A Smart Design and Feedback Optimization System and Method for Mine Roadway Support

Abstract

This invention provides an intelligent design and feedback optimization system and method for mine roadway support, relating to the field of mine support technology. It includes nine subsystems: rock mass quality grading, automatic design of support parameters, knowledge graph case reasoning and AI optimization, rapid verification of proxy models, 3D visualization support design and digital twin modeling, BIM / GIS integration interface, multi-source monitoring and early warning, dynamic feedback optimization and surrounding rock parameter inversion, design report generation, and multi-roadway progress dashboards. Through multi-source data fusion analysis, it achieves intelligent generation and optimization of support schemes based on machine learning algorithms. It utilizes digital twin technology to establish a 3D visualization model of the roadway, and drives dynamic adjustments to the support scheme through real-time monitoring data. This effectively improves the scientific and economical nature of support design, effectively resists the risk of complex geological instability, realizes intelligent management of the entire support lifecycle, and ensures safe and efficient mine production.

Description

Technical Field

[0001] This invention relates to the field of mine support technology, specifically to an intelligent design and feedback optimization system and method for mine roadway support. Background Technology

[0002] As mineral resource extraction extends to deeper levels and geological conditions become increasingly complex, the surrounding rock of mine roadways generally exhibits characteristics such as high ground stress, strong heterogeneity, well-developed structural surfaces, and significant time-varying properties. During tunneling and service, roadways are prone to disasters such as roof subsidence, sidewall convergence, floor heave, and local instability, directly threatening construction safety and production efficiency. As a core technical link in ensuring safe mine production, the scientific design and adaptability of roadway support directly determine the stability and construction safety of the roadway.

[0003] Currently, mine roadway support design mainly relies on surrounding rock classification methods such as RQD, RMR, Q system, and BQ method, combined with national and industry standards for parameter selection, supplemented by engineering analogies and empirical judgment. However, due to the non-uniform distribution of surrounding rock space, the dynamic evolution of construction disturbances and mining effects, existing methods are mostly based on static design assumptions, making it difficult to accurately reflect the dynamic changes throughout the entire service life of the roadway. This often leads to the need for multiple revisions of support schemes in practical applications, resulting in problems such as response lag, increased engineering costs, and accumulated safety risks.

[0004] In recent years, some technologies have attempted to introduce computer-aided design, intelligent algorithms, or monitoring systems to optimize roadway support. For example, they have achieved automatic matching of support parameters through neural networks and multi-objective optimization, or analyzed the stress of anchor bolts (cables) and deformation of surrounding rock through monitoring platforms. Although these technologies have improved the level of design automation to a certain extent, they still have obvious limitations: (1) Most systems focus on parameter recommendation or optimization in the design stage, lacking in-depth linkage with post-construction monitoring data, and cannot form a closed-loop mechanism of design-verification-feedback-optimization; (2) Existing technologies related to full life cycle management focus on specific support forms or construction process management, and fail to systematically integrate surrounding rock classification, standardized design, rapid verification, and dynamic feedback optimization; (3) When monitoring shows that the support performance is insufficient, existing technologies mostly rely on manual experience to make judgments and adjustments, lacking a mechanism for inverting surrounding rock parameters and optimizing support schemes based on monitoring data.

[0005] In view of this, there is an urgent need to develop an intelligent design system and method for mine roadway support that integrates multi-method rock mass classification, automatic design of support schemes, intelligent optimization decision-making, and dynamic feedback of monitoring. Summary of the Invention

[0006] In view of the technical problems existing in the background art, the present invention provides an intelligent design and feedback optimization system and method for mine roadway support, which can realize the automatic design and intelligent optimization of roadway support schemes, as well as the dynamic feedback improvement of support effects after construction, thus ensuring safe production in the mine. The system fully integrates national standards and historical engineering experience to achieve optimized design in the initial design stage. During the roadway's service life, it collects data on the surrounding rock and support status, evaluates the support effect, and automatically generates optimized reinforcement schemes once potential failure trends are identified. Through this intelligent closed-loop design and feedback mechanism, the scientific nature, reliability, and economy of roadway support design can be significantly improved, accident hazards can be reduced, and a strong guarantee for safe and efficient mine production can be provided.

[0007] In a first aspect, embodiments of the present invention provide an intelligent design and feedback optimization system for mine roadway support, comprising: The data input layer is used to acquire multi-source geological data of the surrounding rock of the tunnel. The design processing layer connects to the data input layer, integrating a rock mass quality grading subsystem, an automatic support parameter design subsystem, and a knowledge graph case reasoning and AI optimization subsystem. These three subsystems are sequentially connected. First, the rock mass quality grading subsystem intelligently grades the input geological data and outputs a surrounding rock stability level index. Then, the automatic support parameter design subsystem automatically matches support types based on a standard database and generates initial parameter schemes. Finally, the knowledge graph case reasoning and AI optimization design subsystem integrates historical case experience and multi-objective optimization algorithms to intelligently optimize the support scheme and mitigate risks. The output display layer connects to the design processing layer, integrating a 3D visualization support design and digital twin modeling subsystem, a BIM / GIS integration interface subsystem, and a design report generation and multi-tunnel progress dashboard subsystem. The 3D visualization support design and digital twin modeling subsystem generates 3D models of optimized support schemes. The BIM / GIS integration interface subsystem seamlessly connects design data with the mine's digital platform, enabling data export and spatial information integration in standard formats such as IFC. The design report generation and multi-tunnel progress dashboard subsystem automatically aggregates data from the entire process, generating standardized design reports that conform to industry standards. The feedback optimization layer connects to the output display layer and the design processing layer, integrating a multi-source monitoring and early warning subsystem and a dynamic feedback optimization and surrounding rock parameter inversion subsystem. The multi-source monitoring and early warning subsystem continuously collects sensor data on surrounding rock deformation and support stress in the roadway, and uses a multi-level threshold mechanism to achieve early identification and early warning triggering of abnormal states. The dynamic feedback optimization and surrounding rock parameter inversion subsystem uses monitoring data to invert the actual mechanical parameters of the surrounding rock, and combines this with knowledge graph retrieval of similar failure cases to generate targeted reinforcement schemes. AI optimization algorithms are used to adjust support parameters in real time, and the optimization results are fed back to the design processing layer to form an iterative optimization cycle.

[0008] As a further improvement of the present invention, a rock mass quality classification subsystem is used to receive and process input geological exploration data, working face or core images, and output surrounding rock stability classification results. The automatic support parameter design subsystem is connected to the rock mass quality classification subsystem and is used to automatically generate an initial support design scheme based on the surrounding rock stability classification results and the tunnel design parameters. The knowledge graph case reasoning and AI optimization design subsystem is connected to the automatic support parameter design subsystem. It is used to retrieve historical cases based on the knowledge graph and optimize the initial support design scheme using artificial intelligence optimization algorithms to generate an optimized support scheme. A 3D visualization support design and digital twin modeling subsystem is connected to the knowledge graph case reasoning and AI optimization design subsystem, and is used to perform 3D modeling of the optimized support scheme and generate a visualization support model. The BIM / GIS integration interface subsystem connects to the 3D visualization support design and digital twin modeling subsystem, and is used to realize data interaction and integration between the visualization support model and the external BIM / GIS platform; The multi-source monitoring and early warning subsystem is used to collect monitoring sensor data in the roadway and process the data to provide multi-level early warnings. The dynamic feedback optimization and surrounding rock parameter inversion subsystem is connected to the multi-source monitoring and early warning subsystem and the knowledge graph case reasoning and AI optimization design subsystem, respectively. It is used to invert the surrounding rock mechanical parameters based on the monitoring sensor data and feed the inversion results back to the knowledge graph case reasoning and AI optimization design subsystem to trigger dynamic optimization and adjustment of the support scheme. The design report generation and multi-tunnel progress dashboard subsystem connects the 3D visualization support design and digital twin modeling subsystem with the BIM / GIS integration interface subsystem, and is used to automatically generate design reports and display the construction progress of multiple tunnels. The various subsystems, through the aforementioned connection methods, form a closed-loop control system encompassing data acquisition, intelligent design, verification and optimization, and monitoring and feedback.

[0009] As a further improvement of the present invention, the design processing layer also includes a proxy model rapid verification subsystem, which is connected to the knowledge graph case reasoning and AI optimization subsystem to replace traditional time-consuming mechanical simulation software and realize rapid verification and multi-round iterative optimization of the support scheme; the proxy model rapid verification subsystem includes: The numerical simulation database, serving as the basic data pool for surrogate model training, stores large-scale tunnel support simulation data generated by finite element / discrete element software, providing benchmark training samples covering different geological conditions, support schemes, and mechanical responses for the surrogate model training module. The surrogate model training module has built-in XGBoost surrogate models, LightGBM surrogate models, or lightweight neural network surrogate models pre-trained using simulation data from the numerical simulation database. It can automatically match the optimal model based on the type of input parameters. The rapid prediction module deploys the trained agent model as a real-time inference service to achieve rapid prediction and iterative verification of the mechanical response of the support scheme. The automatic parameter adjustment module automatically optimizes the support parameters in reverse based on the comparison results between the mechanical response output by the rapid prediction module and the safety standards; it generates parameter adjustment schemes through optimization algorithms and feeds the adjusted parameter set back to the surrogate model for iterative verification.

