A deep learning-based surrounding rock model generation method and system

Abstract

This invention discloses a method and system for generating surrounding rock models based on deep learning, relating to the field of tunnel engineering technology. First, geological data, exploration data, monitoring data, and historical engineering case data of the target area are acquired to construct a multi-dimensional dataset. After preprocessing, a three-dimensional surrounding rock generation model based on a neural network is established and trained. Then, real-time data of the area to be predicted is input into the model to generate a three-dimensional surrounding rock model. This invention can efficiently integrate multi-source data. Leveraging the powerful feature extraction and nonlinear modeling capabilities of deep learning, the generated three-dimensional surrounding rock model more accurately reflects the characteristics of the surrounding rock. It solves the problems of low data utilization, poor generalization ability, and difficulty in multi-source data fusion in traditional methods, providing more reliable data support for underground engineering design and construction.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering technology, and more specifically to a method and system for generating surrounding rock models based on deep learning. Background Technology

[0002] Currently, in underground engineering projects such as tunnels and mines, surrounding rock models are the core basis for assessing project stability and formulating support schemes. Traditional methods for generating surrounding rock models mainly rely on geological exploration data (such as borehole and sonic testing) and empirical formulas, which have the following drawbacks: 1. Low data utilization: Relying only on limited discrete point data, it is difficult to capture the spatial continuity and nonlinear distribution of surrounding rock parameters; 2. Poor model generalization ability: Insufficient adaptability to complex geological conditions (such as faults and fracture zones), requiring manual parameter tuning, which is inefficient; 3. Difficulty in multi-source data fusion: It is difficult to effectively integrate geological exploration and monitoring data (such as displacement and stress) and historical engineering cases, resulting in limited model accuracy. With the development of deep learning technology, it has demonstrated powerful feature extraction and nonlinear modeling capabilities in fields such as image recognition and spatiotemporal sequence analysis.

[0003] Therefore, how to provide a systematic deep learning framework for generating surrounding rock models, so as to efficiently and accurately construct digital models of the mechanical properties and structural morphology of surrounding rock, and provide data support for underground engineering design and construction, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of this, the present invention provides a method and system for generating surrounding rock models based on deep learning to solve the problems existing in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A deep learning-based method for generating surrounding rock models includes: Acquire geological data, exploration data, monitoring data, and historical engineering case data of the target area to construct a multi-dimensional dataset; Preprocess the collected multidimensional dataset; Establish a three-dimensional surrounding rock generation model based on neural networks; The neural network-based 3D surrounding rock generation model was trained using the preprocessed multi-dimensional dataset to obtain a trained neural network-based 3D surrounding rock generation model. Real-time data of the area to be predicted is input into a trained neural network-based 3D surrounding rock generation model to generate a 3D surrounding rock model.

[0006] Optionally, the preprocessing includes obtaining labeled data corresponding to the multi-dimensional dataset, wherein the labeled data represents the surrounding rock at the time when a set of data is collected.

[0007] Optionally, the establishment of the three-dimensional surrounding rock generation model based on neural networks specifically includes: Determine the number of nodes in a single hidden layer of the neural network model, select the activation function, and randomly initialize the input weights and biases of each hidden layer node to obtain the output matrix H of each hidden layer node; A similarity metric function is selected, and a graph Laplacian matrix L is constructed based on the matrix composed of the input data of the training samples; Selecting regularization parameters C 0. Calculate the penalty matrix C; where the penalty matrix C is a diagonal matrix, and each diagonal element corresponds to a training sample with labeled data, and its i-th element... i The element in row i and column i is C i = C 0 / N y , and when C i The labeled data in the corresponding training samples represent the surrounding rock at the time of data collection. N y The labeled data represents the number of training samples of the surrounding rock at the time of collection; According to the equation Solving for the output weights of the hidden layers yields the neural network model. Where I is the identity matrix, For the pre-selected manifold regularization parameters, ; ; in, , The input data for the neural network model is N, where N is the number of input data points. The number of hidden layer nodes. ( x ) represents the activation function of the hidden layer, and m represents the dimension of the input data. These are the input weights of each node in the hidden layer of the neural network. This represents the number of hidden layer nodes. , …… These represent the biases of each node in the hidden layer of the neural network; H This represents the output of each node in the hidden layer. The output weights of the hidden layer are obtained by solving; This is the output data of the neural network model.

