High-energy geological environment surrounding rock classification and decision-making method based on digital twinning and multi-source feedback

Abstract

This invention belongs to the field of intelligent construction of tunnels and underground engineering and information technology of geotechnical engineering. It discloses a method for classifying and deciding on surrounding rock in high-energy geological environments based on digital twins and multi-source feedback, solving the problems caused by the difficulty of existing technologies in dynamically perceiving the state of surrounding rock, fusion classification of multi-source data, and closed-loop linkage of construction decisions in high-energy geological environments. First, this invention constructs a three-dimensional geological-structural digital twin of the tunnel based on initial exploration data. During construction, IoT technology is used to collect multi-source data such as geology, construction disturbance, and surrounding rock response in real time, and maps them into the digital twin to achieve synchronous updates between the virtual and real data. Subsequently, a deep learning-parameter inversion hybrid model is constructed based on the multi-source fused data, outputting the Dynamic Surrounding Rock Classification Index (DRCI) and key mechanical parameters, which are then input into a multi-objective optimization module to provide an adaptive drilling and blasting scheme. Finally, the model is corrected through construction feedback results to achieve closed-loop self-learning.

Description

Technical Field

[0001] This invention belongs to the field of intelligent construction of tunnels and underground engineering and information technology of geotechnical engineering, specifically involving a high-energy geological environment surrounding rock classification and decision-making method based on digital twins and multi-source feedback. Background Technology

[0002] As infrastructure construction, including transportation, water conservancy, and energy, expands into deeper underground spaces, the number of projects involving deep-buried long tunnels, high-temperature tunnels, water-rich fractured zones, and areas with frequent mining and seismic activity is increasing. The geological environment of the surrounding rock in these projects is gradually evolving from "conventional" to "high-energy," forming typical high-energy geological environments. These environments simultaneously possess the core characteristics of high ground stress, high ground temperature, high osmotic pressure, and strong disturbances (frequent blasting impacts and mechanical vibrations). Under these conditions, the surrounding rock often exhibits strong nonlinearity, strong anisotropy, and strong spatiotemporal effects. Traditional methods for classifying and designing surrounding rock based on static exploration data are no longer sufficient to meet engineering requirements.

[0003] In high-energy geological environments, the surrounding rock is prone to stress concentration, localized fracturing, rapid opening of structural surfaces, or shear slippage under the combined effects of excavation unloading, blasting disturbance, and groundwater-stress coupling. This can lead to severe over- and under-excavation, large deformation of the surrounding rock, instability of the surrounding rock-support system, and excessive blasting vibration, seriously affecting construction safety and schedule control. Therefore, how to achieve real-time, dynamic, and visual perception and classification of the surrounding rock condition in high-energy environments, and adaptively optimize drilling and blasting schemes accordingly, has become an important scientific and engineering problem in the field of tunnel and underground engineering.

[0004] The classic surrounding rock classification methods widely used in current engineering, such as the Q system, RMR system, and BQS, are mostly based on initial exploration data and limited field information, and have the following shortcomings:

[0005] (1) It is highly static and lacks the ability to evolve over time.

[0006] Traditional classification methods typically determine the surrounding rock grade once before construction or in the early stages of excavation, and only make minor corrections when obvious anomalies are found. This method cannot reflect the dynamic evolution of the surrounding rock under continuous blasting disturbance, unloading relaxation, and support effects, and it is slow to respond to the rapid deterioration or abrupt changes in the properties of the surrounding rock in high-energy environments.

[0007] (2) Data from multiple sources is fragmented and information is not fully utilized.

[0008] The construction process generates a large amount of heterogeneous data, including drilling parameters (thrust, torque, and advance rate), blasting parameters (hole pattern and charge structure), and monitoring data (surrounding rock displacement, convergence, blasting vibration, over-excavation and under-excavation laser scanning, microseismic events, etc.). However, these data are often collected and stored independently by different systems, making it difficult to align and integrate them in time and space on a unified platform, resulting in a large amount of potential information being wasted.

[0009] (3) Disconnect between surrounding rock classification and construction decision making

[0010] Traditional surrounding rock classification results are only used as a reference in the design stage. There is no automatic correlation mechanism between them and blasting parameter design, charge structure optimization, and support parameter adjustment. Surrounding rock classification cannot directly drive construction decision optimization and still relies heavily on the experience judgment of on-site technicians, resulting in insufficient objectivity and stability of decision-making.

[0011] (4) Ignoring high-energy characteristics and spatiotemporal coupling effects

[0012] Existing methods typically assume that the geothermal field, water pressure field, and stress field are approximately stable over a short period of time, making it difficult to describe the coupled effects of heat, permeability, force, and disturbance in the surrounding rock in high-energy geological environments. For example, key influencing factors such as transient temperature rise caused by blasting, microcrack opening, pore water pressure fluctuations, and stress redistribution are difficult to incorporate into traditional static classification systems.

