| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 3.39 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
No presente trabalho é feito um estudo paramétrico de variações do método de Conjuntos
Análogos. O método, que é originalmente utilizado para a realização de pósprocessamento,
é aplicado neste estudo para realizar a previsão de séries temporais. Séries
temporais conhecidas são reconstruídas com o intuito de avaliar e identificar os parâmetros
ótimos que alcancem a minimização de erros obtidos da comparação entre as séries
temporais reais e as previstas. Ao todo foram utilizados nove anos de dados, sete anos
para a fase de treinamento e dois para a comparação com as previsões feitas. São utilizados
dados de três estações meteorológicas, onde o método é aplicado de forma que a
estação A tenha os seus dados previstos a partir das estações B e C.
O estudo demonstrou que é possível identificar valores e padrões de parâmetros que
são úteis para a minimização de erros. No entanto, também mostrou é difícil encontrar
parâmetros globais que minimizem os erros de qualquer variável em qualquer cenário
trabalhado. O estudo de desempenho realizado demonstrou que, na maioria dos casos,
utilizar clusterização diminui consideravelmente o tempo gasto pelo processo de previsão,
alcançando ainda previsões bastante precisas.
In the present work a parametric study of variations of the Analogs Ensembles method of is conducted. The method, that is originally used to perform post-processing, is applied in this study to the forecast of time series. Known time series are reconstructed in order to evaluate and identify the optimal parameters that achieve the minimization of errors obtained from the comparison between the real and predicted time series. Altogether, nine years of data were used, seven years for the training period and two years for comparison with the predictions made. Data from three weather stations are used, where the method is applied so that station A has its data predicted from stations B and C. The study showed that it is possible to identify values and patterns of parameters that are useful for the minimization of errors. However, it also showed that it is difficult to find global parameters that minimize the errors of any variable in any scenario worked on. The performance study carried out showed that, in most cases, using clustering considerably reduces the time spent in the forecasting process and still achieves very accurate forecasts.
In the present work a parametric study of variations of the Analogs Ensembles method of is conducted. The method, that is originally used to perform post-processing, is applied in this study to the forecast of time series. Known time series are reconstructed in order to evaluate and identify the optimal parameters that achieve the minimization of errors obtained from the comparison between the real and predicted time series. Altogether, nine years of data were used, seven years for the training period and two years for comparison with the predictions made. Data from three weather stations are used, where the method is applied so that station A has its data predicted from stations B and C. The study showed that it is possible to identify values and patterns of parameters that are useful for the minimization of errors. However, it also showed that it is difficult to find global parameters that minimize the errors of any variable in any scenario worked on. The performance study carried out showed that, in most cases, using clustering considerably reduces the time spent in the forecasting process and still achieves very accurate forecasts.
Description
Mestrado IPB-ESTG
Keywords
Conjuntos análogos Parametrização Séries temporais Previsão meteorológica