English：

In response to the issue of insufficient prediction accuracy for tunnel shield cutter wear,particularly the significant impact of the cutterhead installation radius effect on traditional methods and theweak generalization ability of prediction models, a collaborative prediction model based on segmentalregression and machine learning is proposed. This model integrates 315 sets of multidimensional dynamicdata covering geological conditions, construction parameters, and tool performance parameters, andconstructs the model based on the rock⁃breaking volume wear rate. The model differentiates between thecentral, middle, and edge zones of the cutterhead to effectively eliminate the interference of theinstallation radius. Random forest, BP neural network, and XGBoost models are used to predict thecutter wear, and the performance of these models is compared and analyzed. The research results showthat the segmental regression model quantifies the wear characteristics of the cutter in different zones. TheXGBoost model achieves a coefficient of determination of 0.92 on the test set,with an average absolutepercentage error of only 7.5%, outperforming both the random forest and BP neural network models, anddemonstrating better generalization ability. Core factors influencing cutter wear include inherent rockproperties such as uniaxial compressive strength, CAI value, and quartz content.