Research On Integrated Control Method Of Highway Bottleneck Area In Cooperative Vehicle Infrastructure System

Due to the factors such as interweaving of vehicles,reduction in number of lanes,or occurrence of traffic accidents,some sections of highway are prone to form bottleneck areas,resulting in a decrease of driving speed and road capacity.In order to improve the traffic efficiency of road network,the integrated control method based on ramp metering and variable speed limit of main line can effectively alleviate traffic congestion in bottleneck area by optimizing the ramp flow and the main line speed distribution.In existing researches,traditional integrated control method is constrained by the mathematical model and has certain limitations.With the application of artificial intelligence in transportation,data-based methods have received more attention.At the same time,Cooperative Vehicle Infrastructure System,as an integral part of Intelligent Transportation System,plays an important role in improving traffic safety and operating efficiency.Based on the above,under the cooperative vehicle infrastructure environment,Multi-Agent Deep Deterministic Policy Gradient algorithm is used to combine the on-ramp metering and variable speed limit control methods.Adopting the actor-critic framework and following the principle of "decentralized execution and centralized training",this integrated control method can avoid the instability of environment caused by the independent training of single agent and enable these two methods to cooperate with each other and have certain self-learning abilities.With the goal of minimizing the total travel time of vehicels,it also can reasonably formulate onramp metering and variable speed limit control schemes.Aiming at the differences between cooperative vehicle infrastructure environment and traditional traffic environment and the changes in driver behavior,a demand analysis of highway-integrated control system under cooperative vehicle infrastructure environment is carried out from three aspects: information perception,information interaction and data processing.The on-ramp metering subsystem,variable speed limit control subsystem and integrated control subsystem based on cooperative vehicle infrastructure environment are designed.Taking part of Nanjing city highway as an experimental scenario and using Py Charm platform for the secondary development of SUMO simulation software,real-time data acquisition module,control strategy output module,vehicle-road interaction module and evaluation index output module are designed to build the cooperative vehicle infrastructure simulation environment of highway.An integrated control method based on Multi-Agent Deep Deterministic Policy Gradient algorithm is proposed,which transforms the integrated control problem of highway bottleneck area into a Markov decision process.Using cooperative vehicle infrastructure simulation environment,ramp metering agent and variable speed limit agent are built.Giving the status values of external environment,these two agents can output a reasonable ramp signal phase and main line speed limit value of different lanes to receive the difference between the corresponding outflow and inflow as reward value.Through continuous training,this method can make two agents work collaboratively to minimize the total travel time of vehicles.Under the same road network conditions,five different control strategies are designed,namely no control method,ALINEA ramp metering method,feedback variable speed limit control method,feedback integrated control method,and the integrated control method based on deep reinforcement learning under cooperative vehicle infrastructure environment.The experimental results show that,compared with traditional integrated control method,the cooperation between ramp metering and variable speed limit control based on deep reinforcement learning under cooperative vehicle infrastructure environment increases the average speed of bottleneck area by 28%,reduces the average occupancy rate by 18%,increases the evacuation flow by 12% and reduces the total travel time of vehicles by 21%.