Abstract: Parameter identification of a concrete dam is the key to evaluating its behaviors. To further improve the efficiency and accuracy of parameter inversion, this paper develops a novel inversion strategy for concrete dam parameters based on the multi-output twin support vector regression (MTSVR) and the improved snow ablation optimization (ISAO). A nonlinear relationship of the parameters to be inverted versus the hydrostatic pressure component of displacement is simulated by training the MTSVR model, in place of complex finite element calculations; ISAO is used for optimizing the inversion of the target parameters. Analysis of engineering examples shows that the results of this surrogate model basically agree with those of the finite element calculations, ISAO has faster convergence and higher accuracy than the traditional meta-inspired optimization, and the computational cost of single parameter inversion is lower. This verifies that our new inversion strategy effectively improves computational efficiency while maintaining the computational accuracy, and that the method is effective, practical and useful for the parameter identification of actual projects.

Key words: concrete dam, parameter inversion, multi-output twin support vector regression, improved snow ablation optimization, surrogate model