Parameter inverse analysis of high rockfill dams considering material uncertainty based on the EJaya-SESM model

Abstract

Uncertainties of rockfill material can significantly impact the structural behavior of rockfill dams. Identifying such uncertainties is thus essential to rockfill dam behavior analysis. Traditional material uncertainty identification using Monte Carlo-Stochastic Finite Element (MC-SFE) calculation is extremely tedious and time-consuming. To alleviate the computational burden, this work presents an efficient machine learning method for inverse calculation of rockfill dam material uncertainties based on the Stacking Ensemble Surrogate Model (SESM) and Enhanced Jaya (EJaya) algorithm. In this methodology, a multi-parameter and multi-zone random field is firstly introduced to describe the spatial heterogeneity of rockfill material in the dam body field. Then, MC-SFE is replaced by the stacking ensemble learning-based surrogate model to explore the complex mapping relationships between rockfill material parameters and dam settlement responses. Subsequently, a novel optimization algorithm called EJaya is developed to minimize the objective function for material parameters' inversion calculations. The implementation of the EJaya-SESM model is demonstrated on a real-world high rockfill dam in service as an illustrative example. Through comprehensive forward analysis, the effectiveness and rationality of inversion calculations are further verified. The numerical results show that the stacking ensemble strategy can greatly improve the surrogate model's accuracy with reduced computational time versus MC-SFE calculation, and the inversion outcomes derived from the EJaya algorithm demonstrate superior precision compared to those attained by several commonly used metaheuristic techniques. This study provides an advanced means to achieve excellent performance in parameter inverse analysis of rockfill dams at a low computation cost.