Before we continue, its necessary to introduce the benchmark
method for WSF, the Persistence model. PM states that due the high
autocorrelation underlying WS behavior, any wind speed future
value is equal to its last known value [28]. Despite its simplicity, PM
achieves very good results in the WSF problem and is used to
compare the quality of new forecasting approaches [2,29]. Typically,
PM predicts a future WS value as bxtþh ¼ xt, where bx stands for
the forecasted value, t for the current time step, and h for the
forecasting horizon. In the case where the day-ahead forecasting is
required, a persistence method called Day-to-Day (D2D) is used
[28]. D2D method forecast a future value as
bxðdþ1;hÞ ¼ xðd;hÞ; h ¼ 1; …; 24 , where d stands for the current day.