Intelligent prediction and control of deformation induced by a servo-strutted deep excavation adjacent to existing tunnels

To enable intelligent prediction and control of excavation-induced deformations, including wall deflection, ground surface settlement, and nearby tunnel displacements, this paper proposes an integrated approach combining in-situ test-based numerical modelling, Bayesian-optimised deep neural networks (BO-DNNs), and a DNN-based Newton-Raphson (DNN-NR) algorithm. The proposed framework serves as a decision-support tool for pre-construction planning of a deep excavation adjacent to existing tunnels. Specifically, the verified numerical models generate the training dataset for the BO-DNN model, which achieves high predictive accuracy for maximum deformations under varying servo-force combinations and excavation geometries. The BO-DNN analysis reveals that servo forces significantly influence deformation patterns and can even alter the direction of wall deflection and ground settlement. Leveraging this surrogate model, the DNN-NR algorithm efficiently identifies optimal servo forces to minimise deformations. The applications demonstrate that the DNN-NR-derived forces effectively restrict deformations within allowable limits. Furthermore, the algorithm quantifies the relative importance of each servo strut in deformation control and provides allowable axial force thresholds, facilitating adaptive force adjustments during the excavation. • In-situ test-based numerical model for simulating deformations in a servo-strutted deep excavation system. • Bayesian-optimised neural networks for predicting excavation-induced deformations with various servo forces. • Developed DNN-NR algorithm to find the optimal solutions among multi-parameters large space. • Proposed the concept of optimal servo forces for controlling excavation-induced deformations.