Addressing the issues of single model algorithm, low accuracy, and poor generalization in existing shield tunneling speed prediction methods, this study proposes a shield tunneling speed prediction approach to improve prediction accuracy based on Variational Mode Decomposition (VMD), Dung Beetle Optimizer (DBO), and Stacking ensemble learning. Firstly, to obtain more effective data, VMD is applied to decompose and reconstruct the original data to obtain denoised construction parameter data for subsequent model prediction. Secondly, based on the ensemble learning strategy, Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) models are selected as base learners, while Gaussian Process Regression (GPR) is chosen as the meta-learner to construct a Stacking ensemble learning prediction model with higher prediction accuracy and stronger generalization ability. Thirdly, to further enhance prediction accuracy, DBO is employed to optimize the hyperparameters of the ensemble learning model. Finally, this prediction method is applied to the shield tunneling construction of a water diversion tunnel project in Henan Province and compared with other prediction methods. Compared to other single models (SVR, RF, XGBoost), the results indicate that the proposed method achieves higher prediction accuracy, with average accuracy improvements of 7.76%, 6.70%, and 4.97%, respectively, providing a new approach for shield tunneling speed prediction.