Application of a hybrid method for forecast time series in concrete deformation in a counterstruct dam
Palavras-chave:
Time series, Deep Learning Neural Networks, Box & Jenkins modelsResumo
The concrete deformations have been researchers study object in the structural safety of dams process.
These deformations, which occur over time, are influenced by various physical and environmental factors. One of
the environmental factors that affect the deformations of concrete is the ambient temperature. In this paper, a
hybrid method called SARIMAX-NEURAL is presented for prediction of concrete deformations that are
influenced by ambient temperature. This hybrid method was defined as a linear combination of predictions from
Box & Jenkins methodology models and Deep Learning neural network models with Long Short-Term Memory
architecture. Historical data of concrete deformations were measured by rosettes strain installed in a buttress block
in the Itaipu dam for a period of 34 years. The proposed hybrid method, which considered the effect of ambient
temperature on the deformations of concrete, effective results presented in comparison with the individual methods
in which the effect was not considered to ambient temperature. The predictive accuracy gains were between 25%
and 60%.