[0008] Optionally, the method further includes using labeled data corresponding to the multi-dimensional dataset as multiple training samples to train a neural network model, thereby obtaining a three-dimensional surrounding rock generation model based on the neural network; wherein the multi-dimensional data in each training sample serves as the input data of the model, and the labeled data of the surrounding rock at the time of collection serves as the output data of the model. Optionally, the preprocessing also includes data cleaning and normalization: Data cleaning: Outliers in multi-dimensional data are identified and removed using box plots, and missing values ​​are filled using the K-nearest neighbor algorithm; Data normalization: The min-max normalization method is used to map data of different dimensions to the [0,1] interval. The formula is as follows:

[0009] in x This is the original data. x' For normalized data, , These represent the minimum and maximum values ​​of the data in this dimension, respectively. Optionally, the mean squared error is used as the loss function during model training, with the following formula: ; Where N is the sample size. To mark the true value of the data, The model predicts the value; the training iteration stops when the loss function value decreases by less than 1e-5 over 50 consecutive iterations, or when the number of iterations reaches a preset upper limit.

[0010] Optionally, the three-dimensional surrounding rock model includes the spatial distribution characteristics and mechanical parameter field of the surrounding rock, specifically including: Structural morphology: rock mass layering, joint distribution, fault location, and fracture zone extent in three-dimensional space; Mechanical parameter field: three-dimensional spatial distribution of elastic modulus, Poisson's ratio, cohesion, and internal friction angle; Model output format: Export as a BIM model or a visual mesh model, supporting engineers to view and analyze it intuitively.

[0011] A deep learning-based system for generating surrounding rock models includes: The data acquisition module obtains geological data, exploration data, monitoring data, and historical engineering case data of the target area to construct a multi-dimensional dataset; The data preprocessing module preprocesses the collected multi-dimensional datasets; The model building module establishes a three-dimensional surrounding rock generation model based on neural networks; The model training module uses the preprocessed multi-dimensional dataset to train the neural network-based 3D surrounding rock generation model, resulting in a trained neural network-based 3D surrounding rock generation model. The results output module takes the real-time data of the area to be predicted and inputs it into the trained neural network-based 3D surrounding rock generation model to generate a 3D surrounding rock model.

[0012] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a method and system for generating surrounding rock models based on deep learning. First, geological data, exploration data, monitoring data and historical engineering case data of the target area are acquired to construct a multi-dimensional dataset. After preprocessing, a three-dimensional surrounding rock generation model based on neural network is established and trained. Then, real-time data of the area to be predicted is input into the model to generate a three-dimensional surrounding rock model. The present invention can efficiently integrate multi-source data. With the powerful feature extraction and nonlinear modeling capabilities of deep learning, the generated three-dimensional surrounding rock model more accurately reflects the characteristics of the surrounding rock. It solves the problems of low data utilization, poor generalization ability and difficulty in multi-source data fusion of traditional methods, and provides more reliable data support for underground engineering design and construction. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0014] Figure 1 This is a schematic diagram of the method flow provided by the present invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] This invention discloses a method for generating surrounding rock models based on deep learning, such as... Figure 1 As shown, it includes: Acquire geological data, exploration data, monitoring data, and historical engineering case data of the target area to construct a multi-dimensional dataset; Preprocess the collected multidimensional dataset; Establish a three-dimensional surrounding rock generation model based on neural networks; The neural network-based 3D surrounding rock generation model was trained using the preprocessed multi-dimensional dataset to obtain a trained neural network-based 3D surrounding rock generation model. Real-time data of the area to be predicted is input into a trained neural network-based 3D surrounding rock generation model to generate a 3D surrounding rock model.

[0017] The construction of the multi-dimensional dataset includes: Geological data includes stratigraphic lithology distribution, geological structure maps (fault strike and dip), and hydrogeological parameters (permeability coefficient and groundwater level), which are derived from geological exploration reports and remote sensing interpretation data. Exploration data: geophysical data (seismic reflection waves, electromagnetic inversion results), drilling data (core compressive strength), etc. Monitoring data: Surrounding rock displacement (such as crown settlement and clearance convergence) and stress-strain data are collected by fiber optic sensors and total stations, with a sampling frequency of 1 time / hour; Historical engineering case data: classification results of surrounding rock, support parameters and stability assessment reports under similar geological conditions.

[0018] In one specific embodiment, preprocessing includes acquiring labeled data corresponding to the multi-dimensional dataset, wherein the labeled data represents the surrounding rock at the time when a set of data is collected.

