Supervised parameter updating of deformation analyses for rockfill dams using prior knowledge

Abstract

Accurate and reliable numerical simulation is crucial for the safe construction and operation of infrastructure such as rockfill dams. Model parameter updating through inverse analysis based on monitoring data is key to improving analysis accuracy. However, existing parameter updating methods for dams often neglect parameter correlations, resulting in discrepancies between the joint distribution of updated parameters and experimental data. Besides, conventional parameter updating methods exhibit considerable randomness, resulting in non-unique updated parameters. These factors limit the improvement of analysis accuracy and even lead to the failure of analysis convergence. Thus, this study proposes a parameter updating method for deformation analysis of rockfill dams based on surrogate-assisted optimization. Innovatively, the multivariate distribution of experimental data of model parameters is incorporated as prior knowledge to supervise the parameter updating. Specifically, a multivariate distribution model of experimental data from 48 rockfill dams worldwide is constructed using multivariate copula function. Then the joint probability density function is integrated into the optimization process through population preselection mechanism and penalty function, guiding the updated parameters to align with the experimental joint distribution. The application to an ultra-high rockfill dam demonstrated that this scheme effectively identified multiple optimal parameters from the constitutive model. With the supervision of prior knowledge, the updated parameters K, n, Kb, and m of the Duncan–Chang E-B model showed strong consistency with the multivariate joint distribution derived from experimental data. This scheme improved the accuracy of the deformation analysis model by 16%, thereby providing critical support for dam safety assessment.