A surrogate-assisted stochastic optimization inversion algorithm: Parameter identification of dams

Abstract

Dynamic monitoring data plays an essential role in the structural health monitoring of dams. This study presents a surrogate-assisted stochastic optimization inversion (SASOI) algorithm, a novel technique for static and dynamic parameter identification. This algorithm is based on probabilistic finite element simulations and Bayesian inference theory. It combines the advantages of low computational cost in surrogate models and fast convergence in the Bayesian algorithm. Taking four cases of different complexity, this paper verifies the effectiveness of the SASOI algorithm and validates its practicality for large dams. Surrogate models consider several alternatives, including polynomial chaos expansion (PCE), Kriging, polynomial chaos Kriging, and support vector regression. Implementation of the SASOI algorithm on dams shows that PCE outperforms other techniques. This algorithm improves the accuracy and efficiency of the static parameter identification methods by nearly 27 times compared to the classical inversion methods. Furthermore, the accuracy of dynamic parameter identification is higher than that of static one. The SASOI algorithm is applicable to other large-scale infrastructures.