Abstract: To address the issue of relying heavily on the experience of construction personnel when setting surface deformation targets for each stage of shield tunneling, the relevance of surface deformation at each stage to the efficiency of shield tunneling and the impact on the surrounding environment under different geological conditions are analyzed. Based on this analysis, a stage target planning method for surface deformation is proposed. The proposed method utilizes machine learning to establish an evaluation model for the efficiency of shield tunneling and a prediction model for the final surface deformation. The predicted results are then comprehensively evaluated, considering tunneling efficiency, environmental impact, and construction cost, to obtain an engineering comprehensive benefit evaluation. Optimization algorithms are used to calculate surface deformation control values for each stage, aiming to achieve optimal comprehensive benefits. The model was integrated into the ZhiYu selfdriving shield system and applied to the Nanjing metro line 5 tunneling project. The results of the engineering application show that the model has effectively improved the safety and efficiency of selfdriving shield tunneling by optimizing the control targets for each stage of surface deformation.

Key words: shield tunnel, surface deformation stage, control target planning, selfdriving, machine learning; optimization algorithms