Skip to Main Content
This paper presents a regression-based adaptive weather sensitive short-term load-forecasting algorithm, which has been developed and implemented in Electric Power Utility of Serbia. The proposed methodology consists of two main steps. The total daily energy is independently forecasted in the first step while hourly loads are predicted in the second step. All model parameters are automatically calculated and updated using realized data in the identification period. This period is a set of patterns from the database with weather and load conditions similar to conditions, which are expected, in the forecasted day. The Euclidean distance is used as a measure of similarity. The programming package based on the methodology presented in the paper has been used in the EPS to forecast 24-h loads one to seven days ahead since 1991. Results obtained in exhausted case studies are presented, analyzed, and compared with other results reported in literature.