01627659

Component

Disseminating accurate travel times can benefit travelers and provide traffic equilibrium with less congestion. The rapid acceptance and adoption of connected vehicle technology provides unique opportunities for acquiring more accurate travel time predictions. A sustainable active traffic management approach could potentially utilize connected vehicles as processing nodes for reporting information to Traffic Management Centers. The aim of this study is twofold: (1) estimating travel times anywhere on freeway network at five-minute intervals; (2) employing an eXtreme Gradient Boosting (XGB), cutting edge machine learning algorithm, for predicting travel times at five-minute intervals over a 120-minute horizon. To achieve these goals, Basic Safety Messages (BSM), generated by the Safety Pilot Model Deployment conducted in Ann Arbor, Michigan were used. These data are available on the USDOT Research Data Exchange website. Nearly two billion messages were processed using six conventional computers, simulating the integration of six vehicles in the processing phase. About 9.6 km freeway stretch was used for evaluating the XGB together with another five algorithms. Sensitivity analyses were performed at different stages of building the model to optimize key parameters and select the best model. Experimental results showed that XGB is superior to all other algorithms, followed by the Gradient Boosting. XGB travel time predictions were accurate and consistent with variations due to accidents, with mean absolute error in prediction about 13 and 20 seconds for 5-minute and 20-minute horizons, respectively. Corresponding values of mean percentage error are 3.8% and 6%, respectively, which is considered a significant achievement.

This paper was sponsored by TRB committee ABJ70 Standing Committee on Artificial Intelligence and Advanced Computing Applications.

01618707

17-04695

English

18p

2017

Transportation Research Board 96th Annual Meeting

Digital/other

Figures; References; Tables

Algorithms; Machine learning; Mobile communication systems; Sensitivity analysis; Traffic forecasting; Travel time

Ann Arbor (Michigan)

Data and Information Technology; Highways; Operations and Traffic Management

Transportation Research Board Annual Meeting 2017 Paper #17-04695

TRIS, TRB, ATRI

Dec 8 2016 11:47AM