Adaptive Weather-Sensitive Short Term Load Forecast

This paper introduces an adaptive, weather sensitive, short term load forecast algorithm that has been developed for two South Carolina Power Systems: CEPCI (Central Electric Power Cooperatives, Inc., Central for short) and Combined System. The model is based on a State Space formulation specially tailored for this application. A detailed correlation study is performed to identify the most relevant weather variables. Different models are used for Summer and Winter, since different weather variables are found to be relevant in both seasons. Adaptivity is attained through careful usage of Kalman filtering and Bayesian techniques. An appropriate methodology is developed to identify and correct anomalous load data and to model nonconforming loads. A new technique is introduced for "filling in" weather forecasts. The model has been sucessfully implemented using state-of-the-art data-base and man-machine techniques. Implementation results are reported. This model benefits from the experience gained using a variety of tools and represents important enhancements over existing methods.