A hybrid approach combining UD and GA-CV-SVM to evaluate shear performance in high asphalt concrete core

Abstract

The shear effect on high asphalt concrete core is significant. However, studies on the reliability of 100-meter-scale cores against shear damage remain limited. A key challenge in this research field is establishing the control criteria for the core and improving the computational efficiency of implicit limit state function (LSF). Additionally, the impact of material parameter uncertainty on the shear failure reliability of the core during the dam construction and impoundment stages remains unclear. To address this, a safety evaluation method based on time discretization was proposed, combining uniform design (UD), K-fold cross-validation (K-CV), and genetic algorithm (GA) to optimize the support vector machines (SVM). The core parameters of 52 asphalt concrete-core rockfill dams (ACCRDs) were analyzed, with the statistical values of the basic variables considered in determining the reliability index. The theoretical derivation of the critical shear failure safety index established a stability formula to assess the safety state of the dam core. The significance parameters were identified, and the sample points were generated at each stage using UD, and the Support Vector Regression (SVR) was applied to reconstruct the LSF, and reliability was calculated through the checking point method.