A tunnel rockburst prediction method based on numerical simulation and neural network

Abstract

The present invention provides a tunnel rockburst prediction method based on numerical simulation and neural network, comprising the following steps: collecting relevant geological parameter data and TBM operation parameters, establishing a database; setting the element parameters of the tunnel model, dividing the grid and adding boundary conditions, and using the above-collected relevant geological parameter data for numerical simulation; introducing damage and fracture mechanics models, performing numerical calculations and simulating the excavation process; training the neural network by simulating the operation parameters of the TBM during the excavation process; and using the trained neural network to output the probability and range of rockburst in actual excavation. The present invention can continuously output prediction results, thereby providing real-time safety warnings for tunnel construction.

Description

Technical Field

[0001] The invention belongs to the technical field of rock burst detection, and in particular relates to a tunnel rock burst prediction method based on numerical simulation and neural network. Background Art

[0002] Rockburst is a common geological disaster in underground engineering. Due to its suddenness and strong destructiveness, it has attracted more and more attention. Because of the huge damage caused by rockburst to personnel and facilities, how to accurately and effectively predict the occurrence of rockburst and propose scientific and effective solutions according to specific circumstances has become an urgent problem to be solved. The current rockburst research methods include: theoretical research, field detection numerical simulation and indoor experiments. Theoretical research cannot thoroughly and effectively reflect rockburst disasters because it relies on theoretical derivation for analysis. Although field detection is the best means to study rockburst, it is expensive, has a high risk of personnel safety and equipment loss, and the number of research projects is limited, so it cannot reflect various geological problems. Although indoor experiments can conveniently set up experimental conditions and instruments, they cannot accurately simulate on-site excavation conditions. The prediction method that is in line with the actual situation and adds reasonable parameters based on numerical simulation is more effective. There are many factors affecting rockburst, including geological conditions, physical and mechanical properties of rocks, and ground stress levels. Many quantitative judgment bases of the rockburst prediction methods currently used are based on one of them for prediction, lacking a systematic, objective and complete prediction method. Summary of the invention

[0003] The purpose of the invention is to provide a tunnel rockburst prediction method based on numerical simulation and neural network.

[0004] The technical solution of the present invention is as follows:

[0005] A tunnel rockburst prediction method based on numerical simulation and neural network includes the following steps:

[0006] Collect relevant geological parameter data and TBM operation parameters and establish a database;

[0007] Set the element parameters of the tunnel model, divide the grid and add boundary conditions, and use the relevant geological parameter data collected above to perform numerical simulation;

[0008] Introducing damage and fracture mechanics models, performing numerical calculations and simulating the excavation process;

[0009] The neural network is trained by simulating the operating parameters of the TBM during excavation;

[0010] The trained neural network is used in actual tunneling to output the probability and range of rockburst.

[0011] Furthermore, the relevant geological parameter data include geological conditions of the rock mass, rock layer structure, fault distribution and groundwater conditions; the operating parameters of the TBM include cutter head torque, propulsion force and rotation speed.

[0012] Furthermore, the element parameters of the tunnel model include Poisson's ratio, elastic modulus, density, friction angle, flow stress ratio, expansion angle, uniaxial tensile strength and compressive strength of rock.

[0013] Furthermore, the damage and fracture mechanics model describes the degree of damage of the rock mass by introducing a damage variable D, wherein the relationship between the change rate of D and the equivalent stress and damage threshold stress of the rock mass is:

[0014] ,

[0015] in, represents the time change rate, 1-D represents the proportion of the undamaged part of the rock mass, Indicates the extent to which the equivalent stress exceeds the damage threshold stress.

[0016] Furthermore, when training the neural network, a convolutional neural network is used to extract features from the exploration images, and after splicing the TBM operating parameters with the image features, the probability and impact range of the rockburst are output through the fully connected layer.

[0017] Furthermore, during the actual tunneling process, the TBM operating parameters and the forward exploration map are used as inputs to the trained neural network, which outputs the probability and range of rockburst.

[0018] Furthermore, the rockburst probability is calculated by a Sigmoid activation function, the impact range of the rockburst is calculated by a regression network, and the impact radius or specific coordinates of the dangerous area are output.

[0019] Furthermore, dynamic learning rate adjustment and optimization algorithms are used in the training process of the neural network to improve the convergence speed and prediction accuracy of the network, and the generalization ability of the network is verified by a cross-validation method.

