High ground stress tunnel extrusion prediction and evaluation method and system
By optimizing the LightGBM model using SSA and combining it with tunnel and surrounding rock characteristic parameters, the problems of multi-classification accuracy and parameter optimization in high-stress tunnel compression prediction were solved, achieving high-precision tunnel compression prediction and engineering risk management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA RAILWAY FIRST SURVEY & DESIGN INST GRP
- Filing Date
- 2025-12-10
- Publication Date
- 2026-05-19
AI Technical Summary
Existing technologies suffer from low accuracy in multi-classification prediction of compression in high-stress tunnels, difficulty in parameter optimization, poor data adaptability, and a lack of in-depth analysis of the importance of compression influencing factors, which limits the safety and construction efficiency of tunnel projects.
The Sparrow Search Algorithm (SSA) is used to optimize the LightGBM model. Combined with tunnel and surrounding rock feature parameters, a multi-class prediction model is constructed. The K-nearest neighbor interpolation method and SMOTE technology are used to handle data missing and imbalance problems, so as to achieve global optimization and efficient training of the model.
It significantly improves the accuracy of tunnel compression prediction, realizes automated parameter adjustment and intelligent decision support, forms a closed-loop risk management system, and provides scientific engineering construction suggestions.
Smart Images
Figure CN122065145A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of tunnel engineering technology, specifically to a method and system for predicting and assessing compression in high-stress tunnels. Background Technology
[0002] With the continuous expansion of tunnel construction in the complex and challenging mountainous areas of western China, tunnel engineering is increasingly moving towards ultra-deep burial, ultra-long distances, and extreme environments. In areas with high ground stress and soft rock, the stress redistribution of the surrounding rock after tunnel excavation often exceeds its strength limit, triggering large-scale plastic flow and deformation, i.e., the problem of large compression deformation. The International Society for Rock Mechanics (ISRM) defines compression as "large time-dependent deformation" and "essentially related to creep caused by exceeding the ultimate shear stress." Tunnel compression not only leads to engineering problems such as support structure failure, lining cracking, and invert arch heave, but may also cause major safety accidents such as collapses and encroachment, seriously affecting the smooth progress of project construction. Therefore, accurately predicting the degree of compression in soft rock tunnels is of great significance for ensuring the safety, construction efficiency, and successful achievement of project schedule goals in tunnel engineering.
[0003] Currently, tunnel compression prediction methods mainly include empirical methods, semi-empirical methods, and theoretical numerical methods. However, these empirical methods can only serve as rough estimates in the early exploration stage, with limited prediction accuracy. With the development of numerical calculation methods, numerical simulation methods such as the finite element method and the discrete element method have been applied to study tunnel compression mechanisms. However, due to the difficulty in accurately obtaining geological information during construction, the reliability and accuracy of numerical simulation methods have considerable uncertainty.
[0004] In recent years, machine learning methods have made some progress in tunnel squeezing prediction, with models such as Support Vector Machine (SVM), Random Forest (RF), and XGBoost being applied to squeezing prediction tasks. However, existing research still has the following limitations:
[0005] (1) The accuracy of multi-class prediction (i.e., prediction of the severity of crushing) is low, which limits its value in engineering applications;
[0006] (2) Model parameter optimization relies on human experience and lacks systematic optimization methods;
[0007] (3) There is insufficient research on prediction methods under conditions of incomplete data;
[0008] (4) Lack of in-depth analysis of the importance of factors affecting extrusion.
[0009] Against this background, this invention proposes a method and system for predicting high-stress tunnel compression based on LightGBM optimized by the Sparrow Search Algorithm (SSA). SSA is an optimization algorithm capable of effectively solving parameter optimization problems in complex nonlinear problems. In recent years, SSA has demonstrated good optimization performance in geotechnical engineering fields such as slope stability prediction, dam deformation analysis, and TBM tunneling adaptability evaluation due to its advantages of simple structure and fast convergence speed. However, its application to multi-level classification prediction of tunnel compression has not yet been observed. LightGBM, as an efficient gradient boosting decision tree algorithm, excels at handling high-dimensional data and nonlinear relationships, and has great potential in the field of deformation prediction under complex geological conditions in tunnels and geotechnical engineering. This invention combines SSA and LightGBM to construct a prediction model based on SSA-LightGBM, which not only improves the model's prediction accuracy but also enhances its adaptability to complex geological conditions, forming a new, efficient, and robust intelligent prediction method for the degree of compression in high-stress tunnels, providing scientific basis and technical support for tunnel engineering design and construction. Summary of the Invention
[0010] This invention provides a method and system for predicting and evaluating the compression of tunnels under high ground stress, in order to solve the problems of low accuracy of multi-classification prediction, difficulty in parameter optimization, and poor data adaptability in the existing technology.
[0011] According to a first aspect, one embodiment provides a method for predicting and assessing high-stress tunnel compression, the method comprising:
[0012] Collect case data of high-stress tunnel engineering projects, extract characteristic parameters of tunnels and surrounding rock, and label them with compression grading labels to construct the original dataset;
[0013] The original dataset is preprocessed and divided into training and test sets proportionally.
