This talk explores recent advances in graph-based deep learning for power systems, which model networks as spatiotemporal graphs to capture both spatial structure and measurement dynamics.