A method for reconstructing and inverting overburden rock and surface crack network for similar simulation experiment

Abstract

The application discloses a kind of overburden and surface fissure network reconstruction and inversion method for similar simulation experiment, it is related to network reconstruction and inversion technical field, including: carrying out simulation test, laying experimental model;Carrying out test, collecting three-dimensional point cloud data;Data is preprocessed, and feature vector is extracted;Based on feature vector reconstruction point cloud, get fissure feature, reconstruct fissure network;Optimize fissure network, carry out three-dimensional visualization and analysis;Model is imported into simulation software, iteratively optimized parameter, and the evolution process of fissure is inverted.The application uses three-dimensional point cloud imaging technology to realize high-precision data acquisition and processing, accurately extracts fissure feature;Based on fault image or point cloud data, three-dimensional reconstruction algorithm is used to reconstruct the three-dimensional model of fissure network, and multi-scale modeling is supported;Through numerical simulation, the evolution process of fissure is inverted, combined with machine learning algorithm to optimize inversion model, improve prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network reconstruction and inversion, more particularly to a method for reconstructing and inverting overburden and surface fracture networks in similar simulation experiments. BACKGROUND

[0002] In mining, tunneling and geological hazard monitoring, the development and evolution of overburden and surface fractures have important influences on engineering safety. In similar simulation experiments, the reconstruction and inversion of overburden and surface fracture networks is a key technology for studying rock deformation, surface subsidence and fracture development patterns. This technology is mainly applied in mining, geological hazard prediction, underground engineering stability assessment and other fields. By simulating actual geological conditions, reconstructing fracture networks and inverting their evolution processes, it provides theoretical basis and technical support for engineering practice.

[0003] Similar simulation experiments are a physical simulation method based on similarity theory, which simulates rock deformation and fracture development processes under actual geological conditions through a scaled-down model. The experiment usually uses similar materials (such as gypsum, sand, cement, etc.) to construct the model, simulating the mechanical behavior of overburden and surface. Overburden refers to the rock layer covering the ore body or engineering structure, whose deformation and failure directly affect the stability of the surface. Fracture network is a system of cracks in rock mass formed due to stress changes, mining activities or geological processes, whose distribution and evolution have important influences on rock mass mechanical properties. Through experimental observation or numerical simulation, the geometry, distribution characteristics and evolution patterns of fracture networks are obtained, and a fracture model consistent with actual geological conditions is constructed. Based on experimental data or field monitoring results, the causes, evolution process and mechanical mechanism of fracture networks are inversely deduced through mathematical methods.

[0004] However, traditional fracture monitoring methods rely heavily on field observation and numerical simulation, making it difficult to accurately reflect the three-dimensional spatial distribution and dynamic evolution process of fractures. As an effective physical simulation method, similar simulation experiments can better simulate the fracture development process of overburden and surface, but there is a lack of a systematic method to reconstruct and invert fracture networks in existing technology, limiting the in-depth analysis and application of experimental data. In addition, traditional methods have problems of insufficient precision and low efficiency in fracture three-dimensional modeling.

[0005] Therefore, how to propose a method for reconstructing and inverting overburden and surface fracture networks in similar simulation experiments, improve the precision and efficiency of fracture three-dimensional modeling, and accurately reflect the three-dimensional spatial distribution and dynamic evolution process of fractures is a problem that needs to be solved by those skilled in the art. SUMMARY

[0006] Therefore, the application provides a method for reconstructing and inverting overburden and surface fissure networks in a similar simulation experiment, improves the accuracy and efficiency of three-dimensional modeling of fissures, and accurately reflects the three-dimensional spatial distribution of fissures and the dynamic evolution process thereof.

[0007] A method for reconstructing and inverting overburden and surface fissure networks in a similar simulation experiment comprises the following steps.

[0008] A similar simulation experiment is designed to lay an experimental model.

[0009] The experimental model is tested according to the experimental model, and three-dimensional point cloud data in the experimental process are collected.