[0014] A LightGBM multi-class prediction model was constructed, and the sparrow search algorithm was used to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination.
[0015] The LightGBM multi-class prediction model is configured with optimal hyperparameters, and the model is trained and tested using the training and test sets.
[0016] The tunnel and surrounding rock feature parameters of the tunnel section to be predicted are input into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction results.
[0017] Based on a pre-built risk response rule base, the tunnel compression classification prediction results are mapped to the corresponding early warning levels and engineering construction recommendations.
[0018] Furthermore, characteristic parameters of the tunnel and surrounding rock are extracted, specifically including:
[0019] The tunnel and surrounding rock characteristic parameters include the surrounding rock strength stress ratio SSR, tunnel burial depth H, rock mass quality index BQ, tunnel equivalent diameter D, and support structure stiffness K.
[0020] The strength-stress ratio (SSR) of surrounding rock is the ratio of the uniaxial compressive strength of the rock to the maximum principal stress. It is a core indicator for assessing the risk of crushing. The formula is as follows:
[0021]
[0022] in: The saturated uniaxial compressive strength of the rock. This represents the maximum ground stress.
[0023] Tunnel burial depth H: reflects the self-weight stress level of the overlying rock mass;
[0024] Rock mass quality index BQ: Calculated according to the "Engineering Rock Mass Classification Standard", it comprehensively reflects the integrity and strength of the rock mass. The formula is as follows:
[0025]
[0026] Where: Q is the Q value of the surrounding rock calculated according to the Q method;
[0027] Equivalent tunnel diameter D: Used to characterize the effect of tunnel cross-sectional dimensions, the formula is as follows:
[0028]
[0029] Where: A is the tunnel opening area;
[0030] Support structure stiffness K: reflects the initial support's ability to resist deformation, and the formula is as follows:
[0031]
[0032] in: The equivalent stiffness of shotcrete. For the equivalent stiffness of the steel frame, This is the equivalent stiffness of the anchor bolt.
[0033] Furthermore, the extrusion grading label includes:
[0034] The relative deformation ε is classified into four categories: 0—non-extrusion, 1—slight extrusion, 2—moderate extrusion, and 3—severe extrusion.
[0035] in, Corresponding label 0, Corresponding to label 1, Corresponding to label 2, For label 3, ε = maximum radial deformation of the tunnel e / tunnel excavation span D.
[0036] Furthermore, the original dataset is preprocessed and divided into training and test sets proportionally, specifically including:
[0037] Missing values were filled using the K-nearest neighbor imputation method, the class distribution was balanced using the SMOTE oversampling technique, and the dataset was divided into training and test sets in an 8:2 ratio using a stratified sampling strategy.
[0038] Furthermore, a LightGBM multi-class prediction model is constructed, specifically including:
[0039] Setting the objective function to "multiclass" explicitly tells the model that this is a multi-class classification task;
[0040] The evaluation metric is set to "multi_logloss", which is multi-class log loss;
[0041] The number of categories is set to 4, corresponding to four extrusion levels.
[0042] Furthermore, a sparrow search algorithm is used to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination, specifically including:
[0043] Evaluation metrics: The macro-average F1 score was selected as the core evaluation metric for model performance.
[0044] Validation strategy: Five-fold stratified cross-validation is used to evaluate the stability of a set of parameters and avoid randomness caused by different data partitions;
[0045] Objective function: Construct an evaluation function that takes the hyperparameter vector to be optimized as input, performs five-fold cross-validation internally, and returns a negative average macro F1 score;
[0046] Hyperparameters to be optimized and their range: Define key hyperparameters and their search boundaries;
[0047] The key hyperparameters and their search boundaries are as follows:
[0048] num_leaves: The number of leaf nodes in a tree, with a value / optimization range of [10, 100].
[0049] learning_rate: learning rate, with a value / optimization range of [0.01, 0.3];
[0050] feature_fraction: Feature sampling ratio, with a value / optimization range of [0.5, 1.0];
[0051] bagging_fraction: The sampling ratio, with a value / optimization range of [0.5, 1.0].
[0052] lambda_l1: L1 regularization, with a value / optimization interval of [0,10];
[0053] lambda_l2: L2 regularization, with a value / optimization interval of [0,10].
[0054] Furthermore, a sparrow search algorithm is used to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination, specifically including:
[0055] Algorithm initialization: Create a Sparrow Search algorithm optimizer instance, pass it to the evaluator function, and set the initial dimension n_dim, population size pop_size, maximum number of iterations max_iter, and the search boundary of the defined key hyperparameters;
[0056] Runtime optimization: The optimization process is started by calling the ssa.run() method. The sparrow search algorithm simulates the foraging and anti-predation behavior of sparrows and performs a global search in the parameter space.
[0057] Obtaining the optimal solution: After optimization, the optimal parameter set best_params that minimizes the objective function value is obtained.
[0058] Furthermore, the tunnel and surrounding rock characteristic parameters of the tunnel cross-section to be predicted are input into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction results, specifically including:
[0059] The tunnel and surrounding rock feature parameters of the tunnel section to be predicted are combined into a feature vector and input into the trained LightGBM multi-class prediction model. The output probability vector is then used to select the category index with the highest probability as the final predicted compression grading label through the argmax function.
