Stochastic Modeling And Hierarchical Optimization Of Urban Traffic Network

With the rapid development of the urbanization and motorization in China,the problem of urban traffic congestion has increasingly become severe.Establishing effective traffic control methods to improve the efficiency of road network is an urgent problem to be solved.But due to the large-scale,nonlinear,and random features of the traffic system in nature,the feasibility and effectiveness of traffic control methods are often limited by following factors: 1)Road network model: The accuracy of traffic network model and the complexity of optimization problem are two key points that need to be trade-off for real-time traffic control.Although there are many effective deterministic traffic flow models,the deviation between the traffic flow model and the actual traffic state still exists due to the nature of the traffic randomness.2)Road network scale: The traffic states of different roads in large-scale road networks are coupled with each other,leading to a large number of control variables and putting great pressure on the online computing.3)Detection information of road network: For a long time,the infrastructure based sensors are the main sources of traffic information for urban traffic control system,such as the in-pavement or video based loop detectors,which only can obtain the traffic flow information of specific positions,incapable of grasping the traffic conditions of the entire road network from a global perspective.In recent years,with the advances in wireless communication technology,the equipped vehicles can communicate with each other as well as the roadside infrastructures by V2 I or V2 V technology.Thus the previous unobtainable vehicular information such as individual vehicle's maneuver,origin/destination and trajectory can be collected and shared,providing a much more complete picture of the traffic states near an intersection.However,at the same time,the traffic control problems become more complicated and the traffic control measures more diversified.Therefore,in view of the above problems and the background of the Internet of Vehicles,our main research work in the thesis can be summarized as follows:· Propose the isolated intersection control based on the state transition probability model.Define the traffic state of a link and calculate its state transition probability matrix with the random traffic demand.Extend it to an intersection and formulate the traffic signal control problem of isolated intersections as Markov Decision Process.The sensitivitybased policy iteration algorithm is applied to dynamically find the optimal strategy of Markov Decision Process.· Propose the stochastic traffic signal control of a subregion with several coupled intersections.In order to avoid a sharp increase in state dimension,the input links of subregions are merged into several virtual links,and the links inside this subregion are viewed as an virtual node that the number of vehicles entering and leaving depend on the signal setting of boundary intersections.Based on the revised link-based state transition probability matrix,the traffic signal control of boundary intersections in a subregion is described as Markov Decision Process,where the state,action,probability and reward function are defined.The sensitivity-based policy iteration algorithm is applied to dynamically find the optimal strategy of Markov Decision Process.The algorithm has great advantages in computational efficiency and can meet real-time requirements.· Propose the coordination of traffic flows among subregions for a large-scale traffic network.Considering the stochastic modeling method based on the link state transition probability matrix leads to the dimension disaster as the traffic network size grows,the coordination of traffic flows among subregions is proposed to optimize the traffic flow in large-scale traffic network.Based on the random macroscopic fundamental diagram model of each subregion in the large-scale network,an adaptive state transition probability model of the entire traffic network is established,which can describe the impact of traffic flows exchanged between subregions on the state of the entire traffic network.The policy iterative algorithm is used to dynamically find the optimal strategy for the Markov decision process of traffic flow coordination.· Propose a two-level hierarchical control structure for large-scale traffic network signal control based on the mixed back pressure control.The upper layer is responsible for the coordination of traffic flow among subregions from a global perspective,while the lower layer optimizes the traffic signals inside each subregion independently.For the traffic signal control of a subregion with general structure,a mixed back pressure index using the vehicular route information is proposed,where the one-step mixed back pressure control is proved to inherit the property of stability of back pressure scheme and multi-step back pressure control developed to overcome the erratic switching among stages by applying model predictive control methodology.Both the one step and multi-step mixed back pressure control are equivalently formulated as mixed integer linear programming problem for computational efficiency.In the two-level hierarchical control,the exchanged traffic flows among subregions optimized in the upper level are added to the mixed back pressure control problem as constraints in the lower level.· Propose a game-based method to simultaneously optimize the signal settings and vehicle routes in Internet of Vehicles(Io V).This method is based on a two-level hierarchical structure: the upper layer is mainly responsible for planning a feasible route space for each vehicle controlled in the lower layer according to the current conditions of the road network and route evaluation index;the lower layer determines the signal setting and the vehicle route selections through the game optimization between intersection signal controller and the delayed vehicles.During the game,each delayed vehicle selects one route from their own feasible route space to reduce the intersection delay time,while the signal controller adjusts the signal setting to meet the trafc demand of the delayed vehicles as well as maximize the throughput.The game can be terminated with end conditions in nite rounds.