Research On Variable Speed Limit Control System Of Main Line Of Freeway Based On Deep Reinforcement Learning

Based on the previous research,this paper proposes a variable speed limit control strategy based on Deep Q-Learning Network(DQN),and develops a simulation software based on cellular automata to verify strategy's effectiveness in improving traffic efficiency.Based on the proposed strategy and simulation environment,a variable speed limit control system with front and rear ends separated is developed.Since the existing microscopic traffic simulation software cannot support the interface provides the image input of the real-time traffic state for the DQN network required by the research,this paper firstly develop the cellular automata simulation software which integrates NaSch model,VE model,discretionary lane changing model and cellular automaton model for bottleneck flow(CABF)using Python language.Then,this paper adjusts the model and calibrates parameters according to real conditions and data of the Beijing-Kunming freeway.The validity of the model is verified by the analysis of the traffic simulation data and the driving track.In addition,the simulation data also effectively simulates the sudden decline of the road capacity in the work zone,providing a simulation basis for the research of variable speed limit control at freeway bottlenecks.Based on the simulation software,this paper attempts to apply the DQN algorithm to the variable speed limit control for the first time,and implements the algorithm based on the TensorFlow framework.The main advantage of the algorithm is to input the position information of the vehicles and bottleneck on the road directly into the convolutional neural network as an image.Compared with other variable speed limit control strategies based on reinforcement learning or deep reinforcement learning algorithms,this approach avoids the manual design of complex traffic states.Therefore,this strategy can better solve congestion problems of temporary bottlenecks caused by work zone or traffic accident than previous algorithms.In order to evaluate the effectiveness of the proposed strategy,the simulation experiment is carried out on the work zone under the conditions of no speed limit,fixed speed limit and variable speed limit calculated by control strategy proposed in this paper.By comparing the experimental results,it is proved that the DQN-based variable speed limit control strategy proposed in this paper can effectively improve the traffic efficiency of congested sections.In order to further improve the practical significance of this paper,a variable speed limit control system software separated from the front and back ends was designed and developed using python and JavaScript language based on the backend framework flask and front-end framework vue.js.This software can manage the information of the roads and the bottlenecks in the simulation environment,and send the instruction for training the DQN network to the back end.When the training is completed,the speed limit control can be performed on the road segment by using the dynamic speed limit strategy or setting the speed limit manually.And the real-time traffic flow data and the track data of vehicles generated by the simulation can be stored in the database.