[0060] Furthermore, based on a pre-built risk response rule base, the tunnel compression classification prediction results are mapped to corresponding early warning levels and engineering construction recommendations, specifically including:
[0061] The mapping relationship between tunnel compression classification prediction results, early warning levels, and engineering recommendations is as follows:
[0062] If the classification label is 0, the warning level is green, and it is recommended to carry out construction according to the standard procedure;
[0063] If the classification label is 1, the warning level is yellow, and it is recommended to carry out construction as designed and maintain routine monitoring.
[0064] If the classification label is 2, the warning level is orange, and it is recommended to strengthen the initial support stiffness, optimize the anchor bolt parameters, and maintain daily monitoring.
[0065] If the classification label is 3, the warning level is red, and it is recommended to take strong support measures, shorten the excavation advance by ≤0.5m, close the ring in time, and increase the frequency of deformation monitoring.
[0066] According to a second aspect, one embodiment provides a high-stress tunnel compression prediction and assessment system, the system comprising:
[0067] The dataset construction module is used to collect case data of high-stress tunnel engineering projects, extract characteristic parameters of tunnels and surrounding rock, and label them with extrusion grading labels to construct the original dataset;
[0068] The dataset processing module is used to preprocess the original dataset and divide it into training and test sets according to a certain ratio.
[0069] The model building and optimization module is used to build the LightGBM multi-class prediction model and uses the sparrow search algorithm to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination.
[0070] The model training and testing module is used to configure the LightGBM multi-class prediction model based on the optimal hyperparameters, and to train and test the model using the training set and the test set.
[0071] The prediction module is used to input the tunnel and surrounding rock feature parameters of the tunnel section to be predicted into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction results.
[0072] The graded early warning and assessment module is used to map the tunnel compression graded prediction results to the corresponding early warning level and engineering construction suggestions based on a pre-built risk response rule base.
[0073] According to a third aspect, one embodiment provides an electronic device, the device comprising: a processor and a memory;
[0074] The memory is used to store one or more program instructions;
[0075] The processor is configured to run one or more program instructions to perform the steps of a high-stress tunnel compression prediction and assessment method as described in any of the preceding claims.
[0076] According to a fourth aspect, one embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a high-stress tunnel compression prediction and assessment method as described in any of the preceding claims.
[0077] This invention provides a method and system for predicting and evaluating the compression of high-stress tunnels, which has the following beneficial effects:
[0078] 1. Significantly Improved Prediction Accuracy: Through fine-tuning of LightGBM using SSA, the model's performance on multi-class classification tasks has achieved a qualitative leap. Experiments show that its overall accuracy reaches 92.9%, and the recognition accuracy for each risk level exceeds 83%, far surpassing existing technologies.
[0079] 2. Achieve automated and intelligent parameter tuning: Completely eliminate the manual parameter tuning mode that relies on engineers' experience. SSA can automatically find the near-global optimal parameter combination in a short time, which greatly improves the efficiency and reliability of model development.
[0080] 3. Constructing a closed-loop risk management system: This invention is not only a prediction tool, but also a complete risk assessment decision support system. It transforms abstract prediction results into specific, executable engineering instructions, truly realizing a closed loop from "data" to "decision," and has extremely strong practical engineering value. Attached Figure Description
[0081] Figure 1 A flowchart illustrating a method for predicting and evaluating high-stress tunnel compression, as provided in one embodiment of the present invention;
[0082] Figure 2 This is a loss curve of the model training process in a high-stress tunnel compression prediction and evaluation method provided in one embodiment of the present invention;
[0083] Figure 3 This is a tunnel compression level prediction confusion matrix provided in a high-stress tunnel compression prediction and evaluation method according to an embodiment of the present invention;
[0084] Figure 4 This is a comparison chart of model indices in a high-stress tunnel compression prediction and evaluation method provided in one embodiment of the present invention;
[0085] Figure 5 This is a feature importance diagram in a high-stress tunnel compression prediction and evaluation method provided in one embodiment of the present invention. Detailed Implementation
[0086] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.
[0087] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.
[0088] The first embodiment of this invention provides a method for predicting and assessing the compression of high-stress tunnels. It significantly improves the prediction accuracy of the LightGBM model by automatically optimizing the parameters using the Sparrow Search Algorithm (SSA). The method also enhances the model's adaptability to incomplete data by employing KNN missing value imputation and SMOTE sample balancing algorithms. Furthermore, it achieves a four-level classification prediction of compression severity with an accuracy rate of 92.9%, significantly higher than traditional methods, providing a scientific basis for tunnel engineering safety. The specific theoretical basis is as follows:
[0089] Machine learning theory: LightGBM, as an efficient variant of GBDT, is theoretically based on approximating complex nonlinear functions by integrating multiple weak learners (decision trees). This perfectly matches the complex nonlinear phenomenon of tunnel squeezing, which is driven by multiple coupled factors.
