Intelligent prediction and optimization of ground settlement induced by shield tunneling construction

Abstract

Predicting and optimizing surface settlement induced by shield tunneling is a challenging task. Although various machine learning methods have been applied to this field, they are mostly limited to predicting conventional settlement patterns, with limited effectiveness in addressing abnormal settlement issues or achieving truly adaptive adjustments during shield tunneling. Consequently, the generalization and applicability of these methods remain insufficient. To tackle these challenges, this paper proposes an adaptive, efficient, and data-driven intelligent prediction and optimization method to comprehensively address the demands of surface settlement prediction and control during shield tunneling. The main innovations include: (1) The development of a generalization performance evaluation method based on incremental training set sampling, particle swarm optimization (PSO) for hyperparameter tuning, and 10-fold cross-validation, systematically assessing robustness and generalization capability of the model under varying data volumes and complex construction conditions. (2) The construction of an extremely randomized trees regression (ETR) model, combined with Shapley Additive Explanations (SHAP) for transparent analysis of prediction results, significantly enhancing the model’s adaptability to complex geological conditions. (3) The design of a three-level optimization strategy, including local optimization, global optimization, and manual intervention, which integrates the ETR model with a genetic algorithm (GA) to propose a comprehensive and rational settlement control plan. The results demonstrate that the proposed method exhibits excellent generalization capability and reliability under different engineering scenarios, particularly in identifying and controlling abnormal settlement and adapting to complex geological environments. This provides an innovative solution for surface settlement prediction and optimization in shield tunneling projects.