Tensor Robust Principal Component Analysis with Continuum Modeling of Traffic Flow: Application to Abnormal Traffic Pattern Extraction in Large Transportation Networks

摘要

The study addresses the needs of detection and description of abnormal traffic patterns in large transportation networks formed due to the presence of unexpected disruptions, such as natural or manmade disasters. In order to take into account complex spatiotemporal structure of traffic dynamics and preserve multi-mode correlations, tensor-based traffic data representation is put forward. Further, with the reasonable assumptions on normal or expected traffic dynamics to exhibit similar periodic structure, the problem of abnormal or unexpected traffic patterns detection is treated as a low-rank modeling problem. More precisely, tensor robust principal component analysis is applied for the purpose of discovering distinctive normal and abnormal traffic patterns. For the validation purposes, continuum modeling approach is employed to emulate traffic dynamics, with consideration of the effect of aforementioned disruptions. The results suggested the applicability of proposed approach in order to extract abnormal traffic patterns in large transportation networks.