Traffic Flow Prediction Based On LSTM And GCN

Since the reform and opening up,with the rapid economic growth,the total number of vehicles has increased year by year,leading to the increasing number of vehicles in cities,which has caused traffic jams,traffic accidents frequently.As people spend more and more time on the road,these problems have a great negative impact on People’s Daily life,which requires us to use the existing technology and knowledge to solve these problems.The realization of urban intelligent transportation system management is regarded as one of the effective ways to solve urban traffic problems,which aims at providing traffic management services for people.It allows users to get more comprehensive information about road conditions,so that people can use the transportation network in a safer and more coordinated way.Traffic flow prediction is the core content of intelligent traffic management and control,and also an important basis of traffic information service.It can be used as a key basis for traffic decisionmaking,and can also provide effective route selection information for traffic travel.With the rapid development of machine learning and deep learning technology,more and more methods are being used in traffic flow prediction,including statistical model,machine learning model and deep learning model.Predictions of traffic flow are becoming more accurate.However,these models have some defects,such as not considering periodicity,too many parameters and so on.In view of the defects of the above models,we put forward two models,LSGCN model and LST-GCN model.LS-GCN model combines long term and short term memory network(LSTM)and graph convolution network(GCN)to predict the traffic flow,while considering the daily periodicity and weekly periodicity in the traffic flow data.On the basis of considering periodicity,LST-GCN model embedded the long and short term memory network into the training process of the graph convolutional network.We found that the LST-GCN model could better capture the temporal and spatial correlation existing in the traffic flow data.At the same time,the number of parameters in the model was greatly reduced and the calculation speed of the model was improved.Through a large number of experiments,it is proved that RMSE,MAE and MAPE are used as evaluation indexes respectively on the PEMS04 and PEMS08 data sets.The results of the two models proposed in this paper are better than those of the classical models such as HA,ARIMA and SVR,indicating that the prediction method proposed in this paper can effectively improve the accuracy of traffic flow prediction.