[0010] Data preprocessing is performed, and a feature vector of the three-dimensional point cloud data is extracted.

[0011] Point cloud reconstruction is performed based on the feature vector to obtain fissure point cloud data features and reconstruct a fissure network.

[0012] The reconstructed fissure network is optimized, and three-dimensional fissure visualization and analysis are performed.

[0013] The reconstructed fissure network model is imported into numerical simulation software, simulation parameters are iteratively optimized according to a similar simulation scheme, and a three-dimensional fissure dynamic evolution process is inverted.

[0014] Optionally, the three-dimensional point cloud data collected in the experimental process comprises the following steps.

[0015] Each layer is scanned by using a three-dimensional point cloud sensor to obtain initial three-dimensional point cloud data of each layer.

[0016] After the model is laid, the entire model is scanned by using a three-dimensional point cloud sensor to obtain initial three-dimensional point cloud data of the entire model.

[0017] According to a similar simulation experiment scheme, materials at corresponding positions are sequentially excavated, and the entire model is scanned by using a three-dimensional point cloud sensor, and is compared and analyzed with the initial three-dimensional point cloud data.

[0018] After the excavation experiment is completed, the entire model is scanned again to obtain three-dimensional point cloud data of the entire model after the experiment.

[0019] In contrast to the model laying, the model is sequentially removed from top to bottom, each layer is scanned by using a three-dimensional point cloud sensor to obtain three-dimensional point cloud data of each layer after the experiment.

[0020] Optionally, the method further comprises the following step: a marker is placed between each layer.

[0021] Optionally, the data preprocessing comprises: removing noise points of the three-dimensional point cloud data by using statistical filtering and radius filtering methods; reducing the amount of point cloud data by using voxel grid filtering and random sampling methods; scaling the point cloud data to a unified range, and selecting the bottom edge of the similar simulation test platform as the coordinate origin.

[0022] Optionally, the feature vector of the three-dimensional point cloud data comprises: calculating the normal vector and curvature of each point in the point cloud data, and detecting the corner points and edge points of the crack point cloud.

[0023] Optionally, the crack point cloud data feature obtained by reconstructing the point cloud based on the feature vector comprises:

[0024] segmenting the point cloud into different regions based on the normal vector and curvature features of the point cloud;

[0025] matching and registering based on the corner point, edge point and key point features, and globally aligning all the point cloud data;

[0026] performing difference value operation on the three-dimensional point data obtained during the model removal and the three-dimensional point cloud data obtained during the model laying to obtain the difference value three-dimensional point cloud data before and after the experiment;

[0027] converting the difference value three-dimensional point cloud data into a continuous surface model by using Poisson reconstruction or Delaunay triangulation method, and converting the data into a network model.

[0028] Optionally, it further comprises: based on the crack point cloud data feature, using Poisson reconstruction, Delaunay triangulation or Ball Pivoting algorithm for surface reconstruction, and then generating a three-dimensional grid model by using Marching Cubes algorithm.

[0029] Optionally, the crack network reconstruction comprises: connecting adjacent crack models according to the spatial position and geometric features of the cracks to construct a complete crack network; analyzing the topological structure of the crack network to calculate the connectivity and branch number parameters of the cracks; and performing smoothing, denoising and simplification processing on the crack network.

[0030] Optionally, the optimization of the reconstructed crack network comprises:

[0031] performing smoothing, denoising and simplification processing on the generated three-dimensional crack grid; filling the holes or missing parts in the model by using interpolation or repair algorithm; repairing the non-manifold edges and self-intersection problems in the grid;

[0032] visualizing and displaying the crack network model by using three-dimensional modeling software; and exporting the three-dimensional crack network model into STL, OBJ or PLY format.

[0033] Optionally, the three-dimensional fracture visualization and analysis includes: calculating the length, width, direction, volume, surface area geometric parameters of the fracture; analyzing the density, connectivity, directionality characteristics of the fracture network.

