Please use this identifier to cite or link to this item:
|Title:||ANÁLISE E PREVISÃO DE TEMPERATURA MÁXIMA MENSAL DA CIDADE DO RIO DE JANEIRO (RJ) PARA O ANO DE 2016|
Analysis and Forecast for maximum monthly temperature of Rio de Janeiro (RJ) city for 2016
|Publisher:||Editora UFLA - Universidade Federal de Lavras - UFLA|
|Description:||In this paper we have compared the fit of additive models of Holt-Winters and SARIMA in time series of monthly maximum temperature in the city of Rio de Janeiro (RJ) using data from January 2003 to December 2015. The Holt-Winters additive model was chosen because the seasonal behavior of the time series. The SARIMA (1,0,1)×(1,1,1)12 was chosen because provided the best Akaike Information and Schwarz Criteria. The final model was chosen by using mean square error and mean absolute percentual error. These measures were calculated using the fitted model to the historical data. Additionally, a Diebold-Mariano test for predictive accuracy was applied and white noise characteristics of the residues were evaluated. The SARIMA (1,0,1)×(1,1,1)12 was the best model fitted and so it was adopted for maximum monthly temperature forecast for the year 2016.|
In this paper we compare the adjustment of additive models of Holt-Winters and SARIMA the series of monthly maximum temperature in the city of Rio de Janeiro (RJ), with data from January 2003 to December 2015. The additive Holt -Winters model was chosen because the seasonal character of series and SARIMA model by Akaike and Schwarz Information Criteria. The final model was chosen by indicators Mean Square Error indicators and Average Percentage Absolute, calculated by adjusting the historical data. Additionally, white noise characteristics of residuals were evaluated. The SARIMA model better adjusted and so was adopted for maximum monthly temperature forecast for the year 2016.
|Appears in Collections:||Revista Brasileira de Biometria|
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.