[0090] Intelligent Optimization Theory: The Sparrow Search Algorithm (SSA) originates from the simulation of the foraging behavior of sparrow colonies and belongs to the metaheuristic optimization algorithm. Through a role-switching mechanism of discoverer-entrant-watcher, it achieves a good balance between global exploration and local exploitation, making it very suitable for solving high-dimensional, non-convex, and multi-modal hyperparameter optimization problems.
[0091] Engineering Geological Theory: The five input features (SSR, H, BQ, D, K) selected in this invention are all key control parameters clearly pointed out in authoritative documents such as the "Railway Tunnel Design Code" and the "Highway Tunnel Design Details", which ensures the physical meaning and engineering rationality of the model input.
[0092] The following is combined Figure 1 Please provide a detailed explanation.
[0093] like Figure 1 As shown, in step S100, high-stress tunnel engineering case data are collected, tunnel and surrounding rock characteristic parameters are extracted, and compression grading labels are marked to construct the original dataset.
[0094] This step aims to provide high-quality, labeled input data for model training. Specifically, the steps include:
[0095] S110, Data Source: Historical data collected from completed high-stress tunnel engineering projects. This data can be derived from authoritative sources such as geological survey reports, construction logs, and monitoring records.
[0096] S120, Feature Engineering: Extract five core input features (independent variables) from each case, which are the key factors affecting tunnel compression deformation:
[0097] Tunnel and surrounding rock characteristic parameters include the surrounding rock strength stress ratio (SSR), tunnel burial depth (H), rock mass quality index (BQ), tunnel equivalent diameter (D), and support structure stiffness (K).
[0098] The strength-stress ratio (SSR) of surrounding rock is the ratio of the uniaxial compressive strength of the rock to the maximum principal stress. It is a core indicator for assessing the risk of crushing. The formula is as follows:
[0099]
[0100] in: The saturated uniaxial compressive strength of the rock. This represents the maximum ground stress.
[0101] Tunnel burial depth H: reflects the self-weight stress level of the overlying rock mass;
[0102] Rock mass quality index BQ: Calculated according to the "Engineering Rock Mass Classification Standard", it comprehensively reflects the integrity and strength of the rock mass. The formula is as follows:
[0103]
[0104] Where: Q is the Q value of the surrounding rock calculated according to the Q method;
[0105] Equivalent tunnel diameter D: Used to characterize the effect of tunnel cross-sectional dimensions, the formula is as follows:
[0106]
[0107] Where: A is the tunnel opening area;
[0108] Support structure stiffness K: reflects the initial support's ability to resist deformation, and the formula is as follows:
[0109]
[0110] in: The equivalent stiffness of shotcrete. For the equivalent stiffness of the steel frame, This is the equivalent stiffness of the anchor bolt.
[0111]
[0112]
[0113] In the formula, This refers to the elastic modulus of concrete. R is the Poisson's ratio of the concrete; R is the tunnel radius. p is the thickness of the shotcrete; p is the radial support pressure of the steel frame; u is the radial deformation of the steel frame; The spacing of the anchor bolts along the circumference; The radial spacing of the anchor bolts. The diameter of the anchor rod. The elastic modulus of the anchor bolt; is the load-displacement constant.
[0114] S130, Label Definition: Based on the actual extrusion deformation observed on site, each case is classified into four levels according to the standard in Table 1: 0: No extrusion, 1: Slight extrusion, 2: Moderate extrusion, 3: Severe extrusion.
[0115] Table 1. Domestic and International Standards for Classification of Large Deformation
[0116]
[0117] Note: ε is the relative deformation, that is, the ratio of the maximum radial deformation e of the tunnel to the tunnel excavation span D; This is the ratio of strength stress to the surrounding rock of the tunnel.
[0118] S140, Output: The final output is a structured original dataset in the form of an N-row, 6-column table (CSV file), where N is the total number of cases, the first 5 columns are features, and the 6th column is the label (target).
[0119] like Figure 1 As shown, in step S200, the original dataset is preprocessed and divided into training set and test set according to the ratio.
[0120] This step aims to clean and augment the data, improving the stability and generalization ability of the model training. The specific steps include:
[0121] S210, Missing value handling:
[0122] Problem: Some feature values may be missing in the original dataset.
[0123] Solution: Use K-Nearest Neighbor (KNNImputer) imputation. Specifically, for a sample with a missing feature value, the algorithm finds the K (e.g., K=15) complete samples that are most similar to it in other features, and fills the missing value with the mean of these K neighbors in that feature.
[0124] Execution: Call the sklearn.impute.KNNImputer(n_neighbors=15) class for processing.
[0125] S220, Class Imbalance Handling:
[0126] Problem: In the real world, there are far fewer "severe squeeze" cases than "non-squeeze" cases, resulting in an imbalance in the distribution of classes in the dataset, and the model may be biased towards the majority class.
[0127] Solution: Employ the SMOTE (Synthetic Minority Over-sampling Technique). This technique balances the dataset by synthesizing new, reasonable minority class samples in the feature space of the minority class samples.
[0128] Execution: The `imblearn.over_sampling.SMOTE(random_state=42, k_neighbors=4)` class is called to process the samples, ensuring that the number of samples in each category is approximately equal.
