One year long-term electric load forecasting based on multiple regression models and Kalman filtering algorithm

This paper presents a new technique for one-year long-term electric power load forecasting problem. The technique is suitable to forecast daily load profiles with a lead-time from several weeks to a few years. The proposed algorithm is mainly based on multiple simple linear regression models used to capture the shape of the load over a certain period of time (one year), in a two-dimensional layout (24 hours x 52 weeks). The regression models are then recursively used to project the 2D load shape for the next period of time (next year). Load demand annual growth is estimated and incorporated in Kalman filtering algorithm to improve the load forecast accuracy obtained, so far, from the regression models. The results show a one-year load prediction with mean average percentage error less than 2.3% of the actual demand load and a standard deviation of 4.6 MW. The algorithm using historic data for different electric utilities may produce different accuracies. But, in general, for long-term load forecasting, an error level up to 10% is acceptable.