A Weather Foundation Model for the Power Grid

2025-09-28

AI Summary

While existing weather foundation models excel in global forecasting, their capability for ultra-local, asset-level prediction of electricity infrastructure–critical meteorological variables—such as surface temperature, precipitation, hub-height wind speed, wind turbine icing risk, and overhead conductor rime ice accumulation—remains unexplored. Method: This work pioneers the domain adaptation of a 1.5-billion-parameter generative forecasting Transformer (GFT) to the power sector, performing post-training and fine-tuning using data from transmission-line weather stations, wind farm met masts, and icing sensors. Contribution/Results: For 6–72-hour forecasts, the adapted model reduces RMSE by 15% (temperature), 35% (precipitation), and 15% (wind speed); achieves a mean precision of 0.72 for rime ice accumulation detection; and delivers several hours of actionable lead time. By transcending the spatial resolution and physical parameterization constraints of traditional numerical weather prediction, this approach establishes a novel operational paradigm for power grid hazard early warning.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Application Domains: Internet of Things, Sensor Networks & Smart Cities
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

Weather foundation models (WFMs) have recently set new benchmarks in global forecast skill, yet their concrete value for the weather-sensitive infrastructure that powers modern society remains largely unexplored. In this study, we fine-tune Silurian AI's 1.5B-parameter WFM, Generative Forecasting Transformer (GFT), on a rich archive of Hydro-Québec asset observations--including transmission-line weather stations, wind-farm met-mast streams, and icing sensors--to deliver hyper-local, asset-level forecasts for five grid-critical variables: surface temperature, precipitation, hub-height wind speed, wind-turbine icing risk, and rime-ice accretion on overhead conductors. Across 6-72 h lead times, the tailored model surpasses state-of-the-art NWP benchmarks, trimming temperature mean absolute error (MAE) by 15%, total-precipitation MAE by 35%, and lowering wind speed MAE by 15%. Most importantly, it attains an average precision score of 0.72 for day-ahead rime-ice detection, a capability absent from existing operational systems, which affords several hours of actionable warning for potentially catastrophic outage events. These results show that WFMs, when post-trained with small amounts of high-fidelity, can serve as a practical foundation for next-generation grid-resilience intelligence.

Problem

• Fine-tuning weather models for hyper-local power grid forecasts
• Improving prediction accuracy for grid-critical weather variables
• Enabling early detection of rime-ice to prevent outages

Innovation

• Fine-tuned weather foundation model on power grid data
• Achieved hyper-local forecasts for critical grid variables
• Outperformed existing benchmarks with improved accuracy metrics