Mine prestressed steel strand support optimization method and system based on digital twinning

Abstract

The application discloses a mine prestressed steel strand supporting optimization method and system based on digital twinning, and belongs to the technical field of steel strand supporting. In view of the problems of traditional supporting design, such as experience dependence, poor dynamic adaptability, and difficulty in playing the advantages of new steel strands, a high-precision digital twinning model of a mine roadway is constructed, parameters such as steel strand specifications, arrangement modes, and pre-tightening forces are optimized by taking supporting safety, economy, and construction efficiency as comprehensive targets, the model is dynamically calibrated through real-time monitoring data, an LSTM-random forest combined model is used to deduce a health evolution trend of a supporting system and to grade early warning, and standardized construction guidance documents are generated to form a mechanism of design, construction, monitoring, and optimization. The application can fully play the technical advantages of super-high elongation and hollow low-relaxation steel strands, significantly improve the accuracy and safety of supporting design under complex geological conditions, and provide an integrated supporting solution for mines.

Description

Technical Field

[0001] This invention belongs to the field of steel strand support technology, specifically relating to an optimization method and system for mine prestressed steel strand support based on digital twins. Background Technology

[0002] Mine roadway support is a fundamental engineering project for ensuring safe production in underground mines. Its design rationality and operational reliability directly affect the safety of miners and mining efficiency. As mining depth continues to extend deeper, deep mines generally face a complex geological environment characterized by "three highs and one disturbance"—high ground stress, high osmotic pressure, high ground temperature, and strong mining disturbance. Geological disasters such as rock bursts, large deformations of soft rock, and fracturing and instability of surrounding rock occur frequently, posing unprecedented challenges to the safety, adaptability, and economy of roadway support systems. Prestressed steel strand support, as one of the most mainstream support forms for mine roadways, has advantages such as strong load-bearing capacity, convenient construction, and wide applicability, and has been widely used in roadway support projects in various mines.

[0003] Currently, traditional support design is mainly based on empirical formulas and industry-standard specifications, employing analogy and empirical correction methods. This makes it difficult to provide precise, customized designs for the diverse geological conditions of different mining areas and working faces, resulting in widespread issues of "over-support" or "under-support." Over-support leads to material waste, increased construction costs, and extended construction periods, while under-support can cause safety accidents such as large tunnel deformation and roof falls. Furthermore, existing design methods are static, one-off designs, failing to consider the dynamic changes in geological conditions and the cumulative effects of mining disturbances during construction. During tunnel excavation and mining, the stress state and mechanical properties of the surrounding rock change significantly. Once traditional support parameters are determined, they are difficult to adjust in real-time according to actual working conditions, leading to a mismatch between the support system and the dynamic evolution of the surrounding rock, and a significant reduction in support reliability.

[0004] In recent years, new support materials such as ultra-high elongation prestressed steel strands and hollow low-relaxation prestressed steel strands have emerged, exhibiting significant technical advantages in rockburst control and grouting reinforcement. Ultra-high elongation steel strands possess large deformation and high energy absorption properties, effectively dissipating the energy released by rockbursts; hollow low-relaxation steel strands enable integrated support and grouting construction, significantly improving grouting efficiency and surrounding rock reinforcement. However, existing design methods have not established optimization models for the unique mechanical properties of these new steel strands, simply replacing the design parameters of ordinary steel strands. This prevents the full realization of their technical advantages and hinders the widespread application of these new support materials. Furthermore, existing monitoring methods are mostly manual inspections or single-point offline monitoring, resulting in low data collection frequency, poor real-time performance, and limited coverage. Data processing and risk identification heavily rely on human experience, making it difficult to accurately assess the health status of the support system and provide early warnings of potential risks, thus failing to meet the requirements for proactive safety control under complex geological conditions. Summary of the Invention

[0005] To address these issues, this invention provides a digital twin-based optimization method and system for prestressed steel strand support in mines, which solves the problems of insufficient design accuracy, poor dynamic adaptability, underutilization of the advantages of new materials, and lack of full life-cycle management in existing prestressed steel strand support technologies for mines.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a digital twin-based optimization method for prestressed steel strand support in mines, comprising the following steps:

[0007] S1: Collect and standardize multi-source heterogeneous data of mine dimensions in a unified manner to build a support engineering database;

[0008] S2: Based on the static basic data of the support engineering database, construct an initial three-dimensional digital twin model of the mine roadway to replicate the physical roadway geometry, geological structure and support structure, and simulate the coupling and interaction law between the surrounding rock and the support structure.

[0009] S3: Using the digital twin initial model as the simulation calculation carrier, construct a support system design model. The support system design model takes support safety, economy and construction efficiency as comprehensive optimization objectives and outputs the optimal initial design parameter set of prestressed steel strands.

[0010] S4: During the entire process of physical tunnel construction and operation, real-time dynamic status data of surrounding rock and support structure are collected, and the dynamic status data is synchronously injected into the initial digital twin model to complete the dynamic calibration and iterative update of the initial digital twin model, thereby obtaining a digital twin mirror model that is consistent with the physical entity in real time.

[0011] S5: Based on the historical operating data and real-time synchronized data of the digital twin mirror model, the health evolution trend of the support system is inferred through machine learning algorithms, potential instability risks are identified, and graded warnings are issued according to the risk level.

[0012] S6: Based on the optimal initial design parameter set and the dynamically adjusted support parameters, generate a standardized construction guidance document that conforms to mine safety regulations, conduct a full-process quantitative evaluation of the construction quality and support effect of the support system, and feed the evaluation results back to the support system design model to drive the support system design model to continuously learn and iteratively optimize.

[0013] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, in step S1, the standardization and fusion of the multi-source heterogeneous data adopts a minimum-maximum normalization model to map mine support engineering data of different dimensions and magnitudes to the [0,1] interval:

[0014]

[0015] In the formula, For the first Sample of a mine roadway support project The normalized values ​​of each feature parameter; For the first Sample of a mine roadway support project The original collected values ​​of each characteristic parameter include geological parameters, steel strand mechanical parameters, construction parameters, and monitoring parameters; For the first The minimum value of each characteristic parameter in all completed support engineering samples; For the first The maximum value of each characteristic parameter across all completed support engineering samples.

[0016] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, in step S2, during the simulation of the coupling interaction between the surrounding rock and the support structure, an elastoplastic finite element model is adopted. The elastoplastic finite element model uses the Mohr-Coulomb yield criterion to describe the plastic failure characteristics of the surrounding rock, and the static equilibrium equation is:

[0017]

[0018] In the formula, The overall stiffness matrix of the surrounding rock-support system is formed by superimposing the elastic-plastic stiffness matrix of the roadway surrounding rock and the stiffness matrix of the steel strand-anchor support structure. The nodal displacement vector includes the roadway roof subsidence displacement, the sidewall convergence displacement, the floor bulge displacement, and the axial deformation displacement of the steel strands. The nodal load vector includes the original rock stress, steel strand preload, anchor bolt axial force, and construction disturbance load.

[0019] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, in step S3, the comprehensive optimization objective is achieved through a multi-objective optimization function, which simultaneously minimizes support risk, engineering cost, and construction period, as shown in the following formula:

[0020]

[0021] In the formula, The design variable vector contains continuous and discrete variables, which correspond to the specifications, structural form, spatial arrangement, preload value, and anchorage length of the steel strand, respectively. The safety objective function represents the comprehensive safety factor of the support system, which is calculated by weighting the surrounding rock stability safety factor, the steel strand tensile safety factor, and the anchor pull-out safety factor. The objective function is the economic cost, which characterizes the comprehensive cost per unit length of the support project, including the cost of steel strand materials, drilling construction cost, grouting cost, and labor cost. The objective function for construction efficiency is to characterize the standard construction time for roadway support per unit length, including drilling time, steel strand installation time, pre-tightening time, and grouting solidification time.

[0022] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, the multi-objective optimization function uses the analytic hierarchy process (AHP) to determine the weight coefficients of each objective. The weight calculation model is as follows:

[0023]

[0024] In the formula, For the first The weighting coefficients of each optimization objective; For the first The first objective is relative to the second objective. The importance scale values ​​for each objective are calculated using a 1-9 scale, where 1 indicates that both objectives are equally important, and 9 indicates that one objective is extremely more important than the other. To optimize the total number of objectives; after calculating the weight vector, a consistency check is performed, and when the consistency ratio is... At that time, it is considered that the judgment matrix has satisfactory consistency and the weight coefficients are effective.

[0025] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, in step S3, during the construction of the support system design model, an energy matching model is adopted for the optimization design of ultra-high elongation steel strands in roadways with rockburst risk, so that the energy absorption capacity of the support system is greater than the impact energy of the surrounding rock. The formula is as follows:

[0026]

[0027] In the formula, The total deformation energy absorption capacity of all ultra-high elongation steel strands participating in the load-bearing area is the product of the plastic deformation energy absorption of a single steel strand within the allowable elongation range and the effective number of working strands; the ultimate energy absorption of a single steel strand is the integral value of its load-elongation curve within the allowable elongation range. It represents the impact energy of the surrounding rock acting on the support area when a rockburst occurs, and is the sum of the kinetic energy and elastic deformation energy released by the surrounding rock. The safety reserve factor characterizes the redundancy of the support system in resisting impact energy and is determined according to the risk level of rockburst.

[0028] For the optimized design of hollow low-relaxation steel strands in grouting reinforced roadways, a spherical permeation and diffusion model of grout under static pressure grouting conditions in homogeneous isotropic surrounding rock is adopted to determine the optimal grouting parameters and steel strand spacing. The formula is as follows:

[0029]

[0030] In the formula, The effective diffusion radius of cement slurry in the fractured zone of the surrounding rock of the roadway; The initial average permeability coefficient of the fractured zone of the surrounding rock in the roadway before grouting; This represents the water head height corresponding to the grouting orifice pressure. This refers to the duration of grouting in a single hole; The average porosity of the fractured zone of the surrounding rock in the tunnel; The effective filling coefficient of the grout represents the proportion of the surrounding rock pores that are actually filled and cemented by the grout.