[0129] S230, Dataset Partitioning:
[0130] Objective: To separate an independent test set to objectively evaluate model performance.
[0131] Method: A stratified sampling strategy was adopted, dividing the preprocessed dataset into a training set (X_train, y_train) and a test set (X_test, y_test) in an 8:2 ratio. Stratified sampling ensures that the proportion of samples of each class in the training and test sets remains consistent with that in the overall dataset.
[0132] Execution: Call the function sklearn.model_selection.train_test_split(..., stratify=y_resampled, test_size=0.2, random_state=42).
[0133] like Figure 1 As shown, in step S300, a LightGBM multi-class prediction model is constructed, and the sparrow search algorithm is used to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination.
[0134] This step builds the basic LightGBM multi-classifier framework. The specific steps include:
[0135] S310, Model Construction: LightGBM (Light Gradient Boosting Machine) was selected as the basic prediction model. LightGBM is an efficient implementation based on gradient boosting decision trees (GBDT), which has advantages such as fast training speed, low memory consumption, and support for categorical features.
[0136] Model configuration:
[0137] ① Objective function: Set to 'multiclass' to explicitly tell the model that this is a multi-class classification task.
[0138] ② Evaluation Metric: Set to 'multi_logloss', which is the standard loss function for multi-class classification tasks.
[0139] ③ Number of categories (num_class): Set to 4, corresponding to four compression levels.
[0140] ④ Other parameters: Initially, the default parameters of LightGBM can be used for preliminary training to provide a baseline for subsequent optimization.
[0141] S320, Predictive Model Optimization
[0142] This step uses the Sparrow Search Algorithm (SSA) to automatically find the optimal combination of hyperparameters for the LightGBM model.
[0143] (1) Optimization objective definition:
[0144] ① Evaluation Metric: The Macro F1-Score was chosen as the core evaluation metric for model performance. This metric assigns equal weight to the F1 score of each class and effectively measures the overall performance of the model on imbalanced multi-class classification tasks.
[0145] ② Validation strategy: Use 5-Fold Stratified Cross-Validation to evaluate the stability of a set of parameters and avoid randomness caused by different data partitioning.
[0146] ③ Objective function: Construct an evaluation function evaluator(params), whose input is the hyperparameter vector params to be optimized. Internally, it performs five-fold cross-validation and returns a negative average macro F1 score (because the optimization algorithm is usually to minimize the objective function).
[0147] ④ Hyperparameters to be optimized and their ranges: Define 6 key hyperparameters and their search boundaries (Lower Bound lb and Upper Bound ub), as shown in Table 2 below.
[0148] Table 2 LightGBM Hyperparameters and Search Range
[0149]
[0150] Given that the LightGBM model contains many hyperparameters, it is difficult to optimize all parameters individually. This embodiment comprehensively considers the impact of each parameter on model performance and selects six key parameters for optimization: learning rate, number of leaf nodes (num_leaves), feature fraction, bagging fraction, L1 regularization coefficient (lambda_l1), and L2 regularization coefficient (lambda_l2). Specifically, learning rate controls the speed of model weight adjustment; n_estimators are crucial to the model's predictive performance; and num_leaves controls the complexity and flexibility of the tree, preventing overfitting.
[0151] (2) SSA optimized execution:
[0152] ① Algorithm initialization: Create an SSA optimizer instance, passing in the evaluator function, dimension n_dim=6, population size pop_size=20, maximum number of iterations max_iter=50, and lb and ub as defined above.
[0153] ② Optimization: The optimization process is initiated by calling the ssa.run() method. SSA simulates the foraging and anti-predation behavior of sparrows, performing a global search within the parameter space.
[0154] ③ Obtain the optimal solution: After optimization, obtain the optimal parameter set best_params that minimizes the objective function value (i.e. maximizes the macro F1 score).
[0155] The hyperparameters of LightGBM were determined through the SSA algorithm and are shown in Table 3 below.
[0156] Table 3. LightGBM hyperparameters obtained via SSA
[0157]
[0158] like Figure 1 As shown, in step S400, the LightGBM multi-class prediction model is configured based on the optimal hyperparameters, and the model is trained and tested using the training set and test set.
[0159] The above steps specifically include:
[0160] S410, reconfigure the LightGBM model parameters using the values in best_params.
[0161] S420 trains the final SSA-LightGBM model on the complete training sets X_train and y_train, and uses the test sets X_test and y_test for early stopping to prevent overfitting.
[0162] The model is further trained using the above hyperparameters, and the loss function curve during the training process is shown in the figure. Figure 2 The model reached convergence after 70 training epochs. The model was validated on the test set, and the resulting confusion matrix is as follows. Figure 3 As shown.
[0163] like Figure 1 As shown, in step S500, the tunnel and surrounding rock feature parameters of the tunnel section to be predicted are input into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction result.
[0164] This step applies the trained model to a real-world engineering project and provides actionable risk assessment conclusions.
[0165] S510 combines the five characteristic parameters of the tunnel cross-section to be predicted into a feature vector.
[0166] S520 is input into the trained SSA-LightGBM model, and the model outputs a probability vector of length 4 [P(0), P(1), P(2), P(3)].
