Data-analog hybrid driven rock deformation and fracture process analysis method

Abstract

This invention relates to the field of underground mine disaster monitoring and early warning technology, and particularly to a data-simulation hybrid-driven method for analyzing rock deformation and fracturing processes. The method includes: collecting monitoring data and preprocessing displacement sequence data; constructing a three-dimensional displacement field of the surrounding rock using a physical constraint optimization fusion algorithm; constructing a three-dimensional geomechanical model based on the monitoring data, constructing a heterogeneous mechanical parameter field based on the basic mechanical baseline parameters, simulating the three-dimensional geomechanical model using the heterogeneous mechanical parameter field, and constructing a simulation database based on the simulation results; training a surrogate model using the simulation database; inputting the basic mechanical baseline parameters of the rock mass into the trained surrogate model to obtain initial mechanical response data; constructing a first objective function; iteratively updating the mechanical parameters using an optimization algorithm, and inputting the updated mechanical parameters into the trained surrogate model until the first objective function converges, obtaining the optimal mechanical parameter field. This invention improves the consistency and interpretability of simulation and prediction.

Description

Technical Field

[0001] This invention relates to the field of underground mine disaster monitoring and early warning technology, and in particular to a data-simulation hybrid-driven method for analyzing rock deformation and fracturing processes. Background Technology

[0002] Under the disturbance conditions of mining operations, the redistribution of surrounding rock stress, slippage of structural planes, and crack propagation often evolve in a coupled manner, easily inducing disaster risks such as large deformation, collapse, and spalling. Interpretable and predictable analysis of the rock mass deformation and fracturing process is an important requirement for safe mine production.

[0003] Existing technologies are generally divided into two categories: one is the monitoring data-driven method, which can reflect the on-site response, but it relies heavily on empirical thresholds or black-box models, making it difficult to explain the disaster mechanism and to be used for scheme optimization and disaster prevention guidance; the other is the numerical simulation method, which can be used for mechanism explanation, but due to the spatial heterogeneity of rock mass parameters, time-varying boundary conditions, and geometric uncertainty of structural surfaces, the model output is difficult to match the on-site situation, resulting in insufficient prediction reliability. In addition, traditional numerical simulation analysis is time-consuming and cannot meet the needs of near real-time analysis. Summary of the Invention

[0004] To address the technical problems of existing monitoring methods' inability to explain disaster mechanisms, inconsistencies between traditional numerical simulations and on-site responses, and long computation times, this invention provides a data-simulation hybrid-driven method for analyzing rock deformation and fracturing processes. This invention primarily utilizes multi-source monitoring data fusion to construct a three-dimensional displacement field of the surrounding rock, builds a numerical simulation database driven by a heterogeneous mechanical parameter field, and trains a surrogate model. Using the reconstructed displacement field and microseismic damage field as constraints, the optimal mechanical parameter field is iteratively inverted through optimization algorithms. This achieves the fusion of the real-time accuracy of monitoring data and the mechanistic interpretability of numerical simulations, enabling near-real-time, highly reliable prediction of rock mass deformation and fracturing processes. This provides an analytical tool for disaster prevention and control with interpretable mechanical parameters and traceable evolution paths.

[0005] The technical means employed in this invention are as follows:

[0006] A data-simulation hybrid-driven method for analyzing rock deformation and fracturing processes includes the following steps: Collect monitoring data, including three-dimensional laser scanning point cloud data, displacement sequence data, microseismic monitoring data, and blasting disturbance monitoring data; preprocess the displacement sequence data. Based on the preprocessed displacement sequence data, a physical constraint optimization fusion algorithm is used to construct the three-dimensional displacement field of the surrounding rock; Obtain the basic mechanical baseline parameters of the rock mass, construct a three-dimensional geomechanical model based on monitoring data, construct a heterogeneous mechanical parameter field based on the basic mechanical baseline parameters, simulate the three-dimensional geomechanical model using the heterogeneous mechanical parameter field, and construct a simulation database based on the simulation results; The agent model is trained using the simulated database; The rock mass foundation mechanical baseline parameters are input into the trained surrogate model to obtain initial mechanical response data, which includes displacement field and damage field. Based on the initial mechanical response data and the three-dimensional displacement field of the surrounding rock, a first objective function is constructed. The mechanical parameters are iteratively updated using an optimization algorithm. The updated mechanical parameters are then input into the trained surrogate model until the first objective function converges. The updated mechanical parameters are then used as the optimal mechanical parameter field.