[0019] In a specific embodiment, establishing a neural network-based three-dimensional surrounding rock generation model specifically includes: Determine the number of nodes in a single hidden layer of the neural network model, select the activation function, and randomly initialize the input weights and biases of each hidden layer node to obtain the output matrix H of each hidden layer node; A similarity metric function is selected, and a graph Laplacian matrix L is constructed based on the matrix composed of the input data of the training samples; Selecting regularization parameters C 0. Calculate the penalty matrix C; where the penalty matrix C is a diagonal matrix, and each diagonal element corresponds to a training sample with labeled data, and its i-th element... i The element in row i and column i is C i = C 0 / N y , and when C i The labeled data in the corresponding training samples represent the surrounding rock at the time of data collection. N y The labeled data represents the number of training samples of the surrounding rock at the time of collection; According to the equation Solving for the output weights of the hidden layers yields the neural network model. Where I is the identity matrix, For the pre-selected manifold regularization parameters, ; ; in, , The input data for the neural network model is N, where N is the number of input data points. The number of hidden layer nodes. ( x ) represents the activation function of the hidden layer, and m represents the dimension of the input data. These are the input weights of each node in the hidden layer of the neural network. This represents the number of hidden layer nodes. , …… These represent the biases of each node in the hidden layer of the neural network; H This represents the output of each node in the hidden layer. The output weights of the hidden layer are obtained by solving; This is the output data of the neural network model.

[0020] In one specific embodiment, the process further includes training a neural network model using labeled data corresponding to the multi-dimensional dataset as multiple training samples to obtain a three-dimensional surrounding rock generation model based on the neural network; wherein, the multi-dimensional data in each training sample serves as the input data of the model, and the labeled data of the surrounding rock at the time of collection serves as the output data of the model. In one specific embodiment, the preprocessing also includes data cleaning and normalization: Data cleaning: Outliers in multi-dimensional data are identified and removed using box plots, and missing values ​​are filled using the K-nearest neighbor algorithm; Data normalization: The min-max normalization method is used to map data of different dimensions to the [0,1] interval. The formula is as follows:

[0021] in x This is the original data. x' For normalized data, , These represent the minimum and maximum values ​​of the data in this dimension, respectively. In a specific embodiment, the mean squared error is used as the loss function during model training, with the formula: ; Where N is the sample size. To mark the true value of the data, The model predicts the value; the training iteration stops when the loss function value decreases by less than 1e-5 over 50 consecutive iterations, or when the number of iterations reaches a preset upper limit.

[0022] In one specific embodiment, the three-dimensional surrounding rock model includes the spatial distribution characteristics and mechanical parameter field of the surrounding rock, specifically including: Structural morphology: rock mass layering, joint distribution, fault location, and fracture zone extent in three-dimensional space; Mechanical parameter field: three-dimensional spatial distribution of elastic modulus, Poisson's ratio, cohesion, and internal friction angle; Model output format: Export as a BIM model or a visual mesh model, supporting engineers to view and analyze it intuitively.

[0023] The following example, using the generation of a surrounding rock model for a railway tunnel project in a mountainous area, will further illustrate the method of this invention.

[0024] Project Background

[0025] A 5.2km long railway tunnel in a mountainous area traverses a complex geological structure zone (including two faults and multiple densely jointed zones). A precise 3D surrounding rock model is required to guide the tunnel excavation and support design. The specific implementation process is as follows: Step 1: Constructing a multidimensional dataset The data acquisition module obtains four types of data from the target area, forming a multi-dimensional dataset: Geological data: taken from regional geological survey reports, including: stratigraphic lithology distribution (interbedded shale and sandstone, burial depth 50-300m); fault parameters (F1 fault strike 310°, dip angle 75°, F2 fault strike 280°, dip angle 60°); hydrogeological data (permeability coefficient 1.2×10⁻⁶). -5 (cm / s, groundwater level depth 20-50m).

[0026] Exploration data: Geophysical data: Rock mass wave velocity distribution inverted by seismic reflection wave method (1800-3500 m / s); Drilling data: Core compressive strength of 12 boreholes (25-80 MPa). Monitoring data: 10 fiber optic sensors were deployed to monitor surrounding rock displacement (daily variation of crown settlement 0.1-2 mm); 5 stress sensors were used to collect surrounding rock stress (0.5-3 MPa), with a sampling frequency of 1 time / hour for 30 days.

[0027] Historical engineering case data: The surrounding rock classification results (mainly Class IV-V), support parameters (shotcrete thickness 20-30cm) and stability reports of three tunnels with similar geological conditions in this area were selected.

