Balancing computational requirements with performance in model predictive traffic control

Simon Stebbins, Mark Hickman, Jiwon Kim, Hai L. Vu

Research output: Chapter in Book/Report/Conference proceedingConference PaperOther


In some form or another, model predictive traffic control has been proposed for some decades. Because it predicts future traffic conditions, rather than reacting to past conditions in hindsight, it has advantages over adaptive traffic control techniques. However, it has not been widely adopted for a couple of reasons. Firstly, there are doubts about its suitability when fine-grained traffic data is unavailable. This objection can be addressed with the promise of vehicle-to-infrastructure (V2I) communication. Secondly, previous implementations have found that the control algorithm’s computational complexity is exponential or worse. This paper addresses this objection by introducing the A* algorithm to significantly decrease computation time. Other methods for reducing computation time include setting a suitable prediction horizon, clustering vehicles into platoons and implementing an incremental version of the control algorithm. With these methods combined, it is shown through simulation, that a real-time model predictive control algorithm is practical because computation time is manageable without having an adverse effect on total delay.

Original languageEnglish
Title of host publicationAustralasian Transport Research Forum 2017 Proceedings
PublisherAustralasian Transport Research Forum
Number of pages17
Publication statusPublished - 2018

Cite this