[0013] In recent years, some studies have attempted to introduce information-based monitoring and intelligent algorithms into surrounding rock classification and risk assessment. For example, laser scanning is used to analyze tunnel outlines and over- or under-excavation, and machine learning is used to cluster and identify monitoring data. However, these solutions still have some shortcomings: they mostly rely on a single data source (such as laser scanning images or displacement monitoring data) and lack coordination between construction disturbance data and geological advance prediction data; although machine learning algorithms are introduced, they are not deeply coupled with three-dimensional geological models or digital twin models, making it difficult to achieve accurate spatial projection and three-dimensional analysis of prediction results; and they mostly remain at the level of auxiliary evaluation or offline analysis, and have not yet formed a closed-loop control system of "perception-classification-decision-feedback".

[0014] In summary, existing technologies cannot simultaneously meet the integrated needs of multi-source dynamic perception, digital twin coupled modeling, deep learning-inversion joint classification, and multi-objective optimization decision-making in high-energy geological environments. There is an urgent need for a solution that can adapt to the complexity of high-energy environments and achieve intelligent linkage between dynamic perception of surrounding rock conditions and construction decision-making. Summary of the Invention

[0015] The technical problem to be solved by this invention is to provide a method for classifying and making decisions about surrounding rocks in high-energy geological environments based on digital twins and multi-source feedback, so as to solve the problems caused by the difficulty of existing technologies in achieving dynamic perception of surrounding rock status, multi-source data fusion classification and closed-loop linkage of construction decisions in high-energy geological environments.

[0016] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0017] A high-energy geological environment surrounding rock classification and decision-making method based on digital twins and multi-source feedback includes the following steps:

[0018] S1. Construct a digital twin of the tunnel and surrounding rock under a high-energy geological environment, wherein the digital twin integrates a three-dimensional geological model, tunnel geometric information and high-energy environmental field variables;

[0019] S2. Real-time collection of geological and exploration data, construction disturbance data, and surrounding rock response data, which are then mapped to the digital twin after spatiotemporal alignment;

[0020] S3. Preprocess and extract features from the collected multi-source data, and integrate them to form a feature vector;

[0021] S4. Input the feature vector into the deep learning-inversion hybrid model, calculate the Dynamic Surrounding Rock Classification Index (DRCI), invert the key mechanical parameters of the surrounding rock, and update the digital twin.

[0022] S5. Based on the Dynamic Surrounding Rock Classification Index (DRCI), determine the surrounding rock category and the corresponding constraint threshold. Based on the key mechanical parameters of the surrounding rock, quantify the correlation between drilling and blasting parameters and over- and under-excavation, surrounding rock damage, and blasting vibration through numerical simulation of digital twins. Construct and solve a multi-objective optimization model. Solve the multi-objective optimization model through a multi-objective optimization algorithm to obtain the optimal drilling and blasting scheme that adapts to the current surrounding rock condition.

[0023] S6. After blasting is carried out according to the optimal drilling and blasting scheme, the actual construction feedback data is collected, and the error index is calculated by comparing it with the prediction results output by the digital twin through numerical simulation. The error index is then used to update the parameters of the deep learning-inversion hybrid model.

[0024] Furthermore, in step S1, the methods for constructing a tunnel-surrounding rock digital twin under high-energy geological conditions include:

[0025] A three-dimensional geological model of the tunnel route was constructed based on the survey data;

[0026] The designed tunnel geometry information is embedded into a three-dimensional geological model to form an initial digital twin integrating the tunnel and surrounding rock;

[0027] High-energy environmental field variables are preset in the initial digital twin of the tunnel-surrounding rock integration, and the spatial distribution of high-energy characteristic regions is recorded to obtain the tunnel-surrounding rock digital twin.

[0028] Furthermore, in step S2, the geological and exploration data includes: geological images of the working face, information on anomalies identified by the advanced geological prediction system, and data on the size and morphology of the slag blocks.

[0029] The construction disturbance data includes drilling parameters such as drilling rig advance speed, thrust, and torque, as well as blasting parameters such as blasting hole spacing, hole depth, charge amount, and detonation network structure.

[0030] The surrounding rock response data includes: surrounding rock displacement monitoring data, blasting vibration parameters, and magnitude, energy, and location coordinates of microseismic monitoring events.

[0031] Furthermore, in step S2, mapping the data to the digital twin after spatiotemporal alignment includes:

[0032] During the data acquisition process, construction mileage station numbers, timestamps, and three-dimensional coordinates inside the tunnel are added to each type of data. Spatiotemporal alignment of multi-source data is achieved through time axis standardization and spatial coordinate matching technology. Subsequently, the aligned data is mapped to the digital twin in real time.

[0033] Furthermore, in step S3, the preprocessing includes: standardizing the time axis of data with different sampling frequencies and achieving time alignment through linear or spline interpolation; projecting the point monitoring data onto the tunnel axis and surrounding rock units and performing spatial interpolation through Kriging or inverse distance weighting methods.

[0034] Furthermore, in step S3, the feature extraction includes: extracting texture features, fracture orientation, and block size distribution from image data; extracting average propulsion resistance and fluctuation coefficient from drilling parameters; and extracting energy spectrum, dominant frequency, and duration from blasting vibration parameters and microseismic monitoring event data.