[0167] S530 uses the argmax function to select the category index with the highest probability as the final predicted label.
[0168] like Figure 1 As shown, in step S600, based on the pre-built risk response rule base, the tunnel compression classification prediction results are mapped to the corresponding early warning level and engineering construction suggestions.
[0169] The above steps specifically include:
[0170] In this embodiment, the system has a built-in risk response rule base that maps prediction labels to specific warning levels and engineering recommendations, as follows:
[0171] ① Predicted label = 3 (Severe crush):
[0172] Warning Level: Red (Highest Level)
[0173] The assessment recommends: "There is an extremely high risk of compression. It is recommended to immediately take strong support measures (such as adding steel arch frames and constructing arch supports), shorten the excavation advance (≤0.5m), close the ring in time, and increase the frequency of deformation monitoring."
[0174] ②Predicted label = 2 (moderate compression):
[0175] Warning level: Orange
[0176] Assessment recommendations: "There is a moderate risk of compression. It is recommended to strengthen the initial support stiffness, optimize the anchor bolt parameters, and maintain daily monitoring."
[0177] ③Predicted label = 1 (slight compression):
[0178] Warning level: Yellow
[0179] Assessment recommendation: "There is a slight risk of compression. It is recommended to construct according to the design and maintain routine monitoring."
[0180] ④ Predicted label = 0 (non-extrusion):
[0181] Warning Level: Green (Safe)
[0182] Assessment recommendation: "The risk of crushing is low, and construction can be carried out according to the conventional method."
[0183] Application example:
[0184] 1. Data preparation: Load the file named tun_dataset.csv, which contains 134 samples, 5 features (SSR, H, BQ, D, K) and 1 label (target).
[0185] 2. Pretreatment:
[0186] Use KNNImputer(n_neighbors=15) to fill in missing values.
[0187] Oversampling was performed using SMOTE(k_neighbors=4) to obtain the balanced dataset.
[0188] The test set was divided into a training set (107 samples) and a test set (27 samples) using the `train_test_split(test_size=0.2, stratify=y)` function.
[0189] 3. SSA optimization:
[0190] Define an evaluator function that performs 5-fold cross-validation and returns a negative macro F1 score.
[0191] Initialize SSA(evaluator, n_dim=6, pop_size=20, max_iter=50, lb=[...], ub=[...]).
[0192] Run ssa.run() to get the optimal parameters, for example: [45, 0.08, 0.75, 0.85, 1.2, 3.5].
[0193] 4. Final Training and Testing:
[0194] Train the final model using the optimal parameters.
[0195] Make predictions on the test set to obtain the predicted label y_pred.
[0196] Calculate and output the following metrics: Accuracy: 0.929, F1-Macro: 0.889, etc.
[0197] Generate a confusion matrix and classification report, and save the model file best_lgb_model.txt.
[0198] Practical engineering applications:
[0199] Scenario: Section YK12+350 of a highway tunnel under construction.
[0200] Input parameters: Based on the analysis of ground-penetrating radar and core drilling, the parameters of this section are: SSR=1.5, H=720m, BQ=280, D=11.8m, K=0.78.
[0201] Prediction process: Input the above parameters into the deployed SSA-LightGBM model.
[0202] Output: The model returns a probability vector [0.05, 0.10, 0.20, 0.65], with a predicted label of 3 (severe crush).
[0203] System Response: The system automatically triggered a red alert and displayed an assessment suggestion on the management platform: "This section has a serious risk of compression!"
[0204] Verification example:
[0205] Dataset: A dataset built based on 134 real tunnel engineering cases.
[0206] Evaluation metrics: Accuracy, F1-Macro score, and Kappa coefficient are used as the three comprehensive metrics.
[0207] Comparison Model: The model was compared with five mainstream models: Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), basic LightGBM, and XGBoost. The results are shown in Table 4. Figure 4 As shown.
[0208] Table 4 Performance of different models
[0209]
[0210] Experimental results:
[0211] Overall performance: The SSA-LightGBM model of this invention achieved an accuracy of 92.9% on the test set, and the F1-Macro score and Kappa coefficient were also the highest among all the comparison models.
[0212] Multi-classification details: In the more challenging multi-classification task, the model's prediction accuracy for each level was as follows: slight crush 94.11%, moderate crush 88.89%, and severe crush 83.33%. Crucially, the model successfully predicted all 41 real crush samples (labels 1, 2, and 3) as crush, without any missed detections (misclassifying crush as non-crushing), which is vital for engineering safety.
[0213] Feature Importance: Since machine learning models are usually regarded as black box systems and their prediction process lacks transparency, in order to improve the interpretability and credibility of the model results, this embodiment introduces the feature importance analysis method to interpret the SSA-LightGBM model. Figure 5 The importance ranking of each feature in the model is presented. SSR and BQ are far more important than other features, which is highly consistent with the core ideas of current tunnel design specifications, proving the rationality and credibility of the model's decision-making logic.