[0031] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, in step S4, the dynamic calibration of the initial digital twin model adopts the extended Kalman filter algorithm, including prediction and update steps, to achieve real-time synchronization between the initial digital twin model and the physical entity, as shown in the following formula:

[0032]

[0033] In the formula, For the first The prior state estimation vector at time t contains the surrounding rock displacement, steel strand stress, anchor bolt axial force, and top plate delamination value; The state transition matrix represents the state evolution law of the surrounding rock-support system under no external disturbance. To control the input matrix, which characterizes the impact of support parameter adjustments on the system state; The input vector is controlled by adjusting the preload of the steel strand, the grouting pressure, and the number of additional anchor bolts. For the first The calibrated state vector at time step; For the first The Kalman gain matrix at time step 1 is used to balance the weights of the predicted and measured values. For the first Vector of actual on-site monitoring data at any given moment; For the first The observation matrix at any given time represents the mapping relationship between system state variables and monitoring data;

[0034] The monitoring data collection frequency adopts an adaptive sampling model based on risk level, increasing the sampling frequency in high-risk areas and decreasing the sampling frequency in low-risk areas, as shown in the following formula:

[0035]

[0036] In the formula, for The frequency of monitoring data collection at any given time; This is the baseline acquisition frequency under normal geological conditions, and the value is determined according to the requirements of routine safety monitoring in mines. The risk sensitivity coefficient is determined based on the geological hazard level of the mine, with a larger value taken for mines with a high risk of rockburst. for The real-time risk level of the support system is measured at any given moment, with a value range of [0,1], where 0 represents no risk and 1 represents extremely high risk.

[0037] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, in step S5, during the process of inferring the health evolution trend of the support system through machine learning algorithms, the health status prediction of the support system adopts an LSTM-random forest combination model, which combines the advantages of time series prediction and feature classification. The formula is as follows:

[0038]

[0039] In the formula, The overall failure probability of the support system within the next 72 hours; The failure probability predicted by the LSTM model based on time series data; This refers to the failure probability predicted by the random forest model based on multi-dimensional feature data. These are the weight coefficients of the LSTM model, which are dynamically adjusted based on historical prediction accuracy, and their values ​​range from 0 to 1.

[0040] As a preferred scheme for the optimization method of prestressed steel strand support in mines based on digital twins, in step S6, during the full-process quantitative evaluation of the construction quality and support effect of the support system, the comprehensive evaluation of the support effect adopts a fuzzy comprehensive evaluation model, and the formula is as follows:

[0041]

[0042] In the formula, This is a vector representing the overall evaluation results, corresponding to four evaluation levels: excellent, good, satisfactory, and unsatisfactory. The weight vector for each evaluation indicator is determined by the analytic hierarchy process, including the weights of indicators such as the pretension force qualification rate of steel strand, the anchorage length qualification rate, and the surrounding rock convergence rate. A fuzzy relation matrix is ​​used to represent the membership degree of each evaluation indicator to different evaluation levels. It is established by combining expert scoring and historical data statistics. It is a weighted average type fuzzy synthesis operator.

[0043] This invention also provides a digital twin-based optimization system for prestressed steel strand support in mines, employing the aforementioned digital twin-based optimization method for prestressed steel strand support in mines, comprising:

[0044] The data integration module is used to uniformly collect and standardize the integration of multi-source heterogeneous data of the dimensions set by the mine to build a support engineering database;

[0045] The digital twin model construction module is used to construct an initial three-dimensional digital twin model of the mine roadway based on the static basic data of the support engineering database, realize the replication of the physical roadway geometry, geological structure and support structure, and simulate the coupling and interaction law between the surrounding rock and the support structure.

[0046] The support system design module is used to construct a support system design model using the digital twin initial model as a simulation calculation carrier. The support system design model takes support safety, economy and construction efficiency as comprehensive optimization objectives and outputs the optimal initial design parameter set of prestressed steel strands.

[0047] The dynamic optimization module is used to collect dynamic status data of the surrounding rock and support structure in real time throughout the construction and operation of the physical tunnel, and to synchronously inject the dynamic status data into the initial digital twin model to complete the dynamic calibration and iterative update of the initial digital twin model, thereby obtaining a digital twin mirror model that is consistent with the physical entity in real time.

[0048] The health monitoring module is used to infer the health evolution trend of the support system based on the historical operating data and real-time synchronized data of the digital twin mirror model, through machine learning algorithms, identify potential instability risks, and issue graded warnings according to the risk level.

[0049] The construction guidance and effect evaluation module is used to generate standardized construction guidance documents that comply with mine safety regulations based on the optimal initial design parameter set and the dynamically adjusted support parameters. It performs a full-process quantitative evaluation of the construction quality and support effect of the support system and feeds the evaluation results back to the support system design model to drive the support system design model to continuously learn and iteratively optimize.

[0050] The present invention has the following advantages:

[0051] This invention abandons the traditional analogy method and static design mode based on experience correction. It constructs a high-precision digital twin model based on multi-source heterogeneous data from all dimensions of the mine, and uses elastoplastic finite element simulation to realistically reproduce the coupling interaction law between the surrounding rock and the support structure. Global optimization is performed with multiple objectives, including support safety, economy, and construction efficiency, achieving customized and precise design for different geological conditions. This method effectively avoids material waste and construction delays caused by experience bias, significantly reduces the overall cost of support engineering, greatly reduces the roadway repair rate, and improves the economic benefits of mining.

[0052] This invention establishes a real-time bidirectional mapping mechanism between physical tunnels and digital twin models. Dynamic data such as surrounding rock deformation and steel strand stress are collected in real time using on-site sensors. An extended Kalman filter algorithm is used to dynamically calibrate the digital twin model, and support parameters are automatically adjusted based on the updated model. This allows the support system to adapt to dynamic changes in geological conditions and the cumulative effects of mining disturbances during construction. Simultaneously, an LSTM-random forest combined model is used to predict the health evolution trend of the support system, enabling early warning of potential instability risks. This transforms traditional passive repair into proactive prevention and control, effectively reducing the probability of geological disasters such as rock bursts, large deformations in soft rock, and roof falls, thus ensuring the safety of mine workers.

[0053] This invention addresses the rockburst risk in roadways by employing an energy matching model to optimize the arrangement of ultra-high elongation steel strands, maximizing their large deformation and high energy absorption characteristics and significantly enhancing rockburst prevention capabilities. For grouting-reinforced roadways, a spherical grout penetration and diffusion model optimizes the grouting parameters of hollow, low-relaxation steel strands, enabling integrated support and grouting construction, significantly improving grouting efficiency and enhancing surrounding rock reinforcement. This technical solution addresses the industry pain point of having products for new support materials but lacking supporting design methods, providing strong technical support for the market promotion of core products.

[0054] The standardized construction guidance documents generated during the construction process of this invention can effectively regulate on-site operations and ensure construction quality; the real-time quantitative evaluation results of the support effect can be fed back to the intelligent design model, driving the model to continuously learn and iteratively optimize. With the continuous accumulation of engineering data, the design accuracy and prediction accuracy of the system will continue to improve, forming a virtuous cycle of data-model-application-data. Attached Figure Description

[0055] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0056] The structures, proportions, sizes, etc. illustrated in this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed herein, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0057] Figure 1 This is a schematic diagram of the optimization method for mine prestressed steel strand support based on digital twin provided in an embodiment of the present invention;

[0058] Figure 2 This is a technical roadmap for the optimization method of mine prestressed steel strand support based on digital twin provided in the embodiments of the present invention;

[0059] Figure 3 This is a diagram of the machine learning algorithm architecture in the digital twin-based optimization method for prestressed steel strand support in mines provided in this embodiment of the invention.

[0060] Figure 4 The figure shows the accuracy verification results of the digital twin model in the optimization method for mine prestressed steel strand support based on digital twin provided in the embodiments of the present invention.

[0061] Figure 5 This is a comparison chart of the intelligent design effects in the optimization method for prestressed steel strand support in mines based on digital twins provided in this embodiment of the invention, wherein (a) support material cost; (b) construction time; (c) construction quality pass rate; and (d) surrounding rock deformation.

[0062] Figure 6This is a comparison of the dynamic optimization effects in the digital twin-based optimization method for prestressed steel strand support in mines provided in this embodiment of the invention; where (a) is the parameter adjustment time; and (b) is the deformation of the surrounding rock after adjustment.

[0063] Figure 7 The diagram shows the health status prediction effect of the optimization method for mine prestressed steel strand support based on digital twin provided in the embodiments of the present invention; wherein, (a) accuracy rate ≥90%, false alarm rate ≤5%, false alarm rate 0%, and the red dashed line is the indicator threshold; (b) average early warning time;

[0064] Figure 8 The images show the application effects of a novel steel strand in the optimization method for mine prestressed steel strand support based on digital twins provided in this embodiment of the invention; wherein, (a) the deformation of the ultra-high elongation steel strand is 3-5 times that of ordinary strands, effectively absorbing energy without breakage; (b) the grouting fullness of the hollow low-relaxation steel strand is increased by >20%, and the grouting time is shortened by 30%;

[0065] Figure 9 This is a diagram of the adaptive control system architecture for the prestressed steel strand stabilization process parameters provided in this embodiment of the invention. Detailed Implementation

[0066] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0067] Example 1

[0068] See Figure 1 and Figure 2 This invention provides an optimization method for prestressed steel strand support in mines based on digital twins, comprising the following steps:

[0069] S1: Collect and standardize multi-source heterogeneous data of mine dimensions in a unified manner to build a support engineering database.

[0070] Specifically, mine support engineering data exhibits heterogeneous characteristics due to its multiple sources, formats, and dimensions. The original data is scattered across geological exploration departments, design units, construction teams, and monitoring systems, resulting in data silos, inconsistent formats, and significant dimensional differences, making it unsuitable for direct use in digital twin modeling and calculation. Step S1 normalizes geological exploration data (borehole columnar sections, rock mechanics parameters, fault distribution), tunnel design parameters (cross-sectional dimensions, excavation methods, support types), support system parameters (steel strand specifications, anchor bolt parameters, preload requirements), construction technology data (borehole depth, grouting pressure, construction progress), and historical engineering data (support failure cases, repair records) through a unified data acquisition interface and standardized processing flow. This eliminates data redundancy and inconsistencies, constructing a unified support engineering database. This database serves as the sole reliable data foundation for the entire system, providing fundamental data support for digital twin model construction, design optimization, and full lifecycle management.

[0071] S2: Based on the static basic data of the support engineering database, construct an initial three-dimensional digital twin model of the mine roadway to replicate the physical roadway geometry, geological structure and support structure, and simulate the coupling and interaction law between the surrounding rock and the support structure.

