@PhDThesis{Anochi:2015:PrClPr,
author = "Anochi, Juliana Aparecida",
title = "Previs{\~a}o clim{\'a}tica de precipita{\c{c}}{\~a}o por redes
neurais autoconfiguradas",
school = "Instituto Nacional de Pesquisas Espaciais (INPE)",
year = "2015",
address = "S{\~a}o Jos{\'e} dos Campos",
month = "2015-11-03",
keywords = "problemas de otimiza{\c{c}}{\~a}o, meta-heur{\'{\i}}stica,
rede neural artificial, previs{\~a}o clim{\'a}tica,
redu{\c{c}}{\~a}o de dados, optimization, meta-heuristic,
artificial neural networks, climate prediction, data reduction.",
abstract = "Previs{\~a}o clim{\'a}tica do campo de precipita{\c{c}}{\~a}o
{\'e} um aspecto chave em meteorologia. Precipita{\c{c}}{\~a}o
{\'e} uma vari{\'a}vel associada a desastres naturais (secas e
enchentes) e safras agr{\'{\i}}colas, com impactos nos setores
de turismo e transporte. Entretanto esta vari{\'a}vel
meteorol{\'o}gica {\'e} de dif{\'{\i}}cil previs{\~a}o,
devido {\`a} grande variabilidade temporal e espacial
(vari{\'a}vel descont{\'{\i}}nua). Neste trabalho, um
m{\'e}todo baseado em Rede Neural Artificial (RNA) {\'e}
aplicado para previs{\~a}o clim{\'a}tica de
precipita{\c{c}}{\~a}o nas regi{\~o}es Sul, Sudeste e Nordeste
do Brasil. {\'E} conhecida a capacidade de redes neurais de
aprendizado e resposta, o que motiva sua aplica{\c{c}}{\~a}o com
sucesso em uma grande variedade de problemas, consolidando-se como
uma t{\'e}cnica de solu{\c{c}}{\~a}o de problemas complexos em
reconhecimento de padr{\~o}es, classifica{\c{c}}{\~a}o,
sistemas de controle, aproxima{\c{c}}{\~a}o de
fun{\c{c}}{\~o}es e modelo preditivo. Redes neurais podem ser
caracterizadas como redes supervisionadas e n{\~a}o
supervisionadas. Em geral, o processo de treinamento de redes
neurais supervisionadas est{\'a} associado {\`a}
determina{\c{c}}{\~a}o dos pesos das conex{\~o}es. A
defini{\c{c}}{\~a}o ou identifica{\c{c}}{\~a}o da arquitetura
{\'o}tima para uma rede neural {\'e} expressa como um problema
de otimiza{\c{c}}{\~a}o, em que cada ponto no espa{\c{c}}o de
busca representa uma topologia diferente. O problema de
otimiza{\c{c}}{\~a}o pode ser formulado por meio de uma
fun{\c{c}}{\~a}o mono-objetivo ou de uma fun{\c{c}}{\~a}o
multiobjetivo. Neste trabalho, a otimiza{\c{c}}{\~a}o
mono-objetivo foi solucionada pelo \emph{Multi-Particle Collision
Algorithm} (MPCA) e o \emph{Non-dominated Sorting Genetic
Algorithm} II (NSGA-II) foi empregado para otimiza{\c{c}}{\~a}o
multiobjetivo. Em meteorologia, dados de diversas fontes
(sat{\'e}lites, esta{\c{c}}{\~o}es de superf{\'{\i}}cie,
boias oce{\^a}nicas, radiossondagens, radar e muitas outras)
s{\~a}o usados nas previs{\~o}es de tempo e clima. Assim,
previs{\~a}o de eventos meteorol{\'o}gicos {\'e} um desafio
complexo, mais ainda deve-se incluir a necessidade de an{\'a}lise
de grande volume de dados. A redu{\c{c}}{\~a}o da dimens{\~a}o
de dados de observa{\c{c}}{\~a}o sem perda de
informa{\c{c}}{\~a}o {\'e} um tema importante de pesquisa. A
Teoria dos Conjuntos Aproximativos, uma t{\'e}cnica de
minera{\c{c}}{\~a}o de dados, foi empregada para identificar as
vari{\'a}veis mais significativas para o processo de
previs{\~a}o clim{\'a}tica. ABSTRACT: Climate precipitation
prediction field is a key aspect in meteorology. The precipitation
is a variable associated with natural disasters (droughts and
floods) agricultural crops and can cause impacts in the sectors of
tourism and shipping. However, this is a meteorological variable
that is difficult to predict because of large spatial and temporal
variability (i.e. variable discontinuous). A method based on
Artificial Neural Network (ANN) is applied to climate prediction
precipitation in the South, Southeast and Northeast regions of
Brazil. It is known the ability of neural network learning and
response, which motivates their successful application in a wide
variety of problems, consolidating its position as a solution
technique of complex problems in pattern recognition,
classification, control systems, proximity functions and
predictive model. Neural networks can be characterized as
supervised and unsupervised networks. In general, the supervised
training process for neural networks is associated with the
determination of the weights of the connections. The definition or
identification of the optimal architecture for a neural network is
expressed as an optimization problem, in which each point in the
search space represents a different topology. The optimization
problem can be formulated by a mono-objective function or a
multiobjective function. The mono-objective optimization was
solved by Multi-Particle Collision Algorithm (MPCA) and
Non-dominated Sorting Genetic Algorithm-II (NSGA-II) was used for
multi-objective optimization. In meteorology, data from various
sources (satellites, ground-based stations, ocean buoys,
soundings, radar and many others) are used in weather and climate
forecasts. Predicting meteorological events is a complex
challenge. The size reduction of the observation data without
losing information is an important subject of research. The Rough
Sets Theory, a data mining technique was used to identify the most
significant variables for the climate prediction process.",
committee = "Guimar{\~a}es, Lamartine Nogueira Frutuoso (presidente) and
Velho, Haroldo Fraga de Campos (orientador) and Shiguemori, Elcio
Hideiti (orientador) and Sandri, Sandra Aparecida and Carvalho,
Solon Ven{\^a}ncio de and Luz, Eduardo F{\'a}vero Pacheco da and
Braga, Antonio de Padua",
copyholder = "SID/SCD",
englishtitle = "Climate precipitation prediction by self-configured neural
networks",
language = "pt",
pages = "159",
ibi = "8JMKD3MGP3W34P/3K98PDP",
url = "http://urlib.net/ibi/8JMKD3MGP3W34P/3K98PDP",
targetfile = "publicacao.pdf",
urlaccessdate = "03 maio 2024"
}