Short-term load forecasting: A power-regression approach

摘要

The short-term load forecasting problem is addressed by means of a power regression approach. Exploiting the highly correlated nature of the explanatory variables, just two loads are deemed sufficiently informative for prediction purposes: one day before and one week before. The notion of “similar day” is then used to extract a meaningful training set from the historical records. The presence of a significant trend throughout the years suggests that invariance should be rather searched across load ratios, an observation that motivates the use of logarithmically transformed load data, thus leading to power regression model. When tested against the Italian national consumption during 2011 and 2012, the new LIST-4 predictor performs better than Sibilla, the forecaster currently used by the Italian TSO.