Methods, equipment, and media for classifying and dynamically designing surrounding rock and support in underground caverns based on multi-source data fusion

Abstract

This application proposes a method, equipment, and medium for the classification and dynamic design of surrounding rock and support in underground caverns based on multi-source data fusion. The method includes: collecting multi-source heterogeneous data during the exploration and construction phases of the underground cavern; aggregating the multi-source heterogeneous data at a feature layer to generate standardized digital twin feature vectors for the surrounding rock; inputting the digital twin feature vectors into a pre-trained spatiotemporal feature extraction network to output the predicted rock level and its confidence level for the current working face in real time; and generating preliminary support parameters based on the predicted rock level and the digital twin feature vectors through deep reinforcement learning, and outputting the final support design parameters after numerical simulation verification. This method enables deep fusion of multi-source heterogeneous data and real-time objective discrimination of the surrounding rock level, improving the accuracy and consistency of classification, ensuring construction safety, and reducing engineering costs.

Description

Technical Field

[0001] This application relates to the fields of geotechnical engineering and underground space development and utilization technology, and in particular to a method, equipment and medium for dynamic design of surrounding rock classification and support of underground caverns based on multi-source data fusion. Background Technology

[0002] Currently, underground caverns, as core structures in water conservancy, hydropower, transportation, and mining engineering projects, directly impact project safety and construction costs through the evaluation of surrounding rock stability and support design. Among related technologies, a traditional surrounding rock evaluation system, represented by the "Engineering Rock Mass Classification Standard," has been constructed through the collaborative work of geological exploration, on-site monitoring, and experience-based grading methods. Specifically, this system covers the entire process from geological sketching and core drilling to geophysical interpretation, including key aspects such as rock mass structure analysis, rock mechanical parameter testing, and pre-design of support parameters, aiming to provide a basis for selecting excavation methods and formulating support schemes.

[0003] However, the rock mass classification and support design methods in related technologies directly employ discrete manual geological sketches and static pre-design models, failing to achieve real-time fusion and dynamic response of multi-source heterogeneous data. This may lead to highly subjective classification results, delayed data acquisition, or a disconnect between the support scheme and the actual site conditions, thus affecting the safety and economy of the project. Specifically, traditional methods struggle to continuously capture abrupt changes in the unexposed rock mass ahead of the tunnel face, and different engineers may provide differing rock mass classifications under the same geological conditions, resulting in arbitrary support decisions. Therefore, there is an urgent need to construct an intelligent system capable of real-time fusion of multi-source geological information and dynamic optimization of support parameters. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art.

[0005] Therefore, the first objective of this application is to propose a method for classifying and dynamically designing the surrounding rock and support of underground caverns based on multi-source data fusion.

[0006] The second objective of this application is to propose an electronic device.

[0007] The third objective of this application is to provide a computer-readable storage medium.

[0008] To achieve the above objectives, the first aspect of this application is to propose a method for classifying and dynamically designing the surrounding rock and support of underground caverns based on multi-source data fusion, comprising the following steps:

[0009] Collect multi-source heterogeneous data during the exploration and construction phases of underground caverns. The multi-source heterogeneous data includes geological exploration data, real-time monitoring data of surrounding rock microseismic activity and deformation, digital image data of full-section rock mass structure, and advanced geological prediction data. The multi-source heterogeneous data are aggregated at the feature layer, and evidence theory is used to eliminate information conflicts and uncertainties to generate standardized digital twin feature vectors of surrounding rock. The surrounding rock digital twin feature vector is input into a pre-trained spatiotemporal feature extraction network, which outputs the prediction result of the surrounding rock level of the current working face and the confidence level of the prediction result in real time. Based on the predicted rock level and the digital twin feature vector of the surrounding rock, preliminary support parameters are generated through deep reinforcement learning. After numerical simulation verification, the preliminary support parameters are output as final support design parameters that meet safety requirements.

[0010] To achieve the above objectives, a second aspect of this application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the dynamic design method for grading and support of surrounding rock in underground caverns based on multi-source data fusion as described in any one of the first aspects above.

[0011] To achieve the above objectives, the third aspect of this application also proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method for dynamic design of surrounding rock classification and support of underground caverns based on multi-source data fusion as described in any of the first aspects above.

[0012] The technical solution provided by the embodiments of this application brings at least the following beneficial effects: This application can achieve deep fusion of multi-source heterogeneous data and real-time objective discrimination of surrounding rock level, improve the accuracy and consistency of classification, and dynamically optimize support parameters through collaborative decision-making of reinforcement learning and numerical simulation, thereby ensuring construction safety and reducing engineering costs. Therefore, this application can predict the surrounding rock level in real time, objectively and accurately, and effectively improve construction safety and economy through dynamic optimization of support parameters.

[0013] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0014] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1A flowchart illustrating a method for classifying and dynamically designing the surrounding rock of underground caverns based on multi-source data fusion, as proposed in an embodiment of this application. Figure 2 This is a schematic diagram of the workflow of a rapid classification and dynamic support design system for surrounding rock of underground caverns based on deep fusion of multi-source data, as proposed in an embodiment of this application. Figure 3 This is a schematic diagram illustrating the workflow of a multi-source heterogeneous data feature layer aggregation module proposed in an embodiment of this application. Figure 4 This is a schematic diagram illustrating the workflow of an intelligent rock grading module proposed in an embodiment of this application. Figure 5 This is a schematic diagram illustrating the workflow of a dynamic decision-making module for support parameters proposed in this application, which combines reinforcement learning and numerical simulation. Detailed Implementation

[0015] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0016] The following description, with reference to the accompanying drawings, describes a method, equipment, and medium for dynamic design of surrounding rock classification and support for underground caverns based on multi-source data fusion, as proposed in the embodiments of this application.

