@InProceedings{LimaCostMartPere:2015:AvPrRa,
author = "Lima, Francisco Jos{\'e} Lopes de and Costa, Rodrigo Santos and
Martins, Fernando R. and Pereira, Enio Bueno",
affiliation = "{Instituto Nacional de Pesquisas Espaciais (INPE)} and {Instituto
Nacional de Pesquisas Espaciais (INPE)} and {Universidade Federal
de S{\~a}o Paulo (UNIFESP)} and {Instituto Nacional de Pesquisas
Espaciais (INPE)}",
title = "Avalia{\c{c}}{\~a}o da previs{\~a}o de radia{\c{c}}{\~a}o com
base em ferramentas de p{\'o}s-processamento aplicadas em
simula{\c{c}}{\~o}es do modelo de mesoescala WRF",
booktitle = "P{\^o}steres",
year = "2015",
organization = "Simp{\'o}sio Internacional de Climatologia, 6. (SIC)",
keywords = "Previs{\~a}o de radia{\c{c}}{\~a}o solar, WRF, Redes Neurais
Artificiais, Regress{\~a}o Linear M{\'u}ltipla, Solar radiation
forecast, WRF model, Artificial Neural Networks, Multiple Linear
Regression.",
abstract = "A previs{\~a}o de curto prazo da radia{\c{c}}{\~a}o solar
incidente {\'e} uma quest{\~a}o importante para as
aplica{\c{c}}{\~o}es deste recurso como fonte de energia. O uso
de modelos num{\'e}ricos de mesoescala, combinados com
ferramentas estat{\'{\i}}sticas de p{\'o}s-processamento podem
aumentar a acur{\'a}cia das simula{\c{c}}{\~o}es de algumas
horas ou mesmo de alguns dias. Neste sentido, apresenta-se a
avalia{\c{c}}{\~a}o de um sistema de previs{\~a}o de
irradia{\c{c}}{\~a}o solar de curto prazo, com base no modelo
meteorol{\'o}gico de mesoescala WRF e em dois m{\'e}todos
estat{\'{\i}}sticos de p{\'o}sprocessamento, a fim de melhorar
o desempenho das estimativas. Foram avaliados resultados obtidos
em simula{\c{c}}{\~o}es do ano de 2009 sobre o Nordeste
Brasileiro (NEB) em dois per{\'{\i}}odos com
caracter{\'{\i}}sticas clim{\'a}ticas distintas na regi{\~a}o,
que s{\~a}o Outono e Primavera e, portanto considerando-se o
per{\'{\i}}odo chuvoso e o per{\'{\i}}odo seco na maior parte
da mesma. O modelo WRF foi integrado com um dom{\'{\i}}nio
externo de resolu{\c{c}}{\~a}o horizontal de 15 km, cobrindo
toda a regi{\~a}o Nordeste, e a partir da{\'{\i}} outros
tr{\^e}s dom{\'{\i}}nios de resolu{\c{c}}{\~a}o horizontal de
5 km foram aninhados. Os resultados das simula{\c{c}}{\~o}es
foram comparados com dados de 121 esta{\c{c}}{\~o}es
meteorol{\'o}gicas autom{\'a}ticas do Instituto Nacional de
Meteorologia (INMET), indicando que o modelo WRF superestima a
irradia{\c{c}}{\~a}o solar nos dois per{\'{\i}}odos simulados,
mas com menores diferen{\c{c}}as no Outono (a hip{\'o}tese
{\'e} a maior nebulosidade na regi{\~a}o). As t{\'e}cnicas de
p{\'o}sprocessamento estat{\'{\i}}stico utilizadas foram as
Redes Neurais Artificiais (RNA) e Regress{\~a}o Linear
M{\'u}ltipla (RLM), que permitiram melhorias significativas nos
resultados das simula{\c{c}}{\~o}es realizadas, verificados a
partir da redu{\c{c}}{\~a}o do BIAS e do RMSE. Dentre estas, as
RNA's tiveram desempenho superior {\`a}s RLM's. Estes resultados
permitem uma an{\'a}lise da confiabilidade de sistemas similares
de previs{\~a}o de irradi{\^a}ncia solar, em termos de sua
disponibilidade de curto prazo e da estimativa da
produ{\c{c}}{\~a}o de energia, indicando algumas melhorias que
podem ser avaliadas e consequentemente implementadas no futuro.
ABSTRACT: The short-term forecast of solar radiation is an
important issue for the applications of this feature as energy
source. The use of mesoscale numerical models combined with
statistical post-processing tools can increase the accuracy of a
few hours or even a few days simulations. In this way, this study
presents the evaluation of a short-term prediction of solar
irradiation system based on the WRF mesoscale meteorological model
with two statistical post-processing technics, in order to improve
the estimates performance. It was evaluated simulations from 2009
on the Brazilian Northeast region (NEB), in two periods with
different climatic characteristics, which are autumn and spring,
and in this way considering the rainy season and the dry season.
The WRF model was integrated with an external domain with
horizontal resolution of 15 km, covering the entire NEB, and from
there other three domains of horizontal resolution of 5 km were
nested. The simulations results were compared with data of 121
automatic weather stations of the National Institute of
Meteorology (INMET), indicating that the WRF model overestimates
the solar irradiation in the two simulated periods, but with minor
differences in autumn (the hypothesis is this case is the
cloudiness increment in region). The statistical post-processing
techniques used were as Artificial Neural Networks (ANN) and
Multiple Linear Regression (MLR), which allowed significant
improvements in the simulations results, indicated by the
reduction of BIAS and RMSE. However, the ANN's outperformed the
MLR's. These results allows an analysis of the reliability of
similar systems of solar irradiance forecast, in terms of his
short term availability and energy production estimates,
indicating some improvements that can be evaluated and
consequently implemented in the future.",
conference-location = "Natal, RN",
conference-year = "13-16 out.",
urlaccessdate = "27 abr. 2024"
}