It is critical to detect and recognize non-recurrent traffic congestion (NRC), which brings unexpected delays to commuters, companies and traffic operators. In this paper, we propose a tensor recovery based non-recurrent traffic congestion recognition (TR-NRC) model to detect and recognize non-recurrent traffic congestion by decomposing the observed travel time tensor into a low-rank tensor and a sparse tensor. A tensor model can fully utilize the intrinsic multiple correlations of travel time data. The sparse tensor represents unexpected congestion. Values of sparse tensors reveal the distribution of unexpected delays compared to expected travel time. The recovered low-rank tensor structure expresses the distribution of general expected travel time as an auxiliary product, which was unattainable in the traditional detection methods. Experimental results show that compared to previous matrix recovery based methods, our proposed method can not only detect unexpected congestion, but can also recognize the congestion patterns more effectively.
Tensor Recovery Based Non-Recurrent Traffic Congestion Recognition
15th COTA International Conference of Transportation Professionals ; 2015 ; Beijing, China
CICTP 2015 ; 591-603
2015-07-13
Conference paper
Electronic Resource
English
Measuring Recurrent and Nonrecurrent Traffic Congestion
Online Contents | 2003
|Measuring Recurrent and Nonrecurrent Traffic Congestion
British Library Conference Proceedings | 2003
|Measuring Recurrent and Nonrecurrent Traffic Congestion
Transportation Research Record | 2003
|Methodology for Measuring Recurrent and Nonrecurrent Traffic Congestion
Transportation Research Record | 2004
|Methodology for Measuring Recurrent and Nonrecurrent Traffic Congestion
British Library Conference Proceedings | 2004
|