During deep excavation, actual soil parameters undergo changes. To enhance the accuracy of soil parameter selection in finite element simulation and improve the precision of finite element analysis, an inversion method for soil parameters based on a PSO-SVM model is proposed. In this method, the particle swarm optimization (PSO) algorithm is utilized to optimize the penalty parameter C and kernel function parameter g of the support vector machine (SVM) model. The optimized PSO-SVM model is employed to establish a nonlinear mapping relationship between the horizontal displacements of retaining structures in deep excavations and soil parameters through orthogonal experimental design and finite element simulation analysis. Subsequently, soil parameters are inverted from monitoring data of horizontal displacements of retaining structures, and the reliability of the parameters is verified. The deformation of the retaining structures during subsequent cases is then predicted. The results demonstrate that the absolute error of the peak maximum horizontal displacements of the retaining structures after inversion is maintained within 1 mm. The maximum relative error is reduced from 18.96% before inversion to 7.63%, indicating that the inverted soil parameters for the deep excavation possess high accuracy. The precision of the finite element simulation for deep excavation is significantly improved, effectively reflecting the actual mechanical properties of the soil during the construction stage. The inverted parameters can be used for the prediction of subsequent retaining structure deformation. During subsequent construction conditions, the predicted maximum horizontal displacement (deformation) of the retaining structure at monitoring point CX1 is 15.66 mm, and that at monitoring point CX2 is predicted to be 14.22 mm. Neither value exceeds the project warning threshold of 30.00 mm.

During deep excavation construction, deformation impacts on adjacent structures are inevitably induced, and field monitoring data are often contaminated by noise that degrades prediction accuracy. To address these issues, this study develops a joint denoising strategy combining CEEMDAN, sample entropy-based adaptive IMF screening, and an improved wavelet threshold (IWT) function, followed by a Bayesian optimization-based extreme gradient boosting (BO-XGBoost) model for surface settlement prediction. The developed method automatically identifies high-noise IMF components via sample entropy and processes them using an improved threshold function that overcomes the discontinuity of hard thresholding and the constant bias of soft thresholding, thereby preserving useful information while suppressing noise. Experimental results on a Wuhan metro deep excavation project demonstrate that the CEEMDAN-IWT method improves SNR by up to 4.09% and reduces RMSE by up to 8.00% compared with conventional CEEMDAN-wavelet threshold denoising. The BO-XGBoost model trained on denoised data achieves an RMSE of 0.09 mm and a MAPE of 3.54%, outperforming BP, LSTM, standard XGBoost, GRU, CNN-LSTM, TCN, and simple regression baselines. Feature importance analysis confirms that the denoised data retain physical interpretability consistent with soil deformation continuity. This framework provides a practical solution with promising accuracy for deformation monitoring and early warning in deep excavation engineering.