Uncertainty-based multi-objective optimization in twin tunnel design considering fluid-solid coupling

This paper presents a multi-objective optimization framework based on uncertainty analysis, focusing on fluid–structure interaction in twin tunnel design. High-quality datasets are generated using three-dimensional fluid–structure interaction theory. Long Short-Term Memory-Attention (LSTM-Attention) models are used to simulate internal forces within the tunnel and ground settlement, improving prediction accuracy. The Snow Ablation Optimizer (SAO) adjusts the hyperparameters of the LSTM-Attention model. The SHapley Additive exPlanations (SHAP) framework is introduced to enhance the model’s transparency and interpretability, aiding in understanding the model’s decision-making process. The hybrid Non-dominated Sorting Genetic Algorithm II (NSGA-II) combined with Particle Swarm Optimization (PSO) is employed for multi-objective optimization. Monte Carlo simulation is used to estimate probability constraints, ensuring that the optimization process yields stable and reliable solutions. A case study analyzes the optimization results under different tunnel radii and uncertainty conditions in detail, validating the method’s effectiveness. The study shows that considering uncertainty significantly enhances the accuracy and stability of the optimization results for internal forces and ground settlement. Additionally, under different tunnel radii and uncertainty conditions, the distribution of optimal solutions is more concentrated. This method provides a novel solution for multi-objective optimization in complex engineering problems and offers theoretical and practical guidance for engineering decision-making and optimization.