Accurate and dynamic identification of surrounding rock grades in TBM tunnels is crucial for ensuring excavation safety and improving construction efficiency. This study proposes a hybrid modeling method based on a physics-data dual-driven approach to achieve high-precision dynamic identification of surrounding rock grades. First, the Isolation Forest model is employed to eliminate outliers from the raw tunneling data, and the key tunneling parameters influencing rock grades are identified using mutual information. Then, the Seasonal and Trend decomposition using LOESS (STL) model is used to perform multimodal decomposition on the dominant tunneling parameters, obtaining the corresponding trend, periodic, and residual components. Subsequently, an Improved Refined Composite Multiscale Sample Entropy (IRCMSE) model is adopted to calculate the feature entropy of each component, forming a dynamic sample database for the data-driven model. Based on this, an improved Convolutional Neural Network - Long Short-Term Memory (CNN-LSTM) model is developed to realize data-driven dynamic identification of TBM tunnel strata. Furthermore, a variation identification formula for surrounding rock grades was proposed based on the principle of geological continuity, enabling physics-driven dynamic identification of surrounding rock grades in TBM tunnels. On this basis, a fusion method combining the physical-driven model and the data-driven model is proposed. The constructed physics-data dual-driven model achieves average precision, recall, F1-score, and accuracy of 98.29 %, 97.98 %, 98.13 %, and 98.30 %, respectively, representing an average improvement of 2.17 % over the data-driven model and 15.07 % over the physical-driven model. Engineering validation results indicate that the overall performance of the model decreases by only 1.74 % and 5.29 % under similar and different geological conditions, respectively, demonstrating strong generalization and robustness, and meeting the requirements of intelligent TBM tunneling under complex geological conditions.

Accurate prediction of TBM tunneling performance is essential for ensuring the safety and efficiency of tunnel construction. This paper addresses the challenge of achieving reliable long-term interval prediction of TBM performance under multiple sources of uncertainty. To tackle this problem, a SVMD-Informer framework is developed, integrating SVMD-based mode decomposition, enhanced Composite Multi-Scale Fuzzy Entropy feature extraction, and Bootstrap-based pseudo-sample generation to quantify both model and random uncertainties. The proposed model achieves excellent accuracy, with an average R2 of 0.9481 and a PICP of 95.3 %, effectively capturing both point and interval characteristics of TBM performance. The results demonstrate that the model provides robust, interpretable, and uncertainty-aware predictions for TBM operators and project managers. These findings advance the development of intelligent tunneling control and motivate future research on hybrid learning frameworks for performance prediction in underground engineering.

Accurate prediction of Tunnel Boring Machine (TBM) tunneling thrust is crucial for optimizing cutterhead force control, reducing cutter wear, and ensuring construction safety and efficiency. This study proposes an innovative multi-channel fusion prediction framework tailored to the complex, nonlinear, and non-stationary characteristics of TBM thrust data. Unlike conventional applications of existing methods, this work introduces several targeted improvements to enhance adaptability and predictive performance under challenging tunneling conditions. Specifically, an Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) algorithm is used to perform signal decomposition, extracting 10 Intrinsic Mode Functions (IMFs) that better capture multi-scale components of thrust signals. These IMFs are further reconstructed into five representative components using an improved Composite Multiscale Attention Entropy (CMATE) model and similarity aggregation, enhancing the interpretability of modal features. Input variables for each component are selected and weighted based on Maximal Information Coefficient (MIC) and feature importance analysis. To model the intricate spatiotemporal dependencies in thrust evolution, a redesigned Spatiotemporal Attention Mechanismbased Transformer (STFormer) architecture is employed, integrating sequence decomposition, Fourier-based autocorrelation, and spatial attention mechanisms. Additionally, a Bidirectional Gated Recurrent Unit (BiGRU)-based correction model is constructed to predict and compensate for residual errors, resulting in the final CMC-STFormer model, which achieves an R2 of 0.9970, MAE of 69.27 kN, MAPE of 0.51%, and RMSE of 87.56 kN, significantly outperforming existing single- and multi-channel models by 76.43% and 61.60% on average, respectively. This study provides a high-precision and robust solution for intelligent TBM thrust prediction in complex tunneling environments.