[0017] Example 1 Figure 1 This is a flowchart illustrating a method for classifying and dynamically designing surrounding rock and support for underground caverns based on multi-source data fusion, as proposed in an embodiment of this application. Figure 1 As shown, the method includes the following steps: Step S101: Collect multi-source heterogeneous data during the exploration and construction phases of the underground cavern. The multi-source heterogeneous data includes geological exploration data, real-time monitoring data of surrounding rock microseismic activity and deformation, digital image data of the full-section rock mass structure, and advanced geological prediction data.

[0018] Specifically, when collecting multi-source heterogeneous data during the exploration and construction phases of underground caverns, the system integrates various sensing methods to acquire multi-dimensional information covering the entire project lifecycle and reflecting the physical and mechanical state and geological conditions of the surrounding rock. This step aims to construct a comprehensive, three-dimensional, and continuous data acquisition system to overcome the limitations of traditional discrete data acquisition methods.

[0019] Specifically, the collected data includes at least geological survey data, real-time monitoring data of surrounding rock microseismic activity and deformation, digital image data of the entire rock mass structure, and advanced geological prediction data. Geological survey data provides the regional geological background and initial mechanical parameters of the rock mass; real-time monitoring data of surrounding rock microseismic activity and deformation continuously captures rock mass fracture events and displacement-strain evolution through a sensor network deployed within the surrounding rock to reflect the dynamic response under construction disturbance; digital image data of the entire rock mass structure utilizes 3D laser scanning and photogrammetry to quickly acquire dense point cloud and texture images of the tunnel face and tunnel walls after each excavation, and automatically extracts the geometric parameters of the rock mass structural surfaces; advanced geological prediction data connects to the interpretation results of various geophysical methods through standardized interfaces to reveal geological anomalies in areas not previously exposed ahead of the tunnel face. While these data types differ in dimensions, precision, and spatiotemporal resolution, they collectively form the basis for subsequent fusion analysis.

[0020] As one possible implementation, the system may specifically include a geological exploration data unit, a high-density Internet of Things (IoT) monitoring network, a full-section digital scanning unit, and an advanced geological prediction data interface. The IoT monitoring network may further include devices such as microseismic sensors, multi-point displacement gauges, fiber optic grating sensors, and anchor bolt axial force gauges. The full-section digital scanning unit may be based on a 3D laser scanner and a high-definition camera equipped with real-time positioning and mapping technology.

[0021] Thus, through this step, the system can achieve synchronous, continuous, and automated acquisition of multi-source heterogeneous data throughout the entire process from exploration to construction, significantly improving the timeliness and spatial coverage density of data acquisition, and providing a rich and reliable data foundation for the accurate identification of the surrounding rock condition and the dynamic optimization of support parameters.

[0022] Step S102: Aggregate multi-source heterogeneous data at the feature layer, use evidence theory to eliminate information conflicts and uncertainties, and generate standardized digital twin feature vectors of surrounding rock.

[0023] Specifically, after acquiring multi-source heterogeneous data covering the entire exploration and construction process, it is necessary to perform feature-level aggregation processing to eliminate information conflicts and uncertainties introduced by differences in data sources, dimensions, and precision. The core of this method lies in fusing data from different sensing methods and with different confidence levels through evidence theory, thereby generating a digital twin feature vector that can comprehensively and standardizedly reflect the current state of the surrounding rock.

[0024] Specifically, all collected multi-source heterogeneous data are first mapped to a spatiotemporal coordinate system based on the engineering axis, thereby aligning the data in spatial location and time series.

[0025] Subsequently, the various data types were transformed into basic probability assignments for different evidence categories, where each data category was considered independent evidence for a specific attribute of the surrounding rock (such as integrity, degree of damage, or geological anomaly). Using the synthesis rules of evidence theory, the orthogonal sum of each piece of evidence was calculated to obtain the fused comprehensive reliability function. This function effectively handles conflicts between different pieces of evidence and quantifies the uncertainty of the fusion result.

[0026] Finally, based on the comprehensive confidence function, a digital twin feature vector of the surrounding rock containing multi-dimensional aggregated feature parameters is generated. This vector is the unified input for subsequent surrounding rock classification and support decision-making.

[0027] As one possible implementation method, microseismic monitoring data, digital scanning data, and advanced prediction data can be constructed as independent evidence bodies characterizing surrounding rock damage, integrity, and geological anomalies, and then fused using the Dempster-Shafer evidence theory synthesis rules to generate a feature vector containing parameters such as geological strength index, rock mass integrity coefficient, and forward anomaly probability.

[0028] Therefore, this step, by introducing evidence theory to fuse the data of the feature layer, effectively solves the common problems of information conflict and uncertainty among multi-source heterogeneous data, significantly improves the accuracy and robustness of the surrounding rock condition characterization, and provides a reliable and standardized data foundation for the subsequent objective and intelligent identification of the surrounding rock level.

[0029] Step S103: Input the digital twin feature vector of the surrounding rock into the pre-trained spatiotemporal feature extraction network, and output the prediction result of the surrounding rock level of the current working face and the confidence level of the prediction result in real time.

[0030] Specifically, the digital twin feature vector of the surrounding rock is input into a pre-trained spatiotemporal feature extraction network. This network is configured to identify and output the predicted level of the surrounding rock at the current working face and its corresponding confidence level in real time from multidimensional aggregated data characterizing the rock mass state. The core of this network lies in its ability to simultaneously handle the non-Euclidean correlation of the rock mass structure in space and the dynamic evolution of the monitoring data over time.

[0031] Specifically, the network first constructs the input surrounding rock digital twin feature vector into graph structure data, where the nodes of the graph correspond to rock mass units at different spatial locations, and the node features are the multidimensional aggregation parameters of the unit; the edges of the graph are defined according to the structural surface network or mechanical relationship within the rock mass, in order to characterize the spatial topological connections and interactions between different rock mass units.

