Page 122 - Advances in Renewable Energies and Power Technologies

6. Forecasting Methods for Different Forecast Horizons 95

• Time seriesebased methods: this set holds for approaches based on statistical
models solely ground on past measurements;
• NWP-based methods: this set holds for approaches based on weather forecasts
provided by a specialized provider (“NWP” is the acronym for “Numerical
Weather Forecast”);
• Satellite imageryebased methods: methods based on images of the earth taken
by satellite;
• Sky imagesebased methods: based on observations of cloud cover from the
ground with an in situ camera.
Strictly speaking, the different categories are not rival and each method is effi-
cient within the range of its own time horizon and even sometimes they can be
used together in hybrid methods.
Power prediction of PV systems usually involves several modeling steps to
obtain the required forecast information from different kinds of input data [58].
Forecasts may apply to a single PV system or to an aggregation of systems spread
over an extended geographic area. The aggregate effect, repartition of ISRES over a
large territory, allows to average and to smooth their production; thus, it is easier to
predict the production of a country than the production of a region, just as it is easier
to predict a production of a region than that of a single PV plant.
Forecasts may focus on the output power of systems or on its rate of change
(ramp rate). Forecasting methods also depend on the tools and information available
to forecasters, such as data from weather stations and satellites, PV system data, and
outputs from NWP models [59] (Fig. 3.9).
Directly forecasting the PV power from an historical PV production data series
can be more complicated than predicting solar radiation incident in the PV plant for
several reasons:
• in various models as statistical and artificial intelligence ones, a large data set is
necessary for the training, which implies that the PV plant has to be installed for
a very long time before being able to train the model, which is a rare occurrence;
• the installed power of the PV plant can be increased, some malfunctions may
occur (or only a part of the PV plant runs); thus, the historical data set is not
more reliable nor homogenous [60].
The main variables influencing PVoutput power are the irradiance in the plane of
the PV array (Global one for nonconcentrating collector and normal beam for
concentrated collector), the temperature at the back of the PV modules, and to a
lesser extent, the wind speed at the PV modules level.
The PV power forecast is obtained by introducing the irradiance forecast into a
PV simulation model. Generally, two models are used in this step: One to calculate
the direct current power output and another for modeling the inverter characteristics.
Such models are easily available with more or less complexity, but even for simple
models, their accuracy is higher than the accuracy of the irradiance forecast.