Research On Hybrid Optimization Algorithm Of Trunk Traffic Based On Deep Reinforcement Learning

In recent years,with the acceleration of urbanization,the traffic flow on urban main roads surges and congestion occurs from time to time,which brings inconvenience to travel and seriously affects the quality of work and life.Aiming at the problem of traffic congestion on main roads,traditional traffic control strategies such as timing control strategies cannot adapt to the rapid changes of traffic flow and have poor flexibility.The researchers tried to upgrade the system intelligently by applying deep reinforcement learning technology to the control of traffic lights on the main road.The algorithm based on reinforcement learning greatly improves the delay optimization performance of the model,but increases the computational cost and the difficulty of model convergence.The research on control algorithm of main road lacks modeling analysis in complex lane scenarios.To solve the above problems,this paper combines traditional algorithms with deep reinforcement learning,and designs a hybrid algorithm for trunk traffic optimization based on deep reinforcement learning.The main research contents of this paper are as follows:Aiming at the problems of low flexibility of traditional traffic control algorithms and high complexity of reinforcement learning algorithm,a hybrid driving control algorithm of trunk line based on green wave and reinforcement learning was proposed.Firstly,for the green wave problem of the main road,the mixed integer linear programming was established,and the constraint conditions were established with the goal of consistent and maximum bandwidth of the green wave band.The period and phase difference of the main road were solved by the Gurobi optimizer,and the basic control factors of the coordination of the main road were determined.Secondly,the road congestion state is taken as input,and the green signal ratio is adjusted by using deep reinforcement learning.In order to realize multi-agent cooperation,a new reward function is introduced for deep reinforcement learning.The SUMO simulation platform was built according to the experimental scenarios,and simulation tests were carried out on the algorithm under different traffic flow scenarios.The experiments proved that the algorithm has good control performance on the main road.In view of the problem that the vehicle density increases and the vehicle running state is complicated in the complex multi-lane scenario,the vehicles in the turning lane in the branch direction are queued up and congested to the main direction,thus affecting the traffic of the main road,the main road is further divided in detail to distinguish the left,straight and right turn lanes,and the queuing characteristics of vehicles are considered.Firstly,the arrival rules of vehicles are analyzed,and a constraint model is established by classifying the delays caused by queuing under different traffic flows.Furthermore,the existing control strategies are expanded to combine the constraint model with reinforcement learning.Finally,the performance of the main road traffic control algorithm is compared under different traffic flows.The simulation results show that the proposed algorithm has a good performance in reducing the average waiting delay and queue length of vehicles.