Forecast of time series earned by the Piezometer through Method Multiple Kernel Sarima Support Vector Regression Wavelet
Palavras-chave:
structural health monitoring, hybrid method, statistical techniques, machine learning techniquesResumo
In this study time series predictions were made from measurements of the instrument called piezometer
(PS), located in the key block (I10) of stretch I of the Itaipu hydroelectric dam. The results show that the forecasting
performance attained by the method called SARIMA Support Vector Regression Wavelet of Multiple Kernels
(SSVRWMN) was notably superior to predictive methods SARIMA, SVR, and SARIMA-SVR combined.
Comparing it to the second-best result (namely, the SVR method), the relative reduction was approximately 39.1%
in the mean square error (MSE) accuracy measure.
