Research On Data-driven Urban Road Network Dynamic Capacity Estimation Method

Road network capacity is the largest network flow under a certain service level and traffic demand,and it is a direct quantitative form of the regional capacity of urban road networks.Affected by many dynamic and static factors such as road network structure,traffic composition,environment,and the difficulty of obtaining network traffic flow data and data quality,road network capacity,especially dynamic road network capacity,is difficult to accurately quantify in reality.The concept of Macroscopic Fundamental Diagram(MFD)provides a new method for capacity analysis of road network.However,due to the assumption of road network homogeneity in MFD theory,it is necessary to study the same proton area division method under signal control for large-scale open road network.For a large-scale urban road network,vehicle travel often presents certain sub-regional characteristics,that is,traffic travel within a certain area is relatively closely connected,but the connection with outside the area is weak.Such characteristics are the basis for dividing the entire urban road network into homogeneous areas and conducting independent research.Aiming at the current research on the division of homoproton regions under signal control only from the perspective of relevance,the empirical rule method is used to complete the correlation strength characterization between homoproton nodes,and the automatic extraction and fusion of multi-dimensional features of nodes is not considered by means of machine learning.Insufficient consideration of the degree of homogeneity of operation characteristics of intersections in the sub-area,the diversification of capacity under the unstable operation state of macroscopic traffic flow,and the lack of a dynamic estimation method of road network capacity.Based on the structured data of vehicle trajectories,this paper uses graph convolution neural network combined with traditional community detection algorithm to study the homogeneous subzone road network dynamic capacity estimation.The specific research contents are as follows.First,in view of the unknown travel completion status in the original track of vehicle identity and location service data obtained by license plate recognition equipment,this paper completes the reliable division of single trip section in the original track of vehicle through the discrimination method of directed intersection OD based on Gaussian mixture model(GMM).Then,based on the shortest path principle and the structured process of time difference algorithm,the spatio-temporal information of the whole hierarchical travel chain of the road network is obtained.Finally,the structural data of vehicle trajectory is used to obtain the operation characteristic data of the path flow at the associated intersection,and the similarity and spatial distribution of the operation characteristic of the path flow at the associated intersection are qualitatively analyzed.The structural processing of vehicle trajectories and the operation characteristics analysis of path flow at associated intersections provide basic data support for the division of urban road homogeneous subzones and the study of road network capacity.In view of the existing research on the division of homogenous sub-zones under signal control,there are problems such as sub-zone division only from the perspective of the correlation between intersections,and insufficient consideration of the degree of homogeneity of the intersections in the sub-zones.Based on the analysis of the operation characteristics of the path flow at the associated intersection,the multidimensional operation characteristics of the intersection are extracted according to the urban road network topology model.At the same time,the correlation strength between the intersections in the urban road network topology model is determined by the operation characteristics of the path flow between the associated intersections.Taking the intersection traffic operation characteristics,spatial scale characteristics,intersection correlation strength and urban road network topology information data set as the input of the model,the traffic control sub area of urban road network is divided by GCN kmeans unsupervised network model framework oriented to the optimal modular community division task of undirected weighted network.At the same time,the community detection evaluation index modularity of undirected weighted network is used as the optimal objective function of the model,and the model optimization process is completed through the forward propagation of signal and the reverse transmission of error.By comparing this method with the traditional community detection algorithm of Louvain,the results show that for the urban road homogeneous subzone division,this method uses deep learning method to learn the degree of multi-dimensional characteristics homogeneity of intersections,and combines with the traditional community detection algorithm theory,taking into account the correlation strength between intersections,In the overall division effect of the homogeneous subzone,it is better than the traditional Louvain algorithm in community detection,and has stronger interpretability and universality.It provides a more reasonable regional division scheme for the subsequent dynamic estimation of homogeneous subzone road network capacity based on the macro basic graph theory.In this paper,the construction method of road network MFD driven by structured data of vehicle trajectory is used to analyze the historical operation status of macro traffic flow in the signal control sub-area.According to the operation law of macro traffic flow,the relationship between the cumulative number of regional vehicles and the number of completed regional vehicle trips Quantitative relationship,through the n-th degree polynomial function for modeling and fitting analysis,and on this basis,complete the static capacity estimation of the road network.Finally,the historical data of MFD state variables are used to divide the decision boundary of road network macro traffic flow unstable operation state based on Gaussian mixture model(GMM).On the basis of road network static capacity estimation results,the change rate of state transition between MFD state point and road network capacity critical state estimation point under macro traffic flow unstable operation state is taken as the decision basis,Through the dynamic estimation algorithm of road network capacity based on Kalman filter,the critical state estimation points of road network capacity are updated dynamically.Based on this,the dynamic estimation of road network capacity on weekdays and weekends in the study area is verified by an example.