Soft computing techniques for short term load forecasting

Short term load forecasting is a recurrent topic in the operative planning activities of companies dedicated to the distribution and trade of energy around the world due to the competitive electricity market, in which an advantage in the previous knowledge of demand could mean the difference between obtaining big benefits or incur in economic losses. In this paper a novel method for short term load forecasting is proposed based on the similar day approach and the use of soft computing techniques. This approach is founded on the search for the most similar day in history, to the forecasted day, based on the explanatory meteorological variables forecast for the load of this day. Once the "similar" day is found, the load forecast will be the same of that day with an adjustment for load growth. LAMDA-fuzzy-clustering techniques, regression trees, CART classification and fuzzy inference for the peak power, daily energy and load curve forecast are used. The validation of the proposed method is made with meteorological and load data from a Colombian city.