[0007] Furthermore, the construction of the three-dimensional displacement field of the surrounding rock using the physical constraint optimization fusion algorithm includes: Calculate the surface displacement of the surrounding rock based on 3D laser scanning point cloud data; Construct the second objective function:

[0008] in, The second objective function is... For surface displacement observation operators, The reconstructed three-dimensional displacement field of the surrounding rock. This represents the displacement of the surrounding rock surface. For deep hole displacement observation operators, This is the preprocessed displacement sequence data. The weights of the smoothing term, For smoothing terms, The weights of the physical constraint terms, For physical constraint terms, The formula for calculating the smoothing term is:

[0009] in, For spatial weighting functions, Let be the displacement gradient tensor. For the computational domain, It is the Frobenius norm. The formula for calculating the physical constraint term is:

[0010] in, The weights of the constraint terms in the balance equation. These are the constraint terms of the equilibrium equation. The weights of the compatibility constraint terms, As a compatibility constraint term, the penalty term is calculated using the following formula:

[0011] in, For physical constraint weighting functions, For stress tensor, For divergence, This is a physical activity. The formula for calculating the compatibility constraint term is as follows:

[0012] in, For compatibility constraints, For strain tensor; The second objective function is solved using the finite memory quasi-Newton method to obtain the three-dimensional displacement field of the surrounding rock.

[0013] Furthermore, the generation step of the heterogeneous mechanical parameter field includes: Centered on the baseline parameters of rock mass foundation mechanics, the random fluctuation range of each parameter is set, and random parameter samples are generated by Latin hypercube sampling. The random parameter samples are mapped to a spatial grid to form a heterogeneous mechanical parameter field. The baseline parameters of rock mass foundation mechanics include elastic modulus, Poisson's ratio and cohesion.

[0014] Furthermore, the elastobrittle damage constitutive model is used to describe the rock mass damage and fracture process in the three-dimensional geomechanical model, and FLAC3D is used to simulate random parameter samples. The output of the three-dimensional geomechanical model is displacement field, stress field, strain field, damage field and energy term.

[0015] Furthermore, the proxy model is established based on the 3D U-Net three-dimensional encoder-decoder network. The input of the proxy model is a non-homogeneous mechanical parameter field, and the output of the proxy model is a displacement field, a stress field, a strain field, a damage field, and an energy term.

[0016] Furthermore, the calculation method for the rock mass damage and observation points includes: The computational domain is discretized into voxel elements using a spatial grid consistent with the proxy model, and the rock mass damage and observation sub-values ​​of all voxels are set to 0. Acquire microseismic monitoring data, which includes a set of microseismic events inside the rock mass, including the event location, occurrence time, and affected area; The microseismic event is defined as a sphere with the unknown event location as its center and the area of ​​influence as its radius. A simplified binary classification model is adopted, and the rock mass damage and observation sub-value of all voxels within the sphere are assigned to 1; After traversing all microseismic events, output the rock mass damage and observation data.

[0017] Furthermore, the formula for calculating the first objective function is as follows:

[0018] in, Let the first objective function be... For the weighting coefficients of the displacement field, For the three-dimensional displacement field of the surrounding rock, For displacement field, These are the weighting coefficients for the damage field. For rock mass damage, For the observation sub, For damage field, These are rock mass mechanical parameters.

[0019] Furthermore, the baseline mechanical parameters of the rock mass foundation were obtained through indoor uniaxial compression, Brazilian splitting, and triaxial tests.

[0020] Furthermore, the stress tensor adopts express, For the equivalent elasticity matrix, For displacement field The calculated small deformation strain tensor, obtained from the displacement field U, is calculated using the following formula: .

[0021] Compared with the prior art, the present invention has the following advantages: This invention improves the quality and reliability of monitoring data through multi-source data alignment and anomaly detection and repair; it achieves continuous three-dimensional displacement field reconstruction by introducing physical constraints such as equilibrium equations and strain compatibility conditions, which facilitates inversion and mechanism analysis; it replaces the high-cost iteration of traditional numerical simulation with a surrogate model to meet the needs of near-real-time analysis; and it achieves dynamic calibration of mechanical parameters by jointly constraining the monitoring displacement field and microseismic damage field, thereby improving the consistency and interpretability of simulation and prediction.