[0020] Compared with the prior art, the present invention has the following advantages:

[0021] The present invention provides a dynamic method that can accurately predict the probability and scope of rock bursts in tunnels by combining numerical simulation with neural networks. First, numerical simulation is used to analyze the rock mass response during tunnel excavation, and the probability and scope of rock bursts are calculated by combining geological parameters and TBM operating parameters. Then, based on the data in the simulated excavation process and the rock burst prediction results, the neural network model is trained to achieve real-time prediction of the probability and scope of rock bursts. During the actual tunnel excavation process, the trained neural network can continuously output prediction results, thereby providing real-time safety warnings for tunnel construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the description and claims, serve to illustrate the embodiments of the invention. Where appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0023] Figure 1 A schematic diagram of the process of the present invention is shown;

[0024] Figure 2 A schematic diagram of the neural network flow of the present invention is shown. DETAILED DESCRIPTION

[0025] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0026] like Figure 1-Figure 2 As shown, the present invention provides a tunnel rockburst prediction method based on numerical simulation and neural network, which specifically includes the following steps:

[0027] Step 1: Obtain and collect relevant geological parameter data, collect basic parameters of TBM, establish a database, and conduct preliminary screening;

[0028] Specifically, relevant geological parameter data from various previous cases are collected, including the geological conditions of the rock mass, the mechanical properties of the rock mass, the rock layer structure, the fault distribution, the groundwater conditions, etc.

[0029] The basic parameters of TBM, including specific characteristics such as cutterhead torque and total thrust, were collected to establish a database.

[0030] The database is constructed based on various past rockburst occurrence cases at home and abroad and the engineering cases of the initial condition groups at that time.

[0031] Step 2: Setting the element parameters of the tunnel model, dividing the grid and adding boundary conditions, and setting the tunnel conditions for numerical simulation using the data of the relevant geological parameters collected above;

[0032] The inputs of the constructed numerical simulation are Poisson’s ratio, elastic modulus, density, friction angle, flow stress ratio, expansion angle, and uniaxial tensile and compressive strength of rock.

[0033] Step 3: Introduce damage and fracture mechanics models, perform numerical calculations and simulate the excavation process

[0034] The damage and fracture mechanics model can comprehensively consider the integrity of the rock mass, the extension of microcracks inside the rock, the evolution of damage, the degradation of the mechanical properties of the rock, etc., mark the rock burst data, and calculate the probability and range of rock burst.

[0035] Through dynamic iteration to simulate various situations of tunnel excavation, the specific formula is as follows:

[0036] The damage variable D is defined as a measure of the reduction in rock mass stiffness. When the rock mass is undamaged, D=0; when the rock mass is completely damaged, D=1.

[0037] Change rate of damage variable D Equivalent stress to rock mass and damage threshold stress related.

[0038] Equivalent stress It is a quantity that comprehensively considers stresses in different directions in the rock mass and is usually given by the following formula: ,

[0039] in, is the principal stress.

[0040] Damage Threshold Stress It is the stress threshold at which rock mass begins to be damaged, which depends on the properties of the rock mass.

[0041] The change rate of the damage variable D is proportional to the part of the equivalent stress of the rock mass that exceeds the damage threshold stress. Specifically:

[0042] ,

[0043] here, represents the time change rate, 1-D represents the proportion of the undamaged part of the rock mass, Indicates the extent to which the equivalent stress exceeds the damage threshold stress.

[0044] when hour, , which means that the rock mass is not damaged.

[0045] when hour, , which means that the rock mass begins to be damaged and the degree of damage increases with time.

[0046] Step 4: Use numerical simulation to train the neural network

[0047] The neural network model inputs two features, namely, the cutter head torque and the total thrust, and inputs the exploration image in front of the tunnel face into a convolutional neural network (CNN).

[0048] The specific implementation steps are as follows:

[0049] 1. Feature input and processing:

[0050] Input of cutter head torque and total thrust: First, the cutter head torque and total thrust are extracted as numerical input data. These data will be normalized (such as Z-score normalization) to ensure that the numerical range of the input features is consistent to avoid the model being too sensitive to a certain feature during training.

[0051] Exploration images in front of the tunnel face: Exploration images in front of the tunnel face (such as underground laser scanning images, geological radar images, ultrasonic images, etc.) are used to describe the geological structure characteristics of the tunnel face and the area in front of it. These images are pre-processed (such as denoising, cropping, normalization, etc.) and then input into the neural network model together with numerical features.