[0035] Furthermore, in step S4, the deep learning-inversion hybrid model includes a CNN-LSTM deep learning model and a PSO-NN inversion model;

[0036] The CNN-LSTM deep learning model takes feature vectors as input and outputs the Dynamic Rock Classification Index (DRCI) and rock category level corresponding to the current excavation cycle. Its training objective function is:

[0037] ;

[0038] Where N is the number of training samples; and These are the predicted value and the actual value of the k-th sample, respectively; and These represent the predicted and actual rock categories for the k-th sample, respectively. The cross-entropy loss function; These are the weighting coefficients;

[0039] The objective function of the PSO-NN inversion model is:

[0040] ;

[0041] in, The objective function value; This is the set of constitutive parameters of the surrounding rock to be inverted. For cohesion, It is the internal friction angle. For deformation modulus, Poisson's ratio; Let be the measured response vector for the k-th operating condition; For parameters The predicted response vector is obtained through numerical simulation using a digital twin.

[0042] Furthermore, in step S4, the calculation formula for the Dynamic Surrounding Rock Classification Index (DRCI) is as follows:

[0043] ;

[0044] in, The surrounding rock stability index; The explosiveness index of the surrounding rock; It is a dynamic risk index; , and Let be the weight coefficient, and satisfy... ;

[0045] ;

[0046] Among them, U c U represents the convergence of the surrounding rock; c,lim This represents the limit of convergence allowed by the corresponding surrounding rock grade or specification; U v U represents the amount of settlement of the vault. v,lim The allowable ultimate settlement of the vault; N s For the internal forces of the support structure; N s,lim The allowable internal forces for the support structure design; α1, α2, and α3 are weighting coefficients, satisfying α1 + α2 + α3 = 1;

[0047] ;

[0048] Among them, F p F represents the average drilling thrust or equivalent drilling resistance. p,ref For reference thrust value; d 50 The median particle size of the slag; d 50,ref For target or reference particle size; v p V represents the peak value of the blast vibration velocity. p , ref The reference values ​​are for allowable or design vibrations; β1, β2, and β3 are weighting coefficients that satisfy β1 + β2 + β3 = 1.

[0049] ;

[0050] Among them, E mThe cumulative energy of microseismic events per unit time or unit distance; E m,ref n is the reference energy threshold. m The frequency of microseismic events; n m , ref H is the reference frequency threshold. e These are high-energy geological environmental factors; γ1, γ2, and γ3 are weighting coefficients, satisfying γ1 + γ2 + γ3 = 1;

[0051] ;

[0052] Where, σ H ρ is the maximum principal stress; T is the surrounding rock temperature; p is the pore water pressure; D is the construction disturbance intensity index. , , , These are the corresponding reference values ​​for the maximum principal stress, surrounding rock temperature, pore water pressure, and construction disturbance intensity index, respectively; δ1, δ2, δ3, and δ4 are weighting coefficients, satisfying δ1+δ2+δ3+δ4=1.

[0053] Furthermore, in step S5, the constraint thresholds include a minimum threshold for surrounding rock stability and a permissible threshold for blasting vibration;

[0054] The objective function of the multi-objective optimization model is:

[0055] ;

[0056] The constraints include:

[0057] Surrounding rock stability constraints: ;

[0058] Vibration control constraints: ;

[0059] Engineering constraints related to safe distances and propellant loading limits;

[0060] in, Let be the drill and blast parameter vector. ,in, The hole spacing is... For row spacing, For the depth of the hole, This refers to the amount of explosives loaded. This represents the minimum threshold for surrounding rock stability. The maximum vibration velocity predicted by numerical simulation using a digital twin under drilling and blasting parameters x; This refers to the permissible threshold for blasting vibration. This could be due to over- or under-excavation or profile deviation. This is an indicator of the surrounding rock damage or fragmentation induced by blasting. This refers to the velocity or energy index of blasting vibration. Unit cycle cost.

[0061] Furthermore, in step S6, the calculation method of the error index includes:

[0062] ;

[0063] in, For error indicators; and The over- and under-excavation volumes are respectively measured and predicted by numerical simulation using a digital twin model; and The test results and numerical simulations using digital twin phantoms are used to predict surrounding rock damage indices, respectively. and These are respectively measured values ​​and numerical simulations using digital twin phantoms to predict microseismic energy or vibration energy; , and Let be the weight coefficient, and satisfy... + + =1.

[0064] The beneficial effects of this invention are:

[0065] (1) Adapt to the complexity of high-energy geological environments and improve the real-time performance and accuracy of surrounding rock condition perception:

[0066] The tunnel-surrounding rock digital twin constructed by this invention integrates high-energy environmental field variables such as high ground stress, high ground temperature, high osmotic pressure and strong disturbance, realizing unified modeling and virtual-real mapping of geological-structural-environmental fields, and can accurately characterize the strong nonlinear and strong anisotropic characteristics of surrounding rock under multi-field coupling.

