Shield excavation face stratum identification method and system based on optimization classification feature selection

Abstract

The invention provides a shield excavation face stratum recognition method and system based on optimization classification feature selection, and relates to the technical field of shield tunnel construction. Shield construction data are collected and preprocessed, and an excavation face stratum recognition database is established; for class imbalance, performing feature importance sorting by using a weight-corrected maximum correlation minimum redundancy algorithm mRMR, selecting first m features corresponding to feature importance curve inflection points to form a roughing data set, and dividing core features, important features and edge features; establishing a multi-target feature selection framework through an NSGA-II-LightGBM algorithm based on a rough selection database with the minimum number of features, the maximum weighted F1 value and the maximum minority class recall rate as targets, outputting a Pareto leading solution, screening the most suitable individual in the Pareto leading solution based on a constraint screening-priority ranking-stability verification strategy, and establishing a fine selection data set; a Bayesian optimization algorithm is adopted to optimize the LightGBM classification model, and rapid and accurate prediction of the excavation face stratum in the shield construction process is achieved.

Description

Technical Field

[0001] This invention belongs to the field of shield tunnel construction technology, and particularly relates to a method and system for identifying the strata at the shield excavation face based on optimized classification feature selection. Background Technology

[0002] With advancements in construction technology and increasing construction requirements, tunnel construction is evolving towards longer distances, larger diameters, higher water pressures, and greater burial depths. Compared to the traditional drill-and-blast method, the shield tunneling method, with its advantages of safety, efficiency, and minimal impact on the surrounding environment, is widely used in tunnel construction. However, due to the increasing diameter and construction distance of the shields, ultra-large diameter shields face increasingly complex geological environments. Unfavorable geological conditions and structures not only increase construction costs and delay progress but can even lead to disasters such as instability at the excavation face.

[0003] Before construction, the spacing between adjacent test points in geotechnical investigation methods ranges from tens to hundreds of meters, resulting in insufficient information between test points and an inability to provide continuous geological information along the construction route. While geophysical exploration techniques such as resistivity methods and seismic wave methods can effectively detect geological information in tunnel sections, they often require additional equipment and necessitate that the tunnel boring machine (TBM) be in a stopped state. Therefore, they can negatively impact construction efficiency and increase construction costs.

[0004] During tunnel boring machine (TBM) construction, hundreds of construction and equipment parameters are collected. By analyzing massive amounts of data and selecting highly relevant parameters, a classification model is established using machine learning. This is an effective way to quickly classify and accurately identify the lithology of the strata ahead of the tunnel face during TBM construction. Establishing a scientifically sound database through data processing and feature selection is a prerequisite and necessary condition for using intelligent algorithms to solve stratum identification problems.

[0005] However, in the research on intelligent tunnel boring machine (TBM) construction (including geological formation identification, tunneling parameter prediction and optimization, etc.), existing studies largely rely on domain knowledge for parameter selection. Further, correlation analysis and multicollinearity analysis are added on top of domain knowledge, but this method only considers the linear relationships between selected input and output features, and between input features. Simple linear analysis cannot reveal the complex correlations between parameters and easily introduces redundant features, leading to overfitting. On the other hand, a combination of two parameters with low correlation can even produce classification results far exceeding those of the parameter with the highest correlation. In summary, parameter selection methods based on domain knowledge and correlation analysis are heavily influenced by researchers' subjectivity, making it difficult to uncover complex nonlinear relationships between parameters and easily filtering out key parameters with important information.

[0006] The uneven distribution of geological strata within the tunnel leads to significant differences in the number of samples across different strata. This imbalance in sample categories severely impacts the performance of classification models. Specifically, during training, the model tends to focus on the class with the larger proportion of samples, neglecting the minority class. This results in a substantial decrease in the recall and classification accuracy of the minority class, ultimately weakening the model's generalization ability and overall classification performance. Summary of the Invention

[0007] To overcome the shortcomings of the existing technologies, this invention provides a method and system for identifying the strata at the tunnel excavation face based on optimized classification feature selection. It adopts a feature selection method integrating mRMR and NSGA-II to achieve scientific parameter selection for the strata identification model, and finally establishes a strata classification and identification model for the tunnel excavation face based on LightGBM, so as to achieve rapid and accurate classification of the strata at the excavation face during construction.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: The first aspect of this invention provides a method for identifying the strata at the tunnel excavation face based on optimized classification feature selection.

