Traffic Flow Prediction Based on Recurrent Independent Mechanisms

Traffic flow prediction is an important issue of the intelligent traffic control and management systems. However, traffic flow data has nonlinear and complex characteristics in both time and space, making it challenging to accurately predict it. In this regard, this paper proposes a Graph temopral recurrent independent mechanisms (G-tRIM) model, which uses Graph attention networks (GAT) to effectively capture the spatial dependencies of traffic flow data, and uses Recurrent independent mechanisms (RIM) to accurately characterize the latent state of traffic flow data. We conduct experiments on the Beijing and Guizhou datasets, and the experimental results show that our proposed G-tRIM outperforms the baseline models on both datasets in terms of MSE and MAE.