STAGformer Boosts Micro Mobility Demand Forecasting with Agent Graph Transformer

Accurate forecasting of demand at individual stations is crucial for the efficient management of micro-mobility services like bike-sharing systems. This task is particularly challenging due to the intricate spatio-temporal relationships within large urban networks. Researchers have introduced STAGformer, a Spatio-Temporal Agent Graph Transformer, to address these complexities. This model is engineered for efficient global modeling, maintaining linear computational complexity, which is a significant improvement over the quadratic costs of standard self-attention mechanisms. STAGformer's core innovation lies in its two-step agent attention mechanism. A small set of learnable spatial and temporal agent tokens first aggregates global information, then broadcasts it back to individual stations and time steps. This effectively captures long-range interactions. The model also integrates modules for spatio-temporal encoding, graph propagation, and temporal convolution, demonstrating superior performance on real-world datasets like NYC Citi-Bike and Chicago Divvy-Bike.