[0009] The method for identifying the geological strata at the tunnel boring machine (TBM) excavation face based on optimized classification feature selection includes the following steps: The tunneling parameters, equipment parameters, and geological parameters during the shield tunneling process are collected to obtain the initial dataset; The maximum relevance minimum redundancy algorithm is used to rank the features in the initial dataset by importance, and the top m relevant features are selected. m is determined by the inflection point of the feature importance curve, which constitutes the coarse selection dataset. The NSGA-II multi-objective optimization algorithm is introduced to select features with the minimum number of features, the maximum weighted F1 value, and the maximum minority class recall. Based on the coarsely selected dataset, feature selection is performed to obtain the Pareto front solution. A combined strategy of constraint screening, priority ranking, and stability verification is adopted to select the most suitable individuals among the Pareto front solutions and construct the final refined dataset; Based on a carefully selected dataset, a Bayesian-optimized LightGBM stratum classification model was constructed to identify the strata at the shield tunneling face.

[0010] A second aspect of the present invention provides a stratum identification system for shield tunnel excavation faces based on optimized classification feature selection.

[0011] A shield tunneling face stratum identification system based on optimized classification feature selection includes: The data acquisition module is configured to collect tunneling parameters, equipment parameters, and geological parameters during the shield tunneling process to obtain an initial dataset. The coarse selection dataset construction module is configured to: use the maximum relevance and minimum redundancy algorithm to rank the features in the initial dataset by importance, select the top m relevant features, where m is determined by the inflection point of the feature importance curve, and form the coarse selection dataset; The multi-objective optimization module is configured to introduce the NSGA-II multi-objective optimization algorithm, with the objectives of selecting the fewest features, maximizing the weighted F1 value, and maximizing the minority class recall. Based on the coarsely selected dataset, feature selection is performed to obtain the Pareto front solution. The dataset selection module is configured to use a combination of constraint filtering, priority sorting, and stability verification to select the most suitable individuals among the Pareto front solutions and construct the final dataset selection. The excavation face stratum classification module is configured to: construct a Bayesian-optimized LightGBM stratum classification model based on a selected dataset to identify the strata at the shield tunnel excavation face.

[0012] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps in the shield tunneling face stratum identification method based on optimized classification feature selection as described in the first aspect of the present invention.

[0013] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the shield tunneling face stratum identification method based on optimized classification feature selection as described in the first aspect of the present invention.

[0014] The above one or more technical solutions have the following beneficial effects: 1. This invention provides a method and system for identifying the strata at a shield tunneling face based on optimized classification feature selection. It employs mRMR and NSGA-II methods to progressively select features and establish a lithology classification and identification model for the shield tunneling face, enabling rapid and accurate prediction of geological conditions. This improves upon the current situation in intelligent shield tunneling research where parameter selection relies solely on experience or correlation analysis, and is of great significance for adjusting and optimizing shield machine operating parameters. Furthermore, by predicting complex construction geology, corresponding engineering measures and preventive measures can be implemented to minimize potential construction risks.

[0015] 2. Shield tunneling construction parameters include tunneling parameters, geological parameters, equipment signals, etc., and the dimensions can reach hundreds or even thousands (let's say n dimensions), then we have: There are various feature combination methods, but the combinatorial explosion can lead to a surge in computational costs and make optimization difficult to implement for pure wrapper methods (such as using NSGA-II alone). mRMR (filtering) first coarsely filters the high-dimensional original features, significantly compressing the feature search space of NSGA-II (wrapping), avoiding the computational difficulties caused by the combinatorial explosion of high-dimensional features, and making multi-objective optimization feasible. However, if mRMR is used alone, it is difficult to explore the complementary effects between parameters. Therefore, a feature selection method that combines the two is adopted.

[0016] 3. Considering the imbalance in the stratigraphic classification database, a weighting factor is added to the mRMR mutual information calculation to strengthen the feature-strata correlation and reduce redundancy, laying a solid foundation for subsequent optimization. NSGA-II further optimizes the algorithm towards the goals of "minimum number of features, maximum weighted F1 score, and maximum minority class recall," fully exploring the complementary value of features to effectively improve the identification capability of minority class strata while further simplifying the feature scale. The combination of these two approaches provides a scientific and reusable paradigm for feature selection in tunnel boring machine intelligent algorithms.

