Pacific Southwest Region 9 UTC

MT-18-10

Project Number

MT-18-10

Project Summary

Funding Source(s) and Amounts Provided (by each agency or organization): CT-PSR TO-001 $100,000.00

 

Start-End Dates: 02/01/2019-01/31/2020

 

Brief Description of Project: 

Project Status

In progress

Year

2019

Topic Area

Integrated Freight and Passenger Systems

P.I. Name & Address

Professor and Director of the Information Laboratory (InfoLAB), Department of Computer Science; USC Viterbi School of Engineering
University of Southern California
3737 Watts Way
Charles Lee Powell Hall (PHE) 306A
Los Angeles, CA 90089-0781
United States
shahabi@usc.edu

Funding Source(s) and Amounts Provided (by each agency or organization): CT-PSR TO-001 $100,000.00

 

Start-End Dates: 02/01/2019-01/31/2020

 

Brief Description of Project: 

The main objective of this proposal is to develop a deep-learning method and a system that can process massive amounts of 1) GPS trajectories from public transportation vehicles and 2) realworld traffic sensor datasets archived in our data warehouse to predict traffic flow to then enable the forecasting of a variety of performance metrics of public transportation systems, including travel-time reliability, on-time performance, bus bunching and travel-time estimation. Previously under a METRANS funded project, we built a system that uses real-time and historical bus GPS trajectory datasets over the past two years in the Los Angeles County to generate these performance metrics. While the system is well-received by Caltrans, our approach only considers current and historical data and does not have the capability to forecast the performance metrics, which is important for both the transportation agencies and riders. For example, a rider needs to know if a particular bus will be delayed in the next 30 minutes at a certain stop. With this proposal, we plan to study a deep learning approach for traffic flow prediction in road networks and use the predictions to forecast a variety of public transportation system performance metrics. With the accurate prediction of the traffic flow, we will be able to predict the arrival time of individual public transportation vehicles and hence forecast their performance measurements. The major challenges in predicting traffic flow (i.e., estimating future speed and volume) are due to the spatial and temporal nonlinearities when transitioning between free flow and congestion at rush hours and during non-recurring events (e.g., accidents, weather conditions, sports games). We will conduct fundamental research in developing a novel graph convolutional recurrent network to capture such nonlinearities inherent in road networks. Furthermore, to demonstrate the benefits of our research, we will develop a web application to show our traffic prediction results concerning real-time traffic conditions and to enable the access and visualization of the predicted performance metrics of public transportation vehicles. This research will exploit a real-world big traffic sensor data and California Highway Patrol (CHP) accident logs collected from Regional Integration of Intelligent Transportation Systems (RIITS) in the last six years under Archived Traffic Data Management System (ADMS) project.