[0032] Subsequently, the network aggregates and updates the neighborhood information of each node using graph convolution operators, thereby extracting the spatial correlation features of the rock mass structure and achieving a quantitative characterization of complex topological structures such as joint networks. Building upon this, the network further introduces a recurrent neural network structure to capture the evolution of node features over time, thus incorporating dynamic information such as deformation trends and microseismic activity frequency from the monitoring data into the analysis.

[0033] Finally, the network outputs the probability distribution of the current working face surrounding rock belonging to each preset level through a fully connected layer and a normalized exponential function, and uses the level corresponding to the highest probability as the prediction result. At the same time, it calculates the confidence level of the prediction based on the entropy value or a similar metric of the probability distribution.

[0034] As one possible implementation, the spatiotemporal feature extraction network can adopt a hybrid architecture of graph convolution and gated recurrent units, where graph convolution layers are used for spatial feature extraction and gated recurrent unit layers are used for temporal feature capture. During the training phase, the model can be pre-trained using historical engineering data and then fine-tuned using initial excavation data from specific engineering sites to ensure its adaptability to specific geological conditions.

[0035] Therefore, this step, by constructing a deep learning network that can simultaneously characterize the spatial topology and temporal evolution of rock masses, achieves real-time, objective, and intelligent identification of surrounding rock grades, significantly improving the accuracy and reliability of the grading results. Furthermore, by quantifying the prediction confidence level, it provides an important reference for subsequent risk management and decision-making.

[0036] Step S104: Based on the prediction results of the surrounding rock level and the digital twin feature vector of the surrounding rock, preliminary support parameters are generated through deep reinforcement learning, and the preliminary support parameters are verified by numerical simulation and then the final support design parameters that meet the safety requirements are output.

[0037] Specifically, the system triggers a dynamic decision-making process when the predicted rock mass level changes, the prediction confidence level falls below a preset threshold, or real-time monitoring data exceeds safety limits. The core of this step lies in automatically generating support design parameters that meet engineering safety requirements based on a multi-dimensional characterization of the current rock mass state, through a combination of intelligent optimization and mechanical verification.

[0038] Specifically, the system first obtains the predicted rock mass level and its confidence level output from the aforementioned steps, as well as a digital twin feature vector characterizing the overall state of the surrounding rock. This vector contains multi-dimensional aggregated information reflecting the rock mass strength, integrity, damage level, and probability of geological anomalies ahead. Based on this, the system calls upon a pre-set support parameter knowledge base, which stores historical support cases and their effect feedback under different geological and engineering conditions.

[0039] Subsequently, a parameter optimization engine based on deep reinforcement learning takes the current digital twin feature vector of the surrounding rock and the construction status as input. Through a pre-defined multi-objective reward function, it guides the agent to explore in a discrete and hybrid action space, generating a preliminary combination of support parameters. To ensure the mechanical rationality of the scheme, this preliminary scheme must be verified by a numerical simulation verification unit. This unit automatically establishes and solves a numerical model based on the current face profile, preliminary support parameters, and aggregated rock mechanics parameters, verifying key indicators such as the range of the plastic zone of the surrounding rock, deformation, and internal forces of the support structure. If the simulation results meet the preset safety standards, the scheme is output as the final support design parameters; if not, it is fed back to the optimization engine for iterative iteration until a scheme that meets the requirements is generated.

[0040] As one possible implementation, the state space of the deep reinforcement learning engine can be defined as the current digital twin feature vector of the surrounding rock, the deformation history of the exposed tunnel section, and the current construction sequence. The action space is an adjustable combination of support parameters, and the reward function is a weighted design that integrates multiple objectives such as safety, economy, and construction convenience. The numerical simulation verification unit can integrate finite element or finite difference software to realize a closed-loop process of modeling and solving through an automated interface.

[0041] Therefore, this step combines data-driven rapid optimization with physics-driven mechanical verification to form a two-layer decision-making mechanism. It utilizes the efficient exploration capabilities of deep reinforcement learning in complex parameter spaces and ensures the mechanical reliability and engineering safety of the support scheme through numerical simulation. Thus, it achieves dynamic optimization and accurate output of the support design while ensuring construction safety.

[0042] Example 2 Based on the above embodiments, this embodiment provides a detailed description of the specific implementation of step S101, "collecting multi-source heterogeneous data during the exploration and construction phases of underground caverns".

[0043] In this embodiment, firstly, the physical and mechanical parameters of the borehole core and the regional geostress field information are obtained through the geological exploration data unit. This unit integrates regional geological maps, borehole columnar sections and core test data. The core test data includes uniaxial compressive strength, elastic modulus and Poisson's ratio. These data serve as the input source for the initial geostress field and basic quality indicators of the rock mass.

[0044] Then, a high-density Internet of Things (IoT) monitoring network is used to collect real-time data on the spatiotemporal evolution of microfractures in the surrounding rock, internal displacement, continuous strain field, and stress data of the support structure. This network uses low-power wide-area network technology to deploy wireless smart sensors, including embedded microseismic monitoring sensors to capture microseismic events generated by rock fractures and locate the fracture source, multi-point displacement gauges and multi-point borehole strain gauges to monitor the radial displacement and strain evolution at different depths inside the surrounding rock, fiber optic grating sensors deployed along the tunnel axis to monitor changes in the continuous strain field and temperature field, and anchor bolt axial force gauges and contact pressure cells to monitor the stress state of the initial support structure in real time. The data output by these sensors are spatiotemporally interpolated and correspond one-to-one with geological units.