[0072] Specifically, the core of digital twin technology is to achieve the mapping between physical entities and digital models. The initial model is the static foundation of the entire digital twin system. Step S2 employs parametric modeling technology, constructing three sub-models based on static data from the support engineering database: a geological body model: using Kriging interpolation to spatially interpolate borehole exploration data, generating a three-dimensional geological model of the rock mass surrounding the tunnel, accurately reproducing geological structures such as strata distribution, faults, and folds; a support structure model: generating three-dimensional models of support structures such as steel strands, anchor bolts, and anchor mesh according to design parameters, endowing them with realistic mechanical properties (elastic modulus, yield strength, elongation, etc.); and a surrounding rock-support coupling model: using the finite element method to mesh and couple the geological body model and the support structure model, simulating the interaction between the two. The surrounding rock and support structure are not isolated entities, but rather a collaborative system that shares the load. After excavation, the surrounding rock will undergo stress redistribution, deformation, and apply loads to the support structure; the support structure, through its own stiffness and strength, limits the deformation of the surrounding rock while transferring the loads to the deeper, stable rock mass. The simulation of coupling and interaction laws can realistically reproduce the stress distribution, deformation trend, and stress state of the surrounding rock under different support parameters, providing a reliable simulation basis for design.

[0073] S3: Using the digital twin initial model as the simulation calculation carrier, construct a support system design model. The support system design model takes support safety, economy and construction efficiency as comprehensive optimization objectives and outputs the optimal initial design parameter set of prestressed steel strands.

[0074] Specifically, traditional support design employs a static model based on analogy and empirical correction, which cannot simultaneously address the three objectives of safety, economy, and construction efficiency. Step S3 uses a digital twin initial model as a simulation platform to construct a multi-objective optimization design model, transforming the support design into a mathematical optimization problem. The design variables are the specifications, structural form, spatial arrangement, preload, and anchorage length of the steel strands. The optimization objectives are support safety (maximizing the overall safety factor), economy (minimizing cost per unit length), and construction efficiency (minimizing construction time per unit length). A mixed-integer programming algorithm (capable of handling both continuous and discrete variables) is used for global optimization to obtain a Pareto optimal solution set. A Pareto optimal solution is one where no single solution can improve the performance of a particular objective without compromising the performance of any other objective. Designers can select the most suitable initial design parameter set from the optimal solution set based on actual engineering needs (e.g., prioritizing safety in high-risk mines) to achieve customized design.

[0075] S4: During the entire process of physical tunnel construction and operation, real-time dynamic status data of the surrounding rock and support structure are collected, and the dynamic status data is synchronously injected into the initial digital twin model to complete the dynamic calibration and iterative update of the initial digital twin model, thereby obtaining a digital twin mirror model that is consistent with the physical entity in real time.

[0076] Specifically, the initial digital twin model is constructed based on static data from the exploration phase, which inherently contains uncertainties such as spatial variability of geological conditions and construction errors. During tunnel construction and mining, the stress state and mechanical properties of the surrounding rock undergo significant changes, which the initial model cannot reflect. Step S4 uses fiber optic sensors, stress sensors, and displacement sensors deployed in the tunnel's surrounding rock and support structure to collect real-time dynamic state data such as surrounding rock deformation (roof subsidence, sidewall convergence, floor bulging), steel strand stress, anchor bolt axial force, and roof delamination. An extended Kalman filter algorithm is used to fuse the real-time monitoring data with the predicted values ​​of the initial digital twin model. The model parameters, such as the surrounding rock elastic modulus, cohesion, and internal friction angle, are dynamically calibrated and iteratively updated to reduce errors in the initial model, resulting in a digital twin mirror model that is synchronized with and consistent with the physical tunnel in real time. This mirror model accurately reflects the actual state of the physical tunnel at the current moment, providing an accurate digital carrier for dynamic optimization and health monitoring.

[0077] S5: Based on the historical operating data and real-time synchronized data of the digital twin mirror model, the health evolution trend of the support system is inferred through machine learning algorithms, potential instability risks are identified, and graded early warnings are issued according to the risk level.

[0078] Specifically, the failure of a support system is a gradual process from quantitative to qualitative change, and the evolutionary pattern of its health status is contained in historical and real-time monitoring data. Traditional risk identification methods rely on human experience, which suffers from strong subjectivity and delayed early warning. In this step S5, machine learning algorithms are used to train the historical operating data of the digital twin mirror model to learn the nonlinear mapping relationship between the health status of the support system and the monitoring data. Then, combined with real-time synchronous data, the future health evolution trend of the support system is inferred. The graded early warning mechanism classifies risks into different levels according to the predicted failure probability and takes corresponding prevention and control measures for different levels of risk: low-risk levels only require enhanced monitoring, medium-risk levels require adjustment of support parameters, and high-risk levels require immediate cessation of work and emergency reinforcement measures. This mechanism realizes the transformation from passive repair to proactive prevention and control, and can identify potential instability risks in advance to avoid safety accidents.

[0079] S6: Based on the optimal initial design parameter set and the dynamically adjusted support parameters, generate a standardized construction guidance document that conforms to mine safety regulations, conduct a full-process quantitative evaluation of the construction quality and support effect of the support system, and feed the evaluation results back to the support system design model to drive the support system design model to continuously learn and iteratively optimize.

[0080] Specifically, standardized construction is a crucial step in ensuring the effectiveness of support. The operational skills and adherence to standards by on-site construction personnel directly impact the load-bearing capacity of the support system. Step S6 automatically generates CAD construction drawings and standardized operating instructions that conform to the "Coal Mine Safety Regulations" and the "Technical Specifications for Rock Grading and Support in Mine Roadways" based on optimized support parameters. This clarifies the construction requirements, technical parameters, and quality standards for each process, standardizing on-site operations and reducing human error. Simultaneously, through quality inspection data during construction and support effect monitoring data during operation, a comprehensive quantitative evaluation of the support system's construction quality and support effect is conducted. The evaluation results are fed back to the support system design model as training data for self-learning, continuously refining the model's parameters and algorithms, improving design accuracy and prediction precision. This allows the system to continuously learn and evolve from engineering practice, and its performance will continuously improve with the accumulation of engineering data.

[0081] In one possible embodiment, in step S1, the standardized fusion of the multi-source heterogeneous data adopts a minimum-maximum normalization model to map mine support engineering data of different dimensions and magnitudes to the [0,1] interval:

[0082]

[0083] In the formula, For the first Sample of a mine roadway support project The normalized values ​​of each feature parameter; For the first Sample of a mine roadway support project The original collected values ​​of each characteristic parameter include geological parameters, steel strand mechanical parameters, construction parameters, and monitoring parameters; For the first The minimum value of each characteristic parameter in all completed support engineering samples; For the first The maximum value of each characteristic parameter across all completed support engineering samples.

[0084] Specifically, in mine support engineering, the dimensions and numerical magnitudes of different characteristic parameters vary greatly. For example, the unit of steel strand stress is MPa, and its numerical range is typically 100-1000; while the unit of surrounding rock displacement is mm, and its numerical range is typically 0-500. Without normalization, features with large numerical magnitudes will dominate in subsequent machine learning models and optimization algorithms, masking the influence of features with smaller numerical magnitudes, leading to model training failure or biased optimization results. Min-max normalization is simple and efficient, and it completely preserves the distribution characteristics and relative relationships of the original data without altering its inherent patterns. For each feature parameter, the normalized value is obtained by subtracting its minimum value and dividing by its range (the difference between the maximum and minimum values). Normalized data has a uniform scale, which improves the convergence speed and computational accuracy of subsequent models.

[0085] In one possible embodiment, in step S2, during the simulation of the coupling interaction between the surrounding rock and the support structure, an elastoplastic finite element model is used. This model employs the Mohr-Coulomb yield criterion to describe the plastic failure characteristics of the surrounding rock, and the static equilibrium equation is:

[0086]

[0087] In the formula, The overall stiffness matrix of the surrounding rock-support system is formed by superimposing the elastic-plastic stiffness matrix of the roadway surrounding rock and the stiffness matrix of the steel strand-anchor support structure. The nodal displacement vector includes the roadway roof subsidence displacement, the sidewall convergence displacement, the floor bulge displacement, and the axial deformation displacement of the steel strands. The nodal load vector includes the original rock stress, steel strand preload, anchor bolt axial force, and construction disturbance load.

[0088] Specifically, the finite element method (FEM) is a general numerical method for solving problems in continuum mechanics. Its core idea is to discretize a continuous physical domain into a finite number of interconnected elements. By solving the equilibrium equations of each element, the equilibrium equations of the entire system are obtained, and then the stress and displacement distributions of the system can be solved. The elastoplastic finite element model considers both elastic and plastic deformation of the material. When the stress is below the yield strength, the material undergoes elastic deformation, which is reversible; when the stress exceeds the yield strength, the material undergoes plastic deformation, which is irreversible.

[0089] Mine surrounding rock is a typical elastoplastic material, exhibiting significant plastic deformation after excavation. Therefore, an elastoplastic finite element model can more realistically simulate the mechanical behavior of the surrounding rock. The Mohr-Coulomb yield criterion is the most widely used yield criterion in mining engineering. It considers rock failure to be shear failure, where the shear stress on the failure surface equals the sum of cohesion and frictional resistance generated by normal stress, accurately describing the characteristic that the compressive strength of rock is much greater than its tensile strength. In the static equilibrium equations, the overall stiffness matrix reflects the system's ability to resist deformation and is composed of the superposition of the stiffness matrices of the surrounding rock elements and the support structure elements. The nodal displacement vector is the unknown quantity to be solved; the nodal load vector is the equivalent nodal load of all external forces acting on the system. By solving these equilibrium equations, the stress and displacement distributions of the surrounding rock and support structure under different support parameters can be obtained, providing a basis for support design.

[0090] In one possible embodiment, in step S3, the comprehensive optimization objective is achieved through a multi-objective optimization function, simultaneously minimizing support risk, engineering cost, and construction period, as shown in the following formula:

[0091]

[0092] In the formula, The design variable vector contains continuous and discrete variables, which correspond to the specifications, structural form, spatial arrangement, preload value, and anchorage length of the steel strand, respectively. The safety objective function represents the comprehensive safety factor of the support system, which is calculated by weighting the surrounding rock stability safety factor, the steel strand tensile safety factor, and the anchor pull-out safety factor. The objective function is the economic cost, which characterizes the comprehensive cost per unit length of the support project, including the cost of steel strand materials, drilling construction cost, grouting cost, and labor cost. The objective function for construction efficiency is to characterize the standard construction time for roadway support per unit length, including drilling time, steel strand installation time, pre-tightening time, and grouting solidification time.

