Research On The Key Technologies Of Dynamic Traffic Characteristics Analysis For Urban Road Networks

With the development of smart cities,a large number of traffic-related data are continuously generated.This is due to the changes of urban traffic from the perspective of data dimensions.Advanced analytical techniques are advantageous in assessing the trend of traffic conditions obtained from large amounts of traffic data.Currently,the level of traffic data analysis determines the level of intelligent transportation systems service.Therefore,understanding the characteristics of urban traffic network through traffic data is important in modern intelligent transportation.The dataset of the actual urban traffic network comprises the traffic information of each section and intersection.The change of the overall data information of the road network reflects the mutual containment,which is the interaction of the traffic states of each component of the road network.The change of detailed data information reflects the time-varying law of traffic,which is a representation of the change of traffic state on each road.Each traffic data is the basic element of traffic network information and has attributes of time and space.In this study,the data model is taken as the research starting point,and different extraction objects are used for technical analysis of traffic characteristics.The technology applies the theories of integrated learning,complex network and deep learning to solve the key problems of dynamic characteristics analysis of urban traffic network.The main research contents are as follows:(1)A technical framework for analysing the dynamic characteristics of urban traffic networksThe core object of the technical research is urban traffic network.First,the model of urban traffic network structure is selected according to the characteristics of the detection data.Considering the spatio–temporal attributes of the traffic data,a tensor description method that considers network structure is proposed for traffic data.Different objects extracted from the data model(including data elements,fibres and tensors)are used for the technical layering of traffic characteristics.The technical framework of urban traffic network dynamic characteristics analysis includes discrete traffic data preprocessing layer,time series traffic data analysis layer and spatial–temporal traffic data prediction layer.(2)Discrete traffic data preprocessing based on multiple detection sourcesThe preprocessing layer of discrete traffic data mainly preprocesses discrete traffic data,which are mainly basic detection data acquired by traditional traffic detectors.A multi-source detection data verification method based on optimised random forest and a multi-source detection data fusion method based on filter estimation are proposed.Considering the outlier characteristics of traffic detection,a mechanism to enhance the association of decision trees is introduced to optimise the random forest.In the optimisation,the training decision tree set is used to verify the traffic data by strengthening the perceptual weight of the relevant samples in the training set.In the experiment,multiple groups of detection data in the actual demonstration area are selected for model verification,and the proposed approach is compared with similar methods.Considering the diversity of traffic data sources,a multi-source data estimation approach is used to obtain reliable traffic data.The effectiveness of the fusion model is confirmed by the verified instance data.The results show that the method can provide reliable traffic data for subsequent time series analysis.(3)Feature analysis of road time series traffic data based on complex networkThe analysis layer of the time series traffic data mainly analyses the state characteristics of the traffic time series data,and the analysis object is the fibre data extracted from the traffic data tensor.The phase-space reconstruction method and visibility graph are used to analyse the information space of traffic flow mechanics characteristics.The modularity,average clustering coefficient and degree distribution of the network structure are analysed by constructing a phase-space reconstructed network of traffic flow time series.Examples of traffic flow time series with different parameters are selected for verification.The results show that the degree distribution of the network is Gaussian and the average clustering coefficient is attenuated.In addition,the network constructed has a high modularity.The study proposes a method to decompose traffic flow time series in different states to construct a complex network,considering the difference of traffic flow time series in the different states.The traffic state is divided by clustering large applications algorithm,and the network adjacency matrix of traffic flow time series is obtained via multi-parameter matrix superposition.The actual time series data are selected to validate the model.The visualisation relationship between the actual state change of traffic and the network structure of time series is mined through the model.(4)Prediction of traffic network congestion based on deep learningThe prediction layer of space-time traffic data mainly predicts the space-time state of traffic networks.The object of this layer is traffic data tensor.A spatial–temporal traffic congestion situation prediction method based on GRU-CNN is proposed.Considering the time and space attributes of traffic data,the third-order tensor of traffic data is extracted from the perspective of time domain,and the GRU is used to predict the traffic flow parameters of the traffic network.Then,the third-order tensor of multi-source spatial–temporal traffic data is compressed into traffic data image,combining with spatial structure.Convolutional neural network feature extraction technology is used to extract and identify traffic network congestion features.The actual urban traffic network data are selected for model verification.The multi-step prediction of traffic flow parameters effectively ensures the prediction accuracy.The proposed model is trained by the actual classified dataset.The prediction results of the test set demonstrate the model reliability.