Load forecasting

Load forecasting predicts future electricity demand on a grid, region, or individual feeder, over horizons ranging from the next few minutes to years ahead. Utilities and grid operators rely on it to plan generation dispatch, buy or sell power in wholesale markets, schedule maintenance, and plan long-term infrastructure investment.

Forecasts are typically split by horizon, each with different drivers and methods: short-term load forecasting (hours to a few days ahead) leans heavily on weather, especially temperature, and calendar effects like weekdays and holidays; long-term forecasting (years ahead) incorporates economic growth, electrification trends such as EV adoption, and population change. Modern forecasting combines classical time series forecasting with machine-learning models trained on granular consumption data from advanced metering infrastructure, which gives far more resolution than the aggregated feeder-level data utilities relied on previously. Load forecasting is a specific application of the more general practice of demand forecasting, adapted to electricity's constraint that supply and demand must balance in near real time.

Accuracy matters directly to cost and reliability: underforecasting risks shortfalls or expensive last-minute power purchases, while overforecasting means paying for generation capacity that goes unused. Load forecasts feed grid analytics used for real-time balancing, and a common pitfall is a model that performs well on typical days but fails during extreme weather, precisely when accurate forecasting matters most.