[0010] As a further improvement of the present invention, the rock mass quality classification subsystem includes: The multi-source data input module is responsible for receiving, cleaning, and standardizing multi-source heterogeneous geological data. The geological feature extraction module has a built-in target detection module that automatically extracts key geological feature parameters from multi-source data based on the YOLOv8 target detection algorithm. The multi-method classification calculation module includes parallel RQD calculation module, RMR scoring module, Q value analysis module, and BQ value calculation module, which are used to run four general rock mass classification algorithms, namely RQD method, RMR method, Q value method and BQ method, respectively, for parallel calculation and result fusion. The expert rule engine module has three built-in logics: consistency judgment, majority judgment, and weighted arbitration. It comprehensively arbitrates the results of multiple methods of classification, eliminates abnormal classification results, and outputs a prompt to supplement geological exploration data or manual review when the confidence level is lower than 0.7. Finally, it outputs the surrounding rock stability level with high confidence and the corresponding mechanical parameters. The grading result output module synchronously outputs the arbitration-reviewed surrounding rock stability level, core grading indicators, and mechanical parameters to the automatic support parameter design subsystem, providing core input for subsequent support design.

[0011] As a further improvement of the present invention, the automatic support parameter design subsystem incorporates a database of national standards and industry specifications for roadway support in metal mines and coal mines. Based on the rock mass classification results, it automatically recommends corresponding support types and parameters, including anchor bolt length, spacing, and shotcrete layer thickness, and considers adjustments for special conditions: increasing support density in areas with large burial depths or high stress; adding anchor cables or I-beam arches for extremely soft surrounding rock; adding prestressed anchor cables or retractable supports in areas affected by dynamic pressure; and considering anti-outburst support requirements in high-gas or outburst-prone mines. The automatic support parameter design subsystem includes: A standardized database is used to store and manage national standards, industry specifications, and historical experience data related to roadway support in metal mines and coal mines. The rock mass-support matching module is used to establish the mapping relationship between rock mass stability level and support parameters. It uses rock mass level and roadway category as dual index keys and adopts a rule constraint engine to achieve hierarchical matching. Level I rock mass corresponds to simple support and Level V rock mass corresponds to reinforced support. It also supports fuzzy matching and nearest neighbor retrieval and handles boundary conditions. The tunnel parameter analysis and adjustment module is used to process and analyze tunnel geometry and engineering conditions; identify special geological and engineering conditions; and adaptively adjust the basic scheme. The parameter optimization and scheme generation module is used to verify the rationality of the generated support parameters and perform local optimization; and to integrate the results of each unit to generate a complete support design scheme.

[0012] As a further improvement of the present invention, the knowledge graph case reasoning and AI optimization design subsystem constructs a knowledge graph based on a graph database, including five entities and their relationships: roadway, surrounding rock type, support scheme, failure mode, and reinforcement measures; the AI ​​algorithm adopts a multi-objective genetic algorithm (NSGA-II), with the objective functions of minimizing support cost and minimizing surrounding rock deformation, iteratively optimizing support parameters, and also has the function of automatic case incremental learning and database entry; the knowledge graph case reasoning and AI optimization design subsystem includes: The knowledge graph construction module has a built-in knowledge graph database. Based on the graph database, it defines five core entities: roadway, surrounding rock type, support scheme, failure mode, and reinforcement measures. It sorts out the relationships between these entities and forms a complete knowledge network in the support field. The case reasoning module connects to a knowledge graph database and includes a case retrieval module, a case reuse module, a case correction module, and a case learning module. It transforms the current roadway conditions into a graph query statement, calculates node similarity through a graph neural network, and retrieves the Top-K most similar historical cases. It extracts support experience from similar cases, clarifies reusable parameters through relational path reasoning, and, based on the differences between the current working conditions and historical cases, adjusts the support parameters accordingly to output an optimized support plan. The NSGA-Ⅱ multi-objective optimization module is connected to the case retrieval module, case reuse module, and case correction module respectively. It uses a multi-objective genetic algorithm to construct an optimization model, and uses the parameters of anchor bolts, anchor cables, and shotcrete layer as decision variables to achieve dual-objective iterative optimization of minimizing support cost and minimizing surrounding rock deformation. The GNN similarity calculation module, connected to the NSGA-Ⅱ multi-objective optimization module, uses graph neural network technology to deeply mine the complex topological relationships in the knowledge graph of roadway support cases. It fuses node features and graph structure information into low-dimensional vectors, achieving accurate case matching based on both semantic and structural similarity. This provides a deep, multi-dimensional similarity measurement benchmark that is difficult to achieve with traditional methods for support scheme optimization. The incremental learning automatic data entry module is connected to the case reasoning module and the knowledge graph construction module respectively. By analyzing the data of newly completed roadway support cases in real time, it automatically extracts the correlation between design parameters, monitoring results and support effects, dynamically updates the entity relationships and case features of the knowledge graph, and realizes the system's autonomous iterative optimization to improve the accuracy of case reasoning and AI optimization.

[0013] As a further improvement of the present invention, the multi-source monitoring and early warning subsystem is deployed in the roadway site to collect real-time data on surrounding rock deformation and support stress, including: The multi-source monitoring data acquisition module connects to various monitoring devices to achieve full coverage acquisition of multi-dimensional monitoring data, including anchor bolt axial force sensors, roof separation meters, tunnel convergence meters, and distributed optical fibers. The data fusion and preprocessing module performs noise reduction and anomaly removal on the collected raw multi-source monitoring data, standardizes and normalizes the multi-source monitoring data, and performs data spatial interpolation and fusion to generate a monitoring dataset in a unified format. The graded early warning module has a built-in safety early warning threshold standard for mine roadway support, corresponding to a three-level early warning mechanism: Level 1 trend warning, which prompts an increase in monitoring frequency when the deformation rate exceeds the threshold; Level 2 parameter fine-tuning warning, which triggers local reinforcement measures when stress data exceeds limits; and Level 3 reconstruction warning, which triggers a redesign of the support scheme when deformation continues to deteriorate. The machine learning prediction model, based on fused multi-source monitoring data, uses machine learning algorithms to predict the deformation trend and instability risk probability of the surrounding rock in the roadway in real time.

[0014] As a further improvement of the present invention, the dynamic feedback optimization and surrounding rock parameter inversion subsystem includes: The surrounding rock parameter inversion module has a built-in inversion model that combines a neural network surrogate model with a particle swarm optimization algorithm. Based on real-time monitoring data provided by the multi-source monitoring and early warning subsystem, it automatically performs inversion calculations and solves the true mechanical parameters of the surrounding rock in reverse by combining intelligent optimization algorithms with surrogate models. The finite element / discrete element simulation verification module verifies the surrounding rock parameters and support schemes generated by the surrounding rock parameter inversion module using numerical simulation software. The similar case retrieval module is used to match similar historical cases corresponding to the problem area and extract successful reinforcement experience for reference; The AI ​​reinforcement and optimization module generates local reinforcement solutions for specific problems. When the top slab settlement is too large, it is recommended to add anchor cables and sprayed concrete. When the convergence of the two sides is severe, it is recommended to add steel frame support or back filling material to ensure that the solution is targeted. The re-optimization module, serving as the comprehensive decision-making output of the dynamic feedback optimization and surrounding rock parameter inversion subsystems, is responsible for integrating the output results of the aforementioned modules to form a complete support update scheme. The specific process is as follows: Using the corrected mechanical parameters (elastic modulus, cohesion, internal friction angle, etc.) output by the surrounding rock parameter inversion module as new surrounding rock parameter inputs, and combining successful reinforcement experience extracted by the similar case retrieval module with local reinforcement suggestions generated by the AI ​​reinforcement optimization module, the NSGA-II multi-objective optimization algorithm is re-invoked. With the dual objectives of minimizing support costs and controlling surrounding rock deformation within a safe threshold, the overall or local support parameters are iteratively re-optimized. After the optimization results are verified by the finite element / discrete element simulation verification module, an "Improved Design Recommendation" containing information such as the type of reinforcement measures, specific parameters, and expected control effects is output. After construction is completed, the inversion parameters, reinforcement scheme, and monitoring verification effects are automatically added to the database as new cases, updating the knowledge graph and driving continuous system evolution.

[0015] As a further improvement of the present invention, the three-dimensional visualization support design and digital twin modeling subsystem can automatically generate a three-dimensional model of the surrounding rock and support components of the tunnel, supporting viewpoint rotation, zooming, and key distance measurement, and the model is updated in real time after parameter modification; the digital twin algorithm maps the real-time stress field and plastic zone distribution of the surrounding rock onto the surface of the tunnel model in the form of a cloud map; including: The 3D model automatic generation module is used as the basic modeling layer to convert support parameters into a structured 3D model; The real-time visualization and adjustment module is used for high-performance graphics rendering and visualization, supporting direct manipulation of model components and real-time updates and adjustments of relevant design parameters. The digital twin data mapping module is the core mapping tool used for real-time synchronization of physical entities and digital models.

[0016] As a further improvement of the present invention, the BIM / GIS integration interface subsystem supports reading the roadway spatial coordinates, strike slope and geological model data in the mine BIM platform, and exports the support scheme into IFC, shapefile and DXF standardized formats, so as to achieve seamless connection and data synchronization with the mine BIM model and GIS system.

[0017] Secondly, embodiments of the present invention provide a method for intelligent design and feedback optimization of mine roadway support, which performs intelligent design and feedback optimization based on the above-mentioned system, including the following steps: S1 collects multi-source geological data of the surrounding rock of the tunnel, performs rock mass quality classification based on multi-source data fusion and multiple classification methods, and outputs the surrounding rock stability level. S2. An initial support scheme is generated based on the surrounding rock stability level and specification database. The support parameters are optimized using knowledge graph case reasoning and artificial intelligence optimization algorithms. A surrogate model is used to verify the force of the optimized scheme to form a support scheme. S3, implement the optimized support scheme and simultaneously install surrounding rock and support status monitoring sensors; S4 generates a 3D visualization model of the roadway support and integrates it with the mine's digital platform. S5 collects real-time data on surrounding rock deformation and support stress, calculates the surrounding rock deformation rate and support stress state, compares them with preset safety thresholds, and monitors the status through a multi-level early warning mechanism. S6 uses monitoring data to invert surrounding rock parameters and regenerates reinforcement schemes. New cases are automatically added to the knowledge graph, allowing the system to continuously learn and evolve, achieving closed-loop feedback optimization.

