Abstract
The prediction of travel time is challenging given the sparseness of real-time traffic data and the uncertainty of travel, because it is influenced by multiple factors on the congested urban road networks. In our paper, we propose a three-layer neural network from big probe vehicles data incorporating multi-factors to estimate travel time. The procedure includes the following three steps. First, we aggregate data according to the travel time of a single taxi traveling a target link on working days as traffic flows display similar traffic patterns over a weekly cycle. We then extract feature relationships between target and adjacent links at 30 min interval. About 224,830,178 records are extracted from probe vehicles. Second, we design a three-layer artificial neural network model. The number of neurons in input layer is eight, and the number of neurons in output layer is one. Finally, the trained neural network model is used for link travel time prediction. Different factors are included to examine their influence on the link travel time. Our model is verified using historical data from probe vehicles collected from May to July 2014 in Wuhan, China. The results show that we could obtain the link travel time prediction results using the designed artificial neural network model and detect the influence of different factors on link travel time.
Highlights
The prediction of travel times is challenging because of the intrinsic uncertainty of travel on congested urban road networks as well as the uncertain influence of factors such as rainfall when probe vehicles travel on road networks
Traffic flowthe isthis influenced by extract many factors such as traffic from adjacent areas, expectation, standard deviation of speed, link degree, link length, and weather are closely related and meteorological information to investigate their influence on link travel time prediction
The value of Mean Absolute Percentage Error (MAPE) becomes slightly bigger with an increase in the amount of data from 1 to 5% of the total data
Summary
The prediction of travel times is challenging because of the intrinsic uncertainty of travel on congested urban road networks as well as the uncertain influence of factors such as rainfall when probe vehicles travel on road networks. Zheng et al [7] proposed a three-layer neural network model to estimate complete link travel time for individual probe vehicles traversing a link. This model was discussed and compared with an analytical estimation model developed by Hellinga et al [8]. Jenelius et al [9] presented a statistical model for urban road network travel time estimation using vehicle trajectories obtained from low frequency GPS probes as observation data. In view of the data sparseness, we put forward a three-layer neural network model based on feature relationship between target link and adjacent link to estimate link travel time. Experimental results show that the proposed neural network model can predict link travel time using the relationship between a target and adjacent links. A discussion of the results and some conclusions are outlined at the end
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
