Abstract: The water temperature in front of super-high arch dam gradually stratifies vertically in the initial operation period, resulting in the rebound of internal temperature of the dam concrete. Periodic terms of temperature effect used in conventional deformation prediction models could not well describe the nonlinearity and nonstationarity of environmental and internal temperatures of dam during its initial operation. In view of this, a method of classifying measured ambient and dam temperatures and selecting typical measurement points is presented by integrating principal component analysis and hierarchical clustering. Meanwhile,a combination of time varying effects, including both index term and cycle term,that reflects the valley deformation under the periodic fluctuation of reservoir water level during the initial operation is introduced. On this basis,a multivariate regression model and a support vector machine model based on measured temperature variable are constructed. Case study demonstrates that the selected measured temperature variable well reflects the spatio-temporal characteristics of dam temperature field during initial operation, and the corresponding constructed model is of higher prediction accuracy than traditional models.

Key words: super-high arch dam, initial operation period, deformation, temperature factor, prediction model, principal component analysis, clustering, support vector machine