[0052] 2. Convolutional Neural Network (CNN) Structure:

[0053] After the exploration image is input into CNN, it passes through several convolution layers for feature extraction. The convolution kernel of each convolution layer will gradually extract the spatial features in the image (such as cracks, abnormal rock formations, rock mass fragmentation, etc.) and deep high-level features (such as local geological unstable areas, potential rock burst source points, etc.). Each convolution operation will generate a set of convolution feature maps, which are reduced in dimension and compressed in features through the pooling layer.

[0054] In the process of image feature extraction, the complex geometric structure of the rock strata in front of the tunnel face and potential rock burst triggering factors are captured by setting the convolution kernel size, step size and pooling method.

[0055] 3. Feature concatenation and vectorization:

[0056] The extracted image features (feature maps output by the convolution layer) and numerical features (cutter torque and total thrust) are combined through feature concatenation. The concatenated feature vector contains both geological image information and numerical working condition information.

[0057] In order to further enhance the model's ability to process different features, the feature vectors are processed with a fully connected layer after concatenation to map the high-dimensional features to the output space of rockburst probability and impact range.

[0058] 4. Output and prediction:

[0059] The feature vector processed by the fully connected layer is sent to the final output layer to output the prediction result:

[0060] Probability of rock burst: Through a Sigmoid activation function, a value between 0 and 1 is output, indicating the probability of rock burst.

[0061] Rockburst impact range: A regression network is used to output the possible impact area of ​​a rockburst, usually expressed as an impact radius or a specific coordinate range of a dangerous area. The output range is converted into a specific spatial area through further processing (such as threshold determination).

[0062] 5. Model training and optimization:

[0063] The model is trained using back-propagation and optimization algorithms (such as Adam, SGD, etc.), and the model parameters are continuously adjusted by minimizing the loss function (such as a weighted combination of cross entropy loss and mean squared error loss).

[0064] In order to improve the stability and generalization ability of the model, regularization techniques such as Dropout, L2 regularization, and Batch Normalization are used during training to prevent overfitting.

[0065] Step 5: Use the trained neural network to output the probability and range of rockburst in actual excavation

[0066] After the training of the above neural network is completed, in the actual excavation process, the TBM operating parameters including the cutterhead thrust and torque and the front exploration map are used as inputs to the trained neural network, and the rockburst probability and range are output.

[0067] Step 6: Put the collected data into the case library to improve the accuracy of the dual neural network and realize dynamic identification of rockburst.

[0068] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A tunnel rockburst prediction method based on numerical simulation and neural network, characterized in that: The following steps are involved: Collect relevant geological parameter data and TBM operation parameters and establish a database; Set the element parameters of the tunnel model, divide the grid and add boundary conditions, and use the relevant geological parameter data collected above to perform numerical simulation; Introducing damage and fracture mechanics models, performing numerical calculations and simulating the excavation process, calculating the probability and impact range of rockbursts; The neural network is trained by simulating the operating parameters of the TBM during excavation, the probability and impact range of rock burst, and the exploration map in front of the tunnel face; wherein the operating parameters of the TBM include the cutter head torque and the total thrust; In actual tunneling, the trained neural network is used to output the probability and impact range of rock burst.

2. The method according to claim 1, characterized in that The element parameters of the tunnel model include Poisson's ratio, elastic modulus, density, friction angle, flow stress ratio, expansion angle, uniaxial tensile strength and compressive strength of rock.

3. The method according to claim 1, characterized in that The damage and fracture mechanics model describes the degree of damage to the rock mass by introducing a damage variable D, wherein the relationship between the change rate of D and the equivalent stress and damage threshold stress of the rock mass is: ， in, represents the time change rate, 1-D represents the proportion of the undamaged part of the rock mass, Indicates the extent to which the equivalent stress exceeds the damage threshold stress.

4. The method according to claim 1, characterized in that: When training the neural network, a convolutional neural network is used to extract features from the exploration map in front of the tunnel face. After splicing the TBM operation parameters with the image features, the probability and impact range of rockburst are output through the fully connected layer.

5. The method according to claim 1, characterized in that In the actual excavation process, the TBM operating parameters and the exploration map in front of the tunnel face are used as input to the trained neural network, and the probability and impact range of rock burst are output.

6. The method according to claim 1, characterized in that The probability of rock burst is calculated by Sigmoid activation function, the impact range of rock burst is calculated by regression network, and the specific coordinates of the impact radius or the dangerous area are output.

7. The method according to claim 1, characterized in that During the training process of the neural network, dynamic learning rate adjustment and optimization algorithm are used to improve the convergence speed and prediction accuracy of the network, and the generalization ability of the network is verified by a cross-validation method.

Images