[0022] Based on the above reasons, this invention can be widely applied in fields such as underground mine disaster monitoring and early warning. Attached Figure Description

[0023] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic flowchart of the data-simulation hybrid-driven rock deformation and fracturing process analysis method of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] The method of the present invention consists of the following three parts: (1) Analysis of surrounding rock mechanical response by fusion of multi-source monitoring data: Time alignment of monitoring data, detection and repair of abnormal data, fusion of surface and deep displacement and reconstruction of continuous three-dimensional displacement field; (2) Surrounding rock damage numerical simulation process proxy model: Construct a fast mapping model of mechanism constraints to realize rapid prediction from physical and mechanical parameters and boundary conditions to mechanical responses such as displacement / stress / damage / energy; (3) Analysis of disaster process by monitoring-simulation information fusion: The mechanical parameters are dynamically corrected by reconstructing the displacement field and microseismic damage field, and the key characteristics of disaster and future stage predictions are output based on the corrected parameters.

[0028] like Figure 1As shown, this invention provides a data-simulation hybrid-driven method for analyzing rock deformation and fracturing processes, comprising the following steps: S1. Collect monitoring data, including 3D laser scanning point cloud data, displacement sequence data, microseismic detection data, blasting seismic monitoring data, and blasting disturbance monitoring data, and preprocess displacement sequence data.

[0029] The number of point clouds in a 3D laser scan is the number of point clouds obtained at each scan time through periodic scanning. The displacement sequence data consists of deep displacement data from multi-point displacement gauges, obtained by continuous acquisition at multiple burial depths. Microseismic monitoring data is used to obtain a set of microfracture events within the rock mass. The recorded parameters must include at least the event location. When it happened , range / volume of influence The data is derived using apparent volume. Blasting seismic / disturbance monitoring data is obtained by collecting mining disturbance signals to obtain disturbance sequences. .

[0030] To ensure that all measuring points of the multi-point displacement gauge acquire data simultaneously at the moment of a disturbance event, a timed synchronous acquisition mechanism is employed. For data from different devices, a unified time reference (such as GPS / NTP or system clock calibration) is used to calibrate the timestamps of each device, with fixed time intervals. By sampling data from multiple sources, a unified timescale can be achieved.

[0031] Preprocessing displacement sequence data includes: Step 1: In the sliding time window Within, calculate the dynamic time-warped distance between any two measuring points in the displacement sequence data. Correlation is calculated based on dynamic time-warped distance:

[0032] in, For correlation, For dynamic time-normalized distance, This is the distance scale coefficient. This is used to control the rate at which similarity decreases with increasing distance, and can be set or calibrated based on the monitored noise level and the statistical characteristics of the displacement amplitude. The smaller the value, the faster the similarity decays.

[0033] The second step is to assemble the correlations into a correlation matrix. Calculate the correlation evolution of the correlation matrix at adjacent time points:

[0034] in, For correlation evolution, Let be the correlation matrix at time t. for Correlation matrix at time points, The spatial correlation observation length can be preset to 1 month, or adjusted based on the actual collected data.

[0035] The third step is to retain measurement points with a correlation greater than a preset correlation threshold to obtain preprocessed displacement sequence data. That is, when... If the measurement points have high spatial correlation, they are considered to be grouped together.

[0036] This invention also includes: after collecting the latest monitoring data, using the K-means unsupervised learning algorithm to... Multi-point displacement monitoring data within the time window The data is divided into two categories. If the latest collected data belongs to a single category, it indicates that there is an anomaly in the data evolution. In this case, the abnormal data will be interpreted and repaired. Otherwise, no action will be taken.

[0037] When multiple measurement points within the same highly correlated measurement point group undergo abrupt changes simultaneously, and within a time window... When no external disturbance signals such as blasting disturbance are detected, the data is considered a true mechanical response of the surrounding rock. When a single measuring point shows a sudden change while other measuring points in its highly correlated group do not, it is considered an equipment acquisition anomaly, and the data for that point is deleted. When a large number of measuring points show significant changes within the same time window, and blasting disturbance signals are present within the window, it is considered an event to be observed, and its trend needs to be continuously monitored within an extended time window. If the displacement after the sudden change shows a continuous accumulation or maintenance characteristic, it is considered a damage response induced by blasting disturbance. If the sudden change rapidly falls back and shows random fluctuation characteristics, it is considered sensor noise caused by blasting disturbance, and is accordingly removed. Through the above processing, a higher quality deep displacement dataset can be obtained. .

[0038] S2. Based on the preprocessed displacement sequence data, a physical constraint optimization fusion algorithm is used to construct the three-dimensional displacement field of the surrounding rock.