[0067] Relying on a multi-source dynamic feedback mechanism, heterogeneous data such as geological advance prediction, construction disturbance, and surrounding rock response are spatiotemporally aligned and deeply fused. Combined with the CNN-LSTM deep learning model and the DRCI dynamic classification index, real-time dynamic updates of surrounding rock categories and states are achieved, which can improve the response speed to the deterioration of surrounding rock properties and sudden disasters, and provide reliable support for risk early warning.

[0068] (2) Achieve deep coupling between classification and decision-making to improve the adaptability and scientific nature of construction schemes:

[0069] This invention uses a deep learning-inversion hybrid model to simultaneously output the DRCI index and key mechanical parameters of the surrounding rock, and directly drives multi-objective optimization decision-making. This allows the surrounding rock classification results to directly trigger the update of drilling and blasting parameters, changing the situation in traditional methods where classification results are disconnected from construction decisions and rely on human experience.

[0070] Based on multi-objective optimization algorithms such as NSGA-II, Pareto optimal solutions are sought among multiple objectives such as fragmentation control, disturbance limitation, cost optimization, and safety assurance. The generated drilling and blasting scheme can accurately match the current surrounding rock condition, effectively reduce over- and under-excavation, control excessive blasting vibration and surrounding rock damage, reduce support and repair costs and construction delays, and achieve a balance between construction efficiency and safety risks.

[0071] (3) Construct a closed-loop self-learning system to enhance the system's long-term adaptability and evolutionary capability:

[0072] This invention designs a closed-loop mechanism of perception-classification-decision-feedback-self-learning. By comparing and analyzing the measured data after construction with the prediction results of the digital twin, it calculates error indicators and reverse-corrects the parameters of the deep learning model and the inversion model. This enables the system to continuously optimize the prediction accuracy and decision rationality during long-term construction, achieving the evolutionary effect of "becoming smarter with construction".

[0073] (4) Improve the safety and economy of the project:

[0074] This invention significantly reduces the probability of disasters such as rock bursts, collapses, and water inrushes by accurately predicting the dynamic risks of surrounding rock and effectively controlling blasting vibrations and surrounding rock instability. It provides comprehensive protection for the safety of construction personnel and equipment and solves the industry problem of difficult-to-manage construction safety risks in high-energy geological environments.

[0075] By optimizing drilling and blasting parameters to reduce over- and under-excavation, lowering surrounding rock damage and support costs, and shortening construction delays, the economic benefits of the project have been significantly improved. At the same time, the digital and intelligent technology system has reduced reliance on the experience of high-end technical personnel and reduced the additional costs caused by human decision-making errors, providing technical support for the efficient construction of deep underground engineering projects. Attached Figure Description

[0076] Figure 1 This is a flowchart of the high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback in an embodiment of the present invention. Detailed Implementation

[0077] This invention aims to provide a method for classifying and making decisions about surrounding rock in high-energy geological environments based on digital twins and multi-source feedback. It addresses the challenges posed by existing technologies in achieving dynamic perception of surrounding rock conditions, multi-source data fusion classification, and closed-loop linkage for construction decisions in high-energy geological environments. The core idea is to use digital twins as the core carrier for virtual-real coupling, multi-source dynamic feedback as data support, and intelligent algorithms to achieve dynamic perception, accurate classification, and closed-loop optimization of surrounding rock conditions in high-energy geological environments, thereby constructing an integrated intelligent technology system of "perception-classification-decision-feedback-self-learning."

[0078] More specifically, the technical means employed by this invention to achieve the above-mentioned core ideas include:

[0079] To address the multi-field coupling characteristics of high-energy geological environments (high ground stress, high ground temperature, high osmotic pressure, and strong disturbance), a tunnel-surrounding rock digital twin integrating geological, structural, and environmental fields is constructed. This enables the virtual-real mapping and real-time updating of the surrounding rock's geometric morphology, mechanical properties, and environmental variables, providing a unified virtual platform for subsequent data fusion, simulation, and prediction.

[0080] By synchronously collecting heterogeneous data such as geological exploration, construction disturbance, and surrounding rock response through IoT technology, and forming a unified input through spatiotemporal alignment and feature extraction; an innovative Dynamic Surrounding Rock Classification Index (DRCI) is designed, which combines the CNN-LSTM deep learning model and the PSO-NN inversion model to achieve synchronous prediction and dynamic updating of surrounding rock categories and key mechanical parameters, thus solving the problem of accurate characterization of the rapid evolution of surrounding rock state under high-energy environments.

[0081] By deeply coupling dynamic classification results with multi-objective optimization algorithms, Pareto optimal drilling and blasting schemes are generated under the constraints of objectives such as safety, cost, and disturbance control. Through error feedback between post-construction measured data and digital twin prediction results, model parameters are corrected in reverse, realizing the system's self-learning evolution and ultimately achieving a continuous optimization closed loop of "data-driven model, model-guided decision-making, decision-driven optimization of practice, and practice-feedback model".

