1.中山大学智能工程学院 / 广东省智能交通系统重点实验室, 广东 深圳 518107
2.鹏城实验室, 广东 深圳 518055
王纪禹(1998年生),男;研究方向:智能交通系统;E-mail:wangjy365@mail2.sysu.edu.cn
何兆成(1977年生),男;研究方向:智能交通系统;E-mail:hezhch@mail.sysu.edu.cn
纸质出版日期:2024-09-25,
网络出版日期:2024-07-22,
收稿日期:2024-03-31,
录用日期:2024-04-22
移动端阅览
王纪禹,陈锐祥,何兆成等.基于时空特征共享的多源交通数据修复[J].中山大学学报(自然科学版)(中英文),2024,63(05):167-176.
WANG Jiyu,CHEN Ruixiang,HE Zhaocheng,et al.Joint imputation of multi-source traffic data based on shared multi-dimensional spatiotemporal feature[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2024,63(05):167-176.
王纪禹,陈锐祥,何兆成等.基于时空特征共享的多源交通数据修复[J].中山大学学报(自然科学版)(中英文),2024,63(05):167-176. DOI: 10.13471/j.cnki.acta.snus.ZR20240091.
WANG Jiyu,CHEN Ruixiang,HE Zhaocheng,et al.Joint imputation of multi-source traffic data based on shared multi-dimensional spatiotemporal feature[J].Acta Scientiarum Naturalium Universitatis Sunyatseni,2024,63(05):167-176. DOI: 10.13471/j.cnki.acta.snus.ZR20240091.
本文提出了一种基于多维度特征共享的多层稀疏张量分解模型(MFS-MSTD)。该模型在CP(CANDECOMP/PARAFAC)分解的基础上,对时空因子矩阵施加低秩正则化;并采用共享低秩因子矩阵的机制表达多源交通数据之间的互补性,在非随机缺失或路段级缺失的场景下完成了因子矩阵的梯度更新。实例验证表明:在速度数据非随机缺失场景下,MFS-MSTD在RMSE、MAE和MAPE三个误差指标上相较于基线方法平均降低17%、21%和18%;在流量数据非随机缺失场景下,RMSE、MAE和MAPE平均降低52%、54%和33%。面对更复杂的路段级缺失场景,MFS-MSTD的修复性能优于TGMC-S和MTNTF两个基线模型,能很好地拟合出未观测路段的交通流量变化。
This paper proposes a multi-layer sparse tensor decomposition model based on multidimensional feature sharing(MFS-MSTD). On the basis of CP(CANDECOMP/PARAFAC) decomposition, this model applies low rank regularization to the spatiotemporal factor matrix. The mechanism of sharing a low rank factor matrix is adopted to express the complementarity between multi-source traffic data, and the gradient update of the factor matrix can be completed in non-random missing or segment level missing scenarios. A real experment results show that in the scenario of non-random missing speed data, MFS-MSTD reduces the RMSE, MAE, and MAPE by an average of 17%, 21%, and 18% compared to the baseline method; in the scenario of non-random missing traffic data, RMSE, MAE, and MAPE decreased by an average of 52%, 54%, and 33%. In the face of more complex road segment missing scenarios, the imputation performance of MFS-MSTD is superior to the baseline models TGMC-S and MTNTF, and it can well fit the trend of traffic volume changes in unobserved road segments.
交通数据修复多源数据共享因子矩阵CP分解
traffic data imputationmulti-source traffic datasharing factor matrixCP decomposition
柏跃龙,彭理群,祁钰茜,等,2019.检测数据缺失条件下的交通流估计方法研究[J].交通信息与安全,37(2):99-106.
高远, 2022. 基于稀疏数据的区域路网交通状态识别与预测技术研究[D].北京:北京交通大学.
ACAR E, DUNLAVY D M, KOLDA T G, et al, 2011. Scalable tensor factorizations for incomplete data[J]. Chemomet Intell Lab Syst, 106(1): 41-56.
CHEN X Y, HE Z C, WANG J W, 2018. Spatial-temporal traffic speed patterns discovery and incomplete data recovery via SVD-combined tensor decomposition[J]. Transp Res Part C Emerg Technol, 86: 59-77.
CHEN Y P, ZHANG Y Q, XIA H, et al, 2022. A hybrid tensor factorization approach for QoS prediction in time-aware mobile edge computing[J]. Appl Intell, 7(52): 8056-8072.
LINT J W C, HOOGENDOORN S P, ZUYLEN H J, 2005. Accurate freeway travel time prediction with state-space neural networks under missing data[J]. Transport Res C-Emer,13(5/6):347-369.
LU C B, MEI Y, 2018. An imputation method for missing data based on an extreme learning machine auto-encoder[J]. IEEE Access, 6: 52930-52935.
MA Z F, TIAN H P, LIU Z C, et al, 2020. A new incomplete pattern belief classification method with multiple estimations based on KNN[J]. Appl Soft Comput, 90: 106175.
MEIER J, WEHLAN H, 2001. Section-wise modeling of traffic flow and its application in traffic state estimation[C]// IEEE Trans Intell Transp Syst. IEEE: 440-445.
NI D H, LEONARD J D, GUIN A, et al, 2005. Multiple imputation scheme for overcoming the missing values and variability issues in ITS data[J]. J Transp Eng,12(131): 931-938.
SAID A B, ERRADI A, 2022. Spatiotemporal tensor completion for improved urban traffic imputation[J]. IEEE Trans Intell Transp Syst, 7(23): 6836-6849.
WU Y, TAN H, LI Y, et al, 2019. A fused CP factorization method for incomplete tensors[J]. IEEE Trans Neural Netw Learn Syst, 30(3): 751-764.
XU X Q, LIN M W, LUO X, et al, 2023. HRST-LR: A hessian regularization spatio-temporal low rank algorithm for traffic data imputation[J]. IEEE Trans Intell Transp Syst, 10(24): 11001-11017.
YU G H, WANG L Q, WAN S C, et al, 2021. Tensor factorization with total variation for internet traffic data imputation[J]. Pacific J Optim, 17(3): 486-505.
ZHANG H, CHEN P, ZHENG J F, et al, 2019. Missing data detection and imputation for urban ANPR system using an iterative tensor decomposition approach[J]. Transp Res Part C Emerg Technol, 107: 337-355.
ZHANG Z C, LI M, LIN X, et al, 2020. Network-wide traffic flow estimation with insufficient volume detection and crowdsourcing data[J]. Transp Res Part C Emerg Technol, 121: 102870.
ZHU Y T, WANG J Y, WANG J B, et al, 2022. Multitask neural tensor factorization for road traffic speed-volume correlation pattern learning and joint imputation[J]. IEEE Trans Intell Transp Syst, 23(12): 24550-24560.
0
浏览量
48
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构