Research On Vehicle Equity Oriented Intelligent Traffic Light Control

With the continuous progress of the urbanization and rapid development of economic construction,the number of motor vehicle owned is soaring.While facilitating people’s travel,it also causes severe traffic congestion,impeding the transformation of cities and regions to intelligent cities.At present,the traditional signal control method with fixed phase time is inefficient in general traffic,while most traffic signal control algorithms only focus on improving the overall traffic efficiency.When facing extremely unbalanced traffic density at an intersection,the traffic light control algorithms mentioned above will result in high priority of dense vehicle,thus the lowdensity sparse traffic flow must wait for intolerable long time or even cannot pass the intersection,leading to the “discrimination” of the minorities and affecting the vehicle equity severely.Based on the situation above,this paper proposes two intelligent vehicle equity-oriented traffic light control algorithms in order to adapt to the changes of dynamic traffic flow and take into account the vehicle equity and overall traffic efficiency at the same time.The main contributions of this paper are as follows:1.Based on reinforcement learning,this paper proposes two intelligent traffic light control algorithms which can react to different traffic scenarios dynamically in real time,fitting complex traffic environment perfectly.The traffic light control strategy is learned through the interaction of the agent with different traffic scenarios.The Dueling DQN structure is adopted in model structure design in order to improve the learning effect of the model,and the Double DQN method is used in training process to optimize the training results.2.Targeting at the traffic conditions with extremely unbalanced traffic density at the intersection,this paper proposed algorithms considering the vehicle equity and overall traffic efficiency at the same time.When designing the state,long time waiting vehicles are allocated with higher weight and more attention is paid to long time waiting vehicles when designing the reward function,so as to enhance the vehicle equity.3.Based on China Mobile One NET cloud platform,this paper realized a "cloud-end" collaborative intelligent traffic signal control system which has better real time ability and applicability towards real scenario.The road network environment is simulated through SUMO simulator,and the simulation data is sent to One NET cloud platform as the input of cloud model.After the model calculates the action,the cloud platform sends it to terminal equipment for the execution of phase action.In order to verify the effectiveness and availability of two algorithms proposed above,this paper utilizes the urban traffic simulator SUMO to carry out a series of simulation experiments to verify that the two algorithms can take into account vehicle equity and global traffic efficiency at the same time while being more targeted for different traffic conditions and responding at real time.