Citation

Nuli S, Subbarao SSV. Int. J. Traffic Transp. Eng. (Belgrade) 2018; 8(2): 249-260.

Copyright

(Copyright © 2018, City Net Scientific Research Center, Faculty of Transport and Traffic Engineering, University of Belgrade)

DOI

10.7708/ijtte.2018.8(2).09

PMID

unavailable

Abstract

State-of-the-art adaptive signal control models, commonly used in developed countries, uses upstream or advance detector information for determining control plan. These models work well in lane-based traffic conditions especially when multiple junctions are involved. However, traffic in many cities across the world is heterogeneous, characterised by non-lane based movement and mixed-vehicle type, which causes inaccurate estimation of turning proportions. These difficulties can be addressed to a great extent by placing the detector at the stop-line rather than in advance or upstream location. However, there is no truly adaptive traffic control models exists for such traffic conditions and sensor location. In this study, a signal control model is proposed that is truly adaptive and uses stop-line detector information. The model aims at real-time allocation of green times through actor-critic reinforcement learning; an approach originated from the machine learning community. This approach can learn relationships between signal control actions and their effect on the traffic system while determining optimal control policy. To test the performance of the model a typical four-way four phase intersection with variable flow was simulated using a traffic micro simulator (VISSIM) and interfaced with the proposed model. The performance of the model was compared with the traditional vehicle actuated system. The results using this approach shows significant improvement over traditional control, especially for varying traffic demand.

Language: en