@PhDThesis{Lima:2015:PrIrSo,
author = "Lima, Francisco Jos{\'e} Lopes de",
title = "Previs{\~a}o de irradia{\c{c}}{\~a}o solar no nordeste do
Brasil empregando o modelo WRF ajustado por redes neurais
artificiais (RNAs)",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2015",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2015-06-10",
keywords = "radia{\c{c}}{\~a}o solar, redes neurais artificiais,
previs{\~a}o, an{\'a}lise de agrupamento, nordeste do Brasil,
solar radiation, artificial neural networks, forecast, cluster
analysis, northeast of Brazil.",
abstract = "Este trabalho tem como objetivo avaliar a capacidade de um modelo
num{\'e}rico regional de mesoescala, em representar o escoamento
atmosf{\'e}rico da regi{\~a}o Nordeste do Brasil (NEB),
possibilitando seu uso para previs{\~a}o de
irradia{\c{c}}{\~a}o solar usando um refinamento
estat{\'{\i}}stico para a redu{\c{c}}{\~a}o dos erros
sistem{\'a}ticos inerente ao modelo de mesoescala. A
motiva{\c{c}}{\~a}o deste estudo decorre da import{\^a}ncia da
radia{\c{c}}{\~a}o solar como recurso vital para a
manuten{\c{c}}{\~a}o da vida no planeta e para atividades
humanas tais como agricultura e aproveitamento de energia. A
intensidade da irradia{\c{c}}{\~a}o solar que incide sobre a
superf{\'{\i}}cie {\'e} de natureza vari{\'a}vel,
principalmente devido {\`a}s nuvens e o ciclo diurno. Este
trabalho se prop{\^o}s a desenvolver uma metodologia para
previs{\~a}o de irradia{\c{c}}{\~a}o solar incidente para a
regi{\~a}o Nordeste do Brasil com o uso de Redes Neurais
Artificiais (RNAs), alimentadas por sa{\'{\i}}das do modelo WRF,
visando reduzir as incertezas associadas {\`a} previs{\~a}o de
irradia{\c{c}}{\~a}o solar deste modelo. As vari{\'a}veis de
sa{\'{\i}}da do modelo WRF, representando as
condi{\c{c}}{\~o}es atmosf{\'e}ricas previstas, foram
empregadas como preditores em modelos de RNAs e Regress{\~o}es
Lineares M{\'u}ltiplas (RLM). O m{\'e}todo de an{\'a}lise de
cluster foi utilizado para estabelecer regi{\~o}es de
caracter{\'{\i}}sticas homog{\^e}neas sob o ponto de vista
climatol{\'o}gico da irradia{\c{c}}{\~a}o solar. Os dados
usados neste trabalho foram dados do INMET, para o
per{\'{\i}}odo de sete anos de 2005 a 2011. Diversos
experimentos foram realizados para ajuste e defini{\c{c}}{\~a}o
de preditores e simula{\c{c}}{\~a}o dos modelos de RNAs.
Par{\^a}metros de avalia{\c{c}}{\~a}o de erros, determinados
frente aos dados observacionais de cada esta{\c{c}}{\~a}o de
coleta de dados em superf{\'{\i}}cie foram calculados,
permitindo a compara{\c{c}}{\~a}o de desempenho das RNA e RLM e
da previs{\~a}o de irradia{\c{c}}{\~a}o solar obtida
diretamente do modelo WRF. Visando maximizar o ganho de desempenho
sobre o modelo WRF e minimizar o n{\'u}mero de vari{\'a}veis,
encontrou-se a melhor arquitetura e um grupo de 10 preditores, com
o qual an{\'a}lises mais aprofundadas foram realizadas, incluindo
avalia{\c{c}}{\~a}o de desempenho para o outono e primavera de
2011, per{\'{\i}}odo chuvoso e seco no NEB, principalmente no
norte do NEB. Houve uma diferen{\c{c}}a significativa entre os
modelos de RNA e RLM, mostrando que os modelos de RNAs foram
superiores ao modelo RLM. Por{\'e}m ambos os m{\'e}todos
promoveram redu{\c{c}}{\~a}o do vi{\'e}s e do \emph{RMSE} e
aumento do coeficiente de correla{\c{c}}{\~a}o em
compara{\c{c}}{\~a}o com as sa{\'{\i}}das de
irradia{\c{c}}{\~a}o solar do WRF. ABSTRACT: This work aims to
evaluate the ability of a regional numerical mesoscale model, to
represent the atmospheric flow in the Northeast region of Brazil
(NEB), allowing its use for forecasting solar irradiation using a
statistical refinement to reduce systematic errors inherent in the
mesoscale model. The motivation of this study lies in the
importance of solar radiation as a vital resource for the
maintenance of life on Earth and to human activities such as
agriculture and energy. The intensity of the solar radiation
incident on the surface is variable in nature, mainly because of
clouds and the diurnal cycle. This study aimed to develop a
methodology to forecast the incident surface solar irradiation in
the Northeast of Brazil by using mesoscale WRF model outputs
adjusted by Artificial Neural Networks (ANN to reduce the model
uncertainties. The output variables of the WRF model, representing
the forecast atmospheric conditions, were used as predictors by
RNAs and Multiple Linear Regressions (MLR), (with the inclusion of
a clearsky model), adjusted to calculate the incident solar
irradiation, in four homogeneous regions defined by the Wards
method. The data used in this study cover the period of seven
years from 2005 to 2011. Several predictors were tested in the
adjustment and simulation of the ANN. Error evaluation parameters,
determined by the observational data of each measurement station
were calculated for each simulation, allowing the comparison of
RNA and RLM, and the prediction of solar irradiation directly from
WRF model. To maximize the performance gain of the WRF model and
minimize the number of variables, it was establish the best
architecture and a group of 10 predictors, with which more
in-depth analyzes were performed, including performance evaluation
for fall and spring of 2011 (rainy and dry season at the NEB,
mainly in North of the Northeast). There was a significant
difference between RNA and MLR models, showing that RNA models
were superior to the MLR model. However, both methods produced
lower bias and RMSE, and an increase in the correlation
coefficient in comparison with the solar radiation in the WRF
Model.",
committee = "Ceballos, Juan Carlos (presidente) and Pereira, Enio Bueno
(orientador) and Sansigolo, Cl{\'o}vis Angeli and Martins,
Fernando Ramos and Ruther, Ricardo",
copyholder = "SID/SCD",
englishtitle = "Forecast of solar irradiation in northeast of Brazil using the WRF
model adjusted by artificial neural network (ANN)",
language = "pt",
pages = "250",
ibi = "8JMKD3MGP8W/3JH3BSE",
url = "http://urlib.net/ibi/8JMKD3MGP8W/3JH3BSE",
targetfile = "publicacao.pdf",
urlaccessdate = "27 abr. 2024"
}