Accurate prediction of TBM tunneling loads is essential for enabling intelligent control. This paper proposes an intelligent prediction framework that integrates modal reconstruction with collaborative modeling. An improved Multivariate Variational Mode Decomposition (IMVMD) combined with Refined Composite Multiscale Diversity Entropy (RCMDE) is employed to extract the trend, seasonal, cyclic, and residual components of tunneling load signals. For each component, specialized predictive models, including Transformer, Bidirectional Gated Recurrent Unit (BiGRU), Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM), and Extreme Gradient Boosting (XGBoost), are developed to construct a collaborative hybrid learning architecture. A CNN-LSTM-based error correction strategy is further introduced, resulting in a corrected hybrid learning (CHL) model that achieved an R2 of 0.9972, a MAPE of 0.66 %, and an MAE of 11.73, exceeding traditional models by more than 60 % on average. The proposed method provides reliable technical support for intelligent perception and automated control in TBM tunneling.

To address the issues of the insufficient accuracy and weak generalization capabilities of single models in landslide displacement prediction, this paper proposes a machine learning model fusion prediction method for landslide displacement based on stacking. Taking the landslide displacement data (F) and rainfall (RAINFALL) of the Baishui River landslide in the Three Gorges Reservoir area as the research object, input sequences were constructed through data preprocessing and feature engineering. Prediction models including SVR, XGBoost, Bayesian optimization, and random forest were established. Based on the stacking framework, an integrated landslide displacement prediction model was developed by dynamically weighting the outputs of the base models using prediction accuracy and stability as fusion indicators. The Baishui River landslide, a typical colluvial landslide, was selected as a case study, with typical displacement data from monitoring points ZG118 and XD-01 from December 2006 to December 2012. The results show that the evaluation metrics (R2, ERMSE, and EMAE) for ZG118 and XD-01 demonstrate satisfactory prediction performance. Compared with traditional single models such as a TCN and XGBoost, the proposed integrated model exhibits improved prediction accuracy, providing scientific support for the real-time monitoring and early warning of landslide hazards.

This study proposes a seismic wave image recognition-based model to predict rock mass fracturing ahead of TBM faces, enhancing tunneling safety and efficiency. By integrating the Grey Wolf Optimization (GWO) algorithm with maximum entropy segmentation (optimal at population size 20), the model achieved superior image segmentation. Comparative analysis against GWO-Otsu, GA-Otsu, SSA-Otsu, GA-Kapur, and SSA-Kapur methods revealed that the maximum entropy method outperformed Otsu in PSNR and SSIM metrics, with GWO consistently attaining higher fitness values. Image slicing and feedback area analysis refined predictions, linking geological forecasts to TBM advancement increments. A total feedback area ratio of 0.3 and a positive-to-negative feedback ratio of 0.25–4 indicated poor rock integrity, triggering feature point detection. The ORB algorithm effectively extracted and matched feature points in SAP image feedback masks, enabling fracture classification based on point density. Engineering validation in a tunnel section confirmed model accuracy: post-reinforcement deformations aligned with predictions. The framework supports intelligent TBM tunneling by optimizing fracture prediction, balancing computational efficiency (via GWO) with precision (via maximum entropy), and establishing quantitative thresholds for rock integrity assessment. This approach advances automated geological hazard mitigation in TBM-driven tunnels.