[0082] To facilitate understanding of the technical terms, some of the technical terms involved in the present invention are explained below:

[0083] 1. High-energy geological environment: refers to complex geological conditions that simultaneously possess characteristics of high ground stress, high ground temperature (high T), high osmotic pressure (high P), and strong disturbance (frequent blasting impacts and mechanical vibrations). Typical scenarios include: deeply buried long tunnels, high-temperature tunnels, water-rich fractured zones, and underground engineering environments with frequent earthquakes or mining seismic activity.

[0084] 2. Digital Twin: Constructing a digital model in virtual space that corresponds to the actual tunnel and surrounding rock, and using real-time acquired multi-source data to drive the integration of "virtual-real mapping, real-time updates, predictive simulation, and reverse optimization" into a unified technology system.

[0085] 3. Multi-source dynamic feedback: This refers to the synchronous collection and dynamic fusion of multiple types of heterogeneous data (geological data, construction disturbance data, surrounding rock response data, etc.) in time and space during the construction process, and the feedback results are used to influence the digital twin model and construction decision-making process, thereby forming a closed-loop control.

[0086] 4. Dynamic Rock-mass Classification Index (DRCI): A composite index that comprehensively reflects the stability, blastability, and dynamic risk level of surrounding rock. It is one of the core indicators used in this invention to characterize the comprehensive state of surrounding rock in high-energy geological environments.

[0087] 5. Deep Learning-Inversion Hybrid Model: Composed of Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), and Particle Swarm Optimization-Neural Network (PSO-NN), it achieves joint prediction and dynamic inversion of surrounding rock category and mechanical parameters by performing feature extraction and time series modeling on multi-source monitoring data.

[0088] 6. Multi-objective optimization and NSGA-II: NSGA-II (Non-dominated Sorting Genetic Algorithm II) is a multi-objective evolutionary algorithm based on fast non-dominated sorting and crowding distance. It is used to comprehensively optimize among multiple objectives such as fragmentation control, disturbance control, cost, and safety, and to give the Pareto optimal solution set of drilling and blasting parameters.

[0089] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0090] This embodiment provides a high-energy geological environment surrounding rock classification and decision-making method based on digital twins and multi-source feedback. The overall implementation idea is as follows: First, a three-dimensional geological-structural digital twin of the tunnel is constructed based on initial exploration data. During construction, multi-source data such as geology, construction disturbance, and surrounding rock response are collected in real time using IoT technology and mapped onto the digital twin to achieve synchronous updates between the virtual and real data. Then, a deep learning-parameter inversion hybrid model is constructed based on the multi-source fused data, outputting the Dynamic Surrounding Rock Classification Index (DRCI) and key mechanical parameters, which are then input into a multi-objective optimization module to provide an adaptive drilling and blasting scheme. Finally, the model is corrected through construction feedback results to achieve closed-loop self-learning.

[0091] See Figure 1 This embodiment specifically includes the following implementation process:

[0092] S1. Constructing a digital twin of tunnel and surrounding rock under high-energy geological conditions

[0093] The core of this step is to construct an integrated digital twin that includes geological, structural, and environmental fields, providing a unified spatiotemporal framework and basic input conditions for subsequent numerical simulations, and ensuring that the simulation results are highly consistent with the actual engineering scenario.

[0094] Specifically, in this embodiment, the method for constructing a tunnel-surrounding rock digital twin under a high-energy geological environment is as follows:

[0095] 1. Collect survey data:

[0096] Obtain complete exploration data for the target tunnel project, including geological profiles, borehole columnar sections, geophysical inversion results, geostress test reports, geothermal field monitoring data, and groundwater distribution survey data, ensuring coverage of core information such as lithology, joint structures, faults, weak interlayers, high geostress distribution, high geothermal areas, and high permeability ranges.

[0097] 2. Construct a three-dimensional geological model:

[0098] Based on the aforementioned exploration data, a three-dimensional geological model along the tunnel route was constructed using professional geological modeling software (such as Midas GTS and FLAC3D). The model needs to accurately reproduce geological features such as lithological stratification, fault location and strike, thickness and distribution of weak interlayers, and groundwater occurrence areas, providing realistic geological boundary conditions for numerical simulation. The three-dimensional geological model also includes the initial rock mass mechanical parameters assigned based on the exploration data, forming the initial mechanical parameter field of the digital twin.

[0099] 3. Integrate tunnel geometry information:

[0100] Key geometric parameters are extracted from the tunnel design documents, including tunnel centerline coordinates, slope, cross-section type, and cross-section dimensions. These parameters are then embedded into a three-dimensional geological model using spatial coordinate matching technology to form an initial digital twin integrating the tunnel and surrounding rock, thus clarifying the spatial range and geometric boundaries of the simulation.

[0101] 4. Preset high-energy environmental field variables:

[0102] In the initial digital twin, high-energy environmental field variables are preset based on the exploration data, specifically including: the geostress tensor field σ(x,y,z), which reflects the maximum principal stress, minimum principal stress and stress direction of the rock mass in different regions; the geothermal field T(x,y,z), which reflects the relationship between different burial depths and geothermal gradient distribution; and the pore water pressure field p(x,y,z), which reflects the pore water pressure in different regions.

