Grid graph convolutional network with neighborhood learning for spatio-temporal prediction

This paper addresses the dual challenges of low accuracy and slow speed in spatio-temporal prediction by proposing a Grid Graph Convolutional Network with Neighborhood Learning (GN-GCN). Leveraging the GeoSOT-4D global grid system for discrete spatiotemporal encoding, the model constructs grid-based knowledge graphs and integrates static graph neural networks, neighborhood grid computation, and temporal evolution units to jointly capture semantic, spatial, and temporal dependencies. Enhanced by a High-level Training and Low-level Testing (HTLT) strategy, GN-GCN achieves state-of-the-art performance in various spatio-temporal tasks, significantly outperforming conventional methods in both accuracy and computational efficiency for complex real-world scenarios.