A novel physics–data hybrid approach for slope stability assessment considering future rainfall patterns

Accurately and dynamically assessing slope stability under changing rainfall patterns is essential for mitigating landslide risks in construction projects. Currently, slope stability analysis commonly relies on data-driven or physics-based methods, neglecting the complementary relationship between both approaches. Thus, this research proposes a slope stability prediction approach that combines numerical analysis techniques and data-driven artificial intelligence techniques. Using slope displacement monitoring data, geotechnical strength parameters are inferred via a random finite element analysis based on a neural network surrogate model. Furthermore, a data-driven time-series prediction model is established to forecast changes in slope geotechnical parameters, before using the strength reduction method to calculate the safety factor. The proposed approach is validated by analyzing two real-world slope cases and comparing it with traditional time-series data mapping methods for safety factor prediction. The results demonstrate that, when combined with meteorological forecast data, the proposed approach effectively captures the evolving trends in slope stability. The stability predictions outperform those based on time-series data mapping methods, providing accurate safety factor estimations and supporting the assessment of future slope failure risks in construction projects.