[0103] 5. Mark high-energy feature regions:

[0104] In the initial digital twin, the spatial coordinate range of high-energy characteristic areas such as deep buried sections, geothermal anomaly sections, and water-rich fracture zones is clearly marked, which delineates the boundaries of the key calculation areas for subsequent numerical simulations.

[0105] Based on the above steps, a digital twin of the tunnel and surrounding rock under a high-energy geological environment is obtained.

[0106] S2. Real-time acquisition and spatiotemporal fusion of multi-source heterogeneous data

[0107] The core of this step is to achieve comprehensive acquisition, spatiotemporal alignment, and twin mapping of multi-source data, providing real-time updated input parameters for numerical simulation and ensuring that the simulation results are synchronized with the actual state of the surrounding rock.

[0108] Specifically, in this embodiment, the method of real-time acquisition and spatiotemporal fusion of multi-source heterogeneous data is as follows:

[0109] 1. Deploy a multi-source data acquisition system:

[0110] Based on the actual needs of the project, multi-dimensional data acquisition equipment was deployed at the tunnel construction site, specifically including:

[0111] Geological and exploration data acquisition equipment: High-definition camera at the working face, used to capture geological images of the working face; Advanced geological prediction system (TSP, ground-penetrating radar), used to identify anomalies such as faults, karst caves, and fracture zones ahead; Image acquisition device for slag conveyor belt, used to acquire data on the size and shape of slag blocks.

[0112] Construction disturbance data acquisition equipment: Torque sensors, thrust sensors, displacement sensors, and other equipment installed on the drilling rig to collect real-time data on the drilling speed. ,thrust Torque Drilling parameters; record the spacing between blast holes using a blasting parameter recorder. Row spacing Hole depth , charge amount Blasting parameters such as detonation network structure.

[0113] Surrounding rock response data acquisition equipment: Convergence meters and crown settlement monitoring instruments are deployed to collect data such as surrounding rock convergence deformation, crown settlement, and peripheral displacement; blasting vibration velocity is collected by installing blasting vibration sensors and microseismic monitoring instruments. acceleration And spectral characteristics, and magnitude of microseismic events. ,energy Location coordinates Data such as...

[0114] 2. Unified data collection and spatiotemporal alignment:

[0115] An IoT gateway was built to connect all data acquisition devices to a unified data platform. During the acquisition process, each type of data was marked with three tags: construction mileage station number, timestamp, and three-dimensional coordinates inside the tunnel. Then, through time axis standardization and spatial coordinate matching technology, spatiotemporal alignment of multi-source data was achieved, ensuring that data from the same time and spatial location can be correlated and analyzed.

[0116] 3. Data mapping to digital twins:

[0117] Multi-source data aligned in time and space are mapped to a digital twin in real time via an API interface, updating the twin's real-time status: geological image data is mapped to the virtual location of the tunnel face at the corresponding mileage; drilling parameters and blasting parameters are associated with the virtual equipment model of the corresponding construction process; displacement, vibration, and microseismic data are mapped to the three-dimensional coordinates of the corresponding monitoring points, providing real-time updated boundary conditions and response data for numerical simulation.

[0118] S3. Multi-source data preprocessing and feature extraction:

[0119] The core of this step is to optimize the collected raw data and extract core features, eliminate redundant information and dimensional differences, and form a unified feature vector to support subsequent intelligent model input and numerical simulation data optimization.

[0120] Specifically, the multi-source data preprocessing and feature extraction methods in this embodiment are as follows:

[0121] 1. Data preprocessing:

[0122] Outlier and missing value handling: Outlier data is identified using the 3σ criterion, and missing values ​​are supplemented by adjacent data interpolation.

[0123] Time alignment: Time axis standardization is performed on data with different sampling frequencies, and a unified time step is achieved through linear or spline interpolation. Data alignment below.

[0124] Spatial interpolation: The Kriging interpolation method is used to project discrete point monitoring data (such as displacement and vibration) onto the surrounding rock unit of the digital twin, realizing the spatial continuity of the data and providing continuous input parameters for numerical simulation.

[0125] 2. Feature extraction:

[0126] Image-based data (face images, slag removal images): CNN convolutional neural network was used to extract texture features, crack features, and block size distribution features.

[0127] Drilling parameter data: Extract features such as average propulsion resistance and propulsion speed fluctuation coefficient.

[0128] Vibration and microseismic data: Energy spectrum, dominant frequency, peak frequency, duration and other features are extracted through Fourier transform; additional features such as magnitude distribution, cumulative energy value and event density are extracted from microseismic data.

[0129] Displacement data: Extract features such as convergence rate, cumulative crown settlement, and displacement growth rate.

[0130] 3. Feature normalization and vector integration:

[0131] The Min-Max normalization method is used to eliminate the dimensional differences between different features, and finally all extracted features are integrated to form a feature vector with uniform dimensions. This serves as the input data for subsequent deep learning-inversion hybrid models.

[0132] S4. Dynamic Surrounding Rock Classification and Mechanical Parameter Inversion:

[0133] The core of this step is to construct a deep learning-inversion hybrid model, combine it with digital twin numerical simulation, to achieve dynamic classification of surrounding rock and accurate inversion of mechanical parameters, and use the inversion results to update the twin, providing accurate surrounding rock state data for subsequent optimization decisions.