[0017] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0019] Figure 1 This is a flowchart of a method according to an embodiment of the present invention.

[0020] Figure 2 This is a flowchart of data acquisition and preprocessing in Embodiment 1 of the present invention.

[0021] Figure 3 This is a flowchart of the process of constructing a coarse selection dataset based on mRMR in Embodiment 1 of the present invention.

[0022] Figure 4 This is a flowchart of the process of constructing a fine-tuned dataset based on NSGA-II in Embodiment 1 of the present invention.

[0023] Figure 5 This is a geological map of the Shantou Bay undersea tunnel, according to Embodiment 1 of the present invention. Detailed Implementation

[0024] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0026] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0027] Overall concept: This invention provides a method and system for identifying geological strata at a tunnel boring machine (TBM) excavation face based on optimized classification feature selection, wherein the method mainly includes: Step 1: Collect tunneling parameters, equipment parameters, and formation parameters, and perform data cleaning and normalization preprocessing; classify the construction strata into 8 categories according to different lithological combinations.

[0028] Step 2: Based on the preprocessed data, the maximum relevance minimum redundancy algorithm (mRMR) is used to rank the features in the dataset by importance, and the top m relevant features are selected. m is determined by the inflection point of the feature importance curve, thus forming the coarsely selected dataset.

[0029] Step 3: Construct a stratigraphic classification model combining NSGA-II and LightGBM. Introduce the NSGA-II multi-objective optimization algorithm to perform feature selection based on the coarsely selected dataset. The goal is to select the fewest features, maximize the weighted F1 score, and maximize the minority class recall. Experiments are conducted using the fast and efficient LightGBM classification model to obtain the Pareto front solution.

[0030] Step 5: Based on the research requirements of the stratigraphic classification task, and considering both computational efficiency and classification accuracy, a combined strategy of "constraint screening - priority ranking - stability verification" is adopted to select the most suitable individual from the Pareto front solutions as the final target feature combination. The LightGBM classification model is optimized using the Bayesian optimization algorithm, and the classification performance of the model is evaluated through multiple indicators.

[0031] This invention employs a feature selection method integrating mRMR and NSGA-Ⅱ to achieve scientific parameter selection for the stratum identification model, and finally establishes a stratum classification and identification model for shield tunnel excavation face based on LightGBM, enabling rapid and accurate classification of strata at the excavation face during construction.

[0032] Example 1 This embodiment discloses a method for identifying the strata at the tunnel boring machine (TBM) excavation face based on optimized classification feature selection. Based on mRMR and NSGA-II optimized feature selection, TBM construction data is collected and preprocessed to establish an original database matching the TBM construction data with the lithology of the excavation face strata. Due to class imbalance in the original database, the weighted maximum relevance minimum redundancy algorithm mRMR is used to rank the features in the dataset by importance. The top m features corresponding to the inflection points of the feature importance curves are selected to form a coarse-selected dataset, dividing it into core features, important features, and marginal features. Based on the coarse-selected database, with the objectives of minimizing the number of features, maximizing the weighted F1 score, and maximizing the minority class recall, a multi-objective feature selection framework is established using the NSGA-II-LightGBM algorithm, outputting Pareto front solutions. A constraint-priority ranking-stability verification strategy is used to select the most suitable individuals from the Pareto front solutions, establishing a refined dataset. A Bayesian optimization algorithm is used to optimize the LightGBM classification model, enabling rapid and accurate prediction of the lithology of the excavation face strata during TBM construction.

