融合自适应图与时空Transformer的交通流预测模型OA

[Background]With the advancement of urban modernization,intelligent transportation systems have become an indispensable part of modern cities.The key to the effectiveness of intelli-gent transportation systems lies in reducing urban road congestion through accurate traffic flow pre-diction.[Objective]To comprehensively consider the spatiotemporal characteristics of traffic data,dynamically capture the complex spatial and temporal correlations of traffic data,and effectively im-prove the accuracy of traffic prediction tasks.[Method]We propose a novel DGC-Transformer mod-el that integrates dynamic graph construction(DGC)with Transformer to dynamically capture spatio-temporal correlations in traffic data.The model employs multilayer perceptron(MLP)projection and temporal embeddings to capture periodic temporal patterns.Within the core Sandwich block,a Trans-former encoder first captures long-range temporal dependencies.Subsequently,a DGC module con-structs a data-driven adaptive adjacency matrix to model hidden spatial dependencies.A graph con-volutional network(GCN)module then aggregates spatial information using this adaptive matrix,followed by a second Transformer module that re-models the temporally contextualized features with spatial information.The architecture stacks two Sandwich blocks with residual connections to enhance expressive power and ensure training stability,outputting predictions through an MLP pro-jection layer.[Data]The model's performance was evaluated on four widely used traffic prediction datasets from the California department of transportation's performance measurement system(PeMS).[Result]Across the four public datasets,the DGC-Transformer model consistently outper-formed five baseline and ten state-of-the-art models in terms of mean absolute error(MAE),root mean square error(RMSE),and mean absolute percentage error(MAPE),underscoring the impor-tance of dynamically capturing spatio-temporal correlations for significantly improved traffic flow prediction accuracy.