OutageDiT is a generative foundation model for power‑outage forecasting and scenario simulation. Trained on nationwide outage and weather records, it produces seven‑day outage trajectories at 15‑minute resolution. The forecasting pipeline has two stages: a condition encoder consumes the historical outage series and known future weather covariates once, yielding horizon‑aligned hidden states; a shallow flow decoder reuses these states to sequentially sample a full trajectory. This architecture enables a single model to deliver point forecasts, probabilistic uncertainty quantification, and conditional event simulation. Benchmarks across several outage‑forecasting datasets show that OutageDiT improves accuracy and scenario quality over strong baselines and supports zero‑shot transfer to unseen regions. Consequently, conditional outage simulation bridges the gap from forecasting to operational planning under uncertainty.
Review