UNSUPERVISED SUPERVISED LEARNING CONCEPT FOR 24-HOUR LOAD FORECASTING

An application of artificial neural networks in short-term load foreca sting is described. An algorithm using an unsupervised/supervised lear ning concept and historical relationship between the load and temperat ure for a given season, day type and hour of the day to forecast hourl y electric load with a lead time of 24 hours is proposed. An additiona l approach using functional link net, temperature variables, average l oad and last one-hour load of previous day is introduced and compared with the ANN model with one hidden layer load forecast. In spite of li mited available weather variables (maximum, minimum and average temper ature for the day) quite acceptable results have been achieved. The 24 -hour-ahead forecast errors (absolute average) ranged from 2.78% for S aturdays and 3.12% for working days to 3.54% for Sundays.