| With the rapid development of our economy and society,traffic congestion in cities has become more and more serious.As an important node in the urban road network,the vehicle traffic efficiency at the intersection determines the overall traffic operation of the road network.Due to excessive traffic demand and insufficient supply,traffic spillover phenomenon often occurs at intersections in some cities due to over-saturation of traffic during commuting peak hours.Spillover will cause congestion to spread to the surrounding intersections,resulting in a complex spillover scenario that evolves from spillover at a single intersection to spillover at multiple intersections,resulting in large-scale congestion at the road network level.Macroscopic fundanmental diagram(MFD),as an important way to describe the traffic operation of the road network,can provide help for quantitative analysis of spillover effects.At present,there are few researches on macro basic graph in spillover scenario.It is of great practical significance for the study of traffic operation law,spillover evolution trend analysis and spillover prediction in spillover scenario.In order to quantitatively study the impact of spillover scenarios on road network traffic conditions,this paper obtains traffic flow data before and after spillover through the simulation platform,and studies the macroscopic basic map model of road network under complex spillover scenarios.The main research contents are as follows:(1)Construct the simulation spillover scene and obtain the basic map data of the road network.Based on the SUMO simulation platform,the complex spillover scene of multiple intersections in the road network is constructed.The real-time traffic flow data of each section were obtained through the relevant interfaces of the simulation platform,and the macro traffic flow data at the road network level were finally obtained based on the weighted average method according to the length of the section.At the same time,a non-spillover scenario is constructed for comparative study.(2)Study the macro basic graph model in the spillover scenario.In terms of model applicability,in the spillover scenario,the basic map applicability of each intersection is related to its position in the road network.The more intersections are directly adjacent,the more susceptible they are to spillover,and the applicability is poor.On the road network level,the applicability is greatly improved,among which the Greenberg model has the highest applicability,which verifies the existence of the basic map in the spillover scenario.In terms of spillover evolution trend,the overall spillover diffusion speed and the flow reduction speed before and after the spillover occurred in the road network showed a trend of "first increase,then decrease".In terms of the stability of the basic graph,the conclusion shows that the distribution of the basic graph tends to be gentle with the increase of the density.At the same time,when half of the intersections in the road network spillover,the dispersion of the basic map is greatly improved,and the RMSE value increases by about 161% compared with the data in the period without spillover.(3)Realize adaptive prediction of macro basic graph in spillover scenario.In the spillover scenario,the macro basic map has the characteristics of high dispersion and low stability.Based on this property,a basic graph adaptive prediction model based on Gaussian process under constraints is proposed.The biggest advantage of this model is that it can automatically correct the prediction results according to the observed data and get the prediction distribution of the basic graph,and the model only needs a small amount of historical data for training.At the same time,the upper and lower bound function of the basic graph is taken as the bounded constraint in the model.When 30 minutes of data training is used,the prediction accuracy of the model is improved by 77.9% compared with the traditional Gaussian process model.(4)Construct the spillover scenario prediction model based on the basic graph data.In this paper,an spillover prediction model combining graph convolution network GCN and LSTM is proposed with the input of macro basic graph data.By capturing the temporal and spatial characteristics of the basic graph,the location and time of the spillover can be predicted at the minute level.Compared with the single LSTM and GCN models,the prediction results of the model improved by more than 20% in terms of accuracy,recall rate and accuracy rate.Meanwhile,an spillover control strategy based on increasing full red time is proposed.The specific full red time is determined by the basic graph data of the node.Experimental results show that this method can reduce the impact of spillover. |