[0028] Step 2: Dataset Preprocessing

[0029] The following operations are performed through the data preprocessing module: 1. Data label acquisition: Match the surrounding rock status label at the corresponding time for each set of data (e.g., the surrounding rock category at a depth of 20m for a certain borehole data is Class IV, and the surrounding rock displacement rate at the corresponding time for a certain monitoring data is 0.3mm / d).

[0030] 2. Data cleaning: Identify outliers using box plot method: Remove displacement anomalies in the monitoring data caused by sensor failure, such as "sudden increase of 10mm"; fill in the two missing core compressive strength values ​​in the drilling data using K-nearest neighbor algorithm (K=5).

[0031] 3. Data normalization: min-max normalization is used to map all data to the [0,1] interval.

[0032] Step 3: Establishment of 3D Surrounding Rock Generation Model

[0033] Build neural network models using the model building module: 1. Network structure design: The number of nodes in a single hidden layer is set to 30; the activation function is selected as the ReLU function; the input weights (range [-0.5, 0.5]) and biases (range [0, 0.1]) of the hidden layer are randomly initialized to obtain the output matrix H of the hidden layer.

[0034] 2. Construction of the Graph Laplacian Matrix L: Euclidean distance is selected as the similarity measure function. The adjacency matrix is ​​calculated based on the training sample input matrix (1200×8-dimensional, 8 feature dimensions), and then transformed into a graph Laplacian matrix L.

[0035] 3. Calculation of penalty matrix C: regularization parameter C 0=10, indicating the number of samples in the labeled data representing the "surrounding rock at the time of collection". N y =800, therefore the diagonal elements .

[0036] 4. Output weight calculation: Manifold regularization parameter λ = 0.01, substitute into the formula Solving for the output weights β yields the model. .

[0037] Step 4: Model Training

[0038] Training is performed using the model training module: Loss function: using mean squared error (N is the number of samples, here N=1200).

[0039] Training parameters: The maximum number of iterations is set to 500, and the learning rate is 0.001.

[0040] Stopping condition: When training reaches the 320th round, the loss function value decreases by less than 1e-5 for 50 consecutive rounds (from 0.0021 to 0.00209), thus meeting the stopping condition and completing the model training.

[0041] Step 5: Generation and Application of 3D Surrounding Rock Model

[0042] Input the real-time monitoring data (including the latest 3 days of displacement and stress data) of the area to be predicted (tunnel section K2+100-K2+300) into the trained model to generate a 3D surrounding rock model: Structural features: The section clearly shows the three layers of rock mass (shale layer 5-8m thick, sandstone layer 10-15m thick), two dense joint zones (with an angle of 30° to the tunnel axis), the specific location of the F1 fault in this section (3m to the left of the tunnel centerline), and the range of the fracture zone (20m long and 5m wide).

[0043] Mechanical parameter field: The three-dimensional spatial distribution of output elastic modulus (20-50GPa), Poisson's ratio (0.2-0.3), cohesion (0.5-2MPa), and internal friction angle (25°-35°), with the cohesion of the fault fracture zone as low as 0.5MPa.

[0044] Output format: Export as a BIM model. Engineers can view the model intuitively through BIM software and optimize the support scheme for this section accordingly (increasing the length of the anchor bolts in the fracture zone from 3m to 4.5m).

[0045] This embodiment integrates multi-source data and uses deep learning modeling to generate a three-dimensional surrounding rock model with smaller errors, providing accurate data support for tunnel construction safety.

[0046] A deep learning-based system for generating surrounding rock models includes: The data acquisition module obtains geological data, exploration data, monitoring data, and historical engineering case data of the target area to construct a multi-dimensional dataset; The data preprocessing module preprocesses the collected multi-dimensional datasets; The model building module establishes a three-dimensional surrounding rock generation model based on neural networks; The model training module uses the preprocessed multi-dimensional dataset to train the neural network-based 3D surrounding rock generation model, resulting in a trained neural network-based 3D surrounding rock generation model. The results output module takes the real-time data of the area to be predicted and inputs it into the trained neural network-based 3D surrounding rock generation model to generate a 3D surrounding rock model.

[0047] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0048] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating surrounding rock models based on deep learning, characterized in that, include: Acquire geological data, exploration data, monitoring data, and historical engineering case data of the target area to construct a multi-dimensional dataset; Preprocess the collected multidimensional dataset; Establish a three-dimensional surrounding rock generation model based on neural networks; The neural network-based 3D surrounding rock generation model was trained using the preprocessed multi-dimensional dataset to obtain a trained neural network-based 3D surrounding rock generation model. Real-time data of the area to be predicted is input into a trained neural network-based 3D surrounding rock generation model to generate a 3D surrounding rock model.

