Research On Traffic Flow Prediction Model Based On Recurrent Convolutional Neural Network

In recent years,countries around the world have been vigorously developing intelligent transportation systems to achieve efficient traffic control.Traffic flow prediction is an indispensable part of the intelligent transportation system.Accurate and real-time traffic flow prediction can provide important data support and theoretical support for the traffic management department to reasonably guide and control the traffic flow.This is of great significance for improving the operational efficiency of the urban traffic network and improving the travel experience of residents.However,traffic flow data often have complex dynamic spatiotemporal correlations,which makes the task of traffic flow prediction much more difficult.Existing traffic flow prediction methods also have different degrees of defects in mining the spatiotemporal characteristics of traffic data.To address the above challenges,this paper proposes several traffic flow prediction models based on recurrent convolutional neural networks to mine the spatiotemporal correlations of traffic data in actual traffic road networks.Then,according to the shortcomings and deficiencies of traditional convolutional and recurrent networks,the model is improved with graph convolutional neural networks and variants of recurrent neural networks to enhance the prediction effect of the model.The main research contents of this paper are as follows:First,this paper proposes a convolutional long and short-term memory network model based on temporal attention mechanism.This model uses the temporal attention mechanism to capture the correlation degree of traffic data at different times of the same node,then uses onedimensional convolution operation to extract the spatial features of the time series data,and finally feeds the output time series to the long and short-term memory network to capture time dependence of traffic sequences.This model fully solves the defects that the traditional time series model cannot model spatial correlation and cannot handle nonlinear data.After that,a gated recurrent graph convolutional network model based on spatial attention mechanism is proposed.This model uses spatial attention mechanism to obtain the degree of correlation of different roads in spatial locations,then uses graph convolutional neural network to mine the spatial features in the traffic data,and finally feeds the output time series into a gated recurrent unit to capture the traffic sequence time dependence.Therefore,the defect that the convolutional neural network can only process two-dimensional grid data and cannot process the structure of the traffic road network map is solved.Then this paper constructs a spatiotemporal recurrent network model based on adaptive graph convolution.This model mainly proposes two adaptive modules to enhance the function of graph convolution operation to fully extract the spatial features of graph data.With weighted average operation,the resulting tensor can better capture the spatial dependencies between nodes from multiple aspects,and finally the output time series is fed into a gated recursive unit to capture the temporal dependencies of traffic data.This model mainly solves the problem that the predefined static adjacency matrix cannot fully contain the information about the spatial dependencies between nodes.Finally,the prediction performance evaluation experiments of the above three models are carried out on the real traffic dataset.Compared with some classic baseline models,the traffic flow prediction model based on recurrent convolutional neural network proposed in this paper has achieved good prediction performance.In addition,this paper also performs ablation analysis on each model,which fully verifies the effectiveness of each module in improving the prediction accuracy.