Short-term traffic flow prediction based on adaptive graph convolutional recurrent network under multi-factor fusion

To enhance the precision of short-term traffic flow predictions, this paper proposes a model named EAG-LSTMA, a multi-factor fused spatiotemporal model based on adaptive graph convolutional recurrent architecture. This model comprehensively considers multiple external factors, including weather, holidays, and road conditions, through multi-factor fusion. By constructing an external data integration module (EDIM), the model effectively combines external factors with traffic data. EAG-LSTMA integrates AGCN directly into the gating computations of LSTM, allowing synchronization of spatio-temporal features in traffic flow data. The addition of an attention mechanism further enhances the model's ability to adapt dynamically to various influencing factors, enabling it to identify and leverage key features in spatio-temporal data. Validation conducted on the flow dataset from Longgang District, Shenzhen, China, reveals that EAG-LSTMA significantly improves prediction accuracy compared to both baseline models and existing traditional combination models, demonstrating its effectiveness in capturing dynamic traffic flow features under multi-factor conditions.