[0045] Then, after each blast, the full-section digital scanning unit quickly acquires dense point cloud and texture images of the tunnel face and tunnel walls. This unit uses a 3D laser scanner and a high-definition camera equipped with real-time positioning and mapping technology to quickly scan the tunnel face and exposed tunnel walls after muck removal, acquire point cloud data and simultaneously take high-definition photos. The geometric parameters of the rock mass structural surfaces are automatically extracted through computer vision algorithms, including the number of joint groups, attitude, spacing, opening, roughness and filling conditions, to generate the structural surface network in the digital twin model.

[0046] Finally, the interpretation results of advanced geological prediction systems using the seismic reflection wave method, ground-penetrating radar method, or induced polarization method are standardized and accessed through an interface for advanced geological prediction data. This interface receives information on the location, scale, and type of anomalous geological bodies ahead, which serves as input for subsequent aggregation modules. The output data of each of the above sub-units are uniformly mapped to a spatiotemporal coordinate system centered on the tunnel axis station number, forming data layer alignment and providing standardized input for subsequent feature layer aggregation.

[0047] Therefore, this embodiment, by refining the multi-dimensional stereoscopic perception module into four functionally distinct sub-units, realizes full-cycle data acquisition from geological exploration, real-time monitoring, digital scanning to advanced forecasting, ensuring the integrity, real-time nature, and standardization of multi-source heterogeneous data, providing a high-quality data foundation for subsequent deep integration based on DS evidence theory, thereby improving the objectivity and accuracy of surrounding rock classification and support decisions.

[0048] Example 3 Based on the above embodiments, this embodiment provides a detailed description of the specific implementation of step S102, which involves "aggregating multi-source heterogeneous data at the feature layer, using evidence theory to eliminate information conflicts and uncertainties, and generating standardized digital twin feature vectors of surrounding rock".

[0049] In this embodiment, firstly, the multi-source heterogeneous data feature layer aggregation module receives various types of data from the multi-dimensional stereoscopic perception module, including geological exploration data, microseismic monitoring data, digital scanning data, and advanced geological prediction data. This module maps all data to a spatiotemporal coordinate system centered on the tunnel axis stationing number, thus aligning the data layers.

[0050] Specifically, for point cloud data acquired by 3D laser scanning, precise station coordinates are assigned using real-time positioning and mapping technology. For IoT monitoring data such as microseismic sensors and multi-point displacement gauges, spatiotemporal interpolation is performed based on their deployment locations to ensure a one-to-one correspondence with geological units defined by station numbers, thereby eliminating the spatial and temporal dispersion of the data. After completing the data layer alignment, the module transforms various types of data into basic probability assignments for different evidence bodies.

[0051] As an example, basic probability allocations for evidence of rock mass integrity are constructed based on the number of joint groups, spacing, volume joint number extracted by digital scanning, and core integrity coefficients from borehole data; basic probability allocations for evidence of surrounding rock damage are constructed based on microseismic monitoring parameters, such as microseismic event density, magnitude-frequency relationship values, and energy index; and basic probability allocations for evidence of geological anomalies are constructed based on information such as wave velocity anomalies and resistivity anomalies interpreted by advanced geological prediction systems.

[0052] Subsequently, the Dempster-Shafer evidence theory synthesis rules were used to calculate the orthogonal sum of each piece of evidence, resulting in a comprehensive reliability function. This synthesis process can effectively handle potential conflicts and uncertainties between different pieces of evidence. For example, when digital scanning shows that the rock mass is relatively intact while microseismic monitoring shows increased damage, the synthesis rules automatically weigh the reliability of each piece of evidence.

[0053] Finally, based on the comprehensive confidence function, a standardized digital twin feature vector of the surrounding rock is generated. This vector specifically includes multi-dimensional aggregated features such as geological strength index, rock mass integrity coefficient, microseismic damage factor, and forward anomaly probability, thus providing a unified and reliable input for subsequent intelligent classification and prediction of the surrounding rock.

[0054] Therefore, through the above specific implementation methods, this system achieves deep fusion of multi-source heterogeneous data at the feature level, effectively eliminating the uncertainty and information conflict of a single data source, generating a standardized feature vector that can comprehensively reflect the true state of the surrounding rock, and providing a solid data foundation for the subsequent objective and accurate identification of the surrounding rock level.

[0055] Example 4 Based on the above embodiments, this embodiment provides a detailed description of the specific implementation of step S103 above, which involves "inputting the surrounding rock digital twin feature vector into a pre-trained spatiotemporal feature extraction network and outputting the prediction result of the surrounding rock level of the current working face and the confidence level of the prediction result in real time".

[0056] In this embodiment, the spatiotemporal feature extraction network is a deep learning network based on a hybrid of graph convolution and gated recurrent units. The specific processing procedure is as follows: First, the input layer receives the surrounding rock digital twin feature vector output from the multi-source heterogeneous data feature layer aggregation module. This vector contains multi-dimensional aggregated features such as geological strength indicators, rock mass integrity coefficient, microseismic damage factor, and anomaly probability at the current station location. The system constructs this feature vector into graph structure data, where the nodes of the graph represent rock mass units at different spatial locations, and the feature of each node is the aggregated feature vector corresponding to that location. The edges of the graph are defined according to the connectivity of the joint network to characterize the mechanical connections and structural coupling between adjacent rock mass units.

[0057] Subsequently, the spatial feature extraction layer uses graph convolution operators to aggregate the neighborhood information of each node, and extracts the spatial correlation of the rock mass structure through convolution operations, thereby realizing the quantitative representation of the topological features of the joint network and outputting node features containing spatial structure information.

[0058] Next, the temporal feature extraction layer uses a gated recurrent unit network, taking the node feature sequence output by the spatial feature extraction layer as input, to capture the evolution of monitoring data over time, including the trend of deformation rate changes and the frequency evolution of microseismic events, and outputs latent state features that integrate spatiotemporal information.

