Discussion on 'Electrical load forecasting by exponential smoothing with covariates' by Rainer Gob, Kristina Lurz and Antonio Pievatolo

摘要

First, I would like to congratulate the authors on the idea and topic developed in this article. Accurately predicting electrical load is of vital importance for both power producers and electricity system operators (SOs). If the power producer makes an offer for the next day that he cannot generate, he receives a penalty from the SO and worse, the lack of electricity problem is typically solved by buying it from a foreign electricity market. In this article, the authors wisely used the hourly average temperature as a covariate. My experience with this kind of problem is that one of the weather variables that influence electricity demand in the Iberian Peninsula is the number of days the temperature falls below 0℃. However, this temperature threshold, used in our work, is contradictory to that given by these authors. They take 15℃ as the threshold temperature and concretely state '15℃ is the threshold, below which the electricity consumption is deemed unaffected by the temperature' (in Section 7 of the Manuscript). The choice made by the authors to establish this temperature as the threshold surprises me. In Spain, cold temperatures increase electricity consumption. A possible explanation for these contradictory thresholds is that perhaps, heating in the region of this study, Piedmont (Northern Italy), is primarily generated by coal, oil, or gas and not by electricity. Conversely, we found in our previously cited work that the most influential variables in predicting demand were related to macroeconomics, such as the Industrial Production Index and the interannual variation rate of the Consumer Price Index. I understand that the authors do not use these macroeconomic variables because, in general, they are obtained monthly and the authors work with hourly data. In any case, it is well known that power consumption depends on the financial state of the country. To have economic indicators, at least daily ones, could help obtain better predictions for the needed electrical load.