[0018] Compared with existing technologies, the intelligent design and feedback optimization system for mine roadway support disclosed in this invention has the following beneficial effects: This invention quantifies rock mass classification and strictly follows specifications for selecting support parameters, reducing the subjectivity of design caused by relying solely on experience. This invention utilizes a knowledge graph case library and AI algorithms to automatically generate support schemes, improving design efficiency. It integrates mine roadway support cases and multiple industry standard data, and improves the scientificity and accuracy of support design decisions by leveraging multi-source data and intelligent decision-making.

[0019] This invention, through intelligent optimization, ensures that the support scheme meets safety requirements while avoiding overly conservative design, achieving dynamic matching between support strength and surrounding rock conditions. Furthermore, this invention integrates monitoring and early warning into the entire support design process, shifting from passive reinforcement to proactive intervention, realizing intelligent closed-loop control of design-monitoring-feedback-optimization, and guaranteeing the long-term stability of roadway support.

[0020] This invention standardizes and automates the preparation of design documents, enabling the generation of support design reports that conform to industry standards with a single click. It also provides a visual overview of the support status of all mine roadways and real-time push notifications of multi-level early warning information. Furthermore, through a digital twin mapping module, the internal states such as the stress field and plastic zone distribution of the surrounding rock are presented in the form of cloud maps, facilitating managers to quickly grasp the safety status of the roadways.

[0021] This invention, through parametric design and modular system structure, can be flexibly configured according to different mine types, roadway cross sections and surrounding rock conditions, and has good versatility and scalability, which is conducive to its promotion and application under different mining conditions.

[0022] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0023] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the present invention will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0024] Figure 1 This is a schematic diagram of the overall functional module structure and data flow of the intelligent design and feedback optimization system for mine roadway support provided by the present invention.

[0025] Figure 2 The flowchart of intelligent rock image recognition and multi-method fusion hierarchical process in the intelligent design and feedback optimization system for mine roadway support provided by this invention.

[0026] Figure 3 The flowchart of knowledge graph-driven case reasoning and AI optimization design in the intelligent design and feedback optimization system for mine roadway support provided by this invention.

[0027] Figure 4 This invention provides a schematic diagram of the rapid verification and iterative optimization principle of the proxy model in the intelligent design and feedback optimization system for mine roadway support.

[0028] Figure 5 A schematic diagram of the multi-source fusion monitoring and three-level early warning mechanism in the intelligent design and feedback optimization system for mine roadway support provided by the present invention.

[0029] Figure 6 The flowchart of the closed-loop control of dynamic feedback optimization and surrounding rock parameter inversion in the intelligent design and feedback optimization system for mine roadway support provided by the present invention is shown. Detailed Implementation

[0030] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0031] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the invention, are intended to cover non-exclusive inclusion.

[0032] In the description of the embodiments of this invention, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this invention, "multiple" means two or more, unless otherwise explicitly defined.

[0033] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0034] In the description of the embodiments of this invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0035] In the description of the embodiments of the present invention, the term "multiple" refers to two or more (including two), similarly, "multiple groups" refers to two or more (including two groups), and "multiple pieces" refers to two or more (including two pieces).

[0036] In the description of the embodiments of the present invention, the technical terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of the present invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of the present invention.

[0037] In the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of the present invention according to the specific circumstances.

[0038] To address the technical problems of static, slow-response, and experience-dependent traditional support designs under deep and complex geological conditions, this invention provides an intelligent design and feedback optimization system and method for mine roadway support. By integrating multiple methods and rock mass classification to eliminate subjective human error, and using a triple-driven design approach combining standards, case studies, and AI with proxy model verification, the mechanical adaptability of the support scheme is ensured. Furthermore, a dynamic response through a monitoring-inversion-optimization closed loop mitigates the risk of insufficient support at its source, rapidly addresses potential instability hazards, and significantly reduces the incidence of safety accidents. Multi-objective AI optimization enables zoned differentiated support, significantly reducing material waste from over-support; knowledge graph reuse of historical experience reduces trial-and-error costs; a dynamic feedback mechanism avoids additional investment in secondary corrections; and integrated functions throughout the entire lifecycle simplify the process, comprehensively controlling design, construction, and management costs. It is compatible with multi-source data input and digital platform integration, enabling real-time prediction of support effects using digital twins. It adapts to the time-varying and heterogeneous characteristics of deep and complex geology, as well as the impacts of construction disturbances and mining, achieving a leap from static design to dynamic adaptation. The system integrates all aspects of hierarchical management, design, optimization, monitoring, and operation and maintenance to form a one-stop solution. Standardized reports and progress dashboards meet actual management needs, support multiple types of roadways and custom configurations, greatly reduce the gap between theory and practice, and improve the efficiency and adaptability of solution implementation.

[0039] Example 1 Please refer to Figures 1 to 6 As shown in Embodiment 1 of the present invention, an intelligent design and feedback optimization system for mine roadway support is provided, comprising four main layers: a data input layer, a design processing layer, an output display layer, and a feedback optimization layer. These are further subdivided into nine subsystems: a rock mass quality grading subsystem, an automatic support parameter design subsystem, a knowledge graph case reasoning and AI optimization design subsystem, a proxy model rapid verification subsystem, a 3D visualization support design and modeling subsystem, a BIM / GIS integration interface subsystem, a multi-source monitoring and early warning subsystem, a dynamic feedback optimization and surrounding rock parameter inversion subsystem, and a design report generation and multi-roadway progress dashboard subsystem.

[0040] The data input layer is used to acquire multi-source geological data of the surrounding rock of the tunnel; The design processing layer, connected to the data input layer, integrates a rock mass quality grading subsystem, an automatic support parameter design subsystem, a knowledge graph case reasoning and AI optimization subsystem, and a proxy model rapid verification subsystem. These four subsystems are sequentially connected. First, the rock mass quality grading subsystem intelligently grades the input geological data and outputs a surrounding rock stability level index. Then, the automatic support parameter design subsystem automatically matches support types based on a standard database and generates initial parameter schemes. Finally, the knowledge graph case reasoning and AI optimization design subsystem integrates historical case experience and multi-objective optimization algorithms to intelligently optimize the support scheme and mitigate risks. The output display layer, connected to the design processing layer, integrates a 3D visualization support design and digital twin modeling subsystem, a BIM / GIS integration interface subsystem, and a design report generation and multi-tunnel progress dashboard subsystem. The 3D visualization support design and digital twin modeling subsystem generates 3D models of optimized support schemes. The BIM / GIS integration interface subsystem seamlessly connects design data with the mine's digital platform, enabling data export and spatial information integration in standard formats such as IFC. The design report generation and multi-tunnel progress dashboard subsystem automatically aggregates data from the entire process, generating standardized design reports that conform to industry standards. The feedback optimization layer is connected to the output display layer and the design processing layer, respectively, and integrates a multi-source monitoring and early warning subsystem and a dynamic feedback optimization and surrounding rock parameter inversion subsystem. The multi-source monitoring and early warning subsystem continuously collects on-site sensor data in the roadway and realizes early identification and early warning triggering of abnormal states based on a multi-level threshold mechanism. The dynamic feedback optimization and surrounding rock parameter inversion subsystem uses monitoring data to invert the actual mechanical parameters of the surrounding rock and combines knowledge graph to retrieve similar failure cases to generate targeted reinforcement schemes.

[0041] Please see Figure 2 As shown, the rock mass quality grading subsystem takes the geological parameters of the surrounding rock in the tunnel and the field images as its core inputs. Through multi-method fusion (comparing and comprehensively evaluating the results of multiple methods) and intelligent processing throughout the entire process, it accurately outputs the stability level of the surrounding rock (Level I: Very Good Rock Mass, Level II: Good Rock Mass, Level III: Medium Rock Mass, Level IV: Poor Rock Mass, Level V: Extremely Poor Rock Mass). Its specific settings are as follows: The rock mass quality grading subsystem possesses highly flexible data input capabilities, supporting multiple methods such as manual entry, batch import from Excel, integration with mining geological databases, and automatic image recognition. It can conveniently acquire geological exploration data such as core drilling results, joint and fracture development, and surrounding rock strength. For intelligent feature extraction, the YOLOv8 target detection algorithm is employed to automatically identify geological structures such as fractures, joints, and faults in tunnel face photographs or core images, accurately calculating key grading indicators such as RQD values ​​and joint spacing, eliminating subjective errors from manual interpretation.

[0042] In terms of grading methods, the rock mass quality grading subsystem incorporates multiple internationally accepted standards and industry-specific solutions: it supports the evaluation of rock mass integrity using the RQD method, the scoring and accumulation of rock uniaxial compressive strength, structural surface condition, groundwater, and other elements using the RMR method, and the calculation of Q values ​​related to rock mass blockiness and joint roughness using the BQ system; and it integrates a coal mine surrounding rock stability classification method (based on indicators such as surrounding rock strength, integrity, geostress, and roof strata structure) and Protodyakonov coefficient analysis function, specifically adapted to the needs of coal-bearing strata composite roof stability assessment, taking into account the characteristics of coal mine sedimentary rock bedding development.

[0043] To ensure the reliability of the grading results, the rock mass quality grading subsystem incorporates an expert rule base to achieve intelligent arbitration of results from multiple methods: (1) Consistency judgment: When the grading results of the RMR method, Q system, and BQ method are consistent (e.g., all are Grade III), the grade is directly output with a confidence level of 1.0; (2) Majority judgment: When two of the three methods have consistent results, the grade with the majority consensus is adopted as the final result with a confidence level of 0.85; (3) Weighted arbitration: When the results of the three methods are inconsistent, weights are automatically assigned according to geological conditions to calculate the weighted grading result. When the confidence level is lower than 0.7, the system prompts that geological exploration data needs to be supplemented or manual review is required to ensure that the grading results are scientific and credible.

