The percentiles are based on the historical forecast
accuracy performance for each individual series and are extracted from
conditional error distributions. All percentiles will be depending on the
prediction horizon, since the further ahead in the future, the uncertainty will
increase. Otherwise, the calculation of the percentiles are depending on the
data type property of the forecast series. This property is set in the
config-tab in the Forecast series table.
•Load Data – For series representing consumption, the percentiles will depend on both hour of the day as well as if it is weekday or weekend.
•Wind Data, Hydro Data – For these data types, the percentiles will depend on the forecast values, having skewed error distributions close to zero as well as close to installed capacity, for wind as well also having an increased uncertainty where the power curves are steepest.
•Solar Data – The uncertainty for PV-production should of course vanish during nights. Similarly as for wind and hydro, the percentiles will depend on the current fraction of available global radiation compared to the maximum global radiation (assuming a clear sky). Also, the percentiles are scaled by the maximum daily radiation, thus enabling comparisons between seasons.