[0214] Corresponding to the above-disclosed method for predicting and assessing high-stress tunnel compression, this invention also discloses a system for predicting and assessing high-stress tunnel compression, which specifically includes:
[0215] The dataset construction module is used to collect case data of high-stress tunnel engineering projects, extract characteristic parameters of tunnels and surrounding rock, and label them with extrusion grading labels to construct the original dataset;
[0216] The dataset processing module is used to preprocess the original dataset and divide it into training and test sets according to a certain ratio.
[0217] The model building and optimization module is used to build the LightGBM multi-class prediction model and uses the sparrow search algorithm to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination.
[0218] The model training and testing module is used to configure the LightGBM multi-class prediction model based on the optimal hyperparameters, and to train and test the model using the training set and the test set.
[0219] The prediction module is used to input the tunnel and surrounding rock feature parameters of the tunnel section to be predicted into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction results.
[0220] The graded early warning and assessment module is used to map the tunnel compression graded prediction results to the corresponding early warning level and engineering construction suggestions based on a pre-built risk response rule base.
[0221] It should be noted that for a detailed description of the high-stress tunnel compression prediction and evaluation system provided in the embodiments of the present invention, please refer to the relevant description of the high-stress tunnel compression prediction and evaluation method provided in the embodiments of the present invention, which will not be repeated here.
[0222] In addition, embodiments of the present invention also provide an electronic device, the device comprising: a processor and a memory; the memory for storing one or more program instructions; the processor for running one or more program instructions to perform the steps of a high-stress tunnel compression prediction and evaluation method as described in any of the preceding embodiments.
[0223] It should be noted that for a detailed description of an electronic device provided in the embodiments of the present invention, please refer to the relevant description of a high-stress tunnel compression prediction and evaluation method provided in the embodiments of the present invention, which will not be repeated here.
[0224] In addition, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the high-stress tunnel compression prediction and evaluation method as described in any of the preceding claims.
[0225] It should be noted that for a detailed description of the computer-readable storage medium provided in the embodiments of the present invention, please refer to the relevant description of the high ground stress tunnel compression prediction and evaluation method provided in the embodiments of the present invention, which will not be repeated here.
[0226] Those skilled in the art will understand that all or part of the functions of the various methods in the above embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, which may include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to achieve the above functions. For example, the program can be stored in the memory of a device, and when the program in the memory is executed by the processor, all or part of the above functions can be achieved. In addition, when all or part of the functions in the above embodiments are implemented by computer programs, the program can also be stored in a server, another computer, disk, optical disk, flash drive, or external hard drive, etc., and can be downloaded or copied to the memory of a local device, or the system of the local device can be updated. When the program in the memory is executed by the processor, all or part of the functions in the above embodiments can be achieved.
[0227] The above examples illustrate the present invention only to aid in understanding it and are not intended to limit the scope of the invention. Those skilled in the art can make various simple deductions, modifications, or substitutions based on the principles of this invention.
Claims
1. A method for predicting and assessing the compression of high-stress tunnels, characterized in that, The method includes: Collect case data of high-stress tunnel engineering projects, extract characteristic parameters of tunnels and surrounding rock, and label them with compression grading labels to construct the original dataset; The original dataset is preprocessed and divided into training and test sets proportionally. A LightGBM multi-class prediction model was constructed, and the sparrow search algorithm was used to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination. The LightGBM multi-class prediction model is configured with optimal hyperparameters, and the model is trained and tested using the training and test sets. The tunnel and surrounding rock feature parameters of the tunnel section to be predicted are input into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction results. Based on a pre-built risk response rule base, the tunnel compression classification prediction results are mapped to the corresponding early warning levels and engineering construction recommendations.
2. The method for predicting and evaluating high-stress tunnel compression as described in claim 1, characterized in that, Extracting characteristic parameters of the tunnel and surrounding rock, specifically including: The tunnel and surrounding rock characteristic parameters include the surrounding rock strength stress ratio SSR, tunnel burial depth H, rock mass quality index BQ, tunnel equivalent diameter D, and support structure stiffness K. The strength-stress ratio (SSR) of surrounding rock is the ratio of the uniaxial compressive strength of the rock to the maximum principal stress. It is a core indicator for assessing the risk of crushing. The formula is as follows: in: The saturated uniaxial compressive strength of the rock. This represents the maximum ground stress. Tunnel burial depth H: reflects the self-weight stress level of the overlying rock mass; Rock mass quality index BQ: Calculated according to the "Engineering Rock Mass Classification Standard", it comprehensively reflects the integrity and strength of the rock mass. The formula is as follows: Where: Q is the Q value of the surrounding rock calculated according to the Q method; Equivalent tunnel diameter D: Used to characterize the effect of tunnel cross-sectional dimensions, the formula is as follows: Where: A is the tunnel opening area; Support structure stiffness K: reflects the initial support's ability to resist deformation, and the formula is as follows: in: The equivalent stiffness of shotcrete. For the equivalent stiffness of the steel frame, This is the equivalent stiffness of the anchor bolt.
3. The method for predicting and evaluating high-stress tunnel compression as described in claim 1, characterized in that, The marking of extrusion grading labels specifically includes: The relative deformation ε is classified into four categories: 0—non-extrusion, 1—slight extrusion, 2—moderate extrusion, and 3—severe extrusion. in, Corresponding label 0, Corresponding to label 1, Corresponding to label 2, For label 3, ε = maximum radial deformation of the tunnel e / tunnel excavation span D.