[0039] Since three-dimensional laser scanning is a discrete periodic observation with insufficient timeliness, and multi-point displacement gauges are deep discrete point observations, in order to obtain a continuous, unified three-dimensional displacement field that can be used for inversion.

[0040] A three-dimensional displacement field of the surrounding rock is constructed using a physical constraint optimization fusion algorithm, including: The first step is to calculate the displacement of the surrounding rock surface based on the three-dimensional laser scanning point cloud data.

[0041] Point clouds of adjacent scan cycles and Perform registration and calculate the displacement of the surrounding rock surface. .

[0042] Step 2, set To solve for the displacement field, the surface displacement observation operator is: The deep hole displacement observation operator is By introducing a smoothing regularization term and physical information constraints, a second objective function is constructed:

[0043] in, The second objective function is... For surface displacement observation operators, The reconstructed three-dimensional displacement field of the surrounding rock. This represents the displacement of the surrounding rock surface. For deep hole displacement observation operators, This is the preprocessed displacement sequence data. The weights of the smoothing term, For smoothing terms, The weights of the physical constraint terms, For physical constraint terms, The smoothing term is used to ensure the spatial continuity of the displacement field. It uses the spatial gradient as the basis for calculating the penalty coefficient. The formula for the smoothing term is:

[0044] in, For the weight function, Let be the displacement gradient tensor. For the computational domain, when the location is determined by the density or energy of microseismic events to have a rupture or damage concentration, a smaller weight is applied to allow for local discontinuities or deformation inconsistencies, thereby avoiding over-smoothing of the rupture area.

[0045] The formula for calculating the physical constraint term is:

[0046] in, The weight of the penalty term, As a penalty item, The weights of the compatibility constraint terms, These are compatibility constraints. Using the square integral of the residuals from the equilibrium equations as a penalty term, and assuming the constrained displacement field satisfies mechanical equilibrium, the formula for calculating the penalty term is:

[0047] in, For divergence, For displacement field The derived stress tensor This is a physical activity item.

[0048] Equivalent elasticity relation can be used It means that, among them, Represented as:

[0049] in, For the equivalent elasticity matrix, For displacement field The calculated small deformation strain tensor.

[0050] The formula for calculating the compatibility constraint term is:

[0051] in, For compatibility constraints, This is the strain tensor obtained by taking the symmetrical part of the displacement gradient.

[0052] The third step is to use the finite memory quasi-Newton method to solve the second objective function and obtain the three-dimensional displacement field of the surrounding rock.

[0053] Specifically, the computational domain is discretized into grid nodes, and the three-dimensional displacement of each node is used to form an unknown vector. In each iteration, the observation residual, smoothing residual and physical constraint residual are calculated and the gradient of the objective function is constructed. The finite memory quasi-Newton method (L-BFGS) is used to solve the problem. The search direction is calculated and the step size is determined by the linear search to update the displacement vector. The iteration stops when the objective function changes relatively or the gradient norm is less than a preset threshold, and the three-dimensional displacement field is obtained.

[0054] S3. Obtain the baseline parameters of rock mass basic mechanics, construct a three-dimensional geomechanical model based on monitoring data, construct a heterogeneous mechanical parameter field based on the baseline parameters of basic mechanics, simulate the three-dimensional geomechanical model using the heterogeneous mechanical parameter field, and construct a simulation database based on the simulation results.

[0055] For high-risk areas, the true contour and convergence information of the tunnels are extracted using 3D laser scanning point clouds, and a 3D geomechanical model is constructed by combining it with mine geological data. Baseline rock mass mechanical parameters M0, including elastic modulus, Poisson's ratio, and cohesion, are obtained through indoor uniaxial compression, Brazilian fracturing, and triaxial tests. Stress inversion methods such as bubbling or linear regression are used to obtain the heterogeneous stress field in the high-risk area, which serves as the input boundary conditions for subsequent stochastic simulations and surrogate models.

[0056] The steps for generating a non-homogeneous mechanical parameter field include: Centered on the baseline parameters of rock mass basic mechanics, a random fluctuation range for each parameter is set (e.g., ±30% or set according to geological zones). Latin hypercube sampling is used to generate random parameter samples. The random parameter samples are mapped to a spatial grid to form a heterogeneous mechanical parameter field. The baseline parameters of rock mass basic mechanics include elastic modulus, Poisson's ratio, and cohesion.

[0057] In the three-dimensional geomechanical model, the elastic-brittle damage constitutive model is used to describe the rock mass damage and fracture process. FLAC3D is used to simulate random parameter samples. The output of the three-dimensional geomechanical model is displacement field, stress field, strain field, damage field and energy terms (elastic energy, dissipated energy, etc.).