[0034] Through the above technical solutions, compared with the prior art, the present disclosure provides an overburden and surface fracture network reconstruction and inversion method for similar simulation experiments, which has the following beneficial effects:

[0035] The present disclosure provides an overburden and surface fracture network reconstruction and inversion method for similar simulation experiments, which includes: designing a similar simulation experiment to lay out an experimental model; collecting three-dimensional point cloud data during the experiment according to the experimental model; performing data preprocessing and extracting feature vectors of the three-dimensional point cloud data; reconstructing the fracture point cloud data based on the feature vectors to obtain the fracture network reconstruction; optimizing the reconstructed fracture network and performing three-dimensional fracture visualization and analysis; importing the reconstructed fracture network model into numerical simulation software, iteratively optimizing the simulation parameters according to the similar simulation scheme, and inverting the three-dimensional fracture dynamic evolution process.

[0036] The present disclosure realizes (1) high-precision data acquisition and processing: using advanced three-dimensional point cloud imaging technology (such as laser radar, depth camera, structured light scanner, or three-dimensional laser scanner, etc.) to obtain high-resolution data of the overburden and surface; combining three-dimensional point cloud data processing algorithm to accurately extract fracture features.

[0037] (2) three-dimensional fracture network reconstruction: based on fault images or point cloud data, using three-dimensional reconstruction algorithms (such as Marching Cubes, Poisson reconstruction) to generate three-dimensional models of the fracture network; supporting multi-scale modeling, which can reflect the distribution characteristics of macroscopic and microscopic fractures at the same time.

[0038] (3) fracture network inversion technology: through numerical simulation (such as finite element analysis, discrete element analysis) to invert the evolution process of the fracture, revealing the fracture formation mechanism; combining machine learning algorithms (such as deep learning, support vector machine) to optimize the inversion model and improve the prediction accuracy.

[0039] (4) multidisciplinary integration: combining geomechanics, geotechnical engineering, computer vision, and other multidisciplinary technologies to provide a comprehensive solution to the fracture network analysis. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are only embodiments of the present application, and those skilled in the art can obtain other drawings according to the provided drawings without creative labor.

[0041] Figure 1 A flowchart of a method for reconstructing and inverting overburden rock and surface fracture network for similar simulation experiment is provided. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0043] The embodiments of the present application disclose a method for reconstructing and inverting overburden rock and surface fracture network for similar simulation experiment, as shown in Figure 1 The method comprises the following steps:

[0044] Designing a similar simulation experiment to lay an experimental model;

[0045] Carrying out the experiment according to the experimental model and collecting three-dimensional point cloud data in the experimental process;

[0046] Carrying out data preprocessing and extracting feature vectors of the three-dimensional point cloud data;

[0047] Reconstructing the point cloud based on the feature vectors to obtain features of the fracture point cloud data and reconstructing the fracture network;

[0048] Optimizing the reconstructed fracture network and visualizing and analyzing the three-dimensional fracture;

[0049] Importing the reconstructed fracture network model into numerical simulation software, iteratively optimizing simulation parameters according to the similar simulation scheme, and inverting the three-dimensional fracture dynamic evolution process.

[0050] Further, the collecting of the three-dimensional point cloud data in the experimental process comprises:

[0051] Scanning each layer by using a three-dimensional point cloud sensor to obtain initial three-dimensional point cloud data of each layer;

[0052] After the model laying is completed, the entire model is scanned by using a three-dimensional point cloud sensor to obtain initial three-dimensional point cloud data of the entire model;

[0053] According to the similar simulation experiment scheme, materials at corresponding positions are sequentially excavated, and the entire model is scanned by using a three-dimensional point cloud sensor, and compared and analyzed with the initial three-dimensional point cloud data;

[0054] After the excavation experiment is completed, the entire model is scanned again to obtain three-dimensional point cloud data of the entire model after the experiment;

[0055] Contrary to model laying, the model is dismantled layer by layer from top to bottom, and the upper surface of each layer is scanned by a three-dimensional point cloud sensor to obtain three-dimensional point cloud data of each layer after the test.