Accurate prediction of landslide susceptibility is a key component of disaster risk reduction and early warning systems. Traditional landslide susceptibility prediction methods often face challenges in capturing complex nonlinear and spatio-temporal relationships inherent in geospatial data. In this study, we propose a Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), Spatial Attention Mechanism (SAM) hybrid deep learning model designed for spatial landslide susceptibility prediction. The model is trained on a comprehensive dataset comprising 19,898 samples, constructed from landslide records and 16 influencing factors in Kumamoto Prefecture, Japan. The input dataset is processed in tabular format using Microsoft Excel and includes variables such as topography, meteorology, soil characteristics, and human activity. The proposed model leverages Convolutional Neural Networks (CNN) to extract spatial features, Long Short-Term Memory networks (LSTM) to model temporal dependencies, and a Spatial Attention Mechanism (SAM) to enhance feature weighting dynamically. Experimental results demonstrate that the CNN–LSTM–SAM–Attention model significantly outperforms traditional machine learning approaches in terms of accuracy, precision, recall, F1 score, ROC–AUC, and PR–AUC. This substantial improvement is attributed to the model’s enhanced capability in capturing complex spatio-temporal patterns and dynamically weighting critical spatial features through the integrated Spatial Attention Mechanism (SAM). This study highlights the potential of deep learning-based approaches for improving the reliability of spatial landslide susceptibility prediction in complex terrain and dynamic climatic conditions.

The automatic identification of surrounding rock grade is a key challenge in the TBM construction of deep tunnels. This study aims to develop an automatic identification system based on TBM ascending section tunneling data to provide accurate guidance for stable section tunneling. A total of 6734 m of per-second TBM tunneling data was collected, and a preprocessing method for null and outlier values was proposed. The TBM complete tunneling cycle was divided into empty pushing, ascending, stable, and descending sections. Based on improved ICEEMDAN and HWPE, the T and F curves of the ascending section were decomposed, feature entropy values of IMF components were extracted, and an IMF component HWPE sample database was constructed. A Stacking ensemble framework was developed, and the Stacking-BIGRU model achieved identification accuracies of 95.0 %, 100.0 %, 97.5 %, and 90.0 % for grade II, IIIa, IIIb, and IV rock, respectively, with an overall accuracy of 95.625 %. Compared with traditional EMD and PE, the improved ICEEMDAN and HWPE enhanced the overall accuracy by 13.33 % and 7.75 %, respectively. The proposed system can effectively extract feature information from ascending section tunneling data, enabling accurate surrounding rock grade identification.

The scientific decision-making of TBM boring phase tunneling parameters is of great significance for ensuring safe and efficient TBM tunneling. This study proposes a short-impending decision-making process framework for TBM tunneling control based on loading phase data. Firstly, the Pearson correlation coefficient method was used to calculate the correlation between the tunneling parameters of the loading phase and the boring phase, and the thrust F, torque T, rotational speed N, and penetration p with the highest correlation were obtained as the input parameters for the loading phase; Then, Improved Symmetric Geometric Mode Decomposition (ISGMD) was used to decompose the input parameters of the TBM loading phase, and the Improved Symmetric Geometry Component (ISGC) with high correlation coefficient was obtained; Subsequently, Composite Multiscale Permutation Entropy (CMPE) was used to calculate the characteristic entropy values of the loading phase input parameters’ ISGCs, which were used as the loading phase input variables, and the surrounding rock grade was used as the geological condition constraint input variable, and the boring phase TBM tunneling parameters of the previous tunneling cycle were used as the time-series input variables to construct a TBM tunneling data sample library; Afterwards, the Osprey-Cauchy Sparrow Search Algorithm (OCSSA) was used to globally optimize the hyperparameters of the Improved Temporal Convolutional Network (ITCN) model, obtain the optimal model hyperparameters, and construct the OCSSA-ITCN model; Finally, based on the OCSSA-ITCN model, the stable tunneling parameters Fs, Ts, Ns, and ps of TBM were predicted. The average R2, MAPE, and RMSE of the predicted results were 0.9460, 6.25%, and 235.39, respectively, indicating high prediction accuracy. In addition, the influence of different input variables on the prediction accuracy of the OCSSA-ITCN model was discussed, as well as the improvement efficiency of the OCSSA and ITCN algorithms, and the enhancement performance of the ISGMD and CMPE models. The rationality of the proposed short-term decision-making framework was further validated using the engineering verification dataset. In summary, the proposed TBM tunneling control short-impending decision-making process framework has good engineering applicability and can provide scientific auxiliary decision-making for drivers to select TBM tunneling parameters.