[0093] The multi-objective optimization function uses the analytic hierarchy process (AHP) to determine the weight coefficients of each objective, and the weight calculation model is as follows:

[0094]

[0095] In the formula, For the first The weighting coefficients of each optimization objective; For the first The first objective is relative to the second objective. The importance scale values ​​for each objective are calculated using a 1-9 scale, where 1 indicates that both objectives are equally important, and 9 indicates that one objective is extremely more important than the other. To optimize the total number of objectives; after calculating the weight vector, a consistency check is performed, and when the consistency ratio is... At that time, it is considered that the judgment matrix has satisfactory consistency and the weight coefficients are effective.

[0096] Specifically, the multi-objective optimization problem is to find a set of non-dominated solutions (Pareto optimal solutions) such that no single solution can improve the performance of any one objective without degrading the performance of any other objective. This invention transforms the three objectives of support design into three objective functions, constructing a multi-objective optimization function. By minimizing To achieve synergistic optimization of three objectives, the importance of each objective varies across different engineering scenarios, necessitating the determination of its weight coefficients. A multi-criteria decision-making method combining qualitative and quantitative analysis using the Analytic Hierarchy Process (AHP) is employed. Its core idea is to decompose the complex decision problem into objective, criterion, and alternative layers, determining the relative importance of elements at each layer through pairwise comparisons, ultimately yielding the weight coefficients for each objective. The weight vector is obtained by calculating the nth root of the product of elements in each row of the judgment matrix and then normalizing it. To ensure the logical consistency of the judgment matrix and avoid contradictions such as "A is more important than B, B is more important than C, and C is more important than A," a consistency check is required. The consistency ratio (CR) is the ratio of the consistency index (CI) to the random consistency index (RI). When CR < 0.1, the judgment matrix is ​​considered to have satisfactory consistency, and the weight coefficients are valid; otherwise, the judgment matrix needs to be reconstructed.

[0097] In one possible embodiment, during step S3, in the process of constructing the support system design model, an energy matching model is used for the optimized design of ultra-high elongation steel strands in roadways at risk of rockburst, so that the energy absorption capacity of the support system is greater than the impact energy of the surrounding rock, as shown in the following formula:

[0098]

[0099] In the formula, The total deformation energy absorption capacity of all ultra-high elongation steel strands participating in the load-bearing area is the product of the plastic deformation energy absorption of a single steel strand within the allowable elongation range and the effective number of working strands; the ultimate energy absorption of a single steel strand is the integral value of its load-elongation curve within the allowable elongation range. It represents the impact energy of the surrounding rock acting on the support area when a rockburst occurs, and is the sum of the kinetic energy and elastic deformation energy released by the surrounding rock. The safety reserve coefficient characterizes the redundancy of the support system in resisting impact energy and is determined according to the risk level of rockburst.

[0100] Specifically, rockburst is a common dynamic disaster in deep mines. Its essence is the sudden release of a large amount of elastic deformation energy accumulated in the surrounding rock, generating a strong impact load. Traditional ordinary steel strands have low elongation and are prone to brittle fracture under impact loads, failing to effectively absorb impact energy. Ultra-high elongation steel strands possess large deformation and high energy absorption mechanical properties, capable of dissipating impact energy through their own plastic deformation, making them ideal support materials for rockburst tunnels. The principle of the energy matching model is to ensure that the total energy absorption capacity of the support system is greater than the impact energy released by the surrounding rock, while leaving sufficient safety reserves. The total energy absorption capacity of the support system... It is the sum of the energy absorbed by all ultra-high elongation steel strands participating in the load-bearing within the support area. The energy absorbed by a single steel strand is the integral value of its load-elongation curve within the allowable elongation range, reflecting the energy absorbed by the steel strand during deformation. (Rock impact energy) It is the energy acting on the supported area when a rockburst occurs, including the elastic deformation energy and kinetic energy released by the surrounding rock. Safety reserve factor. To account for the uncertainty in impact energy calculations and the discreteness of the support system, the higher the rockburst risk level, the larger the safety reserve factor.

[0101] Among them, the optimized design of hollow low-relaxation steel strands for grouting reinforced roadways adopts a spherical permeation and diffusion model of grout under static pressure grouting conditions of homogeneous isotropic surrounding rock to determine the optimal grouting parameters and steel strand arrangement spacing, as shown in the following formula:

[0102]

[0103] In the formula, The effective diffusion radius of cement slurry in the fractured zone of the surrounding rock of the roadway; The initial average permeability coefficient of the fractured zone of the surrounding rock in the roadway before grouting; This represents the water head height corresponding to the grouting orifice pressure. This refers to the duration of grouting in a single hole; The average porosity of the fractured zone of the surrounding rock in the tunnel; The effective filling coefficient of the grout represents the proportion of the surrounding rock pores that are actually filled and cemented by the grout.

[0104] Specifically, hollow low-relaxation steel strands integrate the steel strands and grouting pipes, enabling integrated support and grouting construction. Grouting binds the fractured surrounding rock into a cohesive whole, improving its bearing capacity. The effective diffusion radius of the grout is a key parameter in grouting design, directly determining the spacing of the steel strands. If the spacing is too large, the grout diffusion range cannot overlap, resulting in grouting blind spots; if the spacing is too small, it leads to material waste and reduced construction efficiency. The spherical permeation and diffusion model of the grout is derived from Darcy's law, assuming that the grout undergoes laminar flow in a homogeneous and isotropic porous medium, with the grouting hole as a point source, and the grout diffuses outward in a spherical shape. This model describes the effective diffusion radius of the grout. With grouting pressure Grouting time Permeability coefficient of surrounding rock Porosity and slurry filling coefficient The model establishes a quantitative relationship between the grouting pressure and time. It can calculate the optimal grouting pressure and time based on the required diffusion radius, and also calculate the diffusion radius based on given grouting parameters, thereby determining a reasonable spacing for the steel strands to ensure effective grouting reinforcement.

[0105] In one possible embodiment, in step S4, the dynamic calibration of the initial digital twin model employs an extended Kalman filter algorithm, including prediction and update steps, to achieve real-time synchronization between the initial digital twin model and the physical entity, as shown in the following formula:

[0106]

[0107] In the formula, For the first The prior state estimation vector at time t contains the surrounding rock displacement, steel strand stress, anchor bolt axial force, and top plate delamination value; The state transition matrix represents the state evolution law of the surrounding rock-support system under no external disturbance. To control the input matrix, which characterizes the impact of support parameter adjustments on the system state; The input vector is controlled by adjusting the preload of the steel strand, the grouting pressure, and the number of additional anchor bolts. For the first The calibrated state vector at time step; For the first The Kalman gain matrix at time step 1 is used to balance the weights of the predicted and measured values. For the first Vector of actual on-site monitoring data at any given moment; For the first The observation matrix at any given time represents the mapping relationship between system state variables and monitoring data.

[0108] Specifically, the Extended Kalman Filter (EKF) is a recursive state estimation algorithm suitable for nonlinear systems. Its core principle is to perform a Taylor expansion of the nonlinear system at the current state estimation point and take a first-order approximation, transforming the nonlinear system into a linear system. Then, the standard Kalman Filter algorithm is applied for state estimation. The EKF includes two steps: prediction and update.

[0109] Prediction Steps: Based on the state estimate from the previous moment and the system's state transition model, predict the prior state estimate and covariance matrix for the current moment. Update Steps: Compare the measured data with the predicted values ​​at the current moment, calculate the residuals, and then correct the prior estimates based on the Kalman gain matrix to obtain the posterior state estimate and covariance matrix for the current moment. In the dynamic calibration of the digital twin model, the system state vector includes parameters such as surrounding rock displacement, steel strand stress, and anchor bolt axial force. The state transition matrix is ​​determined based on the mechanical model of the surrounding rock-support system, and the control input matrix reflects the impact of support parameter adjustments on the system state. By using the extended Kalman filter algorithm, the predicted values ​​of the digital twin model and the measured values ​​from field sensors can be fused, model parameters can be corrected in real time, errors and uncertainties in the initial model can be eliminated, and real-time synchronization between the digital twin model and the physical entity can be achieved.

[0110] The monitoring data collection frequency adopts an adaptive sampling model based on risk level, with increased sampling frequency in high-risk areas and decreased sampling frequency in low-risk areas, as shown in the following formula:

[0111]

[0112] In the formula, for The frequency of monitoring data collection at any given time; This is the baseline acquisition frequency under normal geological conditions, and the value is determined according to the requirements of routine safety monitoring in mines. The risk sensitivity coefficient is determined based on the geological hazard level of the mine, with a larger value taken for mines with a high risk of rockburst. for The real-time risk level of the support system is measured at any given moment, with a value range of [0,1], where 0 represents no risk and 1 represents extremely high risk.

[0113] Traditional monitoring systems use a fixed sampling frequency, which cannot be dynamically adjusted according to the risk status of the support system. If the sampling frequency is too high, it increases sensor energy consumption and data storage and transmission costs; if the sampling frequency is too low, it cannot capture rapidly changing abnormal signals in time, leading to delayed early warnings. The adaptive sampling model dynamically adjusts the data acquisition frequency based on the real-time risk level of the support system, achieving higher sampling density as the risk increases. This model uses an exponential function to describe the relationship between sampling frequency and risk level. The exponential function has non-linear growth characteristics, enabling it to quickly increase the sampling frequency when the risk level is high, capturing abnormal signals promptly; and decrease the sampling frequency when the risk level is low, saving resources. Risk sensitivity coefficient. This is used to adjust the sensitivity of the sampling frequency to changes in risk level; for mines with higher geological disaster risk, The larger the value, the faster the sampling frequency increases with the risk level.

[0114] In one possible embodiment, in step S5, during the process of inferring the health evolution trend of the support system through machine learning algorithms, the health status prediction of the support system adopts an LSTM-random forest combination model, which combines the advantages of time series prediction and feature classification, as shown in the following formula:

[0115]

[0116] In the formula, The overall failure probability of the support system within the next 72 hours; The failure probability predicted by the LSTM model based on time series data; This refers to the failure probability predicted by the random forest model based on multi-dimensional feature data. These are the weight coefficients of the LSTM model, which are dynamically adjusted based on historical prediction accuracy, and their values ​​range from 0 to 1.

