Spatiotemporal Multi-Task Graph Transformer for Trip-Level Transit Prediction
Researchers have developed a new spatiotemporal multi-task graph transformer model called SMT-GraphFormer for predicting passenger counts on public transit. This model treats trip-level transit prediction as a sequence-to-sequence task, incorporating factors like weather and temporal information. Evaluations on data from Trondheim, Norway, demonstrated that SMT-GraphFormer outperforms existing stop-level benchmarks, particularly in predicting alightings. AI
IMPACT This new model offers improved accuracy for transit planning and operations by better capturing complex spatiotemporal dynamics.