Accurate prediction of TBM tunneling performance is crucial for improving construction efficiency. This paper proposes a fusion prediction method based on multimodal decomposition and multi-Deep Learning. First, tunneling data are preprocessed to build a sample database. Then, an improved ISTL model is developed to decompose tunneling performance into trend, seasonal, cycle, and residual components. Hyperparameters of multiple Deep Learning models are optimized using an improved IWOA algorithm, forming the ISTL-multi-DL model for preliminary prediction. Subsequently, error correction is applied to obtain the CISTL-multi-DL model, achieving MAPE values of 1.89 % and 1.43 % for FPI and TPI predictions, respectively. Comparative analysis shows that the CISTL-multi-DL model outperforms the IWOA-Autoformer, IWOA-Attention-LSTM, IWOA-BiTCN, and IWOA-DeepAR models by an average of over 40 %, and demonstrates superiority over unoptimized and traditional Machine Learning models. The proposed model provides accurate multi-step predictions and valuable support for TBM tunneling construction.

The real-time prediction of surrounding rock grade is of great significance to the safety and efficiency of TBM tunnel construction. Based on a water diversion project in China, the distribution patterns of nine tunneling parameters in different surrounding rock grades were studied using Boxplot. Then, Principal Component Regression (PCR) and Partial Least Squares Regression (PLSR) were used to reduce the dimension of the model input variables, and compared with the dimensionality reduction results of Boxplot to obtain the optimal model input variables. On this basis, five machine learning models for real-time identification of surrounding rock grade were constructed: Random Forest (RF) model, Error Back Propagation Neural Network (BPNN) model, Bayesian Network (BN) model, Support Vector Machine (SVM) model and K-Nearest Neighbor (KNN) model. Through the statistics of the prediction accuracy and running time of the five models, it was found that the prediction accuracy of RF model was the highest (85

The intelligent decision-making of tunnel boring machine (TBM) tunneling parameters is of great significance for the construction of deep buried long tunnels. Providing the TBM main drivers with the optimal quantitative construction scheme under different surrounding rock conditions can significantly improve the tunneling efficiency and geological adaptability of TBM, and reduce the excavation energy consumption of TBM. This study first removes outliers from TBM tunneling data based on the Local Outlier Factor (LOF) algorithm. Then, a TBM tunnel surrounding rock clustering and grading system considering tunneling performance was constructed using Deep Subspace Clustering (DSC) algorithm. On this basis, a strata identification model based on Bayesian Optimization (BO)-Extreme Gradient Boosting (XGBoost) was constructed, with an identification accuracy of 92.5

Inverse problems are typically tackled using deterministic optimization methods that may become trapped in a local minimum or probabilistic methods that can be computationally demanding. In this study, we explore the potential of the back propagation neural network (BPNN) optimized by the genetic algorithm (GA) for onshore transient electromagnetic (TEM) inversion. The GA is employed to optimize the initial parameters of the BPNN, enhancing its global optimization ability. Once the BPNN optimized by GA (GA-BPNN) is properly trained, it can provide the distribution of subsurface electrical conductivity (σ) in 0.1 s. We train the GA-BPNN using synthetic datasets generated by TEM forward modeling and assess its reliability using both synthetic and field data. Theoretical simulations demonstrate that compared with BPNN, the error of GA-BPNN on the inversion results of six samples is reduced by 23.2

