Page 131 - Advances in Renewable Energies and Power Technologies

104 CHAPTER 3 Forecasting of Intermittent Solar Energy Resource

6.3 SOLAR FORECASTING FOR RANGES BETWEEN 6 H AND DAYS
AHEAD

For an efficient planning and operation of solar energy systems and more specifically
for a good management of the electrical grid, forecasts up to 48 h or even beyond
have to be provided. NWP models allow to calculate fields of meteorological data
as wind speed or radiation in the atmosphere. They use current meteorological
data as input in mathematical models based on equations of the atmosphere within
the goal to predict solar irradiance up to several days ahead [95,96].
The most used NWP models are the global model of the European Center for
Medium-Range Weather Forecasts (ECMWF) and the Global Forecast System model
of the National Centers for Environmental Prediction (NCEP) [10]. Because of the
complexity of the models, only 14 NWPs are currently operating worldwide [59].
Generally, these models give a forecasting from two to four times a day; the res-
olution is relatively low because of the limitation of the computing time (in the range
of 40e90 km). A mesoscale model covers a limited geographical area with a higher
resolution and a best consideration of the local meteorological phenomena. The
initial conditions of the mesoscale model are extracted from a global NWP.
Forecast improvements are often performed by comparing outputs to measured
data during a training period which is used to develop corrected forecasts. This
type of approach is often referred to as Model Output Statistics (MOS) [59].
MOS allows to interpret the outputs of the NWP and to produce site-specific fore-
casts; it is used to reduce the error of forecasting.
Perez et al. [97] found that the National Digital Forecast Database (NDFD) [a
derivative of the operational NWP models of the NCEP] had for a 8- to 26-h forecast
horizons an hourly average relative RMSE of 38%, after the application of a local
correction function.
Remund et al. [96] evaluated different NWP forecasts for global solar irradia-
tions in the United States; their RMSE values ranged between 20% and 40% for a
24-h forecast horizon. Similar results were reported by Perez et al. [83], evaluating
NWP-based irradiance forecasts in several places of the United States [98], and they
realized a similar evaluation for NWP-based forecasts in Europe and found nRMSE
(normalized Root Mean Square Error) values of about 40% for Central Europe and
about 30% for Spain.

6.4 RELIABILITY AND ACCURACY OF THE FORECASTING MODELS
Comparing prediction models is a difficult task also because the meteorological con-
ditions are often very different from a location to another. To overcome this problem,
a benchmarking exercise was organized within the framework of “WIRE” with the
purpose of evaluating the performance of state-of-the-art models for short-term
renewable energy forecasting. The exercise consisted in predicting the power output
of two wind farms and two PV power plants, to compare the merits of forecasts
based on different modeling approaches and input data [57,58]. As an example,