[0058] A simulation database is constructed based on the input being a non-homogeneous mechanical parameter field and the output being a displacement field, stress field, strain field, damage field, and energy term. .

[0059] S4. Train the agent model using a simulated database.

[0060] Proxy Model The model is based on a 3D U-Net three-dimensional encoder-decoder network. The input of the surrogate model is a non-homogeneous mechanical parameter field, and the output of the surrogate model is the displacement field, stress field, strain field, damage field, and energy term.

[0061] The training objective is to minimize the error between the surrogate output and the true value in the database, using L2 loss. After training, the surrogate model realizes the transformation from mechanical parameters to the mechanical response Y. sim The rapid mapping improves computational efficiency and provides support for the dynamic correction of mechanical parameters.

[0062] S5. Input the rock mass foundation mechanical baseline parameters into the trained surrogate model to obtain the initial mechanical response data, which includes the displacement field and the damage field.

[0063] The initial mechanical response data are:

[0064] in, It must contain at least a displacement field. Damage field Stress field, energy field, etc.

[0065] The calculation methods for rock mass damage and observations include: Microseismic monitoring output is a set of events. The computational domain Ω is discretized into voxel elements using a spatial grid consistent with the surrogate model. Its center is .

[0066] The first step is to discretize the computational domain into voxel elements using a spatial grid consistent with the proxy model, and set the rock mass damage and observation sub-values ​​of all voxels to 0.

[0067] The second step is to acquire microseismic monitoring data, which includes a set of microseismic events inside the rock mass. The set of microseismic events inside the rock mass includes the location of the events, the time of occurrence, and the extent of their impact.

[0068] The third step is to set the micro-seismic events within the time window as a sphere with the unknown event as the center and the area of ​​influence as the radius.

[0069] Fourth, a simplified binary classification model is adopted, where locations with microseismic events are considered fractured, and all other locations are considered intact. The rock mass damage and observation sub-values ​​for all voxels within the sphere are assigned a value of 1.

[0070] Step 5: After traversing all microseismic events, output the rock mass damage and observation data.

[0071] S6. Based on the initial mechanical response data, the three-dimensional displacement field of the surrounding rock, the rock mass damage, and the observation data, construct the first objective function.

[0072] The formula for calculating the first objective function is:

[0073] in, Let the first objective function be... For the weighting coefficients of the displacement field, For the three-dimensional displacement field of the surrounding rock, For displacement field, These are the weighting coefficients for the damage field. For rock mass damage, For the observation sub, For damage field, These are rock mass mechanical parameters.

[0074] S7. The mechanical parameters are iteratively updated using an optimization algorithm. The updated mechanical parameters are then input into the trained surrogate model until the first objective function converges. The updated mechanical parameters are then used as the optimal mechanical parameter field.

[0075] Iterative updates using the Particle Swarm Optimization (PSO) algorithm In each iteration, the proxy model is used for rapid computation. and This accelerates the calculation of the objective function and yields the globally optimal physical and mechanical parameter field that matches the field conditions. This enables intelligent calibration of physical and mechanical parameters based on real-time monitoring data.

[0076] Distribution of optimal mechanical parameters By incorporating the proxy model, the displacement, stress, strain, damage, and energy evolution of the rock mass deformation and fracturing process are calculated. Key characteristics of the disaster process (damage penetration path, structural slip zone, energy mutation time, etc.) are output and future stage predictions are made. This fully leverages the advantages of monitoring data in accurately reflecting response and timeliness, as well as the advantages of the mechanism model in revealing disaster mechanisms and making predictions.

[0077] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0078] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing rock deformation and fracture process by data-analog hybrid driving, characterized in that, Includes the following steps: Collect monitoring data, including three-dimensional laser scanning point cloud data, displacement sequence data, microseismic monitoring data, and blasting disturbance monitoring data; preprocess the displacement sequence data. Based on the preprocessed displacement sequence data, a physical constraint optimization fusion algorithm is used to construct the three-dimensional displacement field of the surrounding rock. Obtain the basic mechanical baseline parameters of the rock mass, construct a three-dimensional geomechanical model based on monitoring data, construct a heterogeneous mechanical parameter field based on the basic mechanical baseline parameters, simulate the three-dimensional geomechanical model using the heterogeneous mechanical parameter field, and construct a simulation database based on the simulation results; The agent model is trained using the simulated database; The rock mass foundation mechanical baseline parameters are input into the trained surrogate model to obtain initial mechanical response data, which includes displacement field and damage field. Based on the initial mechanical response data and the three-dimensional displacement field of the surrounding rock, a first objective function is constructed. The mechanical parameters are iteratively updated using an optimization algorithm. The updated mechanical parameters are then input into the trained surrogate model until the first objective function converges. The updated mechanical parameters are then used as the optimal mechanical parameter field.

