Key Technology Study Of Expressway Ramp Area Control With Vehicle And Infrastructure Cooperation

With rapid development of social economy and increasing traffic demand,a series of problems have emerged,such as traffic congestion,traffic safety and environmental pollution.Numerous researches have indicated that the freeway ramp area is the bottleneck of the freeway corridor.If not managed effectively,the traffic congestion and accidents are very likely to happen,and thus worsen the traffic conditions and reduce the benefit of the entire freeway.At present,researchers often use the variable speed limit control and ramp metering control methods to optimize the freeway traffic,because they are the most effective approaches to relieve the freeway bottleneck congestion.Hitherto,prevailing variable speed limit and ramp metering control algorithms are model predictive control and feedback control algorithms.Model predictive control algorithms can predict traffic variation and prevent traffic congestion before it happens,but it requires accurate traffic model.Meanwhile,feedback control algorithms are model free,but their hysteretic nature responding to the traffic perturbation is inferior to the proactive approaches.Reinforcement learning has become another hotspot in the traffic control field.Many researches show that by combining the traffic control methods with the reinforcement learning,there can be significantly improvement of control effects.On the other hand,the vehicle to infrastructure technology is attracting more and more attention.It can retrieve traffic data precisely in real time and ensure vehicles to obey the control instructions.Most existing researches on vehicle to infrastructure traffic control are from the microscopic perspective(i.e.single vehicle or small vehicle platoon),studies from the macroscopic level are rare.Therefore,this dissertation intends to integrate the vehicle to infrastructure technology with variable speed limit and ramp metering control,to improve freeway traffic efficiency and safety.In addition,this paper proposed a multiagent reinforcement learning method to control the ramp area traffic.This dissertation focusses on the road level control.A traffic control system consists of the vehicle to infrastructure technology and the reinforcement learning control methods is designed.The control system collects traffic information and estimates the road traffic state in real time with the vehicle to infrastructure technology.Then,it uses variable speed limit and ramp metering control to optimize the onramp area traffic with reinforcement learning methods.To be more specific,the following aspects are studied:First,this study designed the framework of the traffic control system.The system is a closed loop system.It consists of roadside control system and vehicle system.The vehicle system refers to the vehicles with on-board unit,which can send real time vehicle position and velocity to the traffic control system,and execute control instructions from the control system.The control system aggregates the received vehicle data,generates the traffic state of the control area and calculates control instructions.In addition,the system is a distributed control system,which means the traffic control units work synchronously according to their knowledge in a cooperative manner.Hence,the control complexity is significantly simplified,and the system is competent for real time control.In addition,traffic control units can deploy flexibly on the freeway according to different traffic conditions.Second,the study developed the entire control system on an open source microscopic traffic simulation software called MOTUS.The software includes the driver behavior module,the network module(based on the Google KML file),the traffic state extracting module,the traffic control module,the animation model and the result analysis module.The software can simulate the vehicle to infrastructure environment with traffic control methods precisely and evaluates the control effects.In addition,the results can import to MATLAB for further analysis.Third,the study proposed a distributed reinforcement learning control method to solve the regional traffic control problem.Currently,studies on multiagent cooperative traffic control are insufficient,hence it is hard to achieve good control effect on largescale freeway network.The distributed reinforcement algorithm can optimize the mainstream traffic.On the aspect of onramp and mainstream cooperative control,the study first use the ramp metering to adjust onramp traffic,then use the variable speed limit control to help the ramp metering to achieve a more optimized result.Furthermore,this paper proposed a traffic safety objective for the reinforcement learning,based on the concept of “Time to collision”.With this objective,the control agent can considerably stabilize the mainstream traffic.Finally,this dissertation studied the control schemes of the variable speed limit,ramp metering,and the combination of both methods via simulation.To measure the control effects quantitatively,the total travel time of the freeway corridor,the speed difference between adjacent segments and the vehicle stop time is used.The simulation results shown that the control system has a remarkable improvement in both traffic mobility and safety.Besides,the system can stabilize the traffic flow of the entire ramp area and have better resists towards traffic accident risks.