Detecção e tratamento de outliers em séries temporais aplicado ao controle estatístico de instrumentação de barragens
Resumo
Resumo : Na análise de dados de instrumentação de barragem, além da sazonalidade, é comum encontrar a presença de observações influenciadas por eventos externos e de forma não sistemática ou erros de medição. Neste trabalho avalia-se o efeito de outliers na modelagem de algumas séries temporais de instrumentos que monitoram o trecho E da barragem de ITAIPU Binacional, situada no Rio Paraná, entre o Brasil e Paraguai. No estudo, realiza-se o ajustamento de um modelo ARIMA Sazonal a série de observações do instrumento. Para validar os modelos ajustados considerase a capacidade das suas previsões em termo do erro cometido. A avaliação do efeito de outliers na série se dá pela comparação do teste de normalidade e a carta de controle para a média dos resíduos das previsões, com e sem a presença de valores atípicos. Os resultados indicam que outliers influenciam significativamente na normalidade e na identificação de valores fora de controle estatístico dos resíduos das previsões da série temporal do instrumento utilizado. Portanto, é fundamental a identificação e o tratamento de valores atípicos. Abstract : In the analysis of dam instrumentation data, in addition to the seasonality, it is common to find the presence of observations influenced by external events and in a non-systematic way or measurement errors. This work evaluates the effect of outliers on the modeling of the monitoring instruments time series in the section E of the ITAIPU Binacional dam, located on the Paraná River between Brazil and Paraguay. In this study, a SARIMA model is adjusted in some time series of instrument observations. In order to validate the adjusted models, it is considered the capacity of their forecasts in order of the committed error. The evaluation of the effect of outliers in the time series is done by comparing the normality test and the control chart for the mean of the predicted residuals, with and without the presence of atypical values. The results indicate that outliers significantly influence the normality and the identification of values outside statistical control of the time series residuals predictions of the used instrument. Therefore, the identification and treatment of outliers is very important.