A Study On Dual GRU Traffic Speed Prediction Based On Neighborhood Aggregation And Attention Mechanism

With China’s comprehensive entry into a moderately prosperous society,the number of vehicles in China has been continuously increasing,posing more challenges to people’s transportation.The introduction of intelligent transportation systems has raised new requirements for traffic speed prediction.Considering the overreliance of traditional traffic speed prediction on temporal features,this paper focuses on constructing a traffic speed prediction model based on spatiotemporal features of traffic speed data.This paper first introduces the graphs related to traffic speed prediction,traffic road network graph,corresponding five prediction models and their theories,and then derives the method of extracting time features from traffic speed data.In view of the problem that traditional prediction methods ignore the spatial influence on speed,this paper extracts the spatial features of roads by using neighborhood aggregation.Then two GRUs are connected in parallel to analyze the time features and spatial features respectively,and predict the traffic speed on the road in the future time,and propose a double GRU model based on neighborhood aggregation: N-DGRU.After N-DGRU,this paper connects an attention model,which uses attention mechanism to fuse all hidden states output by two parallel GRUs,and makes full use of the information contained in each hidden state,and proposes a double GRU model based on neighborhood aggregation and attention mechanism: NA-DGRU.NA-DGRU uses the time features and spatial features of each historical moment speed to predict,which can reduce the prediction error and improve the reliability of speed prediction.This paper chooses the real-world datasets SZ-taxi and Los-loop to test the performance of the proposed models.Under the given seven sets of model parameters,the analysis is conducted from the dimensions of RMSE,MAE,and Accuracy.The prediction results show that the NADGRU prediction model proposed in this paper achieves the best performance in all aspects,indicating that this prediction model has both theoretical and practical value.The research indicates that leveraging neighborhood aggregation,GRU,and attention mechanism to extract spatiotemporal features can effectively construct the NA-DGRU traffic speed prediction model.However,due to the diversity of traffic environments and demands,this prediction model still has limitations and numerous aspects that require further optimization.