[0056] Further, it also includes placing markers between each layer.

[0057] Further, the data preprocessing includes: using statistical filtering, radius filtering method to remove noise points of three-dimensional point cloud data; using voxel grid filtering, random sampling method to reduce the amount of point cloud data; scaling the point cloud data to a unified range, selecting the bottom edge of the similar simulation test platform as the coordinate origin.

[0058] Further, the feature vector of the three-dimensional point cloud data includes: calculating the normal vector and curvature of each point in the point cloud data, and detecting the corner points and edge points of the crack point cloud.

[0059] Further, the crack point cloud data feature obtained by reconstructing the point cloud based on the feature vector includes:

[0060] Based on the normal vector and curvature features of the point cloud, the point cloud is divided into different areas;

[0061] Based on the corner point, edge point, and key point features, matching, registration, and global alignment of all point cloud data are performed;

[0062] The three-dimensional point data obtained during model dismantling is subjected to difference operation with the three-dimensional point cloud data obtained during model laying, to obtain difference three-dimensional point cloud data before and after the test;

[0063] Using Poisson reconstruction or Delaunay triangulation method, the difference three-dimensional point cloud data is converted into a continuous surface model, and the data is converted into a network model.

[0064] Further, it also includes: based on the crack point cloud data feature, using Poisson reconstruction, Delaunay triangulation or Ball Pivoting algorithm for surface reconstruction, and then generating a three-dimensional grid model through Marching Cubes algorithm.

[0065] Further, the crack network reconstruction includes: connecting adjacent crack models according to the spatial position and geometric features of the cracks to construct a complete crack network; analyzing the topological structure of the crack network to calculate the connectivity and branch number parameters of the cracks; and smoothing, denoising, and simplifying the crack network.

[0066] Further, the optimization of the reconstructed crack network includes:

[0067] Smooth, denoise, simplify the generated three-dimensional fracture grid; use interpolation or repair algorithm to fill the holes or missing parts in the model; repair the non-manifold edges in the grid, and the self-intersection problem;

[0068] Visualize the fracture network model using three-dimensional modeling software; at the same time, export the three-dimensional fracture network model to STL, OBJ or PLY format.

[0069] Further, the three-dimensional fracture visualization and analysis includes: calculating the length, width, direction, volume, surface area geometric parameters of the fracture; analyzing the density, connectivity, directionality characteristics of the fracture network.

[0070] In a specific embodiment, a method for reconstructing and inverting overburden and surface fracture network for similar simulation experiment, the specific process steps include:

[0071] S1: experimental design: according to the purpose of experiment, engineering geological conditions, design similar simulation test, determine the geometric similarity ratio, time similarity ratio, stress similarity ratio and other similarity parameters according to the similar simulation test platform parameters and similarity principle, determine the experimental scheme;

[0072] S2: similar material: according to the experimental requirements, select the similar simulation material similar to the experimental rock and soil layer, and then determine the similar ratio of the bulk density of different rock layers, and prepare the materials used for similar simulation according to the experimental requirements and experimental purposes;

[0073] S3: model laying: according to the stratification of the prototype stratum, lay the simulation material layer by layer, and after laying each layer, it needs to be flattened and compacted, the thickness meets the design size, and the flatness and density are ensured to avoid error accumulation;

[0074] S4: three-dimensional scanning: use three-dimensional point cloud sensor (such as laser radar, depth camera, structure light scanner or three-dimensional laser scanner, etc.) to scan each layer to obtain the initial three-dimensional point cloud data of each layer;

[0075] S5: placing markers: place markers (such as colored sand, mica, fine line, etc.) between each layer to observe the deformation and damage situation later;

[0076] S6: overall scanning: after the model is laid, the whole model is scanned by using three-dimensional point cloud sensor to obtain the initial three-dimensional point cloud data of the whole model;

