FeSTGCN: A frequency-enhanced spatio-temporal graph convolutional network for traffic flow prediction under adaptive signal timing

Traffic flow prediction is the fundamental cornerstone of intelligent urban transportation systems. However, existing research has predominantly focused on exploring spatiotemporal dependencies within the spatial and temporal domains, often overlooking the frequency information present in traffic data. This study aims to address this limitation by simultaneously modelling the temporal, spatial, and frequency domain dependencies of traffic flow, thereby proposing a novel model called the frequency-enhanced spatiotemporal graph convolutional network (FeSTGCN) for enhanced accuracy and interpretability in traffic flow prediction. Specifically, this study devises an approach that utilises a time–frequency transformation method to extract frequency-domain information from traffic flow. Spatiotemporal domain dependencies were captured using an attention-based diffusion graph and temporal convolutions. Extensive experiments were conducted on a real-world road network using adaptive signal timing. The results demonstrate that the FeSTGCN is highly competitive compared with state-of-the-art models. Furthermore, the FeSTGCN exhibits excellent interpretability as frequency information provides novel insights into the composition and intrinsic patterns of traffic flow.