model was used to roughly estimate the missing data and remove the average trend.The LRTC-TNN model was used to complete the residual time series and then restore the average trend,so as to achieve high-precision completion of traffic flow data.The proposed method not only retained the global structure of traffic flow data,but also considered the characteristics of local variation of the data.The research results indicate that the proposed model framework has higher completion accuracy for the continuously missing of data for single and multiple sensors than the filling method based on the original traffic flow data does.

关 键 词：交通工程 智能交通 交通流量填充 向量自回归模型 张量补全 缺失数据 

分 类 号：U491.14[交通运输工程—交通运输规划与管理]