Research And Implementation Of Regional Traffic Signal Coordinated Control Strategy Based On Multiagent Cooperation

With the sharp increase of urban vehicles and the increasing complexity of traffic networks,the problem of traffic congestion in urban areas has become a tricky battle in the field of intelligent transportation,which needs to be settled immediately.On the premise of not changing the road structure,effectively dividing traffic sub-regions and cooperatively controlling traffic signals at intersections in the specific areas contribute to the improvement of the traffic condition of the whole city.Thus,based on multi-agent technology,this paper tries to probe into the method of controlling traffic signals in urban areas,which is conducive to address similar situations in the future.The main contents are enumerated as follows.To begin with,the community detecting algorithm is raised by this paper,focusing on dividing the traffic network into sub-regions to cope with the vexing problem that the increasingly complex traffic network cannot be controlled thoroughly.To do this,first analyze the influencing factors of the correlation between adjacent intersections to establish the model of total correlation,transforming the traffic network into a community network.And then,heightening the importance of the nodes by using the correlation and improving the representations of the node weight and link weight in the community detecting algorithm.At last,the sub-region division model is set up to accomplish the dynamic division of the sub-regions in the respect of traffic control.Fortunately,the proposed algorithm is more efficient and feasible in road network division through the verification of the simulation experiment of the actual road network.What’s more,based on the single-agent reinforcement learning,an optimal algorithm is put by this study for controlling the signal in isolated intersections to tackle the dilemma of the signal control single intersection in the complicated changing traffic environment,which is the traditional control methods fail to resolve.Here is how it works:firstly,the lanes are built as cellular models on the basis of an intelligent agent architecture which is designed in the intersections,moreover,the high-dimensional real-time traffic information is discretely coded and the actions and the award-penalty function of the agent are established.Secondly,an optimal algorithm related to traffic signals is also devised by this research according to the deep reinforcement learning algorithm,and the traffic simulation platform with deep reinforcement learning is constructed by utilizing SUMO(Simulation of Urban Mobility)for algorithm training and traffic evaluation.Thirdly,it is verified that the algorithm performs better in terms of the average cumulative vehicle delay and queuing length.Last but not least,a regional traffic signal coordination algorithm is conceived based on multi-agent reinforcement learning,which is intended to find a way out of the difficulty of controlling intersections with various traffic stream characteristics in the same region.The concrete steps were taken:first of all,the single-agent network is modified by using LSTM(Long Short-Term Memory)and the analysis of the features of the regional traffic stream.Simultaneously,the model of coordinating control of traffic signals is created by applying the distributed structure.Next,the award-penalty function is ameliorated with the foundation of vehicle delay,latency time,and queuing length.Also,a more comprehensive state input to present dynamic information of traffic flows is laid out in order to accelerate the pace of bringing forward the regional traffic signal coordination algorithm.Finally,the simulation of the actual traffic network proves that the proposed algorithm is expert at the average stop frequency,latency time,and so forth.What matters most is that it can facilitate the vehicular traffic rate effectively and efficiently in specific regions.