[0059] Finally, the output layer maps the hidden state features output by the gated recurrent unit to the probability distribution of the surrounding rock at the current station number belonging to each level through a fully connected layer and a Softmax activation function. The level corresponding to the highest probability is used as the prediction result, and the prediction confidence is calculated based on the entropy value of this probability distribution. When this confidence falls below a preset threshold, the system automatically issues a prompt for encrypted monitoring or supplementary exploration to reduce decision-making risk. Furthermore, this spatiotemporal feature extraction network employs a transfer learning strategy during the model training phase. It first pre-trains using historical engineering data containing geological exploration, monitoring, and final surrounding rock identification results, and then fine-tunes the model using data revealed during initial excavation at specific engineering sites, ensuring the model's adaptability to specific geological conditions.

[0060] Therefore, this specific implementation constructs the digital twin feature vector of the surrounding rock into graph structure data and uses a hybrid network of graph convolution and gated recurrent units for spatiotemporal feature extraction. This achieves a comprehensive capture of the spatial topological features of the rock mass structure and the temporal evolution law of monitoring data, significantly improving the accuracy and robustness of surrounding rock level prediction. At the same time, the system's risk warning capability and engineering adaptability are enhanced through confidence output and transfer learning strategies.

[0061] Example 5 Based on the above embodiments, this embodiment provides a detailed description of the specific implementation of step S104, which involves "generating preliminary support parameters through deep reinforcement learning based on the predicted results of the surrounding rock level and the digital twin feature vector of the surrounding rock, and outputting the final support design parameters that meet the safety requirements after numerical simulation verification of the preliminary support parameters".

[0062] In this embodiment, firstly, an ontology-based support parameter knowledge base is constructed. This knowledge base stores support scheme cases of typical underground engineering projects under different surrounding rock levels, burial depths, spans, and groundwater conditions in the form of a knowledge graph. Each case records in detail parameters such as anchor bolt type, length, and spacing, shotcrete strength and thickness, steel arch frame model, spacing, and secondary lining reinforcement, and is associated with corresponding construction feedback and monitoring effect data.

[0063] Subsequently, a parameter optimization engine based on deep reinforcement learning is launched. Its state space is defined as the digital twin feature vector of the surrounding rock output by the multi-source heterogeneous data feature layer aggregation module, the deformation history data of the exposed tunnel section, and the current construction sequence information. The action space is designed as an adjustable combination of support parameters, using a discrete-hybrid action space to adapt to the adjustment needs of different parameter types. The reward function of this engine is a multi-objective weighted function, specifically including a safety reward determined based on the safety factor predicted by numerical simulation, an economic reward determined based on the cost of support materials, and a construction convenience reward determined based on expert rules. The agent outputs a preliminary support parameter adjustment scheme through interaction with the support knowledge base and trial-and-error learning in the simulation environment.

[0064] Subsequently, the preliminary scheme was sent to the numerical simulation verification unit. This unit integrates an automated calling interface for finite element or finite difference software, which can automatically establish a numerical mesh, assign material properties, apply boundary conditions, and solve based on the current face profile, preliminary support parameters, and aggregated rock mechanics parameters (including deformation modulus, cohesion, and internal friction angle). The verification content specifically includes the range of the plastic zone of the surrounding rock, the settlement of the arch and the horizontal convergence value, and the internal forces of the initial support structure. If the simulation results meet the preset safety standards, the scheme is output as the final dynamic support design parameters; if not, it returns to the deep reinforcement learning engine for re-optimization until a scheme that meets the safety requirements is generated, forming a two-layer decision-making mechanism of "reinforcement learning pre-decision plus numerical simulation verification".

[0065] Therefore, this specific implementation method, by constructing an ontology-based knowledge graph and a deep reinforcement learning optimization engine, and combining numerical simulation verification, realizes dynamic decision-making of support parameters from experience dependence to a combination of data-driven and mechanical verification, which significantly improves the safety and economy of the support scheme under complex geological conditions. At the same time, it balances the ease of construction through a multi-objective reward function, enhancing the engineering practicality of the system.

[0066] Example 6 Based on the above embodiments, this embodiment provides a detailed description of the steps after outputting the final support design parameters that meet safety requirements, namely, "collecting actual geological logging data and long-term deformation monitoring data after each excavation cycle is completed, comparing them with the previous prediction results, and constructing error samples; using an online learning algorithm to incrementally update the spatiotemporal feature extraction network to correct discrimination bias; adjusting the weights of relevant cases in the support parameter knowledge base according to the quality of the support effect, and incorporating the new cases implemented this time into the knowledge base."

[0067] In this embodiment, firstly, after the excavation and support of each section of the tunnel is completed, the system collects the actual geological logging data and long-term deformation monitoring data during the construction process of that section. The actual geological logging data includes detailed records by on-site technicians of the exposed rock mass structure, lithology, and weathering degree. The long-term deformation monitoring data comes from the final displacement and strain values ​​recorded by multi-point displacement gauges and fiber optic grating sensors embedded in the surrounding rock. The system compares these actual data item by item with the preliminary prediction results output by the surrounding rock intelligent grading prediction module in step S103, calculates the deviation between the predicted level and the actual exposed level, and the difference between the predicted deformation trend and the actual monitoring curve, thereby constructing an error sample containing multi-dimensional error information.

[0068] Subsequently, the model parameter adaptive adjustment module inputs the aforementioned error samples into the surrounding rock intelligent classification prediction module, and uses an online learning algorithm to incrementally update the spatiotemporal feature extraction network. Specifically, this online learning algorithm is based on the gradient descent principle and uses the deviation signal in the error samples to fine-tune the network weights of the graph convolutional layer and the gated recurrent unit layer, gradually correcting the model's discrimination bias of local rock mass features, so that the model can more accurately predict similar geological conditions in subsequent excavation cycles.

