Research On Coordinated Control Of Intersection Signals In Autonomous Transportation Environment

Urban traffic congestion is becoming increasingly severe,and traditional transportation systems are no longer able to meet travel needs.In order to alleviate traffic congestion and ensure travel safety,Autonomous Transportation System(ATS)with self-organizing operation and autonomous service capabilities has emerged.As important components of ATS,connected autonomous driving vehicles and traffic signal control play a crucial role in its development.Therefore,this thesis takes the intelligent traffic signal control scenario involving connected autonomous vehicles as the research objective,and adopts deep reinforcement learning methods to optimize traffic signal control and regional coordination control,thereby improving the traffic efficiency of the road network.This thesis mainly conducts research from the following aspects:1.To achieve signal control under ATS,this thesis designs a mixed traffic signal control scenario that includes autonomous and manual driving vehicles.Based on autonomous traffic theory and traffic control theory,the construction of traffic signal control scenarios was completed on the SUMO(Simulation of Urban Mobility)traffic simulation platform.2.To solve the signal control problem in complex and ever-changing traffic environments that traditional control methods cannot handle,this thesis designs a signal control algorithm based on weighted reward reinforcement learning.Firstly,the discrete state encoding method is used to transform the intersection traffic information into a matrix composed of vehicle position information,speed information,number of left turning vehicles,and current phase,which is used as input to the neural network.Weighting three traffic efficiency indicators,namely lane queue length,vehicle waiting time,and number of brakes,as rewards for reinforcement learning,and utilizing a priority experience replay strategy to optimize signal light control.Then the edge computing technology under ATS is simulated to explore the impact of different information computing methods on future traffic capacity.Finally,the signal control performance of several control methods was compared under different penetration rates of traffic conditions,and the results showed that the selected algorithms and technologies had good performance in improving road traffic efficiency.3.To solve the problem of regional traffic signal coordination and control,this thesis designs a regional signal coordination and control algorithm based on collaborative multi-agent reinforcement learning.Firstly,by analyzing the control characteristics of regional traffic signals,the traffic signal control under single agent is extended to multi-agent by using the Double Deep Q Network(DDQN),and the regional traffic signal coordination control model is established through collaborative learning.Then,the spatial discount factor is used to improve the reward function under single point signal control,and a state input that can more comprehensively express traffic information is designed,thereby designing a signal coordination control algorithm suitable for regional traffic.Finally,the performance of different control methods was verified through actual traffic scenarios constructed by SUMO,proving that the algorithm used performs better in improving the traffic capacity of the road network.