[0077] S7: experiment and data recording: according to the similar simulation test scheme, the corresponding position materials are excavated in turn according to the design, at the same time, the whole model is scanned by using three-dimensional point cloud sensor, and compared with the initial three-dimensional point cloud data at the same time;

[0078] S8: Overall scanning after experiment: After the completion of the excavation experiment, the overall model is scanned again to obtain the three-dimensional point cloud data of the overall model after the experiment;

[0079] S9: Model dismantling and three-dimensional scanning: In contrast to laying the model, the model is carefully dismantled layer by layer from top to bottom, and the upper layer marker and the similar material of the previous layer are cleaned at the same time. At the same time, the upper surface of each layer is scanned by three-dimensional point cloud sensing to obtain the three-dimensional point cloud data of each layer after the experiment;

[0080] S10: Data preprocessing: statistical filtering, radius filtering and other methods are used to remove noise points from three-dimensional point cloud data; voxel grid filtering, random sampling and other methods are used to reduce the amount of point cloud data and improve processing efficiency; the point cloud data is scaled to a unified range, and the bottom edge of the similar simulation test platform is selected as the coordinate origin;

[0081] S11: Feature extraction: the normal vector and curvature of each point in the point cloud data are calculated, and the key points such as corner points and edge points of the crack point cloud are detected to ensure subsequent data matching and recognition;

[0082] S12: Point cloud segmentation: in order to reduce the amount of point cloud data and the amount of data processing in the later stage, based on the normal vector and curvature characteristics of the point cloud, the point cloud is segmented into different areas; K-mean or DBSCAN clustering algorithm can be used to segment the point cloud into different clusters; or RANSAC method is used to monitor the planar area in the point cloud. Through point cloud segmentation, the three-dimensional coordinates of each layer model in the three-dimensional scene can be more accurately identified, the number of point clouds can be reduced, the difficulty of data processing can be reduced, and the running efficiency of data processing can be improved.

[0083] S13: Point cloud matching: in order to integrate the point cloud data scanned at different times into the same coordinate system, the point cloud features such as corner points, edge points and key points are matched, and then registered; 4PCS or Go-ICP method is used to align all point cloud data globally. Through point cloud matching, point cloud data at different time points and different angles can be integrated, more comprehensive and accurate three-dimensional scene information can be provided, multi-angle analysis of three-dimensional scene can be realized, accuracy and reliability of analysis can be improved, and multiple local point cloud data sets can be aligned and integrated to generate more complete and accurate three-dimensional model.

[0084] S14: Point cloud calculation: difference operation is performed on the three-dimensional point data of each layer obtained during model dismantling and the three-dimensional point cloud data of each layer obtained during model laying to obtain the difference three-dimensional point cloud data before and after the experiment;

[0085] S15: Point cloud reconstruction: Poisson reconstruction or Delaunay triangulation method is used to convert the difference three-dimensional point cloud data into a continuous surface model, and the data is converted into a network model.

[0086] For each layer of the rock surface model laid by the similar simulation test, point clouds can be generated by uniformly sampling on a plane, and then the points are connected into a triangular mesh using a triangulation algorithm, so as to obtain a three-dimensional mesh model of each layer of rock; conversely, the three-dimensional mesh model of the entire similar simulation can be mapped to a two-dimensional plane by a parameterization method, and then the surface equation of a cube is defined on the parameter plane, so as to obtain a surface model.

