Active learning-driven ensemble framework for efficient and robust reliability analysis in geotechnical engineering

Purpose Surrogate models are frequently used to alleviate the computational cost associated with reliability analysis in geotechnical engineering; however, efficiently training high-accuracy surrogate models remains a significant challenge. This paper aims to introduce AEGPR-MCS, a robust framework that integrates the ensemble model, active learning strategy and Monte Carlo simulation (MCS) to perform the reliability analysis for various geotechnical systems. Design/methodology/approach AEGPR-MCS leverages an ensemble model as a substitute for complex, computationally intensive and implicit geotechnical mechanical models. It incorporates active learning to iteratively choose the most important training samples, ensuring that the surrogate model is trained with both precision and efficiency. The added new training samples are strategically positioned near the limit state surface, which significantly improves model efficiency by focusing computational resources on the most critical regions. Moreover, the adaptive ensemble model dynamically adjusts the weights of its components to enhance predictive robustness across various geotechnical systems. Findings Validation through three geotechnical examples demonstrates significant improvements in both computational efficiency and prediction accuracy over other methods. The AEGPR-MCS method provides reliable and precise reliability assessments for different geotechnical engineering problems while significantly reducing computational resource requirements, making it a well-suited and versatile approach for addressing a broad spectrum of geotechnical reliability challenges. Originality/value This study presents a robust methodology where the active learning strategy plays a pivotal role in identifying the most informative training samples, primarily those located near the limit state surface. This strategic approach allows the surrogate model to be developed iteratively, promoting both precision and resource efficiency. Furthermore, AEGPR-MCS’s adaptive ensemble model dynamically adjusts the weights of its components, thus bolstering its robustness across diverse geotechnical systems.