2. The method for generating a surrounding rock model based on deep learning according to claim 1, characterized in that, The preprocessing includes obtaining labeled data corresponding to the multi-dimensional datasets, wherein the labeled data represents the surrounding rock at the time when a set of data is collected.

3. The method for generating a surrounding rock model based on deep learning according to claim 1, characterized in that, The establishment of the three-dimensional surrounding rock generation model based on neural networks specifically includes: Determine the number of nodes in a single hidden layer of the neural network model, select the activation function, and randomly initialize the input weights and biases of each hidden layer node to obtain the output matrix H of each hidden layer node; A similarity metric function is selected, and a graph Laplacian matrix L is constructed based on the matrix composed of the input data of the training samples; Selecting regularization parameters C 0. Calculate the penalty matrix C; where the penalty matrix C is a diagonal matrix, and each diagonal element corresponds to a training sample with labeled data, and its i-th element... i Line number i The elements of the column are C i = C 0 / N y , and when C i The labeled data in the corresponding training samples represent the surrounding rock at the time of data collection. N y The labeled data represents the number of training samples of the surrounding rock at the time of collection; According to the equation Solving for the output weights of the hidden layers yields the neural network model. Where I is the identity matrix, For the pre-selected manifold regularization parameters, ； ； in, 、 The input data for the neural network model is N, where N is the number of input data points. The number of hidden layer nodes. ( x ) represents the activation function of the hidden layer, and m represents the dimension of the input data. These are the input weights of each node in the hidden layer of the neural network. This represents the number of hidden layer nodes. , …… These represent the biases of each node in the hidden layer of the neural network; H This represents the output of each node in the hidden layer. The output weights of the hidden layer are obtained by solving; This is the output data of the neural network model.

4. The method for generating a surrounding rock model based on deep learning according to claim 3, characterized in that, It also includes using labeled data corresponding to the multi-dimensional dataset as multiple training samples to train a neural network model and obtain a three-dimensional surrounding rock generation model based on the neural network; wherein, the multi-dimensional data in each training sample is used as the input data of the model, and the labeled data of the surrounding rock at the time of collection is used as the output data of the model.

5. The method for generating a surrounding rock model based on deep learning according to claim 1, characterized in that, The preprocessing also includes data cleaning and normalization: Data cleaning: Outliers in multi-dimensional data are identified and removed using box plots, and missing values ​​are filled using the K-nearest neighbor algorithm; Data normalization: The min-max normalization method is used to map data of different dimensions to the [0,1] interval. The formula is as follows: in x This is the original data. x' For normalized data, , These are the minimum and maximum values ​​of the data in this dimension, respectively.

6. The method for generating a surrounding rock model based on deep learning according to claim 1, characterized in that, The model training process uses mean squared error as the loss function, and the formula is as follows: ； Where N is the sample size. To mark the true value of the data, The model predicts the value; the training iteration stops when the loss function value decreases by less than 1e-5 over 50 consecutive iterations, or when the number of iterations reaches a preset upper limit.

7. The method for generating a surrounding rock model based on deep learning according to claim 1, characterized in that, The three-dimensional surrounding rock model includes the spatial distribution characteristics and mechanical parameter field of the surrounding rock, specifically including: Structural morphology: rock mass layering, joint distribution, fault location, and fracture zone extent in three-dimensional space; Mechanical parameter field: three-dimensional spatial distribution of elastic modulus, Poisson's ratio, cohesion, and internal friction angle; Model output format: Export as a BIM model or a visual mesh model, supporting engineers to view and analyze it intuitively.

8. A deep learning-based system for generating surrounding rock models, characterized in that, The method for generating a surrounding rock model based on deep learning according to any one of claims 1-7 includes: The data acquisition module obtains geological data, exploration data, monitoring data, and historical engineering case data of the target area to construct a multi-dimensional dataset; The data preprocessing module preprocesses the collected multi-dimensional datasets; The model building module establishes a three-dimensional surrounding rock generation model based on neural networks; The model training module uses the preprocessed multi-dimensional dataset to train the neural network-based 3D surrounding rock generation model, resulting in a trained neural network-based 3D surrounding rock generation model. The results output module takes the real-time data of the area to be predicted and inputs it into the trained neural network-based 3D surrounding rock generation model to generate a 3D surrounding rock model.

Images