Methods of load forecasting

Load forecasting is the prediction of future peak demand (kW) and energy (kWh) of a system or area, needed for planning generation, transmission and distribution. Methods differ by time horizon and approach.

By time horizon

Main methods

1. Trend analysis (extrapolation): past load data are fitted with a curve (linear $P = a + bt$, exponential $P = a e^{bt}$, polynomial, Gompertz S-curve) and extended into the future. Simple and cheap; suitable for large areas with smooth growth. It ignores causes of growth and fails when conditions change.

2. Econometric (regression) method: load is expressed as a function of factors such as GDP, population, per-capita income, electricity price and number of consumers, e.g. $E = a + b\,(GDP) + c\,(Pop)$. Coefficients come from regression on past data. Explains causes but needs forecasts of the economic variables.

3. End-use method: total demand = Σ (number of appliances/consumers × use per appliance × hours). Built up from consumer classes (domestic, commercial, industrial). Good for new areas and demand-side studies; the load-centre method used by NEA (consumer number × consumption × growth × coverage) is an end-use type method. Needs detailed survey data.

4. Time-series methods: autoregressive (AR), moving-average (MA), combined ARMA/ARIMA and exponential smoothing models of hourly/daily load. Mainly for short-term forecasting.

5. Simulation (land-use) method: for small-area/spatial forecasting. Land is divided into small cells; land-use change (residential, commercial, industrial) is simulated and converted to load using load density per land-use class. Gives where and when load will appear.

6. Judgmental / Delphi method: opinions of experts are collected and refined in rounds. Used when data are lacking.

7. Artificial intelligence methods: artificial neural networks, fuzzy logic and machine learning models trained on historical load and weather data; mostly for short-term forecasting.

In practice, a combination is used: trending/econometric for the national system, end-use and simulation for distribution areas.