PatchSTG: Scalable Spatiotemporal Graph Transformers for Traffic Forecasting on Irregular Sensor Networks

PatchSTG is a novel patch-based spatiotemporal graph Transformer designed for efficient traffic forecasting on irregular sensor networks. It introduces a hierarchical spatial representation that organizes sensors into balanced patches, employing a dual attention encoder to reduce computational complexity from quadratic to near-linear scaling. This model demonstrates competitive forecasting performance on real-world traffic datasets while enhancing computational efficiency, making it a significant advancement for practitioners dealing with irregular spatial data in intelligent transportation systems.