Page 126 - Advances in Renewable Energies and Power Technologies

6. Forecasting Methods for Different Forecast Horizons 99

FIGURE 3.13
Short-term forecasting scheme using statistical methods on satellite imagery.
Reprinted from Heinemann D., Lorenz E., Girodo M., 2006. Forecasting of solar radiation. Solar energy resource
management for electricity generation from local level to global scale, with permission from Nova Science
Publishers, Inc.

As seen in Fig. 3.13, a dimensionless cloud index is calculated for each pixel and
a relationship between this cloud index and the ratio of global irradiance to clear-sky
irradiance is established. The basis of the forecasting methods relies upon the deter-
mination of the cloud structures during the previous recorded time steps. Extrapola-
tion of their motion gives rise to cloud position forecasts (considering the other
parameters constant during the time step) and, as a consequence, to the local radia-
tion situation. This method has the advantage to produce a spatial analysis of an area
within certain resolution capabilities. Several deriving methods are used for motion
vectors [79e81]. The general method is well described in Fig. 3.14 [82].
Perez et al. [83] have compared the results with forecasts based on NWP method
and according to their study, they set at approximately 6h the time horizon where
NWP-based methods become more efficient than satellite images-based ones.
They used infra-red images for determining the possible presence of snow on the
ground [84]. Marquez et al. [85] described a hybrid method coupling a time seriese
based (neural network) and satellite imagesebased methods.

6.2.3 Stochastic Learning Methods or Time SerieseBased Method
A traditional method for forecasting solar radiation is based on the time series of so-
lar energy and sometimes of other meteorological parameters. Autoregressive
models (AR) [86], Moving average (MA), Autoregressive Moving Average
(ARMA), or Markov chain models are frequently used; more recently, artificial in-
telligence methods were developed such as ANN, Fuzzy Logic, and other hybrid
methods [56,87].