Traffic signal control is one of the most important factors in improving the efficiency and safety of urban transportation, and various solutions have been proposed to enhance it. This research presents a method based on reinforcement learning to reduce vehicle waiting time at an isolated intersection. Some of the state variables that the learning agent deals with are continuous variables, including the length and density of vehicle queues at the intersection. One solution to such a problem is to discretize the states and use the Q- learning algorithm. The drawback of this method is the loss of information and increased uncertainty in transitioning from one state to another. The solution proposed in this research to overcome this issue is the use of deep Q-learning. The method introduced in [1, 2] has been selected as the baseline for comparison. By conducting simulations and comparing the results obtained from our proposed method, it is shown that the system’s performance significantly improves in terms of reducing vehicle waiting time and computational load compared to the baseline method. The implementation of control algorithms and the generation of traffics data are performed using the SUMO traffic software.

کلمات کلیدی