AI-driven co-tunneler: Automated excavation parameter selection using information entropy-assisted multi-objective optimization

Intelligent control in mechanized shield tunneling requires the simultaneous optimization of safety, efficiency, and energy consumption. This paper proposes an adaptive multi-objective decision-making framework that integrates AI-based forward prediction with inverse optimization for real-time control of shield tunneling parameters. First, a multilayer perceptron (MLP) is employed to construct a forward model, using controllable shield parameters as inputs and the control objectives as outputs. Pareto-optimal solutions are then generated via NSGA-II. On this basis, the entropy weighting method and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) are further integrated to identify the optimal solution. The framework is validated through a case study from Metro Line 7 in Hefei, Anhui Province, China. The proposed approach increases penetration rate by 56.1% and reduces cutterhead torque by 6.4% under safety constraints. Visualization using t -distributed stochastic neighbor embedding (t-SNE) further validates that the proposed framework consistently surpasses conventional empirical operating practices. • Real-time shield tunneling control is formulated as a coupled forward–inverse optimization problem. • An adaptive framework unites MLP prediction, NSGA-II, and entropy-weighted TOPSIS for real-time shield control. • Entropy weighting assigns objective weights automatically, reducing subjectivity. • The method increases penetration rate by 56.1% and reduces cutterhead torque by 6.4%.