Class-imbalanced is a common phenomenon in rockburst data, and the prediction of rockburst intensity through intelligent methods requires a balanced dataset. This fact presents challenges for standard classification algorithms that are designed for class distributions that are well-balanced. This paper develops the modified synthetic minority oversampling technique by K-means cluster (KM-SMOTE) to reduce the imbalance phenomenon in the rockburst dataset. First, the study collects 226 rockburst cases worldwide as the original supporting dataset and selects four indexes to predict the rockburst intensity, namely, the maximum tangential stress of the surrounding rock & sigma;& theta;, the uniaxial compressive strength of rock & sigma;c, the tensile strength of rock & sigma;t, and the elastic energy index Wet. Second, the KM-SMOTE uses a K-means cluster to cluster the minority-class samples and then performs SMOTE oversampling on each cluster to obtain 388 data. To establish a nonlinear correlation between rockburst intensity and its predictors, six machine-learning classifiers are used. The dataset is randomly divided into training and test sets, with 80% of the data used for training. In the data training and testing phases, the original dataset, SMOTE-processed dataset, and KM-SMOTE-processed dataset were put into the machine learning models for predicting rockburst intensity, where KM-SMOTE was 3.3% and 10.5% more accurate than the SMOTEprocessed dataset in predicting rockburst intensity, respectively. In the Jiangbian Hydropower Station engineering application, the KM-SMOTE algorithm can achieve a maximum improvement of 25% in accuracy compared with the data processed by SMOTE. Overall, the proposed modified oversampling algorithm effectively overcomes class-imbalanced in the rockburst dataset and significantly contributes to the intelligent prediction of rockburst by machine learning in engineering.

Rock mass classification is an essential basis for evaluating the stability of surrounding rock in tunnels and for determining reasonable construction methods and supporting parameters. The classification of surrounding rock mass in tunnels has a scale effect and advance demand. The grade of surrounding rock needs to be predicted in advance for different tunnel scales, so as to allow reaction time for support design optimization. Due to the concealment and uncertainty of parameter acquisition, the preliminary classification of surrounding rock in the tunnel exploration is quite different from the actual surrounding rock grade. This paper proposes a fine classification of surrounding rock in tunnel construction. Multiple geological methods such as geological mapping, ground-penetrating radar, tunnel seismic prediction, deepened blast holes, and advanced horizontal drilling are used to obtain the engineering geological response parameters and information relating to the surrounding rock in front of tunnel faces. Based on the sample database of surrounding rock in excavated sections, the influencing factors are analyzed and weighed. Finally, a prediction model for classifying the surrounding rock mass in unexcavated tunnel sections is established using extension theory and validated using actual engineering. Through the correction of the surrounding rock mass grades, the support optimization of the corresponding unexcavated tunnel sections has been completed, and the stability of the tunnel construction has been maintained. This study effectively improves the accuracy and engineering practicality of surrounding rock mass classification for tunnels and provides practical guidance for tunnel construction optimization.

Water inrush disaster in karst tunnels is still a challenging topic, and the reasonable thickness of water-resistant rock mass is of great significance in its prevention. This paper proposes a method for determining the thickness of the water-resistant rock mass and its corresponding application flow. First, we encapsulated the frequently used model to calculate the thickness, i.e., the strength model, catastrophe model, and shear model, and anatomized the corresponding factors governing the thickness. Subsequently, the discreteness and uncertainty of rock mass were contemplated by introducing the rock mass homogeneity degree (RMHD). A corresponding calculating process for RMHD was proposed to quantify the distribution and the probability of rock mass parameters with the Weibull function. After that, the quantified probability parameters were brought into the three models to get the thickness and its corresponding probability. The accuracy and applicability of the proposed method are verified and discussed by two typical engineering cases. The results show that the calculation result of the strength model is more in line with the actual reserved value. The thickness obtained by considering the probability parameters of rock mass are well consistent with the actual reserved values. Finally, suggestions for practical application are put forward to distinguish water inrush risk levels and effectively guide the construction as part of safety management.

Coal mining goaf is unconsolidated material formed by collapse of overlying strata. Water in goaf can flow into and contaminate surface water systems. An integrated geophysical investigation approach combining microseismic monitoring (MSM) and the opposing-coils transient electromagnetic (OCTEM) method was used to evaluate the effectiveness of grouting, using the Suncun Coal Mine as an example. First, MSM was used to obtain seven shear wave velocity depth profiles and 3D imaging, and the main low-velocity anomalies were delineated. Next, 12 resistivity depth profiles and 3D images were obtained by OCTEM to detect low-resistivity anomalies. Finally, the MSM and OCTEM results were verified by four boreholes drilled in the goaf areas. The results showed that the combination of MSM and OCTEM effectively identified the spatial distribution and locations of area with inadequate grouting and water-conducting channels. The geophysical results can be used for parameter selection and to design future grouting and water plugging programs.