DESIGN OF DRL TIMING SIMULATION SYSTEM FOR SIGNAL INTERSECTION INTEGRATING VISSIM-PYTHON AND QT

In view of the problem that the algorithm portability of vissim-Python co-simulation system is not high in the research of DRL timing of signal-control intersection and it is difficult to be applied in practice, this paper designs an intersection timing simulation system. In this system,in order to meet the needs of practical application, a DRL timing model considering both detector data and countdown was proposed. According to the requirements of this model, the necessary Vissim interface was designed by Python and encapsulated as a simulation environment for Gym reinforcement learning. In order to solve the problem of low portability of the algorithm, the deep learning framework PyTorch was used to standardize the interface of the timing algorithm, and PyQt5 was used to design a visual operation interface to flexibly adjust the algorithm parameters. In order to improve the efficiency of the system, four techniques were used to accelerate the simulation process. Taking the intersection of Linxi Road and Xingye Road in Zhuhai city as an example, the simulation test results show that the system works well, which can not only evaluate the actual intersection timing effect, but also serve as a test platform for related algorithm researchers.