[0087] S16: Three-dimensional reconstruction of fissures: based on the characteristics of the fissure point cloud data, a surface reconstruction is performed by using a Poisson reconstruction, a Delaunay triangulation or a BallPivoting algorithm, and a three-dimensional mesh model is generated by using a MarchingCubes algorithm;

[0088] S17: Reconstruction of a fissure network: adjacent fissure models are connected according to the spatial position and geometric characteristics of the fissures to construct a complete fissure network. The specific steps are as follows: near-fissure screening (spatial coarse matching), geometric compatibility verification is performed on each pair of near-fissures (fine matching), the fissures that meet the requirements of the included angle of the normal vectors, the edge distance and the intersection are linked, and a fissure network graph is constructed. The topological structure of the fissure network is analyzed, and parameters such as the connectivity and the number of branches of the fissures are calculated; the fissure network is processed by smoothing, denoising and simplifying, so as to improve the quality of the model;

[0089] S18: Optimization of the three-dimensional fissure network: the three-dimensional fissure mesh generated is processed by smoothing, denoising and simplifying, so as to improve the quality of the model; interpolation or repair algorithms are used to fill in the holes or missing parts in the model; non-manifold edges, self-intersection and other problems in the mesh are repaired, so as to ensure the integrity of the model;

[0090] S19: Visualization and analysis of the three-dimensional fissure: the fissure network model is visualized and displayed by using a three-dimensional modeling software (such as MeshLab, Blender, CloudCompare, etc.); at the same time, the three-dimensional fissure network model is exported in the formats of STL, OBJ, PLY, etc.

[0091] S20: Data analysis: the geometric parameters of the fissures such as the length, the width, the direction, the volume and the surface area are calculated; the characteristics of the fissure network such as the density, the connectivity and the directionality are analyzed;

[0092] S21: Numerical simulation calculation: the reconstructed fissure network model is imported into a numerical simulation software, and the simulation parameters (including the number of nodes, the mesh size, the rock mechanics parameters and the boundary conditions) are continuously optimized according to the similar simulation scheme (the experimental purpose, the lithology parameters, the field measurement results and the similar simulation test results), so as to inverse the three-dimensional fissure dynamic evolution process.

[0093] S22: Fracture network inversion techniques: Inversion of fracture evolution processes through numerical simulation (e.g. finite element analysis, discrete element analysis) to reveal fracture formation mechanisms; combined with machine learning algorithms (e.g. deep learning, support vector machines) to optimize inversion models and improve prediction accuracy.

[0094] ① Deep learning optimization inversion:

[0095] a) Data collection and integration. Collect multi-source data related to fracture inversion problems, ensure data integrity and consistency, and match and align different sources of data in time and space.

[0096] b) Data cleaning. Remove noise, outliers and missing values in the data. For noisy data, use filtering algorithms (such as median filtering, Gaussian filtering) for processing; for outliers, according to the distribution characteristics of the data, use statistical methods (such as 3σ principle) or machine learning algorithms (such as isolation forest algorithm) for identification and elimination; for missing values, according to the characteristics of the data, choose appropriate filling methods, such as mean filling, median filling, interpolation method (such as linear interpolation, spline interpolation) or machine learning-based prediction filling.

[0097] c) Feature engineering. Extract meaningful features from raw data to better describe the characteristics of the inverted fractures. For example, extract texture features (such as gray level co-occurrence matrix features) and shape features (such as boundary descriptors) of images, and normalize or standardize the features to make different features have the same scale, avoiding the dominant influence of some features with large value range on model training.

[0098] d) Deep learning model construction and training. According to the complexity of the fracture inversion problem and the characteristics of the data, choose appropriate deep learning architecture. For data with three-dimensional fracture network structure, use convolutional neural networks to automatically extract local features and spatial hierarchical structure of data using convolutional layers and pooling layers; also can combine attention mechanism, make the model can focus on the features that have greater influence on the inversion result. Divide the preprocessed data into training set, validation set and test set. Use the training set to train the deep learning model, use appropriate loss function (such as mean square error loss function for regression problem, cross entropy loss function for classification problem) and optimization algorithm (such as stochastic gradient descent SGD, Adam optimizer) to minimize the loss function. In the training process, monitor the performance of the model through the validation set to prevent overfitting. When the performance of the model on the validation set no longer improves, stop training in time and save the optimal model parameters.