[0134] Specifically, in this embodiment, the dynamic surrounding rock classification and mechanical parameter inversion are performed as follows:

[0135] 1. Calculation of Dynamic Surrounding Rock Classification Index (DRCI):

[0136] This embodiment integrates information on surrounding rock stability, blastability, and dynamic risk into a Dynamic Classification Index (DRCI), defined as follows:

[0137] ;

[0138] in, The surrounding rock stability index; The explosiveness index of the surrounding rock; It is a dynamic risk index; , and Let be the weight coefficient, and satisfy... .

[0139] ① Surrounding rock stability index :

[0140] The stability of the surrounding rock mainly reflects the deformation control capability of the surrounding rock-support system after excavation. Therefore, in this embodiment, convergence deformation, crown settlement, and support internal forces are selected as the core control quantities. A dimensionless stability index is constructed by normalization and weighted superposition. The calculation formula is as follows:

[0141] ;

[0142] Among them, U c U represents the convergence of the surrounding rock (mm). c,lim U represents the limit convergence (mm) allowed by the corresponding surrounding rock grade or specification. v U represents the crown settlement (mm); v,lim The allowable maximum crown settlement (mm); N s For internal forces in the support structure (such as anchor bolt axial force or steel arch frame internal force, kN); Ns,lim The allowable internal force (kN) for the support structure design; α1, α2, α3 are weighting coefficients, satisfying α1+α2+α3=1.

[0143] when A value approaching 1 indicates high surrounding rock stability and controlled deformation; when If the value approaches 0, it indicates that the surrounding rock is approaching an unstable state.

[0144] ② Rock blastability index :

[0145] The blastability of surrounding rock reflects the ease of rock mass fracturing and blasting response characteristics under drilling and blasting operations. This embodiment comprehensively considers drilling resistance (reflecting rock mass strength), slag size (reflecting fracturing effect), and blasting vibration response (reflecting energy utilization efficiency) to calculate the blastability index of surrounding rock. :

[0146] ;

[0147] Among them, F p F represents the average drilling thrust or equivalent drilling resistance (kN). p,ref For reference thrust value (which can be the average value of similar surrounding rock or the design value); d 50 The median particle size of the slag (mm); d 50,ref Target or reference particle size (mm); v p V represents the peak value of the blasting vibration velocity (cm / s or mm / s); p , ref The reference values ​​are for allowable or designed vibrations; β1, β2, and β3 are weighting coefficients that satisfy β1+β2+β3=1.

[0148] When I b The larger the value, the more "difficult" the surrounding rock is to blast (requiring higher energy); when I b The smaller size indicates that the surrounding rock is easily broken, resulting in higher blasting efficiency.

[0149] ③ Dynamic Risk Index :

[0150] Dynamic risk mainly stems from energy release, disturbance frequency, and high-energy geological environment. Therefore, this embodiment introduces microseismic energy, event frequency, and high-energy environmental factors to calculate the dynamic risk index. :

[0151] ;

[0152] Among them, E m The cumulative energy (J) of microseismic events per unit time or unit distance; E m,ref Reference energy threshold (J); nm is the frequency of microseismic events (number of times / cycle or number of times / meter); n m , ref is the reference frequency threshold; H e is the high-energy geological environment factor (dimensionless); γ1, γ2, γ3 are weight coefficients, satisfying γ1 + γ2 + γ3 = 1.

[0153] ;

[0154] Among them, σ H is the maximum principal stress (MPa); T is the surrounding rock temperature (°C); p is the pore water pressure (MPa); D is the construction disturbance intensity index (such as blasting specific charge or vibration energy); 、 、 、 are the corresponding reference values of the maximum principal stress, surrounding rock temperature, pore water pressure, and construction disturbance intensity index respectively; δ1, δ2, δ3, δ3 are weight coefficients, satisfying

[0155] δ1 + δ2 + δ3 + δ4 = 1.

[0156] The value range of DRCI is 0 < DRCI < 1. The smaller the DRCI value, the better the overall state of the surrounding rock, the higher the stability, and the lower the dynamic risk; conversely, the larger the DRCI value, the worse the state of the surrounding rock, and the higher the risk of instability and dynamic disasters. <00...​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​Lower; as DRCI increases, Gradually decrease, and and The contribution to DRCI gradually increases, reflecting the evolution of the surrounding rock from a stable state to a high-risk state.

[0162] 2. Training and classification of deep learning models:

[0163] The deep learning model adopts a CNN-LSTM combination model, where the CNN part includes convolutional layers and pooling layers to extract spatial features; the LSTM part includes hidden layers to capture temporal features; and finally, a fully connected layer is used to output the DRCI value and the probability of the surrounding rock category.

