Spiking-STGCN: A Lightweight Spiking Spatio-Temporal Graph Convolutional Network Model for Urban Traffic Flow Prediction

Urban traffic flow prediction holds significant application value in Intelligent Transportation Systems (ITS). This paper proposes a Spiking Spatio-Temporal Graph Convolutional Network (Spiking-STGCN) model integrating spatio-temporal modeling with event-driven computation. By introducing a lightweight event-aware spiking encoder, a graph convolutional layer based on multi-head attention, and a spiking neuron module, the model achieves efficient characterization of spatio-temporal dependencies within traffic networks. The architecture employs stacked Spiking-GCN and Spiking-Temporal blocks. Leveraging the sparse activation property of differentiable spiking neurons, by leveraging the sparse activation property of differentiable spiking neurons, the proposed model achieves a $\mathbf{6 2. 5 \%}$ reduction in parameter size and significantly lowers computational complexity. Experimental results on the PeMS08 dataset demonstrate that the proposed model maintains prediction accuracy while exhibiting superior lightweight characteristics and efficient inference performance, showing potential for deployment on embedded platforms.