[0117] Specifically, predicting the health status of a support system is a typical time series forecasting problem, involving multi-dimensional feature data. A single machine learning model struggles to simultaneously capture both the long-term dependencies of the time series and the non-linear relationships between multi-dimensional features. Long Short-Term Memory (LSTM) networks, a special type of recurrent neural network, solve the gradient vanishing problem of traditional recurrent neural networks through gating mechanisms, effectively capturing long-term dependencies in time series data. They are suitable for predicting parameters such as surrounding rock deformation and steel strand stress that change over time. Random forests, an ensemble learning method composed of multiple decision trees, obtain the final result through voting. They have advantages such as strong anti-overfitting ability and the ability to handle high-dimensional non-linear data, making them suitable for classification prediction based on multi-dimensional feature data. The combined model fuses the prediction results of the LSTM and random forest models through a weighted average, fully leveraging the advantages of both. The LSTM model captures the trend features of the time series, while the random forest model captures the non-linear relationships between multi-dimensional features. (Weight coefficients are listed below.) The combined model is dynamically adjusted based on the historical prediction accuracy of the two models, with the model with higher prediction accuracy given greater weight, thereby further improving the prediction accuracy of the combined model.

[0118] See Figure 3 To address the limitation of single machine learning models in simultaneously capturing the temporal evolution of the health status of support systems and the nonlinear correlation of multi-dimensional features, this embodiment employs a combined prediction architecture of dual-channel feature extraction and dynamic weighted fusion. Unlike traditional single models that can only focus on a certain type of feature, this architecture processes temporal monitoring data and multi-dimensional static / statistical features separately through two parallel feature extraction channels, and then fuses the prediction results of the two through a dynamic weighting mechanism, ultimately outputting a comprehensive failure probability and achieving hierarchical early warning.

[0119] The model's input data all comes from the historical operating database and real-time synchronized data stream of the digital twin mirror model, mainly containing three types of information: The first type is time-series monitoring data collected in chronological order, including the subsidence of the surrounding rock roof, the convergence of the two sides, the bulging of the floor, the axial stress of the steel strands, the axial force of the anchor bolts, and the roof delamination value. The sampling frequency will be adaptively adjusted according to the real-time risk level of the support system. The second type is relatively stable static feature data, covering geological parameters (surrounding rock strength, fault distance, burial depth), support parameters (steel strand specifications, preload, anchorage length, and arrangement spacing), and construction quality parameters (preload qualification rate, anchorage length qualification rate, and grouting fullness). The third type is label data used for model training, marking accident records such as large roadway deformation, steel strand breakage, and roof collapse as 1 (failure), and normal operation status as 0.

[0120] To ensure the quality of the input data and avoid interference from noise and missing values ​​in the model prediction results, we first performed a comprehensive cleaning process on the raw data. For time-series data with no more than three consecutive missing sampling points, linear interpolation was used to fill in the missing data; for data with more than three consecutive missing sampling points, the mean of similar monitoring points in the same region and time period was used for filling in the missing data; for missing values ​​of static features, the median of the feature under the same geological conditions was used for filling in the missing values. Outlier identification adopted the 3σ criterion: when a data point deviates from the mean by more than three times the standard deviation, it is judged as an outlier and replaced with the moving average of the five adjacent normal data points. At the same time, a sliding window averaging filter with a window size of 5 was used to smooth the time-series data to eliminate the influence of sensor noise and random fluctuations.

[0121] After data cleaning, feature engineering was further performed to extract feature vectors that effectively characterize the health status of the support system. For time-series monitoring data, in addition to the original monitoring values, basic statistical features such as the mean, maximum, minimum, standard deviation, and rate of change of each parameter over the past 24 hours, 72 hours, and 7 days were extracted, as well as derived features such as the slope of the linear trend, fluctuation amplitude, and fluctuation frequency obtained by least squares fitting. All features were uniformly mapped to the [0,1] interval using the min-max normalization method to eliminate differences in dimensionality and numerical magnitude between different features. Finally, the dataset was divided into a training set (70%), a validation set (20%), and a test set (10%) according to time order to strictly avoid data leakage. The SMOTE oversampling algorithm was used to handle the class imbalance problem in the training set, increasing the number of failure samples and improving the model's ability to identify minority classes.

[0122] For monitoring data such as surrounding rock displacement and steel strand stress that change continuously over time, this invention employs a bidirectional long short-term memory (BiLSTM) network to construct a time series prediction channel. BiLSTM can simultaneously utilize past and future time-series information to effectively capture the long-term dependencies and trend changes in the health status of the support system, which is impossible with traditional feedforward neural networks. In this embodiment, the BiLSTM network adopts a three-layer stacked architecture: the input layer receives a feature tensor of shape (batch_size, 24, 24), where 24 represents the prediction using monitoring data from the past 24 hours, and 24 represents the total number of the six basic monitoring parameters and their derived statistical features; the first BiLSTM layer has 128 hidden units and returns a complete sequence output to capture short-term time-series features; the second BiLSTM layer has 64 hidden units and also returns a complete sequence to capture medium-term features; the third BiLSTM layer has 32 hidden units and only returns the output of the last time step to integrate long-term time-series dependencies. To prevent overfitting, a Dropout layer with a dropout rate of 0.3 was added to the network, followed by a fully connected layer with 16 neurons. Finally, the failure probability predicted by the LSTM model was output through a Sigmoid activation function. The model was trained using the Adam optimizer with an initial learning rate of 0.001, which decreased by a factor of 0.9 every 10 epochs. The loss function was binary cross-entropy, the batch size was 32, the number of iterations was 100, and an early stopping mechanism was employed: training stopped when the validation set loss no longer decreased for 10 consecutive epochs.

[0123] For static and statistical features such as geological conditions, support parameters, and construction quality, this invention employs an improved random forest algorithm to construct feature classification channels. Random forest, as an ensemble learning method, has advantages such as strong anti-overfitting ability and the ability to handle high-dimensional nonlinear data, effectively identifying key influencing factors leading to support failure. This embodiment makes two targeted improvements based on the traditional random forest: first, when constructing each decision tree, features are weighted according to their importance to increase the probability of selecting important features and enhance the model's sensitivity to key risk factors; second, during final classification, different voting weights are assigned to each decision tree based on its accuracy on the validation set, with decision trees having higher accuracy receiving greater voting weight, further improving the model's classification accuracy. The model parameters are set as follows: 200 decision trees, a maximum depth of 15 layers, a minimum number of sample splits of 10, a minimum number of sample leaf nodes of 5, and each split randomly selects the square root of the total number of features. After the model is trained, it will also output the importance ranking of all features, identify key factors affecting the health status of the support system, such as the rate of change of steel strand stress, the rate of roof settlement, and the strength of surrounding rock, and provide a basis for subsequent support parameter optimization and risk prevention and control.

[0124] To combine the advantages of the two sub-models and avoid the problem of fixed weights failing to adapt to different working conditions, this embodiment adopts a dynamic weighted fusion mechanism based on historical prediction accuracy. The weight coefficients are automatically adjusted according to the real-time prediction accuracy of the two sub-models over the past 7 days, and the calculation formula is as follows:

[0125]

[0126] in, The figures show the prediction accuracies of the LSTM and Random Forest models over the past 7 days, respectively. When the support system is in a normal state and the monitoring data changes gradually, the LSTM model has stronger time-series prediction capabilities and higher accuracy, and is therefore given a larger weight. However, when the support system experiences abnormal fluctuations or geological conditions change abruptly, the Random Forest model's advantage in identifying nonlinear correlations of multi-dimensional features becomes prominent, and its weight automatically increases. This dynamic adjustment mechanism ensures that the better-performing model always dominates the fusion result, significantly improving the overall prediction accuracy of the combined model. The combined model operates in two phases: offline training and online prediction. In the offline training phase, we extract historical data from the digital twin database, complete data preprocessing and feature engineering, train the BiLSTM sub-model and the improved Random Forest sub-model separately, fine-tune the parameters on the validation set, and finally evaluate the model performance on the test set and save the optimal parameters. The initial fusion weight coefficient is set to 0.5. After offline training is completed, the model can enter the online prediction stage to simulate the health evolution trend of the support system in real time. First, real-time monitoring data of the physical tunnels is collected and synchronized to the digital twin mirror model. After preprocessing and feature extraction of the real-time data, it is input into two sub-models to obtain their respective predicted failure probabilities. Then, based on the prediction accuracy over the past 7 days, the fusion weight is dynamically calculated to obtain the final comprehensive failure probability. The system will classify the risk level according to the comprehensive failure probability, issue corresponding warning signals, and record the prediction results and actual status in the database to provide data support for model updates.

[0127] To comprehensively evaluate the model's predictive performance, multiple complementary evaluation metrics are employed. Accuracy reflects the overall proportion of correct predictions, precision measures the false positive rate, recall measures the false negative rate, F1 score balances precision and recall, AUC-ROC reflects the model's ability to distinguish between normal and failure states, and early warning time is a particularly important indicator in the field of mine safety, requiring the model to issue a warning at least 24 hours before the actual failure occurs. To ensure the model can adapt to different geological conditions and the long-term evolution of support systems in various mines, a comprehensive model update and iteration mechanism has been established. Every quarter, the model is fully updated using newly accumulated monitoring data and failure cases, and the model parameters are retrained. When a new support failure occurs, the case is immediately added to the training set for incremental learning to improve the model's ability to identify similar risks. If the model's accuracy falls below a set threshold of 80% in 10 consecutive predictions, a model reconstruction process is automatically triggered to reselect features and adjust model parameters, ensuring the model always maintains optimal predictive performance.

[0128] In one possible embodiment, during step S6, the comprehensive evaluation of the construction quality and support effect of the support system is conducted using a fuzzy comprehensive evaluation model, as shown in the following formula:

[0129]

[0130] In the formula, This is a vector representing the overall evaluation results, corresponding to four evaluation levels: excellent, good, satisfactory, and unsatisfactory. The weight vector for each evaluation indicator is determined by the analytic hierarchy process, including the weights of indicators such as the pretension force qualification rate of steel strand, the anchorage length qualification rate, and the surrounding rock convergence rate. A fuzzy relation matrix is ​​used to represent the membership degree of each evaluation indicator to different evaluation levels. It is established by combining expert scoring and historical data statistics. It is a weighted average type fuzzy synthesis operator.

[0131] Specifically, evaluating the effectiveness of support structures is a comprehensive problem involving multiple factors and indicators. It encompasses both quantitative indicators (such as the pretensioning rate of steel strands and the convergence rate of surrounding rock) and qualitative indicators (such as construction compliance and appearance quality). Furthermore, there are no clear boundaries between evaluation levels, exhibiting fuzziness. The fuzzy comprehensive evaluation model, based on fuzzy mathematics theory, can organically combine qualitative and quantitative evaluations to achieve an objective and comprehensive assessment of support effectiveness. The core of this model is the fuzzy relation matrix. and weight vector Fuzzy relation matrix element r ijThe weight vector represents the membership degree of the i-th evaluation indicator to the j-th evaluation level, determined by expert scoring or historical data statistics; This reflects the importance of each evaluation indicator to the overall evaluation result, determined using the analytic hierarchy process (AHP). The weighted average fuzzy synthesis operator considers the influence of all evaluation indicators, obtaining the overall evaluation result vector through weighted averaging. , The elements in the equation represent the membership degree of the support effect to each evaluation level. Finally, the comprehensive evaluation level of the support effect is determined based on the principle of maximum membership degree.

