Road Network Data Analysis And Traffic Development Trend Prediction Based On Spatial-temporal Correlation

Traffic prediction is an important part of intelligent transportation system.Through realtime collection and analysis of traffic data,traffic flow,traffic speed and traffic state are accurately predicted,so as to optimize traffic operation and improve traffic efficiency and traffic management ability.However,the traffic sequence is dynamic and nonlinear,and the spatial information stored in the traffic spatial topology map is small,and the spatio-temporal information fusion process is easy to lose information.These factors make it difficult to guarantee the accuracy,robustness and generalization ability of the traffic prediction model.In order to solve these problems,this paper studies and designs corresponding deep learning models for traffic speed prediction in terms of enhancing road node information representation,multi-scale analysis of context in time series and spatial relationship in road topology.(1)In order to make the model fully capture the specific traffic patterns of each road in the actual traffic conditions,a graph recurrent neural network(En-GRN)model based on enhanced road information representation is proposed.Firstly,aiming at the problem that the traditional spatial information matrix lacks independent road pattern representation,a roadspecific matrix is constructed in the model,and the matrix parameters are determined by data-driven method,which enhances the information representation ability of the road and improves the ability of the model to fully capture spatial information.Secondly,aiming at the problem that the abnormal state of road traffic affects the prediction results,time and peak labels are constructed to capture the time series law.Finally,aiming at the problem that the model parameters are too large due to the increase of road nodes,the node embedding and matrix decomposition methods are used to optimize the model parameters to improve the training speed.Experiments show that the proposed model has better performance in capturing sequence rules,spatiotemporal feature extraction and prediction accuracy.(2)In order to fully exploit the spatio-temporal correlation information that is conducive to traffic prediction modeling in limited data,a graph recurrent neural network(MSSTAGRN)model based on multi-scale spatio-temporal perception is proposed.Firstly,aiming at the problem that the traditional model is difficult to fully capture the context information of traffic data time series,the internal correlation of traffic time series at different scales is captured by decomposing GRU into hidden states at different scales and updating them according to different update frequencies in the model.Secondly,in order to more fully capture the spatial relationship between traffic network nodes,a spatial component is constructed by using a multi-layer series spatial model,which enables the model to perceive spatial information at different scales in a multi-scale manner.Finally,neural networks are used to fuse multi-scale temporal and spatial information to obtain more comprehensive spatiotemporal characteristics.Experiments show that the proposed multi-scale spatiotemporal perception can improve the accuracy,stability and robustness of traffic prediction.