[0069] Meanwhile, based on the effectiveness of the support implementation, this module adjusts the weights of relevant cases in the support parameter knowledge base. If the monitoring data after the implementation of the support scheme shows that the surrounding rock deformation converges and the support structure is stable under stress, the retrieval weight of the corresponding case in the knowledge base is increased, so that it will be prioritized for retrieval by the deep reinforcement learning engine in similar subsequent working conditions; conversely, if the support effect is poor, its weight is reduced.

[0070] Furthermore, the system incorporates the complete information of the new case study, including its surrounding rock grade, support parameters, construction feedback, and monitoring results, into the support parameter knowledge base as a new record, thereby continuously enriching the case coverage of the knowledge base. Through the above steps, the error samples output from the previous sub-step directly serve as input for online learning in the subsequent sub-step, while the evaluation results of the support effect determine the direction of knowledge base weight adjustment, forming a complete closed loop from data collection and error analysis to model and knowledge base updates.

[0071] Therefore, this specific implementation method, by constructing error samples and using an online learning algorithm to incrementally update the spatiotemporal feature extraction network, and dynamically adjusting the weights of knowledge base cases, achieves continuous adaptive optimization of the system to local geological conditions. This significantly improves the prediction accuracy of the surrounding rock classification model during the engineering process and ensures the timeliness and relevance of support scheme retrieval, thereby effectively guaranteeing construction safety and engineering economy.

[0072] Example 7 This embodiment will describe in detail the complete implementation of the method for classifying and dynamically designing the surrounding rock of underground caverns. This method utilizes a rapid classification and dynamic design system for the surrounding rock of underground caverns based on deep fusion of multi-source data proposed in this application. This system includes: a multi-dimensional stereoscopic perception module, a multi-source heterogeneous data feature layer aggregation module, a surrounding rock intelligent classification and prediction module, a support parameter dynamic decision-making module, and a model parameter adaptive adjustment module. For example... Figure 2 The system's overall workflow is shown. Based on deep fusion of multi-source data, the system achieves rapid and accurate identification of surrounding rock quality and dynamic optimization of support parameters through the coordinated operation of five core links: multi-dimensional three-dimensional perception, aggregation of multi-source heterogeneous data feature layers, intelligent classification and prediction of surrounding rock, dynamic decision-making of support parameters, and adaptive adjustment of model parameters.

[0073] Specifically, during the data acquisition phase, the multi-dimensional stereoscopic sensing module is responsible for covering multi-source heterogeneous data throughout the entire exploration, excavation, and support cycle. This module includes a geological exploration data unit, used to integrate regional geological maps, borehole columnar sections, core test data, and borehole television imaging data to obtain the initial geostress field and basic rock mass quality indicators. The core test data includes uniaxial compressive strength, elastic modulus, and Poisson's ratio. This module also deploys a high-density IoT monitoring network, employing low-power wide-area network technology to deploy wireless intelligent sensors, including embedded microseismic monitoring sensors, multi-point displacement gauges and multi-point borehole strain gauges, fiber Bragg grating sensors, and anchor bolt axial force gauges and contact pressure cells. Microseismic monitoring sensors are used to capture microseismic events generated by rock mass fracturing in real time, locate the fracturing source, and calculate the magnitude and energy release rate. Multi-point displacement gauges and multi-point borehole strain gauges are used to monitor the radial displacement and strain evolution at different depths within the surrounding rock. Fiber optic grating sensors are deployed along the tunnel axis to monitor the continuous strain and temperature field changes of the surrounding rock. Anchor bolt axial force gauges and contact pressure cells are used to monitor the stress state of the initial support structure in real time. In addition, the full-section digital scanning unit utilizes a 3D laser scanner and industrial camera installed on excavating equipment or a dedicated mobile platform to quickly acquire dense point cloud and high-definition texture images of the tunnel face and tunnel walls after each muck removal. Computer vision algorithms are used to automatically extract rock mass structural surface information, including the number of joint groups, attitude, spacing, opening degree, roughness, and filling condition, generating a structural surface network in the digital twin model. The advanced geological prediction data interface is used to standardize the interpretation results of advanced prediction systems such as seismic reflection wave method, ground-penetrating radar method, and induced polarization method, including the location, scale, and type of abnormal geological bodies ahead. Preferably, the sensor network can employ self-powered technologies, such as thermoelectric power generation or vibration energy harvesting, to reduce deployment and maintenance costs.

[0074] During the data fusion phase, the multi-source heterogeneous data feature layer aggregation module adopts a hierarchical aggregation architecture to address the significant differences in dimensions, precision, and spatiotemporal resolution among the aforementioned source data. For example... Figure 3The workflow of this module, as shown, first involves data layer alignment, mapping all data to a spatiotemporal coordinate system centered on the tunnel axis station number. Image point cloud data is assigned precise coordinates using real-time positioning and mapping technology. Sensor monitoring data is spatiotemporally interpolated based on deployment locations to ensure a one-to-one correspondence with geological units. Subsequently, feature layer aggregation is performed, employing Dempster-Shafer evidence theory combined with an adaptive weighting algorithm to transform various data types into basic probability assignments for different evidence bodies. Specifically, microseismic activity characteristics such as microseismic event density, magnitude-frequency relationship values, and energy indices serve as evidence of the degree of surrounding rock damage; deformation characteristics such as deformation rate, acceleration, and spatiotemporal distribution patterns serve as evidence of surrounding rock stability; rock mass structural characteristics such as joint density, volumetric joint number, and integrity coefficient serve as evidence of rock mass integrity; and geophysical anomaly characteristics such as wave velocity anomalies and resistivity anomalies serve as evidence of geological changes ahead. By using the DS synthesis rule, the orthogonal sum of each piece of evidence is calculated to obtain the integrated confidence function after fusion, thereby generating a digital twin feature vector of the surrounding rock that can comprehensively reflect the state of the surrounding rock. This vector contains multi-dimensional aggregated features such as the geological strength index, rock mass integrity index, microseismic damage index, and the probability of anomalies ahead at the current station location.