[0044] Please see Figure 2 As shown, the rock mass quality classification subsystem includes: The multi-source data input module is responsible for receiving, cleaning, and standardizing multi-source heterogeneous geological data. The geological feature extraction module has a built-in target detection module that automatically extracts key geological feature parameters from multi-source data based on the YOLOv8 target detection algorithm. The multi-method classification calculation module includes parallel RQD calculation module, RMR scoring module, Q value analysis module, and BQ value calculation module, which are used to run four general rock mass classification algorithms, namely RQD method, RMR method, Q value method and BQ method, respectively, for parallel calculation and result fusion. The expert rules engine module has three built-in logics: consistency judgment, majority judgment, and weighted arbitration. It comprehensively arbitrates the results of multiple methods and removes abnormal results. The grading result output module synchronously outputs the arbitration-reviewed surrounding rock stability level, core grading indicators, and mechanical parameters to the automatic support parameter design subsystem, providing core input for subsequent support design.

[0045] The automatic support parameter design subsystem takes rock mass quality grading results and core roadway parameters (cross-sectional dimensions, burial depth, etc.) as input, and relies on the built-in metal mine and coal mine support standard database to realize the automated and standardized generation of support schemes. Its specific settings are as follows: The core database of the automatic support parameter design subsystem comprehensively compiles national and industry standards such as the "Design Code for Non-ferrous Metal Mining", "Safety Regulations for Metal and Non-metal Mines", "Technical Specification for Anchor Bolt Support in Coal Mine Roadways", and "Coal Mine Safety Regulations". It integrates various engineering experience data and uses rock mass grade and roadway type as search indexes to quickly match and recommend corresponding support types (combinations of anchor bolts / anchor cables / scaffolding, etc.) and key parameters (anchor bolt length, spacing, shotcrete layer thickness, etc.) to form an initial foundation support scheme.

[0046] For complex geology and special working conditions, the automatic design subsystem for support parameters has a built-in dynamic adjustment mechanism: (1) In areas with a burial depth of more than 800m or high ground stress, the support density is automatically increased to enhance the bearing capacity; (2) In extremely soft surrounding rock scenarios, anchor cables or I-beam arches are added to supplement the support strength; (3) In areas affected by dynamic pressure, prestressed anchor cables or retractable supports are configured to adapt to deformation requirements; (4) In high-gas or outburst mines, the relevant specifications for outburst prevention support are strictly followed to optimize the scheme.

[0047] The final output support scheme includes complete details such as support type combination, specifications and parameters of each component, and layout method, which fully meets the engineering construction guidance requirements, ensuring design compliance and greatly improving the adaptability of the support scheme to the actual working conditions.

[0048] The automatic support parameter design subsystem includes: Standardize the database management module to store and manage national standards, industry specifications, and historical experience data on roadway support in metal mines and coal mines; The rock mass-support matching module is used to establish the mapping relationship between rock mass stability level and support parameters. It uses rock mass level and roadway category as dual index keys and adopts a rule engine to achieve hierarchical matching. Level I rock mass corresponds to simple support and Level V rock mass corresponds to reinforced support. It also supports fuzzy matching and nearest neighbor retrieval and handles boundary conditions. The tunnel parameter analysis and adjustment module is used to process and analyze tunnel geometric features and engineering conditions; identify special geological and engineering conditions; and adaptively adjust the basic scheme. The parameter optimization and scheme generation module is used to verify the rationality of the generated support parameters and perform local optimization; and to integrate the results of each unit to generate a complete support design scheme.

[0049] Please see Figure 3As shown, the knowledge graph case reasoning and AI optimization design subsystem is driven by both case reuse and intelligent optimization to optimize the initial support plan, generate a safer and more economical final plan, and has self-evolution capabilities. Its specific settings are as follows: The knowledge graph case reasoning and AI optimization design subsystem constructs a knowledge graph for the tunnel support field based on a graph database. It covers five core entities: tunnel entities (location, cross-sectional dimensions, burial depth, etc.), surrounding rock type entities (lithology, RQD / RMR / Q / BQ values, mechanical parameters, etc.), support scheme entities (support combination, component specifications, cost and construction period, etc.), failure mode entities (failure type, triggering conditions, etc.), and reinforcement measure entities (measure type, applicable conditions, etc.). A complete association network is constructed through relational edges to provide solid data support for case reasoning.

[0050] The case reasoning process of the knowledge graph case reasoning and AI optimization design subsystem realizes the accurate reuse of historical experience. It is mainly reflected in: (1) Case retrieval: the current roadway conditions are converted into graph query statements, and the node similarity is calculated by graph neural network (GNN) to select the Top-K most similar historical cases; (2) Case reuse: the support parameters and design experience of similar cases are extracted, and the parameters that can be directly reused and the content that needs to be adjusted are identified by relational path reasoning; (3) Case correction: the support parameters are corrected in a targeted manner based on the working condition differences between the current roadway and historical cases; (4) Risk avoidance: the failure mode chain of failed cases is identified by path analysis, and the constraints are extracted to avoid the recurrence of similar problems.

[0051] The AI ​​optimization process of the knowledge graph case reasoning and AI optimization design subsystem enables intelligent iterative upgrades of parameters. First, the tunnel environment parameters are converted into standardized input feature vectors. Then, the working condition matching is enhanced by combining the knowledge graph with weighted Euclidean distance or GNN embedding vector cosine similarity algorithm. Next, a multi-objective genetic algorithm (NSGA-II) is used to construct an optimization model. The anchor bolt / cable specifications, spacing, and shotcrete thickness are used as decision variables. The dual objectives of minimizing support cost and minimizing surrounding rock deformation are used to iteratively optimize the model. After simulation verification, the candidate scheme is closed-loop corrected.

[0052] The knowledge graph case reasoning and AI optimization design subsystem has incremental learning and self-evolution capabilities: whenever a new support project is completed, it automatically extracts the design parameters, monitoring data, effect evaluation and other structured information of the case, updates it to the knowledge graph, continuously enriches the case library and rule library, and continuously improves the accuracy of reasoning and optimization as application scenarios accumulate.

[0053] Please see Figure 3 As shown, the knowledge graph case reasoning and AI optimization design subsystem includes: The knowledge graph construction module has a built-in knowledge graph database. Based on the graph database, it defines five core entities: roadway, surrounding rock type, support scheme, failure mode, and reinforcement measures. It sorts out the relationships between these entities and forms a complete knowledge network in the support field. The case reasoning module connects to a knowledge graph database and includes a case retrieval module, a case reuse module, a case correction module, and a case learning module. It transforms the current roadway conditions into a graph query statement, calculates node similarity through a graph neural network, and retrieves the Top-K most similar historical cases. It extracts support experience from similar cases, clarifies reusable parameters through relational path reasoning, and, based on the differences between the current working conditions and historical cases, adjusts the support parameters accordingly to output an optimized support plan. The NSGA-Ⅱ multi-objective optimization module is connected to the case retrieval module, case reuse module, and case correction module respectively. It uses a multi-objective genetic algorithm to construct an optimization model, and uses the parameters of anchor bolts, anchor cables, and shotcrete layer as decision variables to achieve dual-objective iterative optimization of minimizing support cost and minimizing surrounding rock deformation. The GNN similarity calculation module, connected to the NSGA-Ⅱ multi-objective optimization module, uses graph neural network technology to deeply mine the complex topological relationships in the knowledge graph of roadway support cases. It fuses node features and graph structure information into low-dimensional vectors, achieving accurate case matching based on both semantic and structural similarity. This provides a deep, multi-dimensional similarity measurement benchmark that is difficult to achieve with traditional methods for support scheme optimization. The incremental learning automatic data entry module is connected to the case reasoning module and the knowledge graph construction module respectively. By analyzing the data of newly completed roadway support cases in real time, it automatically extracts the correlation between design parameters, monitoring results and support effects, dynamically updates the entity relationships and case features of the knowledge graph, and realizes the system's autonomous iterative optimization to improve the accuracy of case reasoning and AI optimization.

[0054] Please see Figure 4 As shown, the rapid verification subsystem of the proxy model uses a machine learning model to replace traditional mechanical simulation software, realizing rapid prediction and verification of the mechanical response of the support scheme. This solves the pain points of traditional simulation calculations being time-consuming and unable to support real-time iterative optimization. Its specific settings are as follows: The core proxy model of the rapid verification subsystem is trained and constructed based on XGBoost, LightGBM, or lightweight neural network algorithms, with training data sourced from a large-scale mechanical simulation database. This database utilizes finite element / discrete element software such as FLAC3D, Abaqus, and UDEC to construct typical working condition models covering different surrounding rock grades, burial depths, cross-sectional dimensions, and support schemes, ensuring the model's generalization ability and prediction accuracy.

[0055] The model input feature vector comprehensively covers key influencing factors: (1) surrounding rock parameters (elastic modulus, Poisson's ratio, cohesion, internal friction angle, rock mass quality index); (2) roadway geometric parameters (span, height, burial depth, cross-sectional shape coefficient); (3) stress environment parameters (maximum principal stress, minimum principal stress, lateral pressure coefficient); (4) support parameters (anchor bolt length / diameter / spacing / row spacing, shotcrete layer thickness, whether anchor cables are used). The output end can accurately predict core mechanical response indicators such as roof settlement, sidewall convergence, plastic zone depth, maximum axial force of anchor bolts, and safety factor.

