In numerical simulations of underground engineering projects, accurately predicting rock mass parameters is particularly challenging and critical when tunneling through composite strata in proximity to existing structures. This study presents an intelligent inverse analysis approach tailored for shield tunnels in such complex geological settings: the Firefly Algorithm-Convolutional Neural Network (FA-CNN) algorithm. The proposed method leverages local monitoring data, identifies key inversion parameters through sensitivity analysis and orthogonal experiments, and performs parameter inversion using the FA-CNN algorithm. Results indicate that the parameters obtained via this inverse analysis method exhibit high consistency with actual values. Furthermore, the approach has been successfully applied to predict displacements of adjacent structures during tunnel construction, thereby achieving both cost reduction and the maintenance of structural safety. To further validate its effectiveness, a comparative assessment was conducted against other widely used inverse analysis algorithms, including Multilayer Perceptron (MLP), Support Vector Regression (SVR), and standard CNN. The outcomes demonstrate that the FA-CNN algorithm delivers superior prediction accuracy when processing multidimensional and complex datasets. Additionally, in terms of iteration speed and convergence performance, the recommended parameter ranges for the FA component are ρ = 0.3–0.5 and γ = 0.6–0.8. Overall, the FA-CNN algorithm proposed in this work offers substantial theoretical and practical value for inverse analysis of rock mass parameters in shield tunneling through composite strata. Future research will aim to extend the application of this method to other domains of underground engineering and refine the algorithm to address an even broader array of engineering challenges.
Lv et al. (Wed,) studied this question.