Traffic State Identification And Forecasting Based On Macroscopic Fundamental Diagram

The impact of large-scale traffic congestion on urban development is very obvious.Monitoring,analyzing,scientifically evaluating and predicting the traffic operation status of road network is of great significance for the refinement of urban traffic management.Macroscopic fundamental diagram(MFD)is a tool for modeling road network traffic from a macro perspective,which is suitable for traffic operation status judgment of regional network.With the rapid development of intelligent transportation system,a large number of electronic police devices are deployed at intersections and roads,providing an economical and convenient choice for the construction of MFD.Therefore,this paper uses the automatic number plate recognition data as the research data set and MFD as the research tool to identify and forecast urban regional traffic state.First,combined with previous research,a set of preprocessing process is proposed for the automatic number plate recognition data and road network point data used in the research,and the data quality analysis index is defined to analyze the data accuracy in detail.Based on the preprocessed data,the acquisition and cleaning methods of traffic parameters required to construct MFD using automatic number plate recognition data are explored,and the Van Aerde model is used to fit urban road MFD,and the calculation formula and meaning of traffic flow characteristic parameters are expounded.Based on the cross-validation mean square error method,the time interval of MFD parameter pooling is discussed.The experiment shows that5 min is the optimal time window for urban road MFD.Then,according to the applicable condition of MFD,a method is proposed to divide the heterogeneous urban road network into homogeneous sub-region network.Firstly,the tensor decomposition algorithm is used to extract the characteristics of traffic state in the time dimension,and the similarity between roads is calculated.Then,taking the similarity between road segments as the edge weight,an improved Fast-Newman algorithm is proposed to divide the road network into relatively homogeneous sub-regions.Experiment results show that the proposed algorithm can get sub-regions with strong homogeneity,and can ensure the spatial connectivity of the road segments in the sub-regions.On the basis of obtaining an effective MFD,the traffic operation state is qualitatively divided into 5 categories by the spectral clustering algorithm,and the method of VKT distribution of accumulative density ratio of the road is used to calculate the traffic state of sub-region networks.After obtaining a large number of regional traffic state time series data based on MFD,the transfer entropy is used to construct a correlation mining model considering delays to discover the traffic state causal correlation structure between regions.The transfer entropy algorithm is used as a variable selection method in regional traffic state prediction,and a regional traffic state prediction model based on TE-LSTM is constructed combined with LSTM neural network.Through the analysis of predictability and variable interpretability,it shows that as a filtering method,the transfer entropy algorithm can select variables with large contribution and strong interpretability,which has the prospect of practical application in regional traffic state prediction.