How it works
Short-term load forecasting (hours to days) uses extrapolation from historical load curves, adjusted for weather and day-of-week patterns. Medium-term forecasts (weeks to months) apply regression models correlating load with GDP, population, and temperature. Long-term forecasts (years to decades) use trend extrapolation on semi-log plots, correlation methods, or end-use analysis that counts appliances and industrial loads separately. The exponential smoothing formula Ft+1 = α·Dt + (1−α)·Ft (where α is the smoothing constant between 0 and 1) is widely used for short-term work. Regression: P = a + bT + cT², where T is time and coefficients a, b, c are found by least squares.
Key points to remember
Load forecasting accuracy directly determines the reserve margin planned — a typical utility targets 15–20% spinning reserve above forecast peak demand. Short-term forecasting errors of less than 2% are achievable with modern AI tools, but traditional regression methods give 3–5% error. The load duration curve, derived from the load forecast, is the primary tool for planning base, intermediate, and peaking plant capacity. Annual load growth in India has historically been 6–8%, though this varies by state. Reactive load forecasting is equally important because it determines the sizing of shunt capacitor banks at 33 kV and 11 kV buses.
Exam tip
The examiner always asks you to sketch the load duration curve from a given daily load data table and identify the area under it as energy consumed — practice converting load vs time data into a duration curve step by step.