[0056] The rapid verification process of the surrogate model subsystem achieves automated and rapid iteration. First, the parameters of the support scheme to be verified are combined with the surrounding rock conditions into a standardized input feature vector, and the model outputs the mechanical response prediction result. Then, the system compares the predicted value with the preset safety criterion to determine whether the scheme meets the stability requirements. If it does not meet the requirements, the corresponding support parameters are automatically adjusted according to the type and magnitude of the out-of-limit index, and the surrogate model is called again for verification, forming a rapid iterative cycle of prediction, judgment, adjustment, and re-prediction until the scheme meets all safety constraints or reaches the preset iteration limit.

[0057] To ensure the reliability of the project, the proxy model rapid verification subsystem must undergo rigorous accuracy calibration before deployment. It is verified using a reserved test set from the historical mechanics simulation database. The average relative error between the prediction results and the calculation results of traditional simulation software such as FLAC3D and Abaqus must not exceed 10% before it can be officially put into use.

[0058] Please see Figure 4 As shown, the rapid verification subsystem for the proxy model includes: The numerical simulation database, serving as the basic data pool for surrogate model training, stores large-scale tunnel-support-surrounding rock simulation data generated by finite element / discrete element software, providing benchmark training samples covering different geological conditions, support schemes, and mechanical responses for the surrogate model training module. The surrogate model training module has built-in XGBoost surrogate models, LightGBM surrogate models, or lightweight neural network surrogate models pre-trained using simulation data from the numerical simulation database. It can automatically match the optimal model based on the type of input parameters. The rapid prediction module deploys the trained agent model as a real-time inference service to achieve rapid prediction and iterative verification of the mechanical response of the support scheme. The automatic parameter adjustment module automatically optimizes the support parameters in reverse based on the comparison results between the mechanical response output by the rapid prediction module and the safety standards; it generates parameter adjustment schemes through optimization algorithms and feeds the adjusted parameter set back to the surrogate model for iterative verification.

[0059] The three-dimensional visualization support design and digital twin modeling subsystem, with an interactive three-dimensional interface as its core, realizes the visualization operation of support design and the real-time mapping of the surrounding rock condition, balancing design convenience and condition traceability. Its specific settings are as follows: The three-dimensional visualization support design and digital twin modeling subsystem can automatically generate a complete three-dimensional model based on the support design scheme, accurately restore the three-dimensional geometry of the surrounding rock of the roadway, and clearly present the layout, size and specifications of various support components such as anchor bolts, anchor cables, steel frames, and shotcrete layers, so as to realize the intuitive presentation of the design results.

[0060] The 3D visualization support design and digital twin modeling subsystem supports full-range interactive operation. Users can rotate and zoom the view to carefully check the coverage and density of the support layout, accurately measure the distance of key parts, and ensure that there are no support blind spots or component interference conflicts. At the same time, it supports direct adjustment of parameters such as the specifications and spacing of support components in the 3D interface. After modification, the model is automatically updated in real time, which greatly improves the efficiency of design adjustment.

[0061] The three-dimensional visualization support design and digital twin modeling subsystem integrates dynamic digital twin mapping function, which can accurately fit the real-time stress field and plastic zone distribution inside the surrounding rock onto the surface of the roadway model in the form of cloud map. The color of the cloud map changes dynamically with the monitoring data, reflecting the evolution trend of the mechanical state of the surrounding rock, and providing visualization support for support effect evaluation and dynamic optimization.

[0062] The 3D visualization support design and digital twin modeling subsystem includes: The 3D model automatic generation module is used as the basic modeling layer to convert support parameters into a structured 3D model; The real-time visualization and adjustment module is used for high-performance graphics rendering and visualization, supporting direct manipulation of model components and real-time updates and adjustments of relevant design parameters. The digital twin data mapping module is the core mapping tool used for real-time synchronization of physical entities and digital models.

[0063] The BIM / GIS integration interface subsystem focuses on achieving seamless integration between support design data and the mine's digital platform. Through two-way data interaction, it deeply integrates support design schemes and tunnel geological models into the overall mine information system. Its specific settings are as follows: The BIM / GIS integration interface subsystem supports bidirectional interaction with existing mine BIM models and GIS geographic information systems. On the one hand, it can read core information such as roadway spatial coordinates, strike slope, adjacent roadway support parameters, and geological models from the BIM platform, providing comprehensive engineering background data support for support design and ensuring consistency between the design and the overall mine engineering plan. On the other hand, it can export the generated support schemes into standardized data formats such as IFC, shapefile, and DXF, and directly import them into the mine's BIM model or GIS database, enabling rapid reuse and unified management of design results and facilitating digital collaborative operation and maintenance in the mine.

[0064] Please see Figure 5 As shown, the multi-source monitoring and early warning subsystem connects to the online monitoring sensor network of the surrounding rock in the roadway to achieve real-time perception, intelligent judgment, and graded early warning of the support status, promptly identify potential failure risks, and provide data support for dynamic optimization. Its specific settings are as follows: The multi-source monitoring and early warning subsystem fully integrates various monitoring devices to continuously collect core data such as anchor bolt and cable stress, roof subsidence, and convergence of the roadway sides and roof and floor. It monitors axial force changes through anchor bolt and cable stress sensors, captures roof subsidence dynamics using roof separation meters and laser displacement sensors, and measures surrounding rock convergence using convergence meters, achieving comprehensive coverage of the support status.

[0065] The data processing stage of the multi-source monitoring and early warning subsystem has automated cleaning and analysis capabilities, automatically filters abnormal and noisy data, and accurately calculates the real-time values ​​and change rates of key indicators. At the same time, it integrates prediction functions. Based on the machine learning model trained on historical monitoring data, it can predict the development trend of indicators such as roof displacement in the next 24 hours, and assist in judging the trend of support status.

[0066] The data transmission of the multi-source monitoring and early warning subsystem adopts a direct connection architecture that combines wired transmission and local area network. Various sensors (anchor bolt axial force sensor, roof separation instrument, roadway convergence meter, distributed optical fiber) are connected to the underground monitoring substation through cables or mining optical fibers. The monitoring substation uploads the aggregated monitoring data to the ground server in real time via Ethernet local area network, and the server performs data fusion, preprocessing and early warning analysis.

[0067] The multi-source monitoring and early warning subsystem adopts a sensor-based monitoring system. Point-type monitoring instruments such as anchor bolt axial force sensors, roof delamination meters, multi-point displacement meters, and roadway convergence meters are deployed at key cross-sections of the roadway at designed intervals. Through fixed-point continuous data acquisition from preset discrete measuring points, real-time monitoring of core indicators such as surrounding rock deformation and support stress is achieved.

[0068] The multi-source monitoring and early warning subsystem has built-in multi-level thresholds and combined criteria, such as the cumulative subsidence of the roof exceeding 100mm, the axial force of the anchor bolt reaching 80% of the yield strength, and the surrounding rock convergence speed continuously accelerating. Once a single indicator exceeds the limit or a combined criterion is met (such as the roof subsidence rate significantly increasing and exceeding the threshold), it is determined that the support has a potential failure trend. If it is predicted that the deformation of the surrounding rock will continue to intensify and the existing support cannot be controlled, it is determined that the support effect does not meet the requirements.

[0069] The multi-source monitoring and early warning subsystem establishes a three-level early warning mechanism to achieve differentiated handling: Level 1 is trend early warning, prompting for increased monitoring frequency; Level 2 is parameter fine-tuning early warning, triggering local reinforcement measures such as grouting; Level 3 is reconstruction early warning, initiating strong support measures such as additional anchor cables and scaffolding to ensure the timeliness and pertinence of risk handling.

[0070] Please see Figure 5 As shown, the multi-source monitoring and early warning subsystem is deployed in the roadway to collect real-time data on surrounding rock deformation and support stress, including: The multi-source monitoring data acquisition module connects to various monitoring devices to achieve full coverage acquisition of multi-dimensional monitoring data, including anchor bolt axial force sensors, roof separation meters, tunnel convergence meters, and distributed optical fibers. The data fusion and preprocessing module performs noise reduction and anomaly removal on the collected raw multi-source monitoring data, standardizes and normalizes the multi-source monitoring data, and performs spatial interpolation and fusion of multi-source data to generate a monitoring dataset in a unified format. The graded early warning module has a built-in safety early warning threshold standard for mine roadway support, corresponding to a three-level early warning mechanism; The machine learning prediction model, based on fused multi-source monitoring data, uses machine learning algorithms to predict the deformation trend and instability risk probability of the surrounding rock in the roadway in real time.

[0071] Please see Figure 6 As shown, the dynamic feedback optimization and surrounding rock parameter inversion subsystem constructs an intelligent closed loop of monitoring and diagnosis, parameter inversion, scheme optimization, and simulation verification. When monitoring and analysis determine that the support effect has failed or is approaching failure, the optimization process is automatically triggered. The design is corrected through surrounding rock parameter inversion, relevant historical cases and built-in knowledge base are retrieved, and an optimized support and reinforcement scheme is generated by combining AI algorithms. After verifying the effectiveness of the scheme through mechanical simulation, improved design suggestions are output for on-site implementation, realizing dynamic iterative upgrades of support design. Its specific settings are as follows: The dynamic feedback optimization and surrounding rock parameter inversion subsystem has a built-in surrounding rock parameter inversion module. Based on the actual displacement, stress, anchor bolt axial force and other data obtained from monitoring, it uses a combination of neural network surrogate model and particle swarm optimization algorithm to inversely deduce the true mechanical parameters of the surrounding rock. First, a displacement-parameter inverse mapping is established through a pre-trained surrogate model to obtain an initial estimate. Then, the optimal solution is searched in the parameter space with the objective function of minimizing the sum of squared weighted residuals between the measured displacement and the predicted displacement. The inversion results are used to update the design model, realizing a closed-loop correction from assumed parameters to measured parameters.

