Freeway Short-Term Travel Time Prediction Based on Dynamic Tensor Completion

Abstract

Short-term travel time prediction is one of the key technologies of intelligent transportation systems. Reliable systems that are able to provide accurate travel time information are needed for advanced traffic management systems and advanced traveler information systems. Various methods have been proposed and developed to predict travel time. However, travel time prediction is difficult because of its complex multimodal properties in time and space. Making full use of spatial– temporal information to predict travel time accurately is still a problem. To deal with this shortcoming, a method based on dynamic tensor completion is proposed to predict travel time; this method can make full use of the spatial–temporal correlations of travel time by constructing the travel time data into dynamic four-way tensor streams, and real-time prediction through the dynamic tensor completion model can be realized. Experiments with real traffic speed data collected by 40 detectors on I-405 were used to verify the performance of the proposed approach. For evaluation, two strategies of tensor completion were tested on travel time derived from the I-405 freeway speed data. The experiment results showed that dynamic tensor completion outperformed offline tensor completion and two other benchmarks.