Research map / Time series / Long-term forecasting
Generalising Traffic Forecasting to Regions Without Traffic Observations
Xinyu Su, Majid Sarvi, Feng Liu, Egemen Tanin, Jianzhong Qi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add physics (LWR) loss, weather encoder and spatial grouping together to the STSM-style backbone | GenCast · PEMS07 | RMSE: 8.390 -> 8.253 | Improved |
| remove physics constraint (w/o-phy) (ablation) | GenCast-L · PEMS07 and PEMS-Bay | RMSE | Improved |
| add physics constraint to GenCast-L with frozen LLM spatial embeddings (ablation) | GenCast-L · PEMS07 and PEMS-Bay | RMSE | Improved |
| add physics constraint to GenCast-H with trainable GeoHash embeddings (ablation) | GenCast-H · PEMS07 and PEMS-Bay | RMSE | No clear effect |
| add spatial grouping module with entropy loss (ablation) | GenCast · PEMS07 and PEMS-Bay | RMSE | Improved |
| add weather cross-attention encoder (ablation) | GenCast · PEMS07 and PEMS-Bay | RMSE | Improved |
| add spatial grouping entropy loss (ablation) | GenCast-L · PEMS07 | entropy of group assignment weights | Improved |
| use LLM-based spatial embeddings instead of GeoHash-based | GenCast · PEMS-Bay, METR-LA, Melbourne | RMSE: Melbourne RMSE 9.009 (H) vs 9.258 (L) | Hurt |
| use LLM-based spatial embeddings instead of GeoHash-based | GenCast · PEMS07 and PEMS08 | RMSE: PEMS07 8.285 -> 8.253 | Improved |
| use GenCast on ring split of PEMS-Bay | GenCast · PEMS-Bay (ring split) | R2: 0.189 -> 0.241 | Improved |
| apply GenCast without spatial embeddings or physics constraints to solar power data | GenCast-w/wx · NREL | RMSE: 7.733 -> 7.620 | Improved |
| add KITS virtual-node training instead of masking subgraph | KITS · PEMS07 | RMSE: 8.390 (STSM) vs 9.574 (KITS) | Hurt |
Closest papers by meaning
- FlashST: A Simple and Universal Prompt-Tuning Framework for Traffic Prediction (ICML 2024 · similarity 0.83)
- Efficient Traffic Prediction Through Spatio-Temporal Distillation (AAAI 2025 · similarity 0.83)
- WardropNet: Traffic Flow Predictions via Equilibrium-Augmented Learning (ICLR 2025 · similarity 0.82)
- Generalizing Weather Forecast to Fine-grained Temporal Scales via Physics-AI Hybrid Modeling (NeurIPS 2024 · similarity 0.82)
- Spatiotemporal-aware Trend-Seasonality Decomposition Network for Traffic Flow Forecasting (AAAI 2025 · similarity 0.82)
- Prompt to Transfer: Sim-to-Real Transfer for Traffic Signal Control with Prompt Learning (AAAI 2024 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