[0072] The dynamic feedback optimization and surrounding rock parameter inversion subsystem integrates case reuse and AI intelligent generation in its scheme optimization phase. On the one hand, it searches the case database to match similar historical cases corresponding to the problem area (such as local roof subsidence) and extracts successful reinforcement experiences such as secondary grouting and adding arched sheds for reference. On the other hand, it calls the AI ​​optimization algorithm to generate local reinforcement schemes for specific problems. When the roof subsidence is too large, it is recommended to add anchor cables and sprayed concrete. When the convergence of the two sides is severe, it is recommended to add steel frame support or backfill material to ensure that the scheme is targeted.

[0073] The dynamic feedback optimization and surrounding rock parameter inversion subsystem integrates mechanical simulation verification functions, automatically imports reinforcement schemes into finite element / discrete element software, constructs a three-dimensional model of the roadway-support-surrounding rock, simulates and evaluates key indicators such as surrounding rock stress distribution and deformation, and verifies whether the scheme can control the roof settlement rate and anchor bolt axial force within a safe range; at the same time, it has an economic ranking function, which comprehensively scores candidate schemes through multi-attribute decision-making and recommends the solution with the best cost performance.

[0074] The improved design proposal output by the dynamic feedback optimization and surrounding rock parameter inversion subsystem includes the type of reinforcement measures, specific parameters, expected effects, and economic analysis. It can directly guide on-site implementation, solve the problem of lag in the passive correction of traditional support schemes, and achieve dynamic adaptation between support design and the actual state of surrounding rock.

[0075] Please see Figure 6 As shown, the dynamic feedback optimization and surrounding rock parameter inversion subsystem includes: The surrounding rock parameter inversion module has a built-in inversion model that combines a neural network surrogate model with a particle swarm optimization algorithm. Based on real-time monitoring data provided by the multi-source monitoring and early warning subsystem, it automatically performs inversion calculations and solves the true mechanical parameters of the surrounding rock in reverse by combining intelligent optimization algorithms with surrogate models. The finite element / discrete element simulation verification module verifies the surrounding rock parameters and support schemes generated by the surrounding rock parameter inversion module using numerical simulation software. The similar case retrieval module matches similar historical cases corresponding to the problem area (such as local subsidence of the roof slab) and extracts successful reinforcement experiences such as secondary grouting and adding arched sheds for reference; The AI ​​reinforcement and optimization module generates local reinforcement solutions for specific problems. When the top slab settlement is too large, it is recommended to add anchor cables and sprayed concrete. When the convergence of the two sides is severe, it is recommended to add steel frame support or back filling material to ensure that the solution is targeted. The re-optimization module, serving as the comprehensive decision-making outlet for the dynamic feedback optimization and surrounding rock parameter inversion subsystems, is responsible for integrating the output results of the aforementioned modules to form a complete support update scheme. The specific process is as follows: Using the corrected mechanical parameters (elastic modulus, cohesion, internal friction angle, etc.) output by the surrounding rock parameter inversion module as new surrounding rock parameter inputs, and combining successful reinforcement experience extracted by the similar case retrieval module with local reinforcement suggestions generated by the AI ​​reinforcement optimization module, the NSGA-II multi-objective optimization algorithm is re-invoked. With the dual objectives of minimizing support costs and controlling surrounding rock deformation within a safe threshold, the overall or local support parameters are iteratively re-optimized. After the optimization results are verified by the finite element / discrete element simulation verification module, an "Improved Design Recommendation" containing information such as the type of reinforcement measures, specific parameters, and expected control effects is output. After construction is completed, the inversion parameters, reinforcement scheme, and monitoring verification effects are automatically added to the database as new cases, updating the knowledge graph and driving continuous system evolution.

[0076] The design report generation and multi-tunnel progress dashboard subsystem integrates automated aggregation of design results with visualized management of the support status across the entire mine area, balancing design output efficiency with project progress control requirements. Its specific settings are as follows: The design report generation and multi-tunnel progress dashboard subsystem supports one-click generation of support design specifications that conform to industry standards. It has built-in standardized report templates that cover complete chapters such as project overview, geological condition description, rock mass classification results, support scheme design, parameter calculation process, material list and budget, construction process requirements, safety technical measures, and 3D design drawings. It can output multiple common document formats such as Word and PDF, directly meeting the needs of project archiving and construction handover.

[0077] The design report generation and multi-tunnel progress dashboard subsystem provides a visual overview of the support status of the entire mining area. The interface integrates core modules such as tunnel distribution map, design progress statistics, construction progress statistics, early warning information summary, and key indicator trend chart, presenting the design progress, construction implementation, and support safety status of each tunnel. It also supports real-time push of early warning information and multi-level permission management to ensure that relevant personnel receive risk alerts in a timely manner and to ensure that the management process is standardized and orderly.

[0078] The aforementioned subsystems, with data flow as the main thread and complementary functions as support, form a hierarchical, closed-loop interconnected architecture. The connection paths between the subsystems are as follows: The rock mass quality grading subsystem processes multi-source geological data and outputs the surrounding rock stability level to the support parameter automatic design subsystem; the latter generates an initial scheme based on specifications, which is then intelligently optimized by the knowledge graph case reasoning and AI optimization design subsystem in conjunction with historical cases; the optimized scheme undergoes mechanical response prediction and safety verification by the proxy model rapid verification subsystem; the verified scheme generates a visual model by the 3D visualization support design and digital twin modeling subsystem, and is integrated with an external platform through the BIM / GIS integration interface subsystem; after construction, the multi-source monitoring and early warning subsystem collects on-site data in real time, and the dynamic feedback optimization and surrounding rock parameter inversion subsystem inverts the actual surrounding rock parameters based on the monitoring data and triggers further optimization of the scheme; the design report generation and multi-tunnel progress dashboard subsystem aggregates the entire process data to generate standardized reports and visual dashboards. This connection path ultimately drives design iteration through monitoring data, forming a complete adaptive optimization closed loop.

[0079] In some other implementations, the iterative optimization logic of the weighted Euclidean distance algorithm / GNN+multi-objective genetic algorithm is replaced with an optimization model based on deep reinforcement learning. The system input feature vector remains unchanged, and explicit Top-K similar case retrieval is no longer performed. Instead, a dedicated agent is trained using a historical case library as the training set. The action space is defined by support parameters such as anchor bolt length, spacing, and shotcrete layer thickness, and the composite reward function is achieving the required surrounding rock stability and minimizing support costs. The optimal decision-making strategy is formed through trial and error learning.

[0080] After the algorithm was replaced, when generating the initial scheme, the system directly inputs the geological features into the pre-trained model, and the strategy network can output the optimal combination of support parameters at once, eliminating the intermediate steps of case retrieval and iterative optimization, thus significantly improving optimization efficiency.

[0081] Deep reinforcement learning and genetic algorithms are both effective means of multi-objective optimization. Although they take different paths, they can both find the cost-optimal parameter combination under the premise of satisfying safety constraints and realize the core function of intelligent optimization.

[0082] In some other implementations, the proxy model based on XGBoost / LightGBM in the proxy model training module is replaced with a proxy model based on Physical Information Neural Network (PINN) or Graph Neural Network (GNN).

[0083] After replacing the proxy model, the working logic has been optimized: During training, the PINN-based surrogate model embeds physical constraints such as rock mechanics equilibrium equations and constitutive relations into the neural network loss function, forcing the prediction results to conform to mechanical laws and reducing the physical distortion problem of pure data-driven models in scenarios with sparse samples or extrapolation.

[0084] The GNN-based surrogate model models the tunnel-surrounding rock-support system as a graph structure, with spatial discrete units as nodes and the mechanical coupling relationship between adjacent units as edges. It is naturally adapted to irregular cross-section tunnels and locally refined grids, and accurately captures the local stress concentration effect.

[0085] PINN and GNN are essentially machine learning models that take geological and support parameters as input and output mechanical response mappings, consistent with the functional positioning of XGBoost / LightGBM. Through physical constraint embedding and graph structure modeling respectively, they offer advantages in adaptability to complex working conditions and prediction accuracy, while still achieving rapid mechanical response prediction and supporting the core function of rapid iterative optimization.

[0086] In some other implementations, based on Example 1, the monitoring method of the multi-source monitoring and early warning subsystem is refined: distributed fiber optic sensing technology and 3D laser scanning robots are introduced to upgrade the single sensor numerical monitoring to a point-line-surface multi-source heterogeneous fusion monitoring system.

[0087] The refinement significantly increases the amount of data collected. The system needs to first reduce the dimensionality of the laser point cloud, perform spatial interpolation on the fiber optic strain data, and then fuse it with conventional sensor data. Although the data preprocessing time increases slightly, it eliminates the monitoring blind spots between discrete measurement points.

[0088] In Example 1, discrete sensors may miss local damage between two measuring points. After refinement, continuous and full-coverage monitoring of roadway deformation is achieved, which can accurately capture minute and localized failure precursors, greatly improve the sensitivity and coverage of early warning, and provide more accurate basis for early risk management.

[0089] In some other implementations, based on Example 1, the optimization strategy of the dynamic feedback optimization and surrounding rock parameter inversion subsystem is refined, upgrading the globally unified optimization strategy to a zone-specific collaborative control strategy. The system divides the roadway into several micro-segments along the roadway axis. Based on the stress state and deformation trend monitored and fed back from each segment, optimization parameters are calculated independently for each micro-segment. For example, segment K1 (stress concentration segment) only densifies the anchor bolts, segment K2 (weak surrounding rock segment) only implements grouting reinforcement, and segment K3 (stable segment) maintains the original support parameters unchanged.

[0090] The impact of the refinement on the work process is that the optimization calculation changes from finding uniform parameters for the entire tunnel to finding independent parameters for each segment, which increases the computational complexity. However, this impact can be offset by parallel computing, and more targeted reinforcement solutions can be output.