[0033] like Figure 1 As shown, the method for identifying the strata at the shield tunnel excavation face based on optimized classification feature selection includes the following steps: The tunneling parameters, equipment parameters, and geological parameters during the shield tunneling process are collected to obtain the initial dataset; The maximum relevance minimum redundancy algorithm is used to rank the features in the initial dataset by importance, and the top m relevant features are selected. m is determined by the inflection point of the feature importance curve, which constitutes the coarse selection dataset. The NSGA-II multi-objective optimization algorithm is introduced to establish a geological classification model for the LightGBM shield tunnel excavation face with the objectives of minimizing the number of features, maximizing the weighted F1 value, and maximizing the minority class recall rate, and to obtain the Pareto front solution. Based on the research needs of stratigraphic classification tasks, and combining computational efficiency and classification accuracy, a combined strategy of "constraint screening - priority ranking - stability verification" is adopted to screen the most suitable individuals among the Pareto front solutions and construct the final refined dataset. Based on a carefully selected dataset, a Bayesian-optimized LightGBM stratum classification model was constructed to identify the strata at the shield tunneling face.

[0034] This invention provides a method for identifying the geological strata at a shield tunnel excavation face based on optimized classification feature selection, comprising the following steps: Step 1: Data Acquisition and Preprocessing: Collect tunneling parameters, equipment parameters, and geological parameters, and perform data preprocessing and normalization preprocessing; taking the Shantou Bay Subsea Tunnel shield section as an example, the construction strata are divided into 8 categories according to different rock combinations.

[0035] Step 1.1 Data Acquisition: Referring to the geotechnical investigation report, collect shield tunneling construction parameters to establish a raw database containing tunneling parameters, equipment parameters, and geological parameters. Tunneling data mainly includes parameters directly reflecting the tunneling progress, such as cutterhead rotation speed, cutterhead torque, propulsion force, and propulsion speed. Equipment parameters reflect the mechanical equipment of the shield tunneling machine, such as propulsion cylinder stroke and cutterhead current consumption. Geological information mainly includes tunnel depth, geological type, surrounding rock grade, rock compressive strength, cohesion, and internal friction angle.

[0036] Step 1.2 Data Preprocessing: Based on the original database, non-tunneling state data deletion, missing value and outlier processing, and noise elimination are performed in sequence. First, missing values ​​are filtered out by null value detection, outliers are filtered out by box plot, and the mean of adjacent time points is used to correct outliers and missing values. Finally, noise elimination is achieved based on moving average filtering.

[0037] (1) Deletion of non-tunneling state data: The working state during shield tunnel construction is discontinuous. The shield tunneling and segment assembly are carried out in series. When the cutterhead or shield equipment malfunctions, the shield machine also needs to be stopped for inspection. Non-tunneling state data is difficult to reflect the nature and characteristics of the unexcavated strata ahead and does not provide substantial guidance for subsequent modeling. Shield data is collected continuously. The construction time consumed by the downtime states such as segment assembly and equipment failure is even no less than that of the forward tunneling. Introducing a large amount of non-tunneling state data will increase the calculation task of the model and have an adverse effect on the calculation efficiency and accuracy.

[0038] Through on-site investigation, the non-tunneling state of the tunnel boring machine can be determined by parameters such as propulsion speed, cutterhead rotation speed, and cutterhead torque. The state discrimination function is: (1) (2) (3) In the formula: v is the propulsion speed, n is the cutterhead rotation speed, T is the cutterhead torque, X is the original dataset, and G is the dataset after deleting the non-tunneling state; (2) Handling missing and outlier values: Missing values ​​are filtered out by null detection, and outlier data is identified by box plots. The box plot calculation formula is as follows: (4) (5) In the formula: The upper quartile of the box plot. The lower quartile of the box plot; Abnormal data is in In addition, the mean of adjacent time points is used to correct outliers and missing values.

[0039] (3) Noise elimination: During the shield tunneling process, the sensors will inevitably collect noise data. The noise data is identified and processed by the moving average filtering method.

[0040] (6) in, The filtering result at time t For the measurement data at time t, For the measurement data at time ti, Let t+i be the measurement data at time t+i, and n be the radius of the sliding window.

[0041] The above data is scaled to using the min-max normalization method. between: (7) In the formula, This is the result after outlier handling and noise reduction. This is the statistical minimum value of this feature. This is the statistical maximum value of this feature. This is the result after normalization.

[0042] Step 1.3 Data Labels: such as Figure 5 As shown, taking the Shantou Bay Subsea Tunnel shield tunnel section as an example, the construction section mainly contains 9 types of rocks. The rocks are labeled, such as rock label ① representing silt mixed with mud, and label ② representing mud and silty clay. Detailed information is listed in Table 1. Secondly, based on the results of geotechnical investigation, the strata along the tunnel are divided into 8 categories based on various rock combinations, and the strata are labeled. For example, stratum label 1 represents a stratum composed of two types of rocks: silt mixed with mud and mud and silty clay.

