| At present,the problem facing China’s traffic is that urban motor vehicles are increasing year by year,but the road space,especially the capacity of intersections,is limited.The layout of urban areas and the travel mode of residents make urban traffic obviously low and peak periods,which will lead to uneven traffic flow at some intersections,and the resulting traffic congestion problem has become a stumbling block to the sustainable development of major cities.Although the signal control strategy has been improved to a certain extent by methods such as intersection sensing control and adaptive control,due to the complexity of the intersection environment and the variability of traffic flow,some control strategies are limited by various conditions such as environment and equipment,and it is difficult to be widely implemented at actual intersections.In addition,the application of deep reinforcement learning in signal control has improved the perception ability of signal control systems on intersections,but previous research has more inclined to the design of algorithms and ignored the analysis and optimization of signal control parameters.In view of the above problems,in order to further explore the optimization method of intersection signal control,this paper will conduct a series of analysis of its parameters on the basis of the improved induction control strategy,and use the optimal parameter combination to optimize the DQN algorithm of deep reinforcement learning,the specific research content is as follows:Firstly,this paper summarizes and analyzes various methods in the field of signal control,and elaborates on the methods of induced signal control.On this basis,an improved induction control strategy based on the application of Radar and video all-in-one machine is proposed,which can realize the queuing monitoring of each entrance lane at the intersection and vehicle data extraction.The improved inductive control strategy can realize functions such as phase switching,and set the phase switching mechanism to adapt to different sizes of traffic flows,which lays a foundation for subsequent experimental research.Secondly,in order to screen out the optimal parameters for improving the induction control strategy,an orthogonal test process is designed.The key parameters and values of the improved induction control strategy correspond to the factors and level values of the orthogonal test,and a series of orthogonal tests are carried out at the intersection of Beichen West Road and Kehui South Road.The evaluation index of the average delay of vehicles at intersection was taken into consideration,and two methods were employed to achieve optimal analysis: range analysis and variance analysis.Subsequently,the optimal parameter combination for each traffic group was determined.Finally,comparative experiments are used to confirm the effectiveness of the optimal parameter combination.The experimental results show that the influence of each influencing parameter of induction control on the control performance under different flow conditions is different,and the orthogonal test can quickly and effectively screen out the optimal parameter combination.Finally,the optimal parameter combination obtained by the quadrature test is applied to the DQN algorithm to further optimize the signal control.Set the status,action and reward of the DQN algorithm with reference to the optimal parameter combination,so that the algorithm can better match the control strategy under different intersection traffic to obtain the optimal DQN algorithm.On this basis,the preferential experience playback mechanism is used to improve the sampling method of the algorithm and obtain the optimal PER＿DQN algorithm.The experimental results show that it is effective to improve the DQN algorithm by the optimal parameter combination,and on the basis of using the optimal parameter combination,the sampling method can be improved through priority experience playback,which can further improve the training efficiency of the algorithm and achieve the goal of reducing the average delay of intersection vehicles. |