[0132] To verify the technical effectiveness of this invention, the return airway of a coal mine working face was selected as the test site. This mine is a typical deep-mining mine in western my country, with complex geological conditions and broad industry representativeness. The test roadway is 850m deep, with an original rock stress of 22MPa. The surrounding rock is mainly composed of alternating layers of mudstone and sandy mudstone, with a uniaxial compressive strength of only 12-18MPa. It is prone to softening and disintegration when exposed to water, belonging to a typical roadway of weak and fractured surrounding rock. At the same time, the area is affected by regional tectonic stress, with a moderate risk of rockburst. Historically, there have been several accidents involving large roadway deformation and support failure, placing extremely high demands on the safety and adaptability of the support system.

[0133] The test tunnel is 600m long, with a rectangular cross-section, a net width of 5.2m, and a net height of 3.8m. It is excavated using fully mechanized tunneling, with a daily advance of approximately 8m. The tunnel has a service life of 3 years and is primarily used for return air and equipment transportation at the working face. Its support effectiveness directly affects the safe production of the entire working face.

[0134] To ensure the scientific validity and comparability of the test results, this experiment adopted the controlled variable method, dividing the 600m test tunnel into two equal sections along its strike. The geological conditions, cross-sectional dimensions, tunneling technology, and construction teams of the two sections were completely identical, with differences only in the support design method and the type of steel strand used.

[0135] Test group (300m): The entire process of design and control was carried out using the support system design optimization method based on digital twin of this invention. The support materials selected were Φ21.8mm ultra-high elongation prestressed steel strand (elongation ≥12%) and Φ21.8mm hollow low relaxation prestressed steel strand, which were independently developed by our company and were arranged differently according to the geological risk level of different areas.

[0136] Control group (300m): The mine's existing traditional design method of "analogy method + experience correction" was adopted, and the support parameters of adjacent mined working faces were used for design. Ordinary prestressed steel strands of the same specification (elongation ≥3.5%) were selected as the support material, and uniform support parameters were used throughout the section.

[0137] I. Experimental Design

[0138] To ensure the smooth conduct of the experiment, a complete digital twin support system was built in the test tunnel to achieve full-process control from data acquisition, modeling and analysis, intelligent design to dynamic optimization.

[0139] Multi-source data acquisition system: A comprehensive monitoring section is set up every 20m in the test roadway. Each section is equipped with 3 fiber optic displacement sensors (monitoring midpoint subsidence of the roof, midpoint convergence of the sidewalls, and midpoint bulge of the floor), 6 steel strand stress sensors (evenly distributed on the steel strands of the roof and sidewalls), 4 anchor bolt axial force sensors, and 1 multi-point displacement meter (monitoring roof delamination at different depths). All sensors are connected to an edge computing server via industrial Ethernet. The acquisition frequency is adaptively adjusted according to the real-time risk level of the support system, at 1 time / hour under normal conditions and increased to 1 time / minute under high-risk conditions.

[0140] Digital Twin Platform: Deployed on edge computing servers in the mine, it integrates six core modules: data integration, 3D modeling, intelligent design, dynamic optimization, health monitoring, and construction guidance. The platform uses the Unity3D engine to build a 3D visualization interface, capable of displaying the geological structure, support structure status, and monitoring data of the tunnel in real time, and supports access from multiple terminals.

[0141] On-site construction guidance terminal: Two industrial explosion-proof tablet computers are equipped at the tunneling face, each with a dedicated construction guidance APP installed. Construction personnel can use the terminal to view standardized construction drawings, work instructions, and quality acceptance standards in real time, receive support parameter adjustment instructions and risk warning information pushed by the system, and upload construction quality inspection data.

[0142] The control group roadway was equipped with monitoring equipment according to the mine's routine safety monitoring requirements, with a monitoring section set up every 50m. The monitoring data was recorded manually once a day, and the observation frequency was increased when abnormal situations were encountered.

[0143] This trial lasted 12 months, fully covering the entire lifecycle of the tunnel from excavation to the end of face mining, and was divided into three phases:

[0144] Construction preparation and system debugging phase (month 1): Complete the site survey, installation and calibration of monitoring equipment, deployment and debugging of the digital twin platform, and training of construction personnel to ensure the normal operation of the test system.

[0145] Tunneling and Dynamic Optimization Phase (Months 2-4): Tunneling is carried out according to the design plan, and dynamic calibration of the digital twin model, dynamic adjustment of support parameters, and real-time evaluation of construction quality are carried out simultaneously, recording various data during the construction process.

[0146] Long-term monitoring and operation phase (months 5-12): During the long-term mining process, continuously monitor the operation status of the support system, carry out health status prediction and risk warning, and record the impact of mining disturbances on the support system until the long-term mining is completed.

[0147] To comprehensively and objectively compare the support effects of the experimental and control groups, a comprehensive evaluation index system was established, encompassing four dimensions: safety, economy, construction efficiency, and technical performance.

[0148] Safety indicators include: maximum roof subsidence, maximum sidewall convergence, maximum floor bulge, maximum stress in the steel strands, maximum axial force in the anchor bolts, number of support failures, and accident rate. These indicators directly reflect the load-bearing capacity of the support system and its effectiveness in controlling surrounding rock deformation.

[0149] Economic indicators include the cost of support materials per unit length, construction cost per unit length, repair cost per unit length, and total roadway repair rate. By comparing the two sets of economic indicators, the advantages of this invention in reducing engineering costs are quantified.

[0150] Construction efficiency indicators include the construction time per unit length of tunnel support, the labor intensity of construction workers, and the pass rate of construction quality. Construction time is measured from the start of drilling to the completion of grouting, and labor intensity is assessed based on the daily workload and physical exertion of the construction workers.

[0151] Technical performance indicators include: digital twin model prediction error, health status prediction accuracy, average early warning time, and support parameter adjustment response time. These indicators are used to evaluate the performance of the core technical modules of this invention.

[0152] II. Specific test content and implementation steps

[0153] Digital Twin Model Accuracy Verification Test

[0154] The digital twin model is the foundation of the entire system, and its mapping accuracy directly determines the reliability of intelligent design and dynamic optimization. This experiment comprehensively verifies the model from three aspects: geometric replication accuracy, mechanical simulation accuracy, and dynamic calibration accuracy.

[0155] Initial Model Construction and Geometric Accuracy Verification: Based on geological exploration borehole data, tunnel design drawings, and surrounding rock mechanics test reports of the test tunnel, a three-dimensional digital twin initial model of the mine tunnel was constructed according to the method proposed in this invention. During tunnel excavation, every 50m, a three-dimensional laser scanner was used to scan the actual exposed tunnel cross-section and geological structures. The scanning results were compared with the digital twin model, and the geometric dimensional errors and geological structure position errors of the model were calculated.

[0156] Mechanical simulation accuracy verification: After the tunnel excavation is completed, real-time monitoring data of surrounding rock deformation and steel strand stress are collected and compared with the initial simulation prediction results of the digital twin model to calculate the relative errors of various indicators. The focus is on verifying the model's simulation accuracy of the coupling interaction law between surrounding rock and support, and analyzing the prediction effect of surrounding rock stress distribution and deformation trend under different support parameters.

[0157] Dynamic calibration accuracy verification: When the error between the monitoring data and the model prediction exceeds a set threshold, the extended Kalman filter algorithm is activated to dynamically calibrate the model parameters. After calibration, the difference between the model prediction and subsequent measured values ​​is compared again to evaluate the effect of dynamic calibration on improving model accuracy. A comprehensive evaluation of the overall model accuracy is conducted 1 month, 3 months, and 6 months after tunneling completion.

[0158] Design effect comparison experiment

[0159] This experiment aims to verify the advantages of the intelligent design method of the present invention in balancing safety, economy, and construction efficiency, and to solve the problems of over- or under-support in traditional designs.

[0160] Design Parameter Output and Comparison: The intelligent design model of this invention and the traditional empirical design method were used to design the support for the test and control roadways, respectively. Detailed records were made of core parameters for both designs, including steel strand specifications, spacing, row spacing, preload, and anchorage length. The amount of steel strand, anchor bolts, and grouting material used per unit length of roadway was calculated.

[0161] Construction process comparison: During construction, the actual construction time per meter of support for both sets of roadways was recorded, and the daily workload of the construction personnel was calculated. Simultaneously, the construction quality was tested according to the "Quality Acceptance Specification for Mine Roadway Support Engineering," and indicators such as the pre-tightening force qualification rate of steel strands, the anchorage length qualification rate, and the grouting fullness were recorded.

[0162] Comparison of support effects: After the tunnel excavation was completed, the surrounding rock deformation and support stress of the two sets of tunnels were continuously monitored, and curves showing the changes in roof subsidence, sidewall convergence, and steel strand stress over time were plotted. The deformation characteristics and stress states of the two sets of tunnels at different stages were compared and analyzed to evaluate the control effect of the two design methods on the surrounding rock deformation.

[0163] Dynamic optimization effect verification experiment

[0164] Traditional static design cannot adapt to dynamic changes in geological conditions, while the dynamic optimization mechanism of this invention can automatically adjust support parameters based on real-time monitoring data. This experiment focuses on verifying the response speed and effectiveness of this mechanism:

[0165] Simulation and Response to Changes in Geological Conditions: During the tunnel excavation of the test group, when encountering areas with changing geological conditions such as faults and fractured zones, the system automatically identifies abnormal changes in monitoring data, updates the digital twin model in real time, and generates an optimized support parameter scheme within 30 minutes. Construction personnel carry out construction according to the new scheme, recording the adjusted surrounding rock deformation and support stress.

[0166] Manual adjustment comparison with control group: When the control group roadway encountered the same geological condition changes, the mine's technicians made manual adjustments based on their experience, and recorded the response time from the discovery of the problem to the completion of the parameter adjustment, as well as the support effect after the adjustment.

[0167] Dynamic optimization under mining disturbance: During the mining process, the stress of the surrounding rock in the roadway changes drastically due to mining disturbance. The system continuously optimizes support parameters based on real-time monitoring data, and takes timely reinforcement measures such as installing additional anchor bolts and increasing the preload of steel strands in areas with large deformation. The deformation and number of repairs of the two sets of roadways during mining are compared.