[0043] Table 1. Schematic diagram of rock labels and stratigraphic labels

[0044] Step 2: Construct a coarse dataset based on mRMR: Based on the preprocessed data, the maximum relevance minimum redundancy algorithm mRMR is used to rank the features in the dataset by importance, and the top m relevant features are selected to form a coarse dataset.

[0045] The stratigraphic feature selection method based on mRMR aims to balance maximizing the correlation between selected features and the research objective (stratigraphic category) while minimizing the correlation between selected features themselves (minimum redundancy). mRMR primarily determines the correlation between two variables through mutual information. The formula for calculating the mutual information between two continuous variables is as follows: (8) In the formula: Let x and y be the mutual information values ​​between random variables. These represent the density functions of variables x and y, and their joint density function, respectively.

[0046] Considering the imbalance of stratigraphic classification databases, the weighted correlation between feature x and stratigraphic category D is calculated by combining category weights. To avoid filtering out key features of a minority class; similarly, calculate the weighted redundancy among features. This avoids the characteristic that is "redundant for the majority class but complementary for the minority class" being misjudged as redundant.

[0047] Specifically, let the number of samples in a certain category c be... If the total number of samples is N and the total number of stratigraphic categories is K (K=8), then the unnormalized weight of category c is: After normalization, the final weight of category c is: The weighted correlation between feature x and stratigraphic category D is expressed as: ,feature , The weighted redundancy between them is expressed as .

[0048] Feature ranking using the mRMR method includes the following steps: (1) Define the feature set C to contain all features, initialize the selected feature set S as an empty set, and the feature set to be selected as CS; (2) Select features sequentially from feature set C. ,calculate Weighted mutual information between stratigraphic category D Select to make Reaching the maximum value As the first selected feature, let ,at this time, Candidate feature set ; (3) Assumption Selected features As candidate features, starting from the second round of feature selection, we search for features that satisfy... The features (q is the number of features in the selected feature set S); (4) Repeat step (3) until the candidate feature set CS is empty, and record the selected features in order each time; (5) Calculate the importance scores of all features based on the mRMR method, draw the feature importance inflection point graph (the horizontal axis is the feature and the vertical axis is the importance score), determine the curve inflection point where the downward trend of the importance score slows down, and record the number of features corresponding to the inflection point as m.

[0049] (6) Select the top m features by importance score as the coarse feature set. .Record The top 20% of features are considered core features, which are most strongly correlated with stratigraphic categories and can account for both the majority and minority categories; records 20%-70% of the features are important features; important features can help distinguish categories and have low redundancy; records The latter 30% of the features are marginal features, which contribute little to the classification and may be redundant.

[0050] Step 3: Constructing a refined dataset based on a multi-objective feature selection framework: Construct a stratigraphic classification model combining NSGA-II and LightGBM (NSGA-II-LightGBM), and use the NSGA-II multi-objective optimization algorithm to refine the coarse dataset. Feature selection is performed based on this. The goal is to maximize the weighted F1 score and the minority class recall while minimizing the number of features.

[0051] (1) Optimization objectives: The multi-objective optimization framework has three optimization objectives; Objective 1: Minimize the number of features; Objective 2: Maximize the weighted F1 score by calculating the weighted F1 score using the LightGBM classifier; (9) in, This represents the normalized weight of category c. This represents the precision of category c. This represents the recall rate for category c. In general classification problems, precision and recall are calculated using the following formulas: (10) (11) Table 2 Conceptual Table of Confusion Matrix

[0052] Objective 3: Maximize minority class recall (Let class A have the largest number of samples, and class A contains h samples. If the number of samples in a certain class is less than 0.4h, it is called the minority class. Let the total number of samples be h.) (minority classes), and calculate the minority class recall using the LightGBM classifier; (12) (2) Population initialization: The population size is set to 200. In order to ensure the quality of the initial population, 30% of the individuals are randomly selected to select 15 core features and 10-15 important features to ensure basic performance; 50% of the individuals are randomly selected to select 8-12 core features, 10-15 important features, and 1-5 marginal features, with a small number of marginal features explored to balance accuracy and diversity; 20% of the individuals are randomly selected to select 10-30 features to retain random exploration space and avoid local optima.

