Research On Highway Traffic Flow Prediction Based On Attention Mechanism

The accurate prediction of highway traffic flow is of great research significance for travel planning,traffic management and control.Most current traffic flow prediction models fail to accurately capture the spatial and temporal characteristics of traffic flow data,resulting in unsatisfactory prediction performance of the models.This thesis proposes a highway traffic flow forecasting model,VAR-SPA-BiGRU model,which combines Vector Autoregression(VAR model),Bidirectional Gate Recurrent Unit(BiGRU)and Spatial Attention Mechanism(SPA).The improved whale algorithm(CWOA)is used to find the optimal solution of the initial Learning rate,the number of hidden layer neural units and the training batch parameters in the BiGRU neural network,so as to improve the accuracy of highway traffic flow prediction.The main innovation points are as follows:(1)CWOA algorithm: The traditional whale algorithm is easy to fall into the local optimum when finding the global optimal solution,which leads to the poor accuracy of the algorithm solution.To address this problem,this thesis improves the traditional whale algorithm by introducing Circle mapping to initialize the whale population and increasing the initialized location diversity;a nonlinear convergence factor is introduced to improve the global and local search capability;a neighborhood perturbation is performed at the optimal position during the global position update,so that the algorithm can better jump out of the local optimum and find the global optimal solution during the global search.The simulation experiments confirm that the algorithm is better than other algorithms for finding the best results.(2)VAR-SPA-BiGRU model: Highway traffic flow data often have complex spatial and temporal correlations,which makes the task of traffic flow prediction much more difficult.In this thesis,we use the VAR model to process the data and mine the linear features of the traffic flow data,and input the processed data into the BiGRU model and the spatial attention mechanism respectively.The BiGRU model processes the input sequence in both positive and negative directions to deeply mine the temporal features.The spatial attention mechanism then reveals the complex spatial connections between different road sections and extracts the spatial features in the data.Finally,an example analysis was conducted on the UK highway traffic flow data set,using root mean square difference and absolute coefficient as evaluation indicators to evaluate the predictive performance of the VAR-SPA-BiGRU model.The experimental results show that the model proposed in this article performs better and predicts more accurately in highway traffic flow prediction tasks compared to other models.