[0168] Health Status Prediction and Risk Warning Trial

[0169] This experiment verifies the ability of the LSTM-random forest combined prediction model of this invention to predict the health evolution trend of support systems and the accuracy of risk warning:

[0170] Online Model Operation and Early Warning: Throughout the entire test cycle, the system runs a health status prediction model in real time, calculating the comprehensive failure probability of the support system for the next 72 hours every hour. When the failure probability exceeds a set threshold, the system automatically issues an early warning signal of the corresponding level and pushes it to the on-site construction terminal and the mine dispatch room.

[0171] Early warning effectiveness tracking and verification: After the system issues an early warning, designated personnel are assigned to focus on monitoring the warning area, recording the warning time, warning level, and warning location. The actual status of the subsequent support system is tracked and recorded. If abnormalities such as accelerated surrounding rock deformation or sudden increase in steel strand stress occur, timely preventative measures are taken. The accuracy rate, false alarm rate, and missed alarm rate of the statistical model are calculated, and the average advance warning time is determined.

[0172] Manual inspection comparison: The control group used traditional manual inspection methods for risk identification, conducting inspections once a day and recording any anomalies found and their handling time. The differences between the two methods in terms of timeliness and accuracy of risk identification were compared.

[0173] Application effect verification test of new steel strand

[0174] This invention develops specialized optimization design models for the unique mechanical properties of ultra-high elongation steel strands and hollow low-relaxation steel strands. This experiment verifies whether these models can fully leverage the technical advantages of the new materials.

[0175] Effect of ultra-high elongation steel strand on rockburst prevention: In the rockburst risk area of ​​the test roadway, the layout scheme of ultra-high elongation steel strands was optimized using an energy matching model. In the same area of ​​the control group, ordinary steel strands were used for support. The deformation, stress changes, and fracture conditions of the two groups of steel strands under mining influence were compared to analyze the energy absorption effect and rockburst prevention capability of ultra-high elongation steel strands.

[0176] The effect of hollow low-relaxation steel strand grouting reinforcement: In the fractured rock zone of the test roadway, the grouting parameters of hollow low-relaxation steel strand were optimized using a grout penetration and diffusion model. In the same area of ​​the control group, ordinary steel strand was used in conjunction with traditional grouting pipes for grouting reinforcement. The grouting fullness, grouting time, and surrounding rock reinforcement effect of the two groups were compared to analyze the advantages of hollow low-relaxation steel strand in improving grouting efficiency and reinforcement quality.

[0177] III. Data Acquisition and Processing

[0178] Data collection content and methods

[0179] This experiment collected multi-dimensional data covering the entire life cycle of the support structure to ensure the comprehensiveness and reliability of the test results:

[0180] Basic data collection includes geological exploration data, surrounding rock mechanics test data, tunnel design parameters, and support material performance test reports, which are provided by geological departments and laboratories.

[0181] Construction process data collection: Through on-site construction terminals and quality inspection records, construction data such as the installation position, preload value, anchorage length, grouting pressure, and grouting time of each steel strand are collected, as well as management data such as construction personnel and construction time.

[0182] Real-time monitoring data acquisition: The monitoring data of the experimental group was automatically collected by sensors and transmitted to the digital twin platform, while the monitoring data of the control group was recorded manually once a day. All data were stored in a unified database.

[0183] Event data collection: Record all important events that occur during the test, such as changes in geological conditions, adjustments to support parameters, early warning information, support failure events, and repair records, noting the time, location, and specific details of each event.

[0184] Data processing and analysis methods

[0185] To ensure the validity of the experimental data and the scientific rigor of the analysis results, the following methods were used to process and analyze the collected data:

[0186] Data cleaning: First, the raw data is cleaned to remove obvious outliers (such as sudden data changes caused by sensor failures), and missing values ​​are filled using linear interpolation or the mean of adjacent data.

[0187] Data processing: The cleaned data are classified and organized according to the experimental group and the control group. The average value, standard deviation, maximum value and minimum value of each evaluation indicator are calculated, and data change curves and comparison bar charts are drawn.

[0188] Statistical analysis: Independent samples t-test was used to analyze the statistical significance of various indicators in the experimental group and the control group to determine whether the differences between the two groups were statistically significant (P<0.05).

[0189] Comprehensive evaluation: The analytic hierarchy process (AHP) was used to comprehensively evaluate the support effects of the two groups, calculate the comprehensive score, and quantify the overall advantages of the present invention compared with traditional technologies.

[0190] IV. Experimental Results and Analysis

[0191] The following verification results were obtained through this comparative experiment:

[0192] Digital twin model accuracy: See Figure 4 The initial model has a geometric dimension error of no more than 5% and a geological structure position error of no more than 1m. After dynamic calibration, the model's prediction error for surrounding rock deformation and steel strand stress is no more than 10%, which can accurately reflect the actual state of the physical tunnel.

[0193] Intelligent design effect: See Figure 5 Under the premise of ensuring support safety, the cost of support materials per unit length in the experimental group was reduced by 15%-20% compared with the control group, the construction time per unit length was shortened by 10%-15%, and the construction quality qualification rate was increased to over 95%. The deformation of the surrounding rock in the roadway of the experimental group was 30%-40% smaller than that in the control group, effectively solving the problem of over- or under-support in traditional design.

[0194] Dynamic optimization effect: When encountering changes in geological conditions, the system of this invention can complete the adjustment of support parameters within 30 minutes, while the manual adjustment of the control group requires more than 24 hours. After adjustment, the deformation of the surrounding rock in the test group was more than 50% smaller than that in the control group, and no roadway instability accidents occurred due to untimely support.

[0195] Health prediction effect: The health status prediction model of this invention has an accuracy of no less than 90%, a false alarm rate of less than 5%, a false alarm rate of 0, and an average early warning time of no less than 24 hours, which can buy sufficient time for risk prevention and control, and realize the transformation from passive emergency repair to proactive prevention and control.

[0196] Application effects of the new type of steel strand: The deformation of ultra-high elongation steel strand in the risk area of ​​rock burst is 3-5 times that of ordinary steel strand, which can effectively absorb impact energy and no breakage occurs; the grouting fullness of hollow low relaxation steel strand is more than 20% higher than that of traditional grouting methods, the grouting time is shortened by 30%, and the surrounding rock reinforcement effect is significantly improved.

[0197] V. Experimental Conclusions

[0198] This comparative experiment verifies the scientific validity, effectiveness, and practicality of the proposed digital twin-based optimization method and system for prestressed steel strand support in mines. The experimental results show that, through the deep integration of digital twin technology and intelligent optimization algorithms, this invention successfully addresses the pain points of traditional support technologies, such as insufficient design accuracy, poor dynamic adaptability, and underutilization of the advantages of new materials. It can significantly improve the safety and economy of mine roadway support under complex geological conditions.

[0199] Example 2

[0200] See Figure 9 Embodiment 2 of the present invention also provides a digital twin-based optimization system for prestressed steel strand support in mines, employing the digital twin-based optimization method for prestressed steel strand support in mines described in the above embodiments, including:

[0201] Data integration module 100 is used to uniformly collect and standardize the integration of multi-source heterogeneous data of the mine's set dimensions to build a support engineering database;

[0202] The digital twin model construction module 200 is used to construct an initial three-dimensional digital twin model of the mine roadway based on the static basic data of the support engineering database, realize the replication of the physical roadway geometry, geological structure and support structure, and simulate the coupling and interaction law between the surrounding rock and the support structure.

[0203] The support system design module 300 is used to construct a support system design model using the digital twin initial model as a simulation calculation carrier. The support system design model takes support safety, economy and construction efficiency as comprehensive optimization objectives and outputs the optimal initial design parameter set of prestressed steel strands.

[0204] The dynamic optimization module 400 is used to collect dynamic status data of the surrounding rock and support structure in real time during the entire process of physical tunnel construction and operation, and to synchronously inject the dynamic status data into the initial digital twin model to complete the dynamic calibration and iterative update of the initial digital twin model, thereby obtaining a digital twin mirror model that is consistent with the physical entity in real time.

[0205] The health monitoring module 500 is used to infer the health evolution trend of the support system based on the historical operating data and real-time synchronized data of the digital twin mirror model, through machine learning algorithms, identify potential instability risks, and issue graded warnings according to the risk level.

[0206] The construction guidance and effect evaluation module 600 is used to generate standardized construction guidance documents that comply with mine safety regulations based on the optimal initial design parameter set and the dynamically adjusted support parameters. It performs a full-process quantitative evaluation of the construction quality and support effect of the support system and feeds the evaluation results back to the support system design model to drive the support system design model to continuously learn and iteratively optimize.

[0207] It should be noted that the information interaction and execution process between the various units of the above system are based on the same concept as the method embodiment in Embodiment 1 of this application, and the resulting technical effects are the same as those in the method embodiment of this application. For details, please refer to the description in the method embodiment shown above in this application, and it will not be repeated here.

[0208] Example 3

[0209] Embodiment 3 of the present invention provides a non-transitory computer-readable storage medium storing program code for a digital twin-based optimization method for prestressed steel strand support in mines. The program code includes instructions for executing the digital twin-based optimization method for prestressed steel strand support in mines as described in Embodiment 1 or any possible implementation thereof.

[0210] Computer-readable storage media can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

[0211] Example 4

[0212] Embodiment 4 of the present invention provides an electronic device, including: a memory and a processor;

[0213] The processor and the memory communicate with each other via a bus; the memory stores program instructions that can be executed by the processor, and the processor can call the program instructions to execute the digital twin-based optimization method for prestressed steel strand support in mines, as described in Embodiment 1 or any possible implementation thereof.

[0214] Specifically, a processor can be implemented in hardware or software. When implemented in hardware, the processor can be a logic circuit, an integrated circuit, etc. When implemented in software, the processor can be a general-purpose processor that reads software code stored in memory. This memory can be integrated into the processor or located outside the processor and exist independently.

[0215] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0216] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0217] Although the present invention has been described in detail above with general descriptions and specific embodiments, modifications or improvements can be made to it, which will be obvious to those skilled in the art. Therefore, all such modifications or improvements made without departing from the spirit of the present invention fall within the scope of protection claimed by the present invention.