[0053] (3) Population evolution: New individuals are formed through selection, crossover and mutation operations. The crossover probability is set to 0.8, the mutation probability is set to 0.02, and the number of iterations is 200.

[0054] (4) Fitness assessment: For each feature subset (individual), a stratified sampling method is used to divide the dataset after the initial mRMR selection into a training set and a validation set in a 7:3 ratio. The LightGBM model is trained based on the training set, and the weighted F1 score and minority class recall are calculated on the validation set and used as fitness indicators to avoid overfitting due to evaluation with the full dataset. The classification ability of different individuals is tested using the LightGBM algorithm. When the number of individual features k≥5, the weighted F1 score and minority class recall are calculated according to equations (9) and (12). .

[0055] Feature quantity constraint and penalty: Set penalty trigger conditions. When the number of individual features k < 5, calculate the penalty coefficient according to the penalty formula (13). We employ a weighted F1 (Equation 14) that considers the penalty coefficient and a minority class recall (Equation 15) to avoid extreme minority feature subsets.

[0056] (13) (14) (15) (5) Termination conditions: When the average weighted F1 score of the Pareto front solutions fluctuates by less than 0.002 over 20 consecutive generations, and the minority class recall fluctuates by less than 0.003, convergence is considered achieved, and the iteration is terminated early (to reduce invalid computation). If convergence is not achieved, the iteration continues until generation 200. The Pareto front solution is output when the population reaches the maximum number of iterations or when the population fitness converges.

[0057] (6) Construct a refined dataset: Based on the research needs of stratigraphic classification tasks, and balancing computational efficiency and classification accuracy, a combined strategy of "constraint screening - priority ranking - stability verification" is adopted to select the most suitable individuals among the Pareto front solutions for constructing the final refined dataset. .

[0058] First, we set weighted F1 score ≥ 0.85, minority class recall ≥ 0.7, and feature count ≤ 30 as hard constraints to screen feasible solutions. Second, we screened 5-10 candidate solutions according to the priority order of minority class recall > weighted F1 score > feature count. We then used 5-fold cross-validation to analyze the stability of the candidate solutions (5-fold cross-validation divides the dataset into 5 mutually exclusive subsets, and alternately selects 4 subsets for training and 1 subset for validation, finally obtaining 5 sets of weighted F1 scores and minority class recalls). We selected the candidate solution with the largest mean and smallest standard deviation of "weighted F1 score and minority class recall" as the final most suitable feature subset.

[0059] Step 4: Divide the selected dataset into training, validation, and test sets in a 7:2:1 ratio, and establish a geological classification model for the shield tunneling face based on the LightGBM algorithm.

[0060] (1) Divide the dataset into smaller parts according to a ratio of 7:2:1. Using the training, validation, and test sets, the optimal hyperparameter combination for the LightGBM classification model is found using the Bayesian optimization algorithm. Table 3. Range of some hyperparameter values ​​for LightGBM

[0061] (2) Use the test set to evaluate the trained model and determine its generalization ability on new data. The classification effect of the model is evaluated by weighted F1 score, minority class recall, and accuracy. The formula for weighted F1 score is Equation (9), minority class recall is Equation (12), and accuracy is Equation (16).

[0062] (16) Example 2 This embodiment discloses a shield tunnel excavation face stratum identification system based on optimized classification feature selection.

[0063] A shield tunneling face stratum identification system based on optimized classification feature selection includes: The data acquisition module is configured to collect tunneling parameters, equipment parameters, and geological parameters during the shield tunneling process to obtain an initial dataset. The coarse selection dataset construction module is configured to: use the maximum relevance and minimum redundancy algorithm to rank the features in the initial dataset by importance, select the top m relevant features, where m is determined by the inflection point of the feature importance curve, and form the coarse selection dataset; The multi-objective optimization module is configured to introduce the NSGA-II multi-objective optimization algorithm, with the objectives of selecting the fewest features, maximizing the weighted F1 value, and maximizing the minority class recall. Based on the coarsely selected dataset, feature selection is performed to obtain the Pareto front solution. The dataset selection module is configured to: based on the research needs of the stratigraphic classification task, and combining computational efficiency and classification accuracy, use a combination strategy of "constraint screening - priority ranking - stability verification" to select the most suitable individuals among the Pareto front solutions and construct the final dataset selection; The excavation face stratum classification module is configured to: construct a Bayesian-optimized LightGBM stratum classification model based on a selected dataset to identify the strata at the shield tunnel excavation face.

