Data-driven Hierarchical Optimization And Control For Urban Traffic Network System

The urban traffic network system is huge in scale,characterized by complex interconnection and strong nonlinearity,with problems such as difficulty in establishing accurate road network models,determining control objectives from macroscopic level to microscopic level,and unbalanced queue length at intersections.In this thesis,a series of data-driven hierarchical optimization and control methods for urban traffic network systems are proposed to address the above problems for improving the traffic operation efficiency of urban traffic networks.The main research contents and contributions of this thesis are summarized as follows:1.Aiming at the problem of difficulty of modeling and strong nonlinearity of large-scale macroscopic traffic network systems,this thesis proposes a two-level model-free adaptive perimeter control method.First,a macroscopic traffic augmented system is constructed,with the set points of the total number of vehicles in the road network as the input variable,the performance evaluation index of the overall road network as the output variable.The augmented system is transformed into an equivalent data model by the compact-form dynamic linearization technique,and an upper-level set point optimization method based on model free adaptive control(MFAC)is designed.Then,the set points calculated at the upper level are transferred to the lower level and used as the control objective of the macroscopic traffic region.Considering the actual traffic constraints on the perimeter control ratio and the total number of vehicles,the lower-level model free adaptive perimeter control strategy with input/output constraints is designed.Finally,some simulations are made and some typical perimeter control algorithms are compared to verify the effectiveness of the proposed method.2.At the level of mesoscopic urban traffic network systems,aiming at the problem of difficulty of determining the desired number of vehicles within each link and local congestion in the urban traffic network,a two-level hierarchical optimization control method is proposed.First,the mesoscopic traffic augmented system is constructed and transformed into the equivalent data model by the dynamic linearization technique,the set point of the number of vehicles within each link is calculated at intervals based on the MFAC method.Furthermore,for preventing local congestion,the upper-level set point optimization problem with capacity constraints of downstream links is studied.Then,according to the optimized set point of number of vehicles of each link from the upper level,the model predictive control method and MFAC method are used to design the objective function of the lower-level controller and calculate the signal timing of intersections within the road network for different scenarios,respectively.Finally,the proposed method is compared and validated by using the microscopic traffic simulation software VISSIM.3.Aiming at the characteristics of urban traffic network systems with repetitive operation,a two-level hierarchical iterative optimization and decentralized control architecture is constructed,and a data-driven iterative learning optimization and decentralized estimation decentralized MFAC method is proposed.First,a repetitive traffic augmented system is constructed and transformed into an equivalent data model by using the compact-form dynamic linearization technique along the iterative direction,based on which the objective function is designed,and a datadriven iterative learning set-point optimization method is proposed.Then,the whole traffic road network is decomposed into several interconnected subsystems,where the driving vehicles among the subsystems are considered as the interconnection influence among the subsystems.The equivalent data models of subsystems with interconnected influence are developed,the pseudo Jacobian matrix of the data model of each subsystem is decentrally estimated,and the decentralized traffic signal control method is designed.Finally,the effectiveness of the proposed method is validated by numerical simulations.4.A controller-dynamic-linearization-based model free adaptive predictive queue length balance control for the single intersection traffic system is proposed.First,the corresponding control law is given for a class of unknown nonlinear singleinput single-output ideal predictive controllers by using the compact-form dynamic linearization technique.Then,with the help of the equivalent predictive data model for a single intersection system and the input and output data,the automatic tuning of the pseudo Jacobian matrix for the control law is designed by using the steepest descent method.The convergence of the proposed method is proved in the sense of a generic norm.Finally,the effectiveness of the proposed method is verified by numerical simulations.