4. The method for predicting and evaluating high-stress tunnel compression as described in claim 1, characterized in that, The original dataset is preprocessed and divided into training and test sets proportionally, specifically including: Missing values were filled using the K-nearest neighbor imputation method, the class distribution was balanced using the SMOTE oversampling technique, and the dataset was divided into training and test sets in an 8:2 ratio using a stratified sampling strategy.
5. The method for predicting and evaluating high-stress tunnel compression as described in claim 1, characterized in that, The construction of the LightGBM multi-class prediction model specifically includes: Setting the objective function to "multiclass" explicitly tells the model that this is a multi-class classification task; The evaluation metric is set to "multi_logloss", which is multi-class log loss; The number of categories is set to 4, corresponding to four extrusion levels.
6. The method for predicting and evaluating high-stress tunnel compression as described in claim 1, characterized in that, The Sparrow Search algorithm is used to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination. Specifically, this includes: Evaluation metrics: The macro-average F1 score was selected as the core evaluation metric for model performance. Validation strategy: Five-fold stratified cross-validation is used to evaluate the stability of a set of parameters and avoid randomness caused by different data partitions; Objective function: Construct an evaluation function that takes the hyperparameter vector to be optimized as input, performs five-fold cross-validation internally, and returns a negative average macro F1 score; Hyperparameters to be optimized and their range: Define key hyperparameters and their search boundaries; The key hyperparameters and their search boundaries are as follows: num_leaves: The number of leaf nodes in a tree, with a value / optimization range of [10, 100]. learning_rate: learning rate, with a value / optimization range of [0.01, 0.3]; feature_fraction: Feature sampling ratio, with a value / optimization range of [0.5, 1.0]; bagging_fraction: The sampling ratio, with a value / optimization range of [0.5, 1.0]. lambda_l1: L1 regularization, with a value / optimization interval of [0,10]; lambda_l2: L2 regularization, with a value / optimization interval of [0,10].
7. The method for predicting and evaluating high-stress tunnel compression as described in claim 6, characterized in that, The Sparrow Search algorithm is used to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination. Specifically, this includes: Algorithm initialization: Create a Sparrow Search algorithm optimizer instance, pass it to the evaluator function, and set the initial dimension n_dim, population size pop_size, maximum number of iterations max_iter, and the search boundary of the defined key hyperparameters; Runtime optimization: The optimization process is started by calling the ssa.run() method. The sparrow search algorithm simulates the foraging and anti-predation behavior of sparrows and performs a global search in the parameter space. Obtaining the optimal solution: After optimization, the optimal parameter set best_params that minimizes the objective function value is obtained.
8. The method for predicting and assessing high-stress tunnel compression as described in claim 1, characterized in that, The tunnel and surrounding rock feature parameters of the tunnel cross section to be predicted are input into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction results, specifically including: The tunnel and surrounding rock feature parameters of the tunnel section to be predicted are combined into a feature vector and input into the trained LightGBM multi-class prediction model. The output probability vector is then used to select the category index with the highest probability as the final predicted compression grading label through the argmax function.
9. The method for predicting and evaluating high-stress tunnel compression as described in claim 3, characterized in that, Based on a pre-built risk response rule base, the tunnel compression classification prediction results are mapped to corresponding early warning levels and engineering construction recommendations, specifically including: The mapping relationship between tunnel compression classification prediction results, early warning levels, and engineering recommendations is as follows: If the classification label is 0, the warning level is green, and it is recommended to carry out construction according to the standard procedure; If the classification label is 1, the warning level is yellow, and it is recommended to carry out construction as designed and maintain routine monitoring. If the classification label is 2, the warning level is orange, and it is recommended to strengthen the initial support stiffness, optimize the anchor bolt parameters, and maintain daily monitoring. If the classification label is 3, the warning level is red, and it is recommended to take strong support measures, shorten the excavation advance by ≤0.5m, close the ring in time, and increase the frequency of deformation monitoring.
10. A system for predicting and assessing high-stress tunnel compression, characterized in that, The system includes: The dataset construction module is used to collect case data of high-stress tunnel engineering projects, extract characteristic parameters of tunnels and surrounding rock, and label them with extrusion grading labels to construct the original dataset; The dataset processing module is used to preprocess the original dataset and divide it into training and test sets according to a certain ratio. The model building and optimization module is used to build the LightGBM multi-class prediction model and uses the sparrow search algorithm to globally optimize the hyperparameters of the LightGBM multi-class prediction model to obtain the optimal hyperparameter combination. The model training and testing module is used to configure the LightGBM multi-class prediction model based on the optimal hyperparameters, and to train and test the model using the training set and the test set. The prediction module is used to input the tunnel and surrounding rock feature parameters of the tunnel section to be predicted into the trained LightGBM multi-class prediction model to obtain the tunnel compression classification prediction results. The graded early warning and assessment module is used to map the tunnel compression graded prediction results to the corresponding early warning level and engineering construction suggestions based on a pre-built risk response rule base.