Research On Traffic Congestion Identification And Prediction Based On Translational Nested Grid Model

With the development of Chinese urbanization process,the number of urban population and vehicle ownership increase year by year.Thus the urban traffic congestion problem increases critical,which has become one of the main problems in the big cities.Fortunately,with the development of big data mining and information technology,intelligent transportation technology has gradually become an important means of urban traffic management.Indeed,massive information can be mined based on floating car data.The traffic operation state of urban road network can be obtained,while the real-time traffic congestion information can be provided for traffic managers and road users.However,the current urban traffic management method based on floating car data still has many shortcomings.For urban traffic state identification,due to the complexity of traffic network,the methods of traffic state identification are diversified without any unified standard.Moreover,the traffic feature extraction method based on floating car data often needs electronic GIS map for matching.This method not only depends on the reliability of GIS map,but also has low operation efficiency for large-scale traffic network.Therefore,it is urgent to study a new traffic feature extraction method for large-scale road network.In addition,with the continuous updating of clustering methods and the continuous development of artificial intelligence technology,traffic congestion identification and prediction methods also need new technologies to deepen and develop.In order to improve the management level of urban traffic congestion and play the role of big data technology in urban traffic management,this dissertation creatively puts forward the method of translational nested grid model to study the identification and prediction of traffic congestion driven by massive floating data.The main research contents and findings of this dissertation include the following three aspects:(1)Firstly,we establish a map independent translational nested grid model to extract the traffic characteristics of large-scale urban road network.Aiming at the problem of traffic spatial granularity constraint in node pair grid model,this model uses the method of nested grid to reflect the traffic characteristics of node pair in continuous small cells based on massive floating car data.This model realizes the description of traffic path characteristics at fine-grained level by extracting the corresponding road network structure.To solve the problem that the boundary of the grid model coincides with the road network,the established model uses the translational method to separate the grid boundary from the road network,and combines the translated results with the original model results,so as to extract the traffic network structure that coincides with the grid boundary.Finally,the calculation model of traffic performance index PI is established with speed ratio by K-means clustering method.According to the extracted traffic network structure and traffic characteristics,the temporal and spatial distribution of PI in the fifth ring road network of Beijing is analyzed.The results show that the translational nested grid model can effectively extract large-scale road network structure and traffic characteristics.The central translational method can increase the number of small cells by 20.7%.(2)Secondly,we propose the DGST-DBSCAN model to identify the temporal and spatial characteristics of traffic congestion.The model establishes spatio-temporal clustering rules of directed cells,and provides clustering algorithm for spatio-temporal identification of traffic congestion in directed cells.At first,the topological structure of traffic network based on directed cells is extracted by translational nested grid model.Then the vacancy spatio-temporal data of PI is filled.With the PI data,the congestion clusters of different spatio-temporal range are obtained by DGST-DBSCAN model.Furthermore,by extracting the traffic characteristics of the spatio-temporal congestion cluster,the spatio-temporal range of the spatio-temporal congestion cluster is obtained.From the result,we analyze the spatio-temporal distribution of the congestion occurrence points,the congestion duration points,the congestion dissipation points and the congestion bottleneck points.Finally,based on the PCA method,the traffic congestion evaluation model is established.By sorting the evaluation indexes of spatio-temporal congestion clusters,we present the top ten congested spatio-temporal intervals in Beijing.(3)Thirdly,we establish a spatio-temporal congestion prediction model of traffic network based on the fusion deep learning method.The model contains the variables of the time,space,weather and historical state.After processed by clustering and one-hot method,the variables are smoothed and standardized in order to increase the prediction accuracy.To select the main variables,the importance of variables is evaluated based on the random forest method.By comparing the prediction results of the fusion deep learning model with other models in different road types,different spatial locations,different time periods,different congestion degrees,different congestion frequencies and different time steps,it is found that the LSTM2L-GRU model is the best with accuracy in all the prediction models.Besides,the results show that the difficult conditions of prediction include the intersection structure,working day,peak hours,high congestion level and lower congestion frequency.In this dissertation,the structure of translational nested grid model is proposed for the first time.The established model of traffic congestion identification and prediction based on floating car data has important technical value and practical significance for the development of urban intelligent transportation.