Adaptive short-term load forecasting using artificial neural networks

Abstract

A multi-layer artificial neural network (ANN) with an adaptive learning algorithm is used to forecast system hourly loads up to 168 hours for the Public Utilities Board (PUB) of Singapore. The ANN-based load models are trained using hourly historical load data and daily historical maximum/minimum temperature data supplied by the PUB and Meteorological Service Singapore respectively. The models are trained by day types to predict daily peak and valley loads. The hourly forecast loads are computed from the predicted peak and valley loads and average normalized loads for each day type. The average absolute error for a 24-hour ahead forecast using the actual load and temperature data is shown to be 2.32% for Mondays through Sundays and 5.98% for ten special day types in a year.< >