[0099] ② Support vector machine optimization inversion:

[0100] a) The performance of support vector machines largely depends on the choice of kernel function. For linearly separable or approximately linearly separable data, a linear kernel function can be chosen; for nonlinear data, commonly used kernel functions include Gaussian kernel (RBF kernel), polynomial kernel, etc. The kernel function and its parameters that best suit the current inversion problem are selected through cross-validation and other methods. The training set is used to train the SVM model, and the optimal hyperplane is determined to maximize the separation between data points of different classes. During the training process, the regularization parameter C is adjusted to balance the complexity of the model and its generalization ability.

[0101] b) Model ensemble. Deep learning models and support vector machine models are integrated to fully leverage the strengths of both algorithms. Voting, averaging, stacking, and other ensemble strategies can be used. For example, in the stacking method, the prediction results of deep learning models and support vector machine models are used as input features to train a new meta-model (such as a linear regression model or a decision tree model) to obtain the final prediction results.

[0102] c) Hyperparameter tuning. Use grid search, random search, or Bayesian optimization to tune the hyperparameters of the ensemble model. For example, adjust the learning rate, batch size, and network depth of the deep learning model, and the kernel function parameters and regularization parameters of the support vector machine model. By continuously trying different combinations of hyperparameters, find the optimal hyperparameter settings that make the model perform best on the validation set.

[0103] d) Evaluation index selection. Select appropriate evaluation indices based on the specific type of inversion problem. For regression problems, commonly used evaluation indices include mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ) for classification problems, commonly used evaluation indices include accuracy, precision, recall, and F1 score.

[0104] e) Model validation. Use the test set to validate the optimized model, calculate the evaluation indices of the model on the test set, and evaluate the generalization ability and prediction accuracy of the model. At the same time, analyze the differences between the model's prediction results and the true values, find out the possible shortcomings of the model, and provide direction for further optimization.

[0105] f) Model deployment. Deploy the trained model to the actual production environment, so that it can process new data in real time or in batches for inversion prediction. According to actual needs, choose appropriate deployment methods, such as deploying the model on a local server, a cloud computing platform, or an embedded device.

[0106] Continuous monitoring and updating. During the application of the model, the performance of the model is continuously monitored. When new data is continuously accumulated or the environment of the inversion problem changes, the model is periodically updated and retrained to ensure that the model always maintains good prediction accuracy.

[0107] The embodiment proposes a method for reconstructing and inverting overburden and surface fracture networks in similar simulation experiments. It has scientific research value, provides high-precision experimental data and model support for the formation mechanism and evolution law of overburden and surface fractures, and promotes the theoretical development of geomechanics, mining engineering, and geological disaster prevention. It has engineering application value, provides quantitative evaluation of fracture networks in mine exploitation, tunnel engineering, and slope stability analysis, guides engineering design and construction, and provides technical support for the prediction and prevention of geological disasters such as landslides and collapses. It has technical innovation value, promotes the innovation and application of three-dimensional reconstruction, inversion algorithm, and dynamic monitoring technologies, and provides new technical means for similar simulation experiments, improving the scientificity and reliability of experiments. It can be directly applied to mine exploitation to study the evolution law of mining-induced overburden fractures and evaluate the stability of goaf, geological disaster prevention to analyze the fracture development characteristics of landslides and collapses and develop prevention measures, tunnel and underground engineering to evaluate the influence of surrounding rock fractures on tunnel stability and optimize support design, and environmental protection to study the influence of surface fractures on soil erosion and vegetation damage and develop ecological restoration schemes.

[0108] The embodiment realizes accurate reconstruction and evolution law research of overburden and surface fracture networks through high-precision data acquisition, three-dimensional reconstruction, inversion analysis, and dynamic monitoring, and has significant technical advancement and wide application value. It not only promotes scientific research in related fields, but also provides strong technical support for engineering practice and disaster prevention, and has important economic and social benefits.

[0109] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts can be referred to the method part.