[0064] Example 3 The purpose of this embodiment is to provide a computer-readable storage medium.

[0065] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the shield tunneling face stratum identification method based on optimized classification feature selection as described in Embodiment 1 of this disclosure.

[0066] Example 4 The purpose of this embodiment is to provide an electronic device.

[0067] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the shield tunneling face stratum identification method based on optimized classification feature selection as described in Embodiment 1 of this disclosure.

[0068] The steps and methods involved in the apparatuses of Embodiments 2, 3, and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0069] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0070] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for identifying geological strata at a shield tunneling face based on optimized classification feature selection, characterized in that, Includes the following steps: The tunneling parameters, equipment parameters, and geological parameters during the shield tunneling process are collected to obtain the initial dataset; The maximum relevance minimum redundancy algorithm is used to rank the features in the initial dataset by importance, and the top m relevant features are selected. m is determined by the inflection point of the feature importance curve, which constitutes the coarse selection dataset. The NSGA-II multi-objective optimization algorithm is introduced to select features with the minimum number of features, the maximum weighted F1 value, and the maximum minority class recall. Based on the coarsely selected dataset, feature selection is performed to obtain the Pareto front solution. A combined strategy of constraint screening, priority ranking, and stability verification is adopted to select the most suitable individuals among the Pareto front solutions and construct the final refined dataset; Based on a carefully selected dataset, a Bayesian-optimized LightGBM stratum classification model was constructed to identify the strata at the shield tunneling face.

2. The method for identifying the strata at a shield tunnel excavation face based on optimized classification feature selection as described in claim 1, characterized in that, Collect tunneling parameters, equipment parameters, and geological parameters during the shield tunneling process, specifically including: The tunneling data are parameters that directly reflect the construction and tunneling status, including cutterhead rotation speed, cutterhead torque, propulsion force, and propulsion speed; the equipment parameters are parameters that reflect the mechanical equipment of the tunnel boring machine, including propulsion cylinder stroke and cutterhead current consumption; the stratum information includes tunnel burial depth, stratum type, surrounding rock grade, rock compressive strength, cohesion, and internal friction angle information. or, It also includes preprocessing the acquired tunneling parameters, equipment parameters and formation parameters, including deleting non-tunneling state data, handling missing and outlier values ​​and eliminating noise in sequence; or, This also includes classifying and labeling the geological strata along the tunnel route: The rocks in the construction area are classified into several categories, including silt and mud, mud and silty clay, silty clay, medium silty sand, medium sand with silt, fine gravelly soil, completely weathered granite, strongly weathered granite and weakly weathered granite. By using various rock categories, the types of strata are identified and tagged. The tagged rocks can be one rock category or a combination of multiple rock categories.

3. The method for identifying the strata at a shield tunnel excavation face based on optimized classification feature selection as described in claim 1, characterized in that, The specific construction process of the coarsely selected dataset is as follows: Step 1: Define the feature set C to contain all features in the initial dataset, initialize the selected feature set S as an empty set, and the candidate feature set as CS; Step 2: Select features sequentially from feature set C. ,calculate Weighted mutual information between stratigraphic category D Select to make Reaching the maximum value As the first selected feature, let ,at this time, Candidate feature set ; Step 3: Assumption Selected features As candidate features, starting from the second round of feature selection, we search for features that satisfy... The feature is q, which is the number of features in the selected feature set S; Step 4: Repeat Step 3 until the candidate feature set CS is empty, and record the selected features in order each time; Step 5: Calculate the importance scores of all features based on the mRMR method, plot the feature importance curve, and determine the inflection point m where the downward trend of the importance score slows down; Step 6: Select features whose importance scores precede the inflection point m on the curve as the initial feature set. .