Claims

1. A digital twin-based optimization method for prestressed steel strand support in mines, characterized in that, Includes the following steps: S1: Collect and standardize multi-source heterogeneous data of mine dimensions in a unified manner to build a support engineering database; S2: Based on the static basic data of the support engineering database, construct an initial three-dimensional digital twin model of the mine roadway to replicate the physical roadway geometry, geological structure and support structure, and simulate the coupling and interaction law between the surrounding rock and the support structure. S3: Using the digital twin initial model as the simulation calculation carrier, construct a support system design model. The support system design model takes support safety, economy and construction efficiency as comprehensive optimization objectives and outputs the optimal initial design parameter set of prestressed steel strands. S4: During the entire process of physical tunnel construction and operation, real-time dynamic status data of surrounding rock and support structure are collected, and the dynamic status data is synchronously injected into the initial digital twin model to complete the dynamic calibration and iterative update of the initial digital twin model, thereby obtaining a digital twin mirror model that is consistent with the physical entity in real time. S5: Based on the historical operating data and real-time synchronized data of the digital twin mirror model, the health evolution trend of the support system is inferred through machine learning algorithms, potential instability risks are identified, and graded early warnings are issued according to the risk level; support parameters are adjusted for different risk levels. S6: Based on the optimal initial design parameter set and the dynamically adjusted support parameters, generate a standardized construction guidance document that conforms to the mine safety specifications, conduct a full-process quantitative evaluation of the construction quality and support effect of the support system, and feed the evaluation results back to the support system design model to drive the support system design model to continuously learn and iteratively optimize. In step S3, the comprehensive optimization objective is achieved through a multi-objective optimization function, which simultaneously minimizes support risk, engineering cost, and construction period, as shown in the following formula: ， In the formula, The design variable vector contains continuous and discrete variables, which correspond to the specifications, structural form, spatial arrangement, preload value, and anchorage length of the steel strand, respectively. The safety objective function represents the comprehensive safety factor of the support system, which is calculated by weighting the surrounding rock stability safety factor, the steel strand tensile safety factor, and the anchor pull-out safety factor. The objective function is the economic cost, which characterizes the comprehensive cost per unit length of the support project, including the cost of steel strand materials, drilling construction cost, grouting cost, and labor cost. The objective function for construction efficiency is to characterize the standard construction time for roadway support per unit length, including drilling time, steel strand installation time, pre-tightening time, and grouting solidification time. The multi-objective optimization function uses the analytic hierarchy process (AHP) to determine the weight coefficients of each objective. The weight calculation model is as follows: ， In the formula, For the first The weight coefficients of each optimization objective; For the first The first objective is relative to the second objective. The importance scale values ​​for each objective are calculated using a 1-9 scale, where 1 indicates that both objectives are equally important, and 9 indicates that one objective is extremely more important than the other. To optimize the total number of objectives; after calculating the weight vector, a consistency check is performed, and when the consistency ratio is... At that time, it is considered that the judgment matrix has satisfactory consistency and the weight coefficients are effective; In step S5, during the process of inferring the health evolution trend of the support system using machine learning algorithms, the health status prediction of the support system adopts an LSTM-random forest combined model, which combines the advantages of time series prediction and feature classification. The formula is as follows: ， In the formula, The overall failure probability of the support system within the next 72 hours; The failure probability predicted by the LSTM model based on time series data; This refers to the failure probability predicted by the random forest model based on multi-dimensional feature data. These are the weight coefficients of the LSTM model, ranging from 0 to 1, and are calculated using the following formula: ， In the formula, These are the prediction accuracies of the LSTM model and the random forest model over the past 7 days, respectively.

2. The optimization method for prestressed steel strand support in mines based on digital twins according to claim 1, characterized in that, In step S1, the standardized fusion of the multi-source heterogeneous data adopts a min-max normalization model to map mine support engineering data of different dimensions and magnitudes to the [0,1] interval: ， In the formula, For the first Sample of a mine roadway support project The normalized values ​​of each feature parameter; For the first Sample of a mine roadway support project The original collected values ​​of each characteristic parameter include geological parameters, steel strand mechanical parameters, construction parameters, and monitoring parameters; For the first The minimum value of each characteristic parameter in all completed support engineering samples; For the first The maximum value of each characteristic parameter across all completed support engineering samples.

3. The optimization method for prestressed steel strand support in mines based on digital twins according to claim 1, characterized in that, In step S2, during the simulation of the coupling interaction between the surrounding rock and the support structure, an elastoplastic finite element model is used. This model employs the Mohr-Coulomb yield criterion to describe the plastic failure characteristics of the surrounding rock, and the static equilibrium equations are as follows: ， In the formula, The overall stiffness matrix of the surrounding rock-support system is formed by superimposing the elastic-plastic stiffness matrix of the roadway surrounding rock and the stiffness matrix of the steel strand-anchor support structure. The nodal displacement vector includes the roadway roof subsidence displacement, the sidewall convergence displacement, the floor bulge displacement, and the axial deformation displacement of the steel strands. The nodal load vector includes the original rock stress, steel strand preload, anchor bolt axial force, and construction disturbance load.

4. The optimization method for prestressed steel strand support in mines based on digital twins according to claim 1, characterized in that, In step S3, during the construction of the support system design model, an energy matching model is adopted for the optimization design of ultra-high elongation steel strands in roadways at risk of rockburst, so that the energy absorption capacity of the support system is greater than the impact energy of the surrounding rock. The formula is as follows: ， In the formula, The total deformation energy absorption capacity of all ultra-high elongation steel strands participating in the load-bearing area is the product of the plastic deformation energy absorption of a single steel strand within the allowable elongation range and the effective number of working strands; the ultimate energy absorption of a single steel strand is the integral value of its load-elongation curve within the allowable elongation range. It represents the impact energy of the surrounding rock acting on the support area when a rockburst occurs, and is the sum of the kinetic energy and elastic deformation energy released by the surrounding rock. The safety reserve factor characterizes the redundancy of the support system in resisting impact energy and is determined according to the risk level of rockburst. For the optimized design of hollow low-relaxation steel strands in grouting reinforced roadways, a spherical permeation and diffusion model of grout under static pressure grouting conditions in homogeneous isotropic surrounding rock is adopted to determine the optimal grouting parameters and steel strand spacing. The formula is as follows: ， In the formula, The effective diffusion radius of cement slurry in the fractured zone of the surrounding rock of the roadway; The initial average permeability coefficient of the fractured zone of the surrounding rock in the roadway before grouting; This represents the water head height corresponding to the grouting orifice pressure. This refers to the duration of grouting in a single hole; The average porosity of the fractured zone of the surrounding rock in the tunnel; The effective filling coefficient of the grout represents the proportion of the surrounding rock pores that are actually filled and cemented by the grout.

5. The optimization method for prestressed steel strand support in mines based on digital twins according to claim 1, characterized in that, In step S4, the dynamic calibration of the initial digital twin model employs an extended Kalman filter algorithm, including prediction and update steps, to achieve real-time synchronization between the initial digital twin model and the physical entity. The formula is as follows: ， In the formula, For the first The prior state estimation vector at time t contains the surrounding rock displacement, steel strand stress, anchor bolt axial force, and top plate delamination value; The state transition matrix represents the state evolution law of the surrounding rock-support system under no external disturbance. To control the input matrix, which characterizes the impact of support parameter adjustments on the system state; The input vector is controlled by adjusting the preload of the steel strand, the grouting pressure, and the number of additional anchor bolts. For the first The calibrated state vector at time step; For the first The Kalman gain matrix at time step is used to balance the weights of the predicted and measured values; For the first Vector of actual on-site monitoring data at any given moment; For the first The observation matrix at any given time represents the mapping relationship between system state variables and monitoring data; The monitoring data collection frequency adopts an adaptive sampling model based on risk level, increasing the sampling frequency in high-risk areas and decreasing the sampling frequency in low-risk areas, as shown in the following formula: ， In the formula, for The frequency of monitoring data collection at any given time; This is the baseline acquisition frequency under normal geological conditions, and the value is determined according to the requirements of routine safety monitoring in mines. The risk sensitivity coefficient is determined based on the geological hazard level of the mine. for The real-time risk level of the support system is measured at any given moment, with a value range of [0,1], where 0 represents no risk and 1 represents extremely high risk.

6. The optimization method for prestressed steel strand support in mines based on digital twins according to claim 1, characterized in that, In step S6, during the full-process quantitative evaluation of the construction quality and support effect of the support system, the comprehensive evaluation of the support effect adopts a fuzzy comprehensive evaluation model, the formula of which is as follows: ， In the formula, This is a vector representing the overall evaluation results, corresponding to four evaluation levels: excellent, good, satisfactory, and unsatisfactory. The weight vector for each evaluation indicator is determined by the analytic hierarchy process, and includes the weights of the steel strand pretension force qualification rate, anchorage length qualification rate, and surrounding rock convergence rate. A fuzzy relation matrix is ​​used to represent the membership degree of each evaluation indicator to different evaluation levels. It is established by combining expert scoring and historical data statistics. It is a weighted average type fuzzy synthesis operator.

7. A digital twin-based optimization system for prestressed steel strand support in mines, employing the digital twin-based optimization method for prestressed steel strand support in mines as described in any one of claims 1 to 6, characterized in that... include: The data integration module is used to uniformly collect and standardize the integration of multi-source heterogeneous data of the dimensions set by the mine to build a support engineering database; The digital twin model construction module is used to construct an initial three-dimensional digital twin model of the mine roadway based on the static basic data of the support engineering database, realize the replication of the physical roadway geometry, geological structure and support structure, and simulate the coupling and interaction law between the surrounding rock and the support structure. The support system design module is used to construct a support system design model using the digital twin initial model as a simulation calculation carrier. The support system design model takes support safety, economy and construction efficiency as comprehensive optimization objectives and outputs the optimal initial design parameter set of prestressed steel strands. The dynamic optimization module is used to collect dynamic status data of the surrounding rock and support structure in real time throughout the construction and operation of the physical tunnel, and to synchronously inject the dynamic status data into the initial digital twin model to complete the dynamic calibration and iterative update of the initial digital twin model, thereby obtaining a digital twin mirror model that is consistent with the physical entity in real time. The health monitoring module is used to infer the health evolution trend of the support system based on the historical operating data and real-time synchronized data of the digital twin mirror model, through machine learning algorithms, identify potential instability risks, and issue graded warnings according to the risk level. The construction guidance and effect evaluation module is used to generate standardized construction guidance documents that comply with mine safety regulations based on the optimal initial design parameter set and the dynamically adjusted support parameters. It performs a full-process quantitative evaluation of the construction quality and support effect of the support system and feeds the evaluation results back to the support system design model to drive the support system design model to continuously learn and iteratively optimize.

Images