Research On Traffic Flow Forecasting Method Based On Spatio-temporal Graph Convolutional Neural Networks

With the continuous acceleration of urbanization and the increase of travel vehicles in the city,resulting in increasingly prominent traffic congestion problems faced by urban residents.Traffic flow forecasting refers to the use of historical traffic flow data and techniques such as data analysis and machine learning to forecast the future traffic conditions over a period of time.This can provide travel route planning recommendations for traffic managers to optimize urban traffic operations and improve travel efficiency.However,current traffic flow forecasting models often cannot well express the topological structure characteristics of road networks,are difficult to fully capture the spatio-temporal characteristics,and lack the ability to model the spatiotemporal correlation.Therefore,this paper conducts research on the problem of traffic flow forecasting,and proposes two traffic flow forecasting models based on graph convolutional neural networks.The specific research results includes:(1)A traffic flow forecasting model based on Gated Chebyshev Graph Convolutional Recurrent Neural Network(GGCRNN)is proposed,which includes two feature learning mechanisms for traffic flow: spatial feature and temporal feature learning mechanism.In the spatial feature learning mechanism,the model parametetizes the distance matrix and the Laplacian matrix,then designs an adaptive topology network learning matrix,which can better express the topological structure characteristics of the road network.And the model combines the Chebyshev graph convolutional neural network with the gating mechanism to enhance the model’s ability of spatial feature learning in traffic flow data.In the temporal feature learning mechanism,the model uses the gated recurrent neural unit to capure the temporal features of traffic flow after extracting the spatial features.So the model can capture the spatio-temporal features of traffic flow data more effectively.The experimental results show that the model is the best compared to other baseline models.(2)A traffic flow forecasting model based on Attention Mechanism and SpatioTemporal Graph Convolutional Recurrent Neural Network(ASTG-CRNN)is proposed.The model introduces attention mechanism in both spatial and temporal dimensions to improve the ability to model dynamic spatio-temporal correlations in traffic flow data.And the gated Chebyshev graph convolutional neural network is used to capture the spatial features in the traffic flow data.To address the issue of poor prediction performance of the GGCRNN model in long-term traffic flow forecasting,the convolutional neural network and the bidirectional gated recurrent neural unit are combined to automatically learn long-term changes in traffic flow data.The experimental results show that the model has better prediction performance than other baseline models,and the long-term prediction performance is significantly improved compared to the GGCRNN model.