Research On Short-term Traffic State Prediction Method Based On Deep Learning

With the rapid development of the social economy,the number of motor vehicles in the country continues to increase,making the city’s traffic pressure greater and greater and the problem of traffic congestion increasingly serious.As the core content of the intelligent transportation system,short-term traffic state prediction can not only provide travel information for travelers,but also provide technical support to the traffic management department to help alleviate traffic congestion.The traditional shallow prediction method has poor prediction effect due to limited modeling ability.Deep learning has powerful feature learning capabilities,so this paper combines deep learning to make short-term traffic state prediction.In this paper,the short-term traffic state prediction is divided into two parts: short-term traffic flow parameter prediction and traffic state identification.Firstly,the traffic flow parameters are predicted,and then the traffic state is discriminated according to the predicted traffic flow parameters,so as to obtain the traffic state at the future moment.The main research contents are as follows:(1)Aiming at the shortcomings of the existing short-term traffic flow parameter prediction model that cannot fully utilize the spatiotemporal characteristics of traffic flow data,a convolution gated recurrent unit prediction model combined with attention mechanism(ACGRU)is proposed.The model firstly uses the convolutional neural network to extract the spatial features of traffic flow,and then uses the gated recurrent unit combined with attention mechanism to extract the temporal features.At the same time,the traffic flow periodic input matrix is constructed to extract periodic features,and finally all the features are fused to predict.The experimental results show that the ACGRU model can improve the prediction accuracy of traffic flow,and has smaller prediction errors than other prediction models.(2)A traffic state identification model based on KM-FCM and random forest is proposed.Firstly,the FCM algorithm is improved,and a KM-FCM clustering algorithm is proposed.The algorithm uses the optimal clustering center generated by K-means++ algorithm clustering as the initial clustering center of the FCM algorithm,which improves the shortcomings of the FCM algorithm randomly initializing the clustering center.Then use the KM-FCM algorithm to perform cluster analysis on the traffic flow data,divide the traffic state,and finally use the data with traffic state to train a random forest model for identification.The experimental results show that the various traffic states obtained by KM-FCM algorithm clustering conform to the laws of actual traffic conditions,and the selected random forest algorithm has a higher identification accuracy.The short-term traffic flow parameter prediction based on ACGRU and the traffic state identification based on KM-FCM and random forest constitute the short-term traffic state prediction model.This model has a high prediction accuracy and can provide real-time and reliable traffic trip information,which is of great significance for alleviating urban traffic pressure.