[0091] Differentiated control by zone avoids indiscriminate over-support and enables refined management of surrounding rock hazards. Customized measures are adopted for roadway sections with different stress states, which not only accurately solves local safety hazards, but also saves support materials and construction costs, achieving a balance between safety and economy.

[0092] Example 2 Embodiment 2 of the present invention also provides a method for intelligent design and feedback optimization of mine roadway support. The method uses the aforementioned intelligent design and feedback optimization system for mine roadway support to perform intelligent design and feedback optimization, including the following steps: Before proceeding with the specific design, the core database of the system is initialized, including loading the standard specification library, building a historical case knowledge graph, establishing a basic BIM model for the mine, and training the agent model to support subsequent intelligent decision-making. S1, collect geological information of the surrounding rock of the tunnel, and classify the rock mass quality by fusion of multiple methods; S2, matching the standard parameters according to the grading results, and using the knowledge graph-driven case reasoning subsystem to retrieve similar cases, the AI ​​optimization algorithm performs multi-objective optimization of the solution, and the agent model rapid verification subsystem verifies the strength of the optimized solution to form a support solution; S3, implement the optimized support scheme and simultaneously install surrounding rock and support status monitoring sensors; S4 generates a 3D visualization model of the roadway support and integrates it with the mine's digital platform. S5 continuously collects monitoring data, calculates the deformation rate of the surrounding rock and the stress state of the support, and compares it with the preset safety threshold. S6. If the monitoring indicators exceed the limits, the dynamic feedback optimization and surrounding rock parameter inversion subsystem will automatically identify the failure mode, invert the actual surrounding rock parameters based on the current monitoring data, regenerate the reinforcement optimization plan and guide the on-site implementation; new cases will be automatically added to the knowledge graph to update the system, so as to realize the continuous learning and evolution of the system.

[0093] Specifically, this embodiment 2 demonstrates the process of performing full life-cycle support management of a deep transportation roadway located 800 meters underground using the intelligent design and feedback optimization system for mine roadway support of the present invention.

[0094] S0 involves system deployment and data initialization. This system runs on a high-performance server in the mine's ground dispatch center and connects to the underground monitoring base station via the mine's industrial Ethernet. It completes initialization tasks such as loading the standard specification database, constructing a historical case knowledge graph, and training the agent model.

[0095] After the tunnel excavation exposed the surrounding rock in S1, geological technicians took photos of the tunnel face and uploaded them to the system. The intelligent rock image recognition module automatically identified fracture characteristics. Combined with on-site geological data, the rock mass quality grading subsystem automatically calculated the RQD, RMR, Q value, and BQ value. The results of the four methods were consistent, and the stability level of the surrounding rock in this section was finally determined to be Level III (moderately stable).

[0096] S2, the automatic support parameter design subsystem, based on the Class III rock mass and the roadway cross-section (4.5m wide × 4.0m high), automatically generates a basic scheme (Φ20mm resin anchor bolts, 2.0m long, 1.2m × 1.2m spacing, with metal mesh) by searching the built-in mining design specification database. The knowledge graph retrieves three historical successful cases with a similarity > 0.85. An AI optimization algorithm is then used to optimize and adjust the scheme, increasing the anchor bolt length to 2.4m and adjusting the spacing to 1.1m × 1.1m. A proxy model is used for force verification. After the scheme is approved, the 3D visualization support design and modeling subsystem automatically generates a 3D model of the roadway and anchor bolt layout based on the optimized scheme. Designers rotate the view in the interface to confirm that the anchor bolts are positioned correctly on the arch and sides without any blind spots. Subsequently, the BIM / GIS integration interface subsystem exports the design scheme in IFC format and automatically updates it to the overall BIM model of the mine for technical briefing by the construction team.

[0097] S3 was constructed according to the optimized plan, and monitoring equipment such as roof separation meter, anchor bolt axial force gauge, multi-point displacement meter, and distributed optical fiber were installed simultaneously.

[0098] S4 generates a 3D visualization model of the roadway support and integrates it with the mine's digital platform. The S5 multi-source monitoring and early warning subsystem sets the roof subsidence rate threshold to 2 mm / d and the anchor bolt axial force alarm value to 80% of the yield strength (i.e., 120 kN), and begins real-time data acquisition. On the 15th day after the support was completed, the monitoring subsystem detected an anomaly: the roof subsidence rate suddenly increased to 4.5 mm / d, and the anchor bolt axial force in a localized area reached 135 kN (exceeding the 80% threshold). The system determined that the support had entered a potential failure state, with an early warning level of Level II (parameter fine-tuning), and predicted that without intervention in the next 24 hours, the deformation would continue to worsen.

[0099] S6, the dynamic feedback optimization and surrounding rock parameter inversion subsystem is automatically triggered. The surrounding rock parameter inversion module inverts the actual parameters of the surrounding rock based on the monitored displacement data, and retrieves similar failure repair cases of deep high-stress rheological rock masses to generate a reinforcement plan: delayed grouting and prestressed anchor cable reinforcement are implemented in the area of ​​severe deformation of the arch crown. The proxy model verifies that the settlement rate of the top plate will be reduced to 0.5 mm / d after reinforcement. The system outputs an "Improved Design Proposal". After the reinforcement construction is completed, the monitoring data verifies the effectiveness of the plan; this case is automatically added to the knowledge graph for updating.

[0100] In summary, this invention discloses an intelligent design and feedback optimization system and method for mine roadway support, relating to the field of mine support technology, aiming to solve the problems of static, slow response, and reliance on experience in traditional support design under deep and complex geological conditions. The system integrates nine core subsystems: rock mass quality grading, automatic support parameter design, knowledge graph case reasoning and AI optimization, rapid verification of proxy models, 3D visualization and digital twin modeling, BIM / GIS integration, multi-source monitoring and early warning, dynamic feedback optimization, and design report generation, forming a fully intelligent closed loop of grading-design-optimization-verification-monitoring-feedback-optimization. The system automatically identifies surrounding rock structural features using the YOLOv8 algorithm and achieves accurate assessment by integrating multiple methods of rock mass grading; it generates initial support schemes based on national standards and knowledge graph case reasoning, which are then optimized by the NSGA-II multi-objective AI algorithm and rapidly verified using machine learning proxy models; it relies on a multi-source sensor network to collect support status data in real time and identifies potential failure risks through a three-level early warning mechanism; and it automatically generates targeted reinforcement schemes by combining surrounding rock parameter inversion and AI algorithms, outputting implementation suggestions after simulation verification. Meanwhile, the system supports 3D visualization interaction, seamless BIM / GIS data integration, and one-click generation of standardized reports, enabling intelligent and standardized management of the entire lifecycle of support design, construction, monitoring, and operation and maintenance. This invention significantly improves the scientific rigor, adaptability, and economy of roadway support design, effectively resisting the instability risks brought about by complex geological conditions such as high ground stress and strong heterogeneity, reducing safety hazards and engineering costs, and providing strong support for safe and efficient mine production.

[0101] It should be noted that the present invention is not limited to the above-described embodiments. The above embodiments are merely examples, and any embodiments that have the same structure and perform the same effects as the technical concept within the scope of the present invention are included within the scope of the present invention. Furthermore, various modifications that can be conceived by those skilled in the art to the embodiments, and other ways of constructing by combining some of the constituent elements of the embodiments, without departing from the spirit of the present invention, are also included within the scope of the present invention.

Claims

1. A smart design and feedback optimization system for mine roadway support, characterized in that, include: The data input layer is used to acquire multi-source geological data of the surrounding rock of the tunnel. The design processing layer connects to the data input layer and integrates the rock mass quality grading subsystem, the automatic support parameter design subsystem, and the knowledge graph case reasoning and AI optimization subsystem. The three subsystems are connected in sequence. First, the rock mass quality grading subsystem intelligently grades the input geological data and outputs the surrounding rock stability level index. Then, the automatic support parameter design subsystem automatically matches the support type based on the standard database and generates the initial parameter scheme. Then, by integrating historical case experience and multi-objective optimization algorithms through the knowledge graph case reasoning and AI optimization design subsystem, the support plan is intelligently optimized and risks are avoided. The output display layer connects to the design processing layer and integrates a 3D visualization support design and digital twin modeling subsystem, a BIM / GIS integration interface subsystem, a design report generation and multi-lane progress dashboard subsystem. The optimized support scheme is generated into a 3D model through the 3D visualization support design and digital twin modeling subsystem; The BIM / GIS integration interface subsystem seamlessly connects design data with the mine's digital platform, enabling data export and spatial information integration in standard formats such as IFC. The design report generation and multi-tunnel progress dashboard subsystem automatically aggregates data from the entire process to generate standardized design reports that conform to industry standards. The feedback optimization layer connects to the output display layer and the design processing layer, and integrates the multi-source monitoring and early warning subsystem and the dynamic feedback optimization and surrounding rock parameter inversion subsystem. The multi-source monitoring and early warning subsystem continuously collects on-site sensor data in the roadway and realizes early identification and early warning triggering of abnormal states based on a multi-level threshold mechanism. Through the dynamic feedback optimization and surrounding rock parameter inversion subsystem, the actual mechanical parameters of the surrounding rock are inverted using monitoring data, and similar failure cases are retrieved by combining knowledge graphs to generate targeted reinforcement schemes.

