Objective As one of the most common geological hazards in China, landslides frequently cause deformation, rupture, leakage, and even catastrophic explosions along long-distance pipelines. Studying pipeline failures caused by landslides provides a valuable approach for more effectively evaluating the safety status of these pipelines.

Methods Based on X80 pipes selected as the research subject, a finite element model of landslide-pipeline interactions was established using ABAQUS software. The simulation results were then integrated with machine learning techniques to develop a Bald Eagle Search-Extreme Learning Machine (BES-ELM) model. This model was employed to analyze the mechanical responses and predictive performance of buried natural gas pipelines affected by landslides. Through multi-factor analysis, the influence of seven factors on the equivalent stress in the pipes was explored: landslide displacement, landslide width, internal pipeline pressure, pipeline wall thickness, soil cohesion, internal friction angle of soil, and soil type.

Results Under the influence of landslide displacement, the most significant impacts on the pipes were primarily due to axial tension and pressure. When landslide displacements ranged from 2.5 m to 3.0 m, the likelihood of the pipes experiencing maximum equivalent stress exceeding their yield strength increased, consequently heightening the risk of pipeline failures. Within this displacement range, the internal pipeline pressure, soil cohesion, and the friction angle of the soil were positively correlated with the maximum equivalent stress in the pipes, while wall thickness exhibited a negative correlation. A stress concentration area was observed within the landslide widths. As the landslide width gradually increased from 10 m to 50 m, the maximum equivalent stress in the pipes displayed a non-linear change pattern, initially increasing before decreasing. Due to the landslides, different parts of the pipes encountered varying levels of risk, with the highest risk alternating between the front and back faces relative to the sliding direction as displacement and other factors changed.

Conclusion A total of 150 groups of finite element data were selected to compare the predictive performance of the BES-ELM model and the traditional Extreme Learning Machine (ELM) model regarding pipe equivalent stress under landslide influence. The BES-ELM model produced superior prediction results, with a maximum relative error of 1.06％, a coefficient of determination of 0.977, and a root mean square error reduced by 65.18％ compared to the traditional ELM model. This developed model has proven effective as a tool for quickly identifying equivalent stress in pipes. In summary, when selecting routes for long-distance pipelines, it is essential to investigate the geological conditions to bypass the influence range of potential landslides or to implement reinforcement measures to ensure pipeline safety.