Electricity load forecasting under extreme heat and cold waves

Extreme heat and cold waves can push electricity demand beyond the conditions for which power systems are routinely planned, increasing the risk of shortages and blackouts. Accurate load forecasts are therefore critical because they give operators the lead time needed to keep supply and demand in balance. Because such events are rare, forecasting models are trained mostly on ordinary weather conditions and often fail when demand patterns shift during extremes. Here we show that Extreme Synthesis-Disentanglement Forecasting, a data-centered approach that generates realistic examples of extreme weather and filters them to retain reliable demand signals, improves electricity load forecasting under heat and cold waves. Across 80 real-world datasets from six regions, this approach reduces two standard measures of forecasting error by 9.8–33.5% and 7.5–34.5%, respectively, compared with leading models. These improvements offer a practical way to support more reliable power-system operation as climate extremes intensify. A data-focused forecasting framework improves electricity demand forecasting during heatwaves and cold waves, showing that better use of scarce data can help power systems manage extreme-weather risks.