Abstract
Background:
As wireless communication technologies evolve, probe-based travel-time collection systems are becoming popular around the globe. However, two problems generally arise in probe-based systems: one is the outlier and the other is time lag. To resolve the problems, methods for outlier removal and travel-time prediction need to be applied.
Methods:
In this study, data processing methods for addressing the two issues are proposed. After investigating the characteristic of the travel times on the test section, the modified z-score was suggested for censoring outliers contained in probe travel times. To mitigate the time-lag phenomenon, a recurrent neural network, a class of deep learning where temporal sequence data are normally treated, was applied to predict travel times.
Results:
As a result of evaluation with ground-truth data obtained through test-car runs, the proposed methods showed enhanced performances with prediction errors lower than 13% on average compared to current practices.
Conclusion:
The suggested methods can make drivers to better arrange their trip schedules with real-time travel-time information with improved accuracy.
Publisher
Bentham Science Publishers Ltd.
Subject
Transportation,Modelling and Simulation
Reference36 articles.
1. Southwest Research Institute.
Automatic Vehicle Identification Model Deployment Initiative-System Design Document
1998.
2. Clark S D, Grant-Muller S, Chen H.
Cleaning of Matched License Plate Data.
Transportation Res Rec
2002;
(1804):
3. Dion F, Rakha H.
Estimating Dynamic Roadway Travel Times Using Automatic Vehicle Identification Data Trans Res Part B
2006.
4. Ma X, Koutsopoulos H.
Estimation of the automatic vehicle identification based spatial travel time information collected in stockholm.
IET Intell Transp Syst
2010;
4
(4)
5. Boxel DV, Schneider WH IV, Bakula C.
An innovative real-time methodology for detecting travel time outliers on interstate highways and urban arterials.
TRB 2011 Annual Meeting CD-ROM
Washington D.C.
2011.