2. The data-analogue hybrid driven rock deformation and failure process analysis method according to claim 1, characterized in that, The construction of the three-dimensional displacement field of the surrounding rock using a physical constraint optimization fusion algorithm includes: Calculate the surface displacement of the surrounding rock based on 3D laser scanning point cloud data; Construct the second objective function: in, The second objective function is... For surface displacement observation operators, The reconstructed three-dimensional displacement field of the surrounding rock. This represents the displacement of the surrounding rock surface. For deep hole displacement observation operators, This is the preprocessed displacement sequence data. The weights of the smoothing term, For smoothing terms, The weights of the physical constraint terms, For physical constraints, The formula for calculating the smoothing term is: in, For spatial weighting functions, Let be the displacement gradient tensor. For the computational domain, It is the Frobenius norm. The formula for calculating the physical constraint term is: wherein, is a weight of the balance equation constraint term, is a balance equation constraint term, is a weight of the compatibility constraint term, is a compatibility constraint term, The formula for calculating the penalty item is: wherein, is a physical constraint weight function, is a stress tensor, is a divergence, is a body force term, The formula for calculating the compatibility constraint term is as follows: wherein is a compatibility constraint term, is a strain tensor; The second objective function is solved using the finite memory quasi-Newton method to obtain the three-dimensional displacement field of the surrounding rock.

3. The data-analogue hybrid driven rock deformation and failure process analysis method according to claim 1, characterized in that, The steps for generating the heterogeneous mechanical parameter field include: Centered on the baseline parameters of rock mass foundation mechanics, the random fluctuation range of each parameter is set, and random parameter samples are generated by Latin hypercube sampling. The random parameter samples are mapped to a spatial grid to form a heterogeneous mechanical parameter field. The baseline parameters of rock mass foundation mechanics include elastic modulus, Poisson's ratio and cohesion.

4. The data-simulation hybrid-driven rock deformation and fracturing process analysis method according to claim 1, characterized in that, The three-dimensional geomechanical model uses an elastic-brittle damage constitutive model to describe the rock mass damage and fracture process. FLAC3D is used to simulate random parameter samples. The output of the three-dimensional geomechanical model is displacement field, stress field, strain field, damage field and energy term.

5. The data-analogue hybrid driven rock deformation and failure process analysis method of claim 1, wherein, The proxy model is established based on the 3D U-Net three-dimensional encoder-decoder network. The input of the proxy model is a non-homogeneous mechanical parameter field, and the output of the proxy model is the displacement field, stress field, strain field, damage field, and energy term.

6. The data-analogue hybrid driven rock deformation and failure process analysis method of claim 1, wherein, The calculation methods for rock mass damage and observations include: The computational domain is discretized into voxel elements using a spatial grid consistent with the proxy model, and the rock mass damage and observation sub-values ​​of all voxels are set to 0. Acquire microseismic monitoring data, which includes a set of microseismic events inside the rock mass, including the event location, occurrence time, and affected area; The microseismic event is defined as a sphere with the unknown event location as its center and the area of ​​influence as its radius. A simplified binary classification model is adopted, and the rock mass damage and observation sub-value of all voxels within the sphere are assigned to 1; After traversing all microseismic events, output the rock mass damage and observation data.

7. The data-analogue hybrid driven rock deformation and failure process analysis method of claim 1, wherein, The calculation formula of the first target function is: in, Let the first objective function be... For the weighting coefficients of the displacement field, For the three-dimensional displacement field of the surrounding rock, For displacement field, These are the weighting coefficients for the damage field. For rock mass damage, For the observation sub, For damage field, These are rock mass mechanics parameters.

8. The data-analogue hybrid driven rock deformation and failure process analysis method of claim 1, wherein, The rock mass foundation mechanical baseline parameters are obtained through indoor uniaxial compression, Brazilian splitting and triaxial test.

9. The data-analogue hybrid driven rock deformation and failure process analysis method of claim 1, wherein, The stress tensor is adopted express, For the equivalent elasticity matrix, For displacement field The calculated small deformation strain tensor, obtained from the displacement field U, is calculated using the following formula: 。

Images