English：

Accurate prediction of rockburst is essential for ensuring the safety of tunnel construction underhigh geostress conditions. Taking the No. 1 inclined shaft of a plateau railway tunnel as the engineeringbackground, a refined three-dimensional topographic-geometric model of the inclined shaft area wasestablished based on detailed field investigation data to analyze the geostress characteristics. Combinedwith on-site tests and monitoring data, particle swarm optimization algorithm was employed to optimize BPneural network for the inversion of surrounding rock parameters. And a rockburst prediction method wasproposed that integrates geostress inversion and surrounding rock parameter inversion,which wasvalidated against actual rockburst occurrences observed on site. The research results show that thegeostress inversion method based on refined modeling can accurately characterize the geostress distributionin the inclined shaft area. The geostress increases with burial depth, and the variation trend of themaximum principal stress is generally consistent with the relief of the longitudinal profile of the inclinedshaft. The particle swarm optimization algorithm effectively improves the prediction accuracy of the BPneural network model,with a relative error of 2.95% between the predicted and measured surroundingrock strength. Rockburst is more likely to occur in the bottom section of the inclined shaft. The proposedmethod,which combines geostress and surrounding rock parameter inversion, can predict rockburst grademore effectively, achieving a prediction accuracy of 82.9%, and thus provides a useful reference for fieldconstruction and rockburst prevention and control in high geostress tunnels.