| In the context of global climate change and rapid urbanization,uncertainties and unknown risks confronted by our urban systems are keep growing.Traffic system is among the critical infrastructures,which not only provides mobility for people and goods,but also act as the lifeline of emergency rescue.External disruptions,such as extreme weather and the pandemic,are resulting in high risk of cascading failure or large-scale paralysis.Therefore,the requirement of developing resilient traffic system is imperative.The thesis employs “livability,resilience and intelligence” as the overall goal.Investigations mainly concentrates on the modelling,characteristics and control of largescale urban roadway network,which deals with multi-factor coupling modeling,system respondence mechanism,as well as resilience based autonomous cooperative control.To begin with,a Bayesian Causal Network(CBN)is proposed for the difficulties in the modeling of multi-factor coupling mechanism of complicated traffic systems.The Leaky Noisy-OR gate model and dynamic discretization is utilized to simplify the solver.Two issues have been considered,they respectively are,1)definition,modeling and evaluation of cyber-physical resilience of urban roadway network;2)a case study of system resilience analysis with a utmost failure mode.CBN provides a view of “systemof-systems”,which utilize intervention and counterfactual methods to expand the dimension of system issue that can assist the macro decision-making under coupling effect of complex systems.Subsequently,concerning on unclear response mechanism of the large-scale roadway network to disturbance events,a massive RFID data mining is conducted to illustrate its self-absorbing characteristics.In this dataset,the average speed of road network during heavy rain have dropped of approximately 8-9%,and the average flow rate decreased near 2-5%.More than 80% of the links exhibits declined performance,which indicates the disorder perception of drivers’ subjective feeling under rain.Meanwhile,the concept of stress-strain curve is introduced for description of the time invariant and demand independent Network Resilience Curve(NRC).The described NRC promote the dimension from static evaluation to dynamic enhancement,which is also the theoretical basis for the selection of resilience-based perimeter controller’s setpoint.Finally,aiming at the problem of high demands on real-time control for large-scale road network under disturbance events,a resilience-based perimeter control strategy is proposed,which is a data-driven Proportional-Integral(PI)controller.Results show that the described controller could help the system absorb and adapt to the disruption automatically,and the investigated scenarios reveal the controller’s robustness and scalability.Moreover,with regard to the heterogeneity of traffic flow in a large-scale network,a graph attention autoencoder framework considering neighbors is proposed.The parameters are selected by offline training,in which training data is obtained by microscopic simulation.The cooperative controller is chosen online via the described framework,and local max-pressure control is applied in accordance to the online selection.This method could deals with the dimension reduction of directed and high-dimensional graph data,in which various local optimization strategies could be combined for the optimization of the network operation.The research in this thesis is an interdisciplinary problem which combines real-time situational awareness,cooperative control as well as emerging communication technologies.This study has perceptiveness thinking,theoretical significance and value in reaction to uncertain and problematical events of transportation system,as well as considerable needs of cooperative control. |
PDF Full Text Request