Load Forecasting under extreme climatic conditions

Abstract

An artificial ricu~al nctwoik is applied to forccast tiic short tcrin load unclcr cxtrciiic clinintic conditions. Historical (lata collcctcd over a period of 3 ycars (e.g. calendar years 1990, 1991, and 1992) is used for lraining and tesling the proposed A" network. Based on tile kiiowri differenccs among the load responses for thc days of tlic week, aseparatc ANNis uscd forcach day ofthe week: scvcti ANN'S in all. In ttic forecasting stagc, the ANN nctwork is supplied wilh only the input data Ior thc Iorccastcd day and tlie nctwork presents a 24 liour load forccast foi that (lay a t one time. Very accurate results have been obk1incd for all days of the week..