The 20th and 21st joint COTA International Conference of Transportation Professionals (CICTP 2020-21)

Travel time is an inevitable and significant parameter in urban transportation planning and management. One of the most import access to obtain travel time is by analyzing monitoring data, which contains abundant information of travelers. Because of the limitations of equipment layout and data missing, it is difficult to get a complete travel time information in urban road network. In this paper, we treat travel time estimation as a problem of tensor completion, and propose a collaborative block term decomposition model to complete travel time tensor using monitoring data in Ruian City. Due to the multi-dimensional nature of travel time data, we model different drivers’ travel time on different road segments in different time slots with a three dimensional tensor. Meanwhile, a historical travel time tensor is built to help to discovery underlying information and relieve the problem of data sparsity. Then, a geographical matrix, a spatial-temporal matrix and a traveler relationship matrix are extracted to capture the contextual information of travel time. Following these, the three matrixes and a historical travel time tensor are collaborated by the object function to deal with the sparse travel time tensor completion problem. Empirically, rely on the monitoring data collected from 108 road segments within 22 days, our model is applied to demonstrate that it is an effective approach for travel time estimation.

CICTP