-.--------------------------.

Then, to calculate the 3 quarters within each couple of hourly forecasted values, the following auto-regressive model was initially adopted:

F(k + n) = A(n, k) . F(k) + B(n, k) . F(k + 1) being: k = 0, ... 24; n = 1/4, 1/2,3/4

The A(n, k), B(n, k) coefficients are different for each quarter and are calculated with the minimum squared method for:

• Each of the 24 couples of hourly values • Each kind of 12 daily courses representing typical days of the 4 sets: weekdays, days before

holidays, holidays, and days between holidays • The 3 periods of summer, winter, and fall/spring together

This method gave good results and proved to be robust; however, the static classification of curves in the 12 classes proved unable to represent all the possible load demand characteristics. A possible improvement goes through a larger number of classes to be defined automatically. The uncertainty of how to find the most effective number of classes led to an alternative solution which translates