Research On Short-Term Traffic Flow Forecast Based On Temporal And Spatial Features

With the continuous development of society,the number of urban motor vehicles continues to increase,and traffic congestion has become an important issue hindering the development of urban travel.For this reason,many countries have solved the traffic congestion problem by developing intelligent transportation systems,and short-term traffic flow forecasting information provides data support for them,providing travelers with real-time reasonable travel plans,and alleviating the traffic pressure in the city.Therefore,real-time and accurate prediction of traffic flow information has become an important part of the development of intelligent transportation systems.Due to the large scale of traffic flow data,strong temporal and spatial dependence,and obvious social relevance,it is difficult to guarantee the time delay and accuracy of traffic flow prediction.Many scholars have used traditional applied statistical models,neural network models,etc.to predict and analyze future traffic flow information based on historical traffic flow information.These models have not fully considered the temporal and spatial characteristics and long-term dependence of traffic flow data.Therefore,this paper proposes two models that integrate temporal and spatial characteristics to predict traffic flow in order to improve the prediction accuracy of short-term traffic flow.The main work and innovations of this paper are as follows:Taking into account the temporal and spatial characteristics of traffic flow,a graph convolutional network model based on the attention mechanism(RES2GCN)is proposed.The model first modulates the linear graph convolutional network with a nonlinear graph convolutional network,and then stacks the modulated graph convolutional network to form a residual graph convolutional network in order to better extract the spatial variation characteristics of the traffic flow Finally,the attention mechanism is used to introduce weight coefficients to weight the time series to reconstruct the traffic flow series.Further considering the long-term dependence of traffic flow,a spatio-temporal graph convolutional network model(TGCN2S)is proposed.The model is based on the encoder-decoder.The encoding part learns the temporal and spatial correlation characteristics and long-term dependence relationship from the temporal and spatial sequence related to the input traffic flow through T-GCN,and the GRU decoder decodes based on the encoded temporal and spatial vector value to reconstruct the traffic flow sequence.In order to verify the effectiveness of the model proposed in this paper,the public California traffic flow data set and Seattle traffic flow data set were used for experimental verification,and compared with HA,ARIMA,GCN,T-GCN and A3T-GCN models.The experimental results show compared with other models,the prediction accuracy of the model in this paper is improved both in single-step prediction and multi-step prediction,and the error rate is reduced.The results show that the RES2GCN model is preferred when the road network structure is complex or single-step prediction is performed,and the TGCN2S model is preferred for multi-step prediction.