[0075] In the surrounding rock classification prediction stage, the core of the intelligent surrounding rock classification prediction module is a spatiotemporal feature extraction network based on a hybrid of graph convolution and gated recurrent units. For example... Figure 4 The workflow of this module is shown. The input layer of the network constructs a graph structure data by fusing feature vectors. The nodes of the graph represent rock mass units at different spatial locations. The node features include aggregated rock mass parameters. The edges of the graph are defined according to the connectivity of the joint network, representing the mechanical relationships between units. The spatial feature extraction layer uses a graph convolution operator to aggregate the neighborhood information of each node, extracting the spatial correlation of the rock mass structure and realizing the quantitative representation of the topological features of the joint network. The temporal feature extraction layer uses a gated recurrent unit network to capture the evolution of monitoring data over time, including the trend of deformation rate changes and the frequency evolution of microseismic events. The output layer outputs the probability distribution of the surrounding rock at the current working face and within a set range ahead, based on a fully connected layer and a Softmax activation function. The level corresponding to the highest probability is used as the prediction result. At the same time, the prediction confidence is calculated based on the entropy value of the probability distribution. The range ahead is set according to engineering needs. During the model training phase, a transfer learning strategy is employed. Pre-training is first performed using historical engineering data, including geological survey, monitoring, and final surrounding rock identification results. Then, fine-tuning is conducted on-site using data revealed during initial excavation to ensure the model's adaptability to specific geological conditions. As a preferred feature, this module also includes uncertainty quantification functionality. When the output confidence level falls below a preset threshold, the system automatically marks areas requiring attention on the construction drawings and suggests increased monitoring or supplementary exploration for verification.

[0076] During the dynamic decision-making phase of support parameters, the dynamic decision-making process is triggered when the predicted results of the surrounding rock level change, or when real-time monitoring data indicates that the deformation of the surrounding rock exceeds a preset threshold. For example... Figure 5 The workflow of the dynamic decision-making module for support parameters, as shown, begins with a support parameter knowledge base. This knowledge base is an ontology-based knowledge graph that stores support scheme examples for typical underground engineering projects under different surrounding rock levels, burial depths, spans, and groundwater conditions. This includes anchor bolt types, lengths, and spacing; shotcrete strength and thickness; steel arch frame models and spacing; and secondary lining reinforcement. It also links the construction feedback and monitoring results for each example. Subsequently, a parameter optimization engine based on deep reinforcement learning begins operation. Its state space is defined as the current digital twin feature vector of the surrounding rock, the deformation history of the exposed tunnel section, and the current construction sequence. The action space is defined as adjustable combinations of support parameters, employing a discrete-hybrid action space design. The reward function is designed as a multi-objective weighted function, including a safety reward determined based on the safety factor predicted by numerical simulation, an economic reward determined based on the cost of support materials, and a construction convenience reward determined based on expert rules. By introducing a deep reinforcement learning algorithm, the agent learns decision-making strategies through interaction with the support knowledge base and trial and error in a simulated environment, outputting a preliminary support parameter adjustment scheme. Finally, to ensure the mechanical rationality of the scheme, the numerical simulation verification unit integrates automated calling interfaces of finite element or finite difference software, forming a two-layer decision-making mechanism of "reinforcement learning pre-decision plus numerical simulation verification". Based on the current face profile, preliminary support parameters, and aggregated rock mechanics parameters, it automatically establishes a numerical mesh, assigns material properties, applies boundary conditions, and solves the problem. The rock mechanics parameters include deformation modulus, cohesion, and internal friction angle. The verification content includes the range of the plastic zone of the surrounding rock, the settlement of the arch and the horizontal convergence value, and the internal forces of the initial support structure. If the simulation results do not meet the preset safety standards, the system returns to the reinforcement learning engine for re-optimization until the final dynamic support design parameters that meet the safety requirements are generated. Preferably, this module supports a human-computer interaction mode. The final scheme output is executed after confirmation by on-site technicians. Technicians can also manually adjust the parameters and input reasons, and the system uses this adjustment as a new sample for learning.

[0077] During the adaptive optimization phase, the model parameter adaptive adjustment module implements continuous optimization based on construction feedback. After the excavation and support of each tunnel section is completed, the final deformation monitoring data and actual exposed geological conditions of that section are collected and compared with the previous prediction results to construct error samples. On the one hand, the error samples are input into the surrounding rock intelligent classification module, and an online learning algorithm is used to incrementally update the spatiotemporal feature extraction network, gradually correcting the model's discrimination bias. On the other hand, based on the support implementation effect, the case weights in the support parameter knowledge base are adjusted so that effective solutions appear first in subsequent searches, and new cases implemented this time are included in the knowledge base. Through the above mechanism, the system can learn from each excavation cycle, gradually improving the accuracy of the surrounding rock classification model in judging local rock masses as the project progresses, and making the dynamic adjustment of support schemes more precise, thereby effectively ensuring construction safety and project economy.

[0078] To implement the above embodiments, this application also proposes an electronic device, including: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to execute the dynamic design method for grading and support of surrounding rock of underground caverns based on multi-source data fusion as described in any of the first aspect embodiments above.

[0079] To implement the above embodiments, this application also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method for dynamic design of surrounding rock classification and support of underground caverns based on multi-source data fusion as described in any one of the first aspect embodiments above.

