Abstract

In Uppaal Stratego for Intelligent Traffic Lights, Eriksen et. al. showed that there is a great optimization potential in using better control algorithms, which optimize the traffic flow based on the radar inputs for a single isolated crossing. The work presented in this paper, show the effect of deploying a similar controller in four signal-controlled intersections on the same road segment, our controller is fully distributed meaning there is no direct coordination between the intersections. Our controller is tested against the real controller deployed in the intersections, currently three of the intersections are coordinated and the last one is traffic controlled. The controllers have been tested using the micro simulation program VISSIM. The simulation shows that in comparison with the existing controller, this controller provides a reduction of between 30% and 50% in average delays, queues and number of stops. The fuel consumption and total travel time of cars in the coordinated section are reduced by about 20% in the simulation study. All these reductions are achieved without making the situation worse for the side roads.
Original languageEnglish
JournalTransportation Research Procedia
ISSN2352-1465
Publication statusPublished - 2019
EventWorld Conference on Transport Research - WCTR 2019 - Mumbai, India
Duration: 26 May 201931 May 2019

Conference

ConferenceWorld Conference on Transport Research - WCTR 2019
CountryIndia
CityMumbai
Period26/05/201931/05/2019

Fingerprint

Learning systems
traffic
Controllers
simulation
road
learning
Telecommunication traffic
travel
Travel time
Fuel consumption
Radar
Railroad cars
time

Keywords

  • Signalized intersections
  • Optimization
  • Machine learning
  • Reinforcement learning
  • Model checking
  • Uppaal Stratego

Cite this

@inproceedings{cbbf392283f248968c57b60e4b2f526b,
title = "Controlling Signalized Intersections using Machine Learning",
abstract = "In Uppaal Stratego for Intelligent Traffic Lights, Eriksen et. al. showed that there is a great optimization potential in using better control algorithms, which optimize the traffic flow based on the radar inputs for a single isolated crossing. The work presented in this paper, show the effect of deploying a similar controller in four signal-controlled intersections on the same road segment, our controller is fully distributed meaning there is no direct coordination between the intersections. Our controller is tested against the real controller deployed in the intersections, currently three of the intersections are coordinated and the last one is traffic controlled. The controllers have been tested using the micro simulation program VISSIM. The simulation shows that in comparison with the existing controller, this controller provides a reduction of between 30{\%} and 50{\%} in average delays, queues and number of stops. The fuel consumption and total travel time of cars in the coordinated section are reduced by about 20{\%} in the simulation study. All these reductions are achieved without making the situation worse for the side roads.",
keywords = "Signalized intersections, Optimization, Machine learning, Reinforcement learning, Model checking, Uppaal Stratego, Signalized intersections, Optimization, Machine learning, Reinforcement learning, Model checking, Uppaal Stratego",
author = "Eriksen, {Andreas Berre} and Harry Lahrmann and Larsen, {Kim Guldstrand} and Taankvist, {Jakob Haahr}",
year = "2019",
language = "English",
journal = "Transportation Research Procedia",
issn = "2352-1465",
publisher = "Elsevier",

}

Controlling Signalized Intersections using Machine Learning. / Eriksen, Andreas Berre; Lahrmann, Harry; Larsen, Kim Guldstrand; Taankvist, Jakob Haahr.

In: Transportation Research Procedia, 2019.

Research output: Contribution to journalConference article in JournalResearchpeer-review

TY - GEN

T1 - Controlling Signalized Intersections using Machine Learning

AU - Eriksen, Andreas Berre

AU - Lahrmann, Harry

AU - Larsen, Kim Guldstrand

AU - Taankvist, Jakob Haahr

PY - 2019

Y1 - 2019

N2 - In Uppaal Stratego for Intelligent Traffic Lights, Eriksen et. al. showed that there is a great optimization potential in using better control algorithms, which optimize the traffic flow based on the radar inputs for a single isolated crossing. The work presented in this paper, show the effect of deploying a similar controller in four signal-controlled intersections on the same road segment, our controller is fully distributed meaning there is no direct coordination between the intersections. Our controller is tested against the real controller deployed in the intersections, currently three of the intersections are coordinated and the last one is traffic controlled. The controllers have been tested using the micro simulation program VISSIM. The simulation shows that in comparison with the existing controller, this controller provides a reduction of between 30% and 50% in average delays, queues and number of stops. The fuel consumption and total travel time of cars in the coordinated section are reduced by about 20% in the simulation study. All these reductions are achieved without making the situation worse for the side roads.

AB - In Uppaal Stratego for Intelligent Traffic Lights, Eriksen et. al. showed that there is a great optimization potential in using better control algorithms, which optimize the traffic flow based on the radar inputs for a single isolated crossing. The work presented in this paper, show the effect of deploying a similar controller in four signal-controlled intersections on the same road segment, our controller is fully distributed meaning there is no direct coordination between the intersections. Our controller is tested against the real controller deployed in the intersections, currently three of the intersections are coordinated and the last one is traffic controlled. The controllers have been tested using the micro simulation program VISSIM. The simulation shows that in comparison with the existing controller, this controller provides a reduction of between 30% and 50% in average delays, queues and number of stops. The fuel consumption and total travel time of cars in the coordinated section are reduced by about 20% in the simulation study. All these reductions are achieved without making the situation worse for the side roads.

KW - Signalized intersections

KW - Optimization

KW - Machine learning

KW - Reinforcement learning

KW - Model checking

KW - Uppaal Stratego

KW - Signalized intersections

KW - Optimization

KW - Machine learning

KW - Reinforcement learning

KW - Model checking

KW - Uppaal Stratego

M3 - Conference article in Journal

JO - Transportation Research Procedia

JF - Transportation Research Procedia

SN - 2352-1465

ER -