A multi-scale spatiotemporal network traffic prediction method based on spiking neural model

Spiking neural P systems are a class of distributed parallel neural-like computational models inspired by the mechanism of spiking neurons. Traffic prediction is a kind of spatiotemporal series prediction problem with nonlinear, non-stationary and complex multi-scale spatiotemporal dependencies. In order to cope with this complex challenge, this study introduces a novel model to propose a new long-term traffic prediction model, namely MSST-SNP. In the time view, for obtaining more robust traffic temporal features, we utilize spiking neural models to develop a multi-scale temporal feature fusion method. Then, in the spatial view, a spatial feature fusion method is proposed to adaptively capture the static physical and potential dynamic spatial features between road nodes. Finally, a spatial-temporal gated attention fusion module is utilized to extract spatiotemporal features at different levels through multi-view information fusion. Based on experiments conducted with four authentic public traffic flow datasets, our method demonstrates state-of-the-art performance in predicting traffic flow, as evidenced by the latest results obtained.