A collaborative numerical simulation-soft computing approach for earth dams first impoundment modeling

Abstract

Uncertainty quantification plays a crucial role in the design, monitoring, and risk assessment of earth dams. To reduce the computational burden, we employ a combination of finite difference method and soft computing techniques to investigate material uncertainties in earth dams during the initial impoundment stage. The findings of sensitivity analysis with the Tornado diagram indicate that key material properties such as dry density, elasticity modulus, friction angle, and Poisson's ratio significantly influence the displacements and stress analysis. In our study, we explore four variants of extreme learning machines (ELMs): the standalone ELM, hybridized versions with the improved grey wolf optimizer algorithm, ant colony optimization for continuous domains, and artificial bee colony. These methods are assessed across various training sizes to predict multiple parameters, including horizontal and vertical displacements, stresses, and the factor of safety (FoS). The hybridized ELM with the improved grey wolf optimizer algorithm emerges as the superior choice for most of the response variables. A minimum of 200 numerical simulations is required to establish a stable and accurate meta-model with an average prediction error of less than 3% for responses and the FoS.