Calibration improves reliability of wind power forecast uncertainty bands
The Cost of Coverage: Conformalising Native Forecast Uncertainty for GB Wind Power
Machine Learning
Summary
Wind power forecasts come with uncertainty bands that help decide how much backup power is needed, but these bands often don’t cover actual wind conditions well. The authors looked at a method called conformal prediction to adjust these uncertainty bands for better accuracy. They tested six forecasting methods at two wind farms and found that calibration fixes coverage problems overall without making predictions worse, though problems remain during very strong winds. When calibration is done separately for different wind conditions, the forecast bands become more reliable, especially for high winds, reducing unexpected power gaps.
What this means in practice
- •For wind farm operators: Improve day-ahead power scheduling by using calibrated uncertainty bands tailored to wind regimes for more reliable reserve sizing.
- •For energy grid planners: Enhance grid stability by incorporating regime-specific calibrated forecast intervals to better anticipate fluctuations in wind power supply.
Authors
Carl Somers
Abstract
Uncertainty bands on wind forecasts help inform reserve sizing and commitments, but often fail to deliver their stated coverage. We apply conformal prediction to native bands to see what it adds and what it costs, comparing six forecasters, from ARIMA to a tabular foundation model, at two wind farms in Great Britain (GB) under operational conditions. We show that calibration repairs empirical marginal coverage with little or no Winkler Score penalty, but this apparent success conceals under-coverage in high winds. Calibrating by forecast wind regime turns marginal coverage into regime-level coverage, cutting day-ahead out-of-band volume in high winds by over 40% relative to pooled calibration while narrowing bands overall.