2. The intelligent design and feedback optimization system for mine roadway support according to claim 1, characterized in that, The design processing layer also includes a surrogate model rapid verification subsystem, connected to the knowledge graph case reasoning and AI optimization subsystem, used to replace traditional time-consuming mechanical simulation software to achieve rapid verification and multi-round iterative optimization of support schemes; the surrogate model rapid verification subsystem includes: The numerical simulation database, serving as the basic data pool for surrogate model training, stores large-scale tunnel-support-surrounding rock simulation data generated by finite element / discrete element software, providing benchmark training samples covering different geological conditions, support schemes, and mechanical responses for the surrogate model training module. The surrogate model training module has built-in XGBoost surrogate models, LightGBM surrogate models, or lightweight neural network surrogate models pre-trained using simulation data from the numerical simulation database. It can automatically match the optimal model based on the type of input parameters. The rapid prediction module deploys the trained agent model as a real-time inference service to achieve rapid prediction and iterative verification of the mechanical response of the support scheme. The automatic parameter adjustment module automatically optimizes the support parameters in reverse based on the comparison results between the mechanical response output by the rapid prediction module and the safety standards; it generates parameter adjustment schemes through optimization algorithms and feeds the adjusted parameter set back to the surrogate model for iterative verification.

3. The intelligent design and feedback optimization system for mine roadway support according to claim 1, characterized in that, The rock mass quality classification subsystem includes: The multi-source data input module is responsible for receiving, cleaning, and standardizing multi-source heterogeneous geological data. The geological feature extraction module has a built-in target detection module that automatically extracts key geological feature parameters from multi-source data based on the YOLOv8 target detection algorithm. The multi-method classification calculation module includes parallel RQD calculation module, RMR scoring module, Q value analysis module, and BQ value calculation module, which are used to run four general rock mass classification algorithms, namely RQD method, RMR method, Q value method and BQ method, respectively, for parallel calculation and result fusion. The expert rule engine module has three built-in logics: consistency judgment, majority judgment, and weighted arbitration. It comprehensively arbitrates the results of multiple methods of classification, eliminates abnormal classification results, and outputs a prompt to supplement geological exploration data or manual review when the confidence level is lower than 0.

7. Finally, it outputs the surrounding rock stability level with high confidence and the corresponding mechanical parameters. The grading result output module synchronously outputs the arbitration-reviewed surrounding rock stability level, core grading indicators, and mechanical parameters to the automatic support parameter design subsystem, providing core input for subsequent support design.

4. The intelligent design and feedback optimization system for mine roadway support according to claim 1, characterized in that, The knowledge graph case reasoning and AI optimization design subsystem includes: The knowledge graph database is built on graph database, defining five core entities: roadway, surrounding rock type, support scheme, failure mode, and reinforcement measures. It sorts out the relationships between the entities and forms a complete knowledge network in the support field. The case reasoning module connects to a knowledge graph database and includes a case retrieval module, a case reuse module, a case correction module, and a case learning module. It transforms the current roadway conditions into a graph query statement, calculates node similarity through a graph neural network, and retrieves the Top-K most similar historical cases. It extracts support experience from similar cases, clarifies reusable parameters through relational path reasoning, and, based on the differences between the current working conditions and historical cases, adjusts the support parameters accordingly to output an optimized support plan. The NSGA-Ⅱ multi-objective optimization module is connected to the case retrieval module, case reuse module, and case correction module respectively. It uses a multi-objective genetic algorithm to construct an optimization model, and uses the parameters of anchor bolts, anchor cables, and shotcrete layer as decision variables to achieve dual-objective iterative optimization of minimizing support cost and minimizing surrounding rock deformation. The GNN similarity calculation module is connected to the NSGA-Ⅱ multi-objective optimization module. It uses graph neural network technology to deeply mine the complex topological relationships in the knowledge graph of roadway support cases, and fuses node features and graph structure information into low-dimensional vectors to achieve accurate case matching based on both semantic and structural similarity, providing a similarity measurement benchmark for support scheme optimization. The incremental learning automatic data entry module is connected to the case reasoning module and the knowledge graph construction module respectively. By analyzing the data of newly completed roadway support cases in real time, it automatically extracts the correlation between design parameters, monitoring results and support effects, dynamically updates the entity relationships and case features of the knowledge graph, and realizes the system's autonomous iterative optimization to improve the accuracy of case reasoning and AI optimization.

5. The intelligent design and feedback optimization system for mine roadway support according to claim 1, characterized in that, The multi-source monitoring and early warning subsystem is deployed in the roadway to collect real-time data on surrounding rock deformation and support stress, including: The multi-source monitoring data acquisition module connects to various monitoring devices to achieve full coverage acquisition of multi-dimensional monitoring data, including anchor bolt axial force sensors, roof separation meters, tunnel convergence meters, and distributed optical fibers. The data fusion and preprocessing module performs noise reduction and anomaly removal on the collected raw multi-source monitoring data, standardizes and normalizes the multi-source monitoring data, and performs spatial interpolation and fusion of multi-source data to generate a monitoring dataset in a unified format. The graded early warning module has a built-in safety early warning threshold standard for mine roadway support, corresponding to a three-level early warning mechanism: Level 1 trend warning, which prompts an increase in monitoring frequency when the deformation rate exceeds the threshold; Level 2 parameter fine-tuning warning, which triggers local reinforcement measures when stress data exceeds limits; and Level 3 reconstruction warning, which triggers a redesign of the support scheme when deformation continues to deteriorate. The machine learning prediction model, based on fused multi-source monitoring data, uses machine learning algorithms to predict the deformation trend and instability risk probability of the surrounding rock in the roadway in real time.

6. The intelligent design and feedback optimization system for mine roadway support according to claim 2, characterized in that, The dynamic feedback optimization and surrounding rock parameter inversion subsystem includes: The surrounding rock parameter inversion module has a built-in inversion model that combines a neural network surrogate model with a particle swarm optimization algorithm. Based on real-time monitoring data provided by the multi-source monitoring and early warning subsystem, it automatically performs inversion calculations and solves the true mechanical parameters of the surrounding rock in reverse by combining intelligent optimization algorithms with surrogate models. The finite element / discrete element simulation verification module verifies the surrounding rock parameters and support schemes generated by the surrounding rock parameter inversion module using numerical simulation software. The similar case retrieval module is used to match similar historical cases corresponding to the problem area and extract successful reinforcement experience for reference; The AI ​​reinforcement and optimization module generates local reinforcement solutions for specific problems. When the top slab settlement is too large, it is recommended to add anchor cables and sprayed concrete. When the convergence of the two sides is severe, it is recommended to add steel frame support or back filling material to ensure that the solution is targeted. The re-optimization module integrates the mechanical parameters output by the surrounding rock parameter inversion module, the historical reinforcement experience extracted by the similar case retrieval module, and the local reinforcement scheme generated by the AI ​​reinforcement optimization module. Based on the updated surrounding rock parameters as constraints, the NSGA-II multi-objective optimization algorithm is invoked again to further optimize the support parameters of the entire section or part, generating a complete updated support scheme verified by simulation. The optimization results are synchronously fed back to the knowledge graph case reasoning and AI optimization design subsystem, triggering a new round of support scheme iteration and realizing dynamic feedback closed-loop control.

7. The intelligent design and feedback optimization system for mine roadway support according to claim 1, characterized in that, The automatic support parameter design subsystem includes: A standardized database is used to store and manage national standards, industry specifications, and historical experience data related to roadway support in metal mines and coal mines. The rock mass-support matching module is used to establish the mapping relationship between rock mass stability level and support parameters. It uses rock mass level and roadway category as dual index keys and adopts a rule engine to achieve hierarchical matching. Level I rock mass corresponds to simple support and Level V rock mass corresponds to reinforced support. It also supports fuzzy matching and nearest neighbor retrieval and handles boundary conditions. The tunnel parameter analysis and adjustment module is used to process and analyze tunnel geometry and engineering conditions; identify special geological and engineering conditions; and adaptively adjust the basic scheme. The parameter optimization and scheme generation module is used to verify the rationality of the generated support parameters and perform local optimization; and to integrate the results of each unit to generate a complete support design scheme.

8. The intelligent design and feedback optimization system for mine roadway support according to claim 1, characterized in that, The 3D visualization support design and digital twin modeling subsystem includes: The 3D model automatic generation module is used as the basic modeling layer to convert support parameters into a structured 3D model. The real-time visualization and adjustment module is used for high-performance graphics rendering and visualization, supporting direct manipulation of model components and real-time updates and adjustments of relevant design parameters. The digital twin data mapping module is the core mapping tool used for real-time synchronization of physical entities and digital models.

9. The intelligent design and feedback optimization system for mine roadway support according to claim 1, characterized in that, The BIM / GIS integration interface subsystem supports reading the roadway spatial coordinates, strike slope, and geological model data from the mine's BIM platform, and exports the support scheme into standardized formats such as IFC, shapefile, and DXF, achieving seamless integration and data synchronization with the mine's BIM model and GIS system.

10. A method for intelligent design and feedback optimization of mine roadway support, characterized in that, Intelligent design and feedback optimization based on the system according to any one of claims 1 to 9 includes the following steps: S1 collects multi-source geological data of the surrounding rock of the tunnel, performs rock mass quality classification based on multi-source data fusion and multiple classification methods, and outputs the surrounding rock stability level. S2. An initial support scheme is generated based on the surrounding rock stability level and specification database. The support parameters are optimized using knowledge graph case reasoning and artificial intelligence optimization algorithms. A surrogate model is used to verify the force of the optimized scheme to form a support scheme. S3, implement the optimized support scheme and simultaneously install surrounding rock and support status monitoring sensors; S4 generates a 3D visualization model of the roadway support and integrates it with the mine's digital platform. S5 collects real-time data on surrounding rock deformation and support stress, calculates the surrounding rock deformation rate and support stress state, compares it with preset safety thresholds, and monitors the status through a multi-level early warning mechanism. S6 uses monitoring data to invert surrounding rock parameters and regenerates reinforcement schemes. New cases are automatically added to the knowledge graph, allowing the system to continuously learn and evolve, achieving closed-loop feedback optimization.

Images