[0110] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments, characterized in that, include: Design similar simulation experiments to lay out experimental models; Experiments were conducted based on the experimental model, and three-dimensional point cloud data was collected during the experiment. The 3D point cloud data acquired during the experiment includes: Each layer is scanned using a 3D point cloud sensor to obtain the initial 3D point cloud data for each layer; After the model is laid out, a 3D point cloud sensor is used to scan the entire model to obtain the initial 3D point cloud data of the model as a whole. According to the similar simulation test plan, the material at the corresponding positions was removed in sequence according to the design. At the same time, the entire model was scanned with a three-dimensional point cloud sensor and compared and analyzed with the initial three-dimensional point cloud data. After the excavation experiment was completed, the overall model was scanned again to obtain the three-dimensional point cloud data of the overall model after the experiment. In contrast to model laying, the model is dismantled layer by layer from top to bottom, and the surface of each layer is scanned using a 3D point cloud sensor to obtain the 3D point cloud data of each layer after the experiment. Perform data preprocessing to extract feature vectors from the 3D point cloud data; Point cloud data features of fracture point cloud are obtained by reconstructing point cloud based on feature vectors, and fracture network reconstruction is then performed. The features of the fracture point cloud data obtained by point cloud reconstruction based on feature vectors include: Based on the normal vector and curvature features of the point cloud, the point cloud is segmented into different regions; Matching and registration are performed based on corner points, edge points, and key point features, and global alignment is performed on all point cloud data. The difference between the 3D point data of each layer obtained during model dismantling and the 3D point cloud data of each layer obtained during model laying is calculated to obtain the difference 3D point cloud data before and after the experiment. The Poisson reconstruction or Delaunay triangulation method is used to convert the difference 3D point cloud data into a continuous surface model, and at the same time, the data is converted into a network model. The reconstructed fracture network was optimized, and three-dimensional fracture visualization and analysis were performed. The reconstructed fracture network model is imported into numerical simulation software. Based on a similar simulation scheme, the simulation parameters are iteratively optimized to invert the dynamic evolution process of the three-dimensional fracture.

2. The method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments according to claim 1, characterized in that, Also includes: Place markers between each layer.

3. The method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments according to claim 1, characterized in that, The data preprocessing includes: using statistical filtering and radius filtering methods to remove noise points from the three-dimensional point cloud data; using voxel grid filtering and random sampling methods to reduce the amount of point cloud data; scaling the point cloud data to a uniform range and selecting the bottom edge of the similar simulation test platform as the origin of the coordinate system.

4. The method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments according to claim 1, characterized in that, The extraction of feature vectors from 3D point cloud data includes: calculating the normal vector and curvature of each point in the point cloud data, and detecting the corner points and edge points of the crack point cloud.

5. The method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments according to claim 1, characterized in that, Also includes: Based on the characteristics of the fracture point cloud data, Poisson reconstruction, Delaunay triangulation, or BallPivoting algorithms are used for surface reconstruction, and then the Marching Cubes algorithm is used to generate a three-dimensional mesh model.

6. The method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments according to claim 1, characterized in that, The fracture network reconstruction involves connecting adjacent fracture models based on their spatial location and geometric characteristics to construct a complete fracture network; analyzing the topology of the fracture network and calculating the connectivity and branch number parameters of the fractures; and then smoothing, denoising, and simplifying the fracture network.

7. The method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments according to claim 1, characterized in that, The optimization of the reconstructed fracture network includes: The generated 3D fractured mesh is smoothed, denoised, and simplified; interpolation or patching algorithms are used to fill holes or missing parts in the model; and non-manifold edges and self-intersections in the mesh are repaired. The fracture network model is visualized using 3D modeling software; the 3D fracture network model is also exported as an STL, OBJ, or PLY format.

8. The method for reconstructing and inverting overburden and surface fracture networks for similar simulation experiments according to claim 1, characterized in that, The three-dimensional fracture visualization and analysis includes: calculating the geometric parameters of fracture length, width, direction, volume, and surface area; and analyzing the density, connectivity, and directional characteristics of the fracture network.

Images