TFTC Webinar: Dr. Vikash Gayah -- Thursday, February 24th @ 3 PM EST
Rafegh Aghamohammadi
Dear All,
The next TFTC webinar in partnership with the University of Michigan's NGTS seminars will be held on Thursday, February 24th, at 3:00 PM EST. You can save the attached calendar file to your calendar app for a reminder. Please find the detailed information of the seminar below:
Title: Application of deep reinforcement learning for perimeter metering control
Speaker: Dr. Vikash V. Gayah, Pennsylvania State University
Time: Thursday, February 24th, 3:00-4:15 PM EST
Meeting Link: https://umich.zoom.us/j/95590276501 (Passcode: NGTS)
Abstract:
Various perimeter control strategies have been proposed for urban traffic networks that rely on the existence of well-defined relationships between network productivity and accumulation, known more commonly as network Macroscopic Fundamental Diagrams (MFD). Most existing perimeter control strategies require accurate modeling of traffic dynamics with full knowledge of the network’s MFD. However, such information is generally difficult to obtain and subject to error. This talk describes recent efforts to alleviate this using deep reinforcement learning for networks made up of two unique regions. The proposed methods are completely model free in that they do not require knowledge of the network’s MFD. The algorithm learns the consequences of different control actions over time and uses this information to obtain optimal control policies under different situations. Results from numerical experiments show that the proposed method: (a) can stably learn perimeter control strategies under various types of environment configurations; (b) can consistently outperform the state-of-the-art, model predictive control (MPC); (c) demonstrates sufficient transferability to a wide range of traffic conditions and dynamics in the environment; and, (d) exhibits great potential for practical implementation. Furthermore, integration of limited knowledge of congestion dynamics can improve the algorithm considerably, allowing it to be applied on larger, more complicated network structures.
Traffic flow webinars
Meeting Link: https://umich.zoom.us/j/95590276501 (Passcode: NGTS)