[0080] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0081] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0082] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0083] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0084] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0085] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0086] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0087] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for classifying and dynamically designing the surrounding rock and support of underground caverns based on multi-source data fusion, characterized in that, Includes the following steps: Collect multi-source heterogeneous data during the exploration and construction phases of underground caverns. The multi-source heterogeneous data includes geological exploration data, real-time monitoring data of surrounding rock microseismic activity and deformation, digital image data of full-section rock mass structure, and advanced geological prediction data. The multi-source heterogeneous data are aggregated at the feature layer, and evidence theory is used to eliminate information conflicts and uncertainties to generate standardized digital twin feature vectors of surrounding rock. The surrounding rock digital twin feature vector is input into a pre-trained spatiotemporal feature extraction network, which outputs the prediction result of the surrounding rock level of the current working face and the confidence level of the prediction result in real time. Based on the predicted rock level and the digital twin feature vector of the surrounding rock, preliminary support parameters are generated through deep reinforcement learning. After numerical simulation verification, the preliminary support parameters are output as final support design parameters that meet safety requirements.

2. The method according to claim 1, characterized in that, The multi-source heterogeneous data collected during the exploration and construction phases of underground caverns include: The physical and mechanical parameters and regional geostress field information of the borehole cores are obtained through geological exploration data units. Real-time data on the spatiotemporal evolution of microfractures, internal displacement, continuous strain field, and stress of the support structure of the surrounding rock are collected through a high-density Internet of Things monitoring network. After each blast, the dense point cloud and texture images of the tunnel face and tunnel wall are quickly acquired by the full-section digital scanning unit, and the geometric parameters of the rock mass structure are automatically extracted by computer vision algorithms. The interpretation results of advanced geological prediction systems using the standardized interface of advanced geological prediction data are integrated with the seismic reflection wave method, ground-penetrating radar method, or induced polarization method.

3. The method according to claim 1, characterized in that, The process of aggregating the multi-source heterogeneous data at the feature layer, using evidence theory to eliminate information conflicts and uncertainties, and generating standardized digital twin feature vectors for surrounding rock includes: All data are uniformly mapped to a spatiotemporal coordinate system centered on the tunnel axis station number, and data layer alignment is performed; The data are converted into basic probability allocations for different types of evidence. Among them, rock mass integrity evidence is based on digital scanning and borehole data, rock mass damage evidence is based on microseismic monitoring parameters, and geological anomaly evidence is based on advanced prediction interpretation. Using the Dempster-Shafer evidence theory synthesis rules, the orthogonal sum of each piece of evidence is calculated to obtain the comprehensive confidence function, and a digital twin feature vector of the surrounding rock containing geological strength index, rock mass integrity coefficient, microseismic damage factor and anomaly probability is generated.

4. The method according to claim 1, characterized in that, The spatiotemporal feature extraction network is a hybrid network based on graph convolution and gated recurrent units. The real-time output of the surrounding rock grade prediction results and their confidence levels for the current working face includes: The digital twin feature vector of the surrounding rock is constructed as graph structure data, where graph nodes represent rock mass units at different spatial locations, and graph edges are defined according to the connectivity of the joint network; The spatial correlation of rock mass structure is extracted by aggregating the neighborhood information of each node through graph convolution operators. The evolution of monitoring data over time is captured through a gated cyclic unit network; By using a fully connected layer and a Softmax activation function, the probability distribution of the surrounding rock at the current station number belonging to each level is output, and the prediction confidence is calculated based on the entropy value of the probability distribution.

5. The method according to claim 1, characterized in that, The process of generating preliminary support parameters through deep reinforcement learning, and then outputting final support design parameters that meet safety requirements after numerical simulation verification of these preliminary support parameters, includes: A knowledge base for support parameters is constructed based on ontology-based knowledge graphs, storing support scheme cases for typical underground projects under different surrounding rock levels, burial depths, spans, and groundwater conditions, as well as their construction feedback and monitoring effects. Using the current digital twin feature vector of the surrounding rock and historical deformation data as the state, the adjustable support parameter combination as the action, and the multi-objective weighted function of safety, economy and construction convenience as the reward, the algorithm is trained and optimized through deep reinforcement learning to output a preliminary support parameter adjustment scheme. A numerical model is automatically established based on the preliminary support parameters and the aggregated rock mechanics parameters to simulate the mechanical response and verify the safety of the scheme. If the requirements are met, the final dynamic support design parameters are output; otherwise, the algorithm is returned to the deep reinforcement learning algorithm for re-optimization.

6. The method according to claim 1, characterized in that, After the output of the final support design parameters that meet safety requirements, the following is also included: Collect actual geological logging data and long-term deformation monitoring data after each excavation cycle is completed, compare them with the previous prediction results, and construct an error sample; An online learning algorithm is used to incrementally update the spatiotemporal feature extraction network to correct discrimination bias. Based on the effectiveness of the support, the weights of relevant cases in the support parameter knowledge base are adjusted, and new cases implemented this time are included in the knowledge base.

7. The method according to claim 1, characterized in that, Before inputting the digital twin feature vector of the surrounding rock into the pre-trained spatiotemporal feature extraction network, the method further includes: The spatiotemporal feature extraction network was pre-trained using historical engineering data, and then fine-tuned using initial excavation data at the actual engineering site. After the real-time output of the predicted rock grade of the current working face and the confidence level of the predicted results, the following is also included: When the confidence level of the prediction result is lower than a preset threshold, an automatic prompt instruction for encrypted monitoring or supplementary exploration is issued.

8. The method according to claim 1, characterized in that, The reward function of the deep reinforcement learning is a multi-objective weighted function, which includes: a safety reward determined based on the safety factor predicted by numerical simulation, an economic reward determined based on the cost of support materials, and a construction convenience reward determined based on expert rules. The numerical simulation verification includes the following: the range of the plastic zone of the surrounding rock, the settlement and horizontal convergence value of the arch crown, and the internal forces of the initial support structure.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the dynamic design method for the classification and support of underground cavern surrounding rock based on multi-source data fusion as described in any one of claims 1-8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for dynamic design of surrounding rock classification and support of underground caverns based on multi-source data fusion as described in any one of claims 1-8.

Images