Traffic Speed Prediction Of Urban Road Network Based On Deep Hybrid Model

With the rapid development of modern cities,the population and car ownership continue to rise,and the demand for travel has increased sharply,so alleviating the pressure on the road network and reducing traffic congestion have become urgent problems in the development of modern cities.As the main parameter for evaluating traffic conditions,traffic speed contains the spatial information of the road network and the temporal characteristics of travel of traffic participants,the analysis and excavation of the potential laws of traffic speed is helpful to guide traffic planning and alleviate traffic congestion.At the same time,accurately predicting traffic speed is also of great practical significance for the realization of intelligent transportation systems.Most of the existing research methods are based on one of the characteristics of traffic speed,and a hybrid model is the trend to obtain the spatiotemporal depth characteristics of traffic speed.Therefore,this paper focused on both the temporal characteristics of traffic speed and the space of road networks,and two different deep hybrid models were designed to predict urban traffic speed.The main research contents were as follows:Firstly,the traffic speed on the urban road network was defined and analyzed based on the original GPS data and road network.The original GPS data was preprocessed,the abnormal matched GPS data was analyzed;then K Nearest Neighbor algorithm(KNN)and Spatiotemporal Similarity KNN(STS-KNN)were used to predict the abnormal GPS data,and experiments had shown that the STS-KNN can effectively predict the abnormal matched GPS data to improve the availability of GPS data;according to the matched GPS data,the traffic speed of the road segment was defined,and the spatial and temporal characteristics of the traffic speed in the urban road network were analyzed.Secondly,considering the temporal characteristics of traffic speed and the information of road network,A Long Short-Term Memory and Restricted Boltzmann Machine Network(LSTM-RBM)model with fine-tuning strategy was proposed for traffic speed prediction.LSTM-RBM was designed to effectively learn the spatiotemporal characteristics of speed sequences.At the same time,a fine-tuning strategy was introduced to pre-train LSTM-RBM to further improve its prediction accuracy.Experiments showed that the performance of thehybrid model was better than existing deep learning models.Finally,considering the spatial characteristics of the urban road network and the temporal characteristics of traffic speed,graph was used to describe the topology of the road network,and GCN-GRU-GCN based on Graph Convolutional Neural Network(GCN)and Gated Recurrent Unit(GRU)was proposed to predict urban traffic speed.First,GCN was used to capture the spatial characteristics of the road segments that are closer to each other;second,GRU was applied to learn the temporal characteristics of the speed data;then the GCN was used to learn the spatial characteristics of the longer road segments;finally experiments were designed from different angles to prove the effectiveness and superiority of the model.