4. The method for identifying the strata at a shield tunnel excavation face based on optimized classification feature selection as described in claim 3, characterized in that, The weighted correlation between feature x and stratigraphic category D is expressed as: ； in, The weight of stratigraphic category c; This represents the mutual information between feature x and stratigraphic category c; This represents the weighted correlation between feature x and stratigraphic category D; ； in, The weight before normalization is the weight of stratigraphic category c.

5. The method for identifying the strata at a shield tunnel excavation face based on optimized classification feature selection as described in claim 1, characterized in that, The NSGA-II multi-objective optimization algorithm is introduced, aiming to minimize the number of features selected, maximize the weighted F1 score, and maximize the minority class recall. Feature selection is performed on a coarsely selected dataset to obtain the Pareto front solution, which includes: There are three optimization objectives in designing a multi-objective optimization problem: minimizing the number of features selected, maximizing the weighted F1 score, and maximizing the minority class recall. Population initialization: Based on different selections of core features, important features, and marginal features, individuals at different levels are set; Population evolution: the formation of new individuals through selection, crossover, and mutation. Fitness evaluation: For each individual, a stratified sampling method is used to divide the coarse dataset into a training set and a validation set. The LightGBM model is trained based on the training set, and the weighted F1 score and minority class recall are calculated on the validation set and used as fitness metrics. Termination condition: The convergence condition is set based on the weighted F1 score and the minority class recall rate. When the population reaches the convergence condition, that is, when the maximum number of iterations or the population fitness converges, the Pareto front solution is output.

6. The method for identifying the strata at a shield tunnel excavation face based on optimized classification feature selection as described in claim 5, characterized in that, The specific construction process of the selected dataset includes: First, based on the weighted F1 score, minority class recall rate, and number of features, hard constraints are set to select feasible solutions from the Pareto front solutions; Secondly, a set number of candidate solutions are selected from the feasible solutions according to the priority order of minority class recall rate > weighted F1 score > number of features; Five-fold cross-validation was used to analyze the stability of candidate solutions. The candidate solutions with the largest weighted F1 score, the largest mean minority class recall, and the smallest standard deviation were selected as the refinement dataset.

7. The method for identifying the strata at a shield tunnel excavation face based on optimized classification feature selection as described in claim 5, characterized in that, The core features, important features, and marginal features are all obtained based on a coarsely selected dataset: The core features are the top a% of the features in the coarsely selected dataset; The key features are those that represent a%-b% of the coarsely selected dataset; The edge features are b% of the features after the coarse selection of the dataset; Both a% and b% are set values; or, The formula for calculating the weighted F1 score is: ； in, This represents the precision of category c. This represents the recall rate for category c; The formula for calculating the recall rate of minority classes is: ； Let category A have the largest number of samples, containing h samples. If the number of samples in a certain category is less than p*h, it is denoted as the minority class, where p is the critical proportion of the minority class, and the value of p ranges from (0, 1). Let there be a total of A minority category.

8. A shield tunneling face stratum identification system based on optimized classification feature selection, characterized in that, include: The data acquisition module is configured to collect tunneling parameters, equipment parameters, and geological parameters during the shield tunneling process to obtain an initial dataset. The coarse selection dataset calculation module is configured to: use the maximum relevance and minimum redundancy algorithm to rank the features in the initial dataset by importance, select the top m relevant features, where m is determined by the inflection point of the feature importance curve, and form the coarse selection dataset; The multi-objective optimization module is configured to introduce the NSGA-II multi-objective optimization algorithm, with the objectives of selecting the fewest features, maximizing the weighted F1 value, and maximizing the minority class recall. Based on the coarsely selected dataset, feature selection is performed to obtain the Pareto front solution. The dataset selection module is configured to use a combination of constraint filtering, priority sorting, and stability verification to select the most suitable individuals among the Pareto front solutions and construct the final dataset. The excavation face stratum classification module is configured to: construct a Bayesian-optimized LightGBM stratum classification model based on a selected dataset to identify the strata at the shield tunnel excavation face.

9. A computer-readable storage medium having a program stored thereon, characterized in that, When executed by the processor, the program implements the steps in the shield tunnel excavation face stratum identification method based on optimized classification feature selection as described in any one of claims 1-7.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the shield tunnel excavation face stratum identification method based on optimized classification feature selection as described in any one of claims 1-7.

Images