Abstract: The selection of important updating parameters can effectively alleviate the “curse of dimensionality” in high-dimensional Bayesian model updating while maintaining prediction accuracy. However, existing methods for se-lecting important updating parameters based on computational models often fail to directly reflect the influence of input parameters on model prediction errors. To address this issue, the likelihood function is introduced as the evaluation function in the Sobol global sensitivity analysis framework, enabling a quantitative assessment of the influence of input parameters on model prediction errors. Based on this framework, an important parameter selec-tion method is proposed using the cumulative sum of total Sobol indices derived from the likelihood function. Fur-thermore, within a framework that separates the contributions of aleatory and epistemic uncertainties, an important updating parameter selection and Bayesian updating method under the separation of aleatory and epistemic un-certainties is developed. This method can effectively identify important epistemic parameters that dominate the model prediction error and possess reducible uncertainty for updating. Numerical examples demonstrate that, un-der aleatory uncertainty conditions, the proposed method can significantly reduce the number of parameters to be updated while achieving prediction accuracy and uncertainty reduction comparable to those obtained from multi-parameter or even full-parameter updating. Under the condition of separating aleatory and epistemic uncertainties, the proposed method can stably identify important epistemic parameters that influence model prediction errors, thereby avoiding ineffective or redundant updating of parameters associated with aleatory uncertainty. The results further indicate that distinguishing between aleatory and epistemic uncertainties during the model updating process is essential for reducing the updating dimensionality and enhancing the physical interpretability of the updating results.

Key words: parameter selection, bayesian updating, global sensitivity analysis, aleatory uncertainty, epistemic uncertainty