Long-term/mid-term electric load forecasting based on short-term correlation and annual growth

This paper presents a novel approach for long-term/mid-term electric power load forecasting. The strong short-term correlations of daily (24 h) and yearly (52 weeks) load behavior are implemented to predict future load demand. The algorithm is suitable for forecasting weekly average load profiles for 24 h of a day with a lead-time from several weeks to a few years. It successively incorporates alternating daily and weekly simple (1st order) linear regression models of previous (1 year) data augmented with annual load growth to predict future load demand. The results demonstrate successful (one year ahead) load forecast with a mean absolute error of less than 3.8% and with a standard deviation of less than 4.2, which prove superior to other techniques published earlier in the literature.