Research On Short-term Traffic Flow Prediction Of Urban Roads Based On Graph Convolution Network

Managers cannot effectively develop traffic control strategies,making it more difficult for people to get around.Therefore,how to achieve accurate short-term traffic flow prediction has become the focus of more and more scholars.However,due to the limitations of traditional traffic flow acquisition devices,the quality of the collected traffic flow data is not high,and the traditional short-term traffic flow prediction object is often to study a single road,but in urban roads,the traffic flow of a single road is usually affected by the conditions of upstream and downstream roads,so in urban road traffic flow prediction,it is necessary to consider the correlation between intersections,and the processing ability of traditional prediction models on the complex road network of urban roads needs to be improved.These have become urgent problems for short-term traffic flow forecasting methods.In this context,this paper proposes a short-term traffic flow prediction method for urban roads in graph convolutional network based on spatial correlation degree.Firstly,the three methods of simple numeric relationship,single-point threshold relationship and combinatorial logical relationship layer are used to deal with missing values and outliers in the original data.Then,for the missing data,a fault data repair model based on GM(1,N)-GA-RBFNN based on the spatiotemporal characteristics of traffic is proposed,and the speed data of Fuma Road in Fuzhou City is taken as an example to verify the repair effect of the method.In order to build a prediction model of urban road traffic flow and consider the influence of traffic flow in spatially adjacent road sections,we analyze the spatial correlation degree of urban roads to determine the degree of correlation between the target road segment and the surrounding road segment.A more practical and more conforming spatial feature extraction method for urban road traffic network structure is proposed.This method treats the urban road network as a topology map and uses the traffic flow on the road as the characteristics of the edges.Use the line graph transformation method in graph theory to convert the road network topology into a road adjacency topology,and convert the traffic flow on the road into node information features.This method can effectively solve the problem of insufficient traditional spatial feature extraction methods for traffic flow prediction.Then,the GCN is used to realize the reasonable extraction and utilization of the spatial characteristics of traffic flow,and then the retrieval of the periodic information of the target road section is realized with the help of the memory characteristics of Bi LSTM,and a combined model of S-GCNBi LSTM is proposed to realize the prediction of urban road traffic flow.In order to verify the effectiveness of the proposed model,this paper takes the road network in the central urban area of Fuzhou as a research example,selects the short-term traffic flow prediction model of Fuma Road traffic flow data to the S-GCN-Bi LSTM deep network for verification,and compares and analyzes the prediction results of classical short-term traffic flow prediction models such as ARIMA,DCRNN and T-GCN through the comparative experiments of the spatiotemporal module and the prediction results of classical short-term traffic flow prediction models such as ARIMA,DCRNN and T-GCN.The urban road traffic flow prediction model based on the S-GCN-Bi LSTM model can reduce about 12,3-16 on weekdays and weekends,respectively,on weekdays and weekends,and by about 1-8 and 3-5 on weekdays and weekends,respectively;the prediction effect is poor with 30-minute time granularity,but it can be reduced by 3-6 and 2-9on weekdays and weekends,respectively.It can be seen from the experimental results that the prediction performance of S-GCN-Bi LSTM proposed in this paper is greatly improved compared with the classical prediction model.The model can not only learn the trend of traffic volume better,but also make the prediction results of the model more accurate by paying attention to the spatial correlation of traffic volume.