Data-Driven Framework For Dynamic Urban Road Capacity Estimation

One of the leading causes of urban traffic congestion is the mismatch between limited road supply resources and high traffic demand.The soaring number of private cars has made this problem increasingly serious.Although the area of urban roads in China has been increasing year by year,urban traffic congestion has not been resolved,and the increase in road supply resources is still constantly inducing new traffic demand.In order to solve the issues above,the current strategy of urban traffic management in many domestic cities is shifting from uni-laterally increasing supply to demand regulation.Understanding the supply of the urban traffic network is an important prerequisite for demand regulation.However,the existing road ca-pacity definition and the corresponding estimation methods lack consideration of the dynamics of road supply conditions? hence it is difficult to adapt them to the requirements of real-time traffic management.This research focuses on the problem of urban road capacity estimation.We propose the concept of road dynamic capacity,and,utilizing emerging machine learning technologies,a data-driven estimation framework as well.This research first summarizes existing definitions of road capacity.The shortcomings of road capacity in existing domestic and foreign standards are analyzed.The concept of dynamic road capacity is therefore proposed by taking into account the dynamic changes of road supply conditions.In addition,in view of the doubts about the rationality and generalization of the adjustment coefficients in existing capacity estimation methods,a data-driven road capacity es-timation method is designed,which can be decomposed into two steps,including traffic volume estimation and capacity estimation.Secondly,this research defines the problem of traffic volume estimation,and analyzes the influencing factors of traffic volume from two aspects,including traffic flow demand-side factors such as speed and density,as well as road supply conditions such as road width,signal timing,and weather.At the same time,it sorts out the definitions of three parameters of classic traffic flow,analyzes the connection and equivalence between different definitions under the assumption of the traffic flow equilibrium state.The rationality of the basic assumption of the traffic flow equilibrium state is also discussed.Subsequently,based on the analysis of the traffic volume estimation problem,a data-driven macroscopic traffic flow model is proposed,which relaxes the traffic flow equilibrium assump-tion of classical traffic flow models,and can simultaneously incorporate multi-dimensional traf-fic flow demand-side variables and road supply-side variables.It expands the dimensionality of traffic flow analysis.With the help of Gaussian process regression model,the accuracy of traffic volume estimation can be improved.Next,this research introduces the preliminaries of Gaussian process regression model,and makes an in-depth study of its kernel function and hyperparameter optimization problem.To address the overfitting phenomenon that occurs during the hyperparameter optimization process of Gaussian process regression model,this research proposes a truncated Newton method such that the robustness of hyperparameter optimization can be improved and the risk of overfitting can be reduced.This research also discusses the adaptation of Gaussian process regression model given noises in the input data.Finally,the proposed model is validated through simulation experiments.In the experi-ment,performances of models with different feature combinations and different noise parame-ters are compared.Comparison is also made between the proposed model and three fundamental diagram models.Results show that the proposed model can effectively improve the accuracy of traffic volume estimation.Road capacity under different road supply conditions can also be successfully obtained.This research is effective in improving the accuracy of macroscopic traffic flow analysis.It can effectively support the analysis of urban traffic network capacity,and assist traffic man-agement departments for real-time traffic management and control.It is of great significance for understanding the supply of urban road resources and alleviating urban traffic congestion.