Citation

Chen K, Wu G, Peng S, Zeng X, Ju L. Int. J. Veh. Perform. 2023; 9(3): 250-271.

Copyright

(Copyright © 2023, Inderscience Publishers)

DOI

10.1504/IJVP.2023.131974

PMID

unavailable

Abstract

This paper proposes a speed strategy based on reinforcement learning on the basis of double traffic lights. This strategy can ensure that vehicles can pass traffic lights without stopping or with little stopping. First of all, Prescan software is used to build traffic lights, roads, and vehicles and other scenario models. Simulink software is used for vehicles, traffic lights control, and other models. Secondly, the double traffic lights scenario has analysed in detail. And then, the improved Q-learning algorithm is used to build the vehicle speed decision model and train the Q table. Q table is used for subsequent real vehicle tests and simulation verification. Finally, the feasibility of the strategy is verified in a variety of conditions, and the results show that the strategy can guarantee fuel economy and get through the double traffic lights as smoothly as possible.

Keywords: Q-learning; double traffic lights; vehicle speed decision; reinforcement learning; Markov model; fuel consumption model.

Language: en