| At present,for the detection and imputation of traffic outlier,traditional methods such as time series analysis not only ignore the longitudinal data features of traffic data,but also do not apply to the massive data characteristics in the big data context,and lack effective unified algorithms for the detection and imputation of traffic data outliers.In addition,in the process of identifying critical intersections of urban traffic network,the core lies in extracting the traffic network data features and considering the correlation on consecutive time steps,which can be effectively solved by combining the road network traffic data with the temporal network model.In this paper,we conduct the outlier detection and imputation unified model of traffic data,and the recurrent congestion detection model based on road network critical intersection identification are studied as follows.In the analysis of traffic monitoring data,the magnitude of traffic density plays an important role in determining the degree of traffic congestion.In this study,an spatioFunctional Principal Component Analysis(s-FPCA)data imputation method is proposed,which combines the detection of traffic density anomalies at target intersection,the confirmation of anomalous values and imputation.Firstly,the anomaly detection is based on the binary principal component scores obtained from Functional Principal Component Analysis(FPCA),and the outliers are identified by the threshold method.Secondly,a method is proposed to estimate the traffic missing data based on upstream and downstream.Finally,simulations are performed on the actual traffic density data,and the imputation accuracy of s-FPCA is improved by 8.28%,8.91% and 7.48%compared with FPCA for daily traffic density data missing rate of 5%,10% and 20%,respectively,which proves the superiority of the method.Intersection information in urban traffic networks is crucial to the safe and efficient operation of road traffic.By controlling the critical intersections during the peak traffic period,the controllability of the road network can be increased.On the one hand,we build a time-series network model based on the Directed Supra-Adjacency Matrix(DSAM).The results show that the DSAM model has the characteristics of phase continuity in ranking the importance of intersections,and the results are more stable at different time granularity,which is better than the static key intersection identification results.On the other hand,based on the DSAM model to identify critical intersections in the road network,the short-term traffic flow prediction with the help of road network adjacency matrix shows that the Chebyshev Graph Convolutional Neural Network(Cheb Net)performs best and can be used to detect recurrent congestion in the urban road network,and reduce the possibility of widespread congestion by making corresponding control measures in a shorter time step.In summary,the s-FPCA method is superior for detecting and imputation traffic data outliers and can be applied to other longitudinal data analysis with periodic fluctuations.The DSAM model can identify critical intersections in the road network in the temporal network and perform recurrent congestion detection based on the identification of critical intersections. |