[0164] The unified feature vector formed in step S3 Using DRCI true values ​​and surrounding rock categories as inputs, training parameters are set, and the objective function used for training is:

[0165] ;

[0166] Where N is the number of training samples; and These are the predicted value and the actual value of the k-th sample, respectively; and These represent the predicted and actual rock categories for the k-th sample, respectively. Cross-entropy loss function is used to constrain the accuracy of surrounding rock category prediction. In other words, the smaller the CE, the closer the category probability given by the model is to the true category. is the weighting coefficient, used to balance the DRCI regression error and the category classification error.

[0167] When performing model inference, the feature vector of the current excavation cycle is input into the trained model, and the DRCI prediction value and the corresponding surrounding rock category level can be output.

[0168] 3. Mechanical parameter inversion:

[0169] Based on the monitored surrounding rock response and numerical simulation results in the digital twin, the cohesion was realized using the PSO-NN inversion model. internal friction angle Deformation modulus Poisson's ratio Inversion of equal mechanical parameters.

[0170] By inputting the combined mechanical parameters to be inverted into a digital twin, the surrounding rock response is simulated using numerical simulation software, and the predicted response vector is calculated. ; using the measured response vector With simulated response vector The deviation is the objective function:

[0171] ;

[0172] in, The objective function value; This is the set of constitutive parameters of the surrounding rock to be inverted. For cohesion, It is the internal friction angle. For deformation modulus, Poisson's ratio; Let be the measured response vector for the k-th operating condition; For parameters The predicted response vector is obtained through numerical simulation using a digital twin.

[0173] The optimal parameter space is searched using the Particle Swarm Optimization (PSO) algorithm. , so that the objective function Minimize and finally output the key mechanical parameters of the surrounding rock obtained by inversion.

[0174] 4. Digital twin updates:

[0175] The mechanical parameters obtained from the inversion The parameters are updated to the mechanical property library of the surrounding rock unit in the digital twin according to the spatial coordinates. The linear interpolation method is used to achieve smooth transition of parameters for adjacent surrounding rock units, ensuring the continuity of the mechanical field of the twin and providing accurate mechanical parameter input for the numerical simulation of subsequent drilling and blasting scheme optimization.

[0176] S5. Multi-objective optimization and decision output of drilling and blasting scheme:

[0177] The core of this step is to construct a multi-objective optimization model based on DRCI, inverted mechanical parameters, and digital twin numerical simulation, and solve for the optimal drilling and blasting scheme that is suitable for the current surrounding rock condition.

[0178] Specifically, in this embodiment, the multi-objective optimization and decision output of the drilling and blasting scheme are performed as follows:

[0179] 1. Determine the optimization constraints:

[0180] Based on the DRCI and surrounding rock category output in step S4, determine the constraint thresholds for multi-objective optimization:

[0181] Surrounding rock stability constraints: Set a minimum stability index threshold based on the surrounding rock type. ; needs to meet ;

[0182] Vibration control constraints: Based on the type of surrounding rock and the requirements of surrounding structures, set the allowable threshold for blasting vibration. ; needs to meet ;in, The maximum vibration velocity predicted by numerical simulation using a digital twin under drilling and blasting parameters x;

[0183] Engineering constraints: Engineering specifications require specific parameters such as the spacing between blasting holes, hole depth, charge per hole, and safety distance.

[0184] 2. Construct a multi-objective optimization model:

[0185] Design variables: based on the drill and blast parameter vector For design variables.

[0186] Objective function: Constructing a four-objective minimization model: ;

[0187] in, This could be due to over- or under-excavation or profile deviation. This is an indicator of the surrounding rock damage or fragmentation induced by blasting. This refers to the velocity or energy index of blasting vibration. Unit cycle cost (including drilling, explosives, support, etc.).

[0188] 3. Optimization algorithm solution:

[0189] By employing multi-objective optimization algorithms, such as the NSGA-II algorithm, to solve the model, a set of Pareto-optimal drill-blast schemes (i.e., drill-blast parameter vectors) corresponding to the current rock category can be obtained. Combining engineering preferences and safety requirements, the final implementation scheme is selected and executed by system or engineering calculation personnel.

[0190] S6. Closed-loop feedback and model self-learning update:

[0191] The core of this step is to calculate the error and update the model parameters by comparing the actual construction data with the results of the digital twin simulation, so as to realize the self-learning evolution of the system and ensure that the accuracy continues to improve during long-term construction.

[0192] Specifically, in this embodiment, the closed-loop feedback and model self-learning update methods are as follows:

[0193] 1. Collect construction feedback data:

[0194] After the drill-and-blast plan is implemented, the actual construction effect data will be collected using the following equipment:

[0195] Laser scanning equipment: scans the tunnel excavation outline and calculates the measured over- and under-excavation volume. ;

[0196] Rock damage detector: detects the depth and extent of rock damage after blasting and calculates the measured rock damage index. ;

[0197] Vibration monitoring instrument and microseismic monitoring instrument: collect actual blasting vibration energy and microseismic energy, and calculate the measured energy. ;

[0198] 2. Calculate the error index:

[0199] Retrieve the predicted data from the digital twin under the executed drill-and-blast parameters through numerical simulation: predicted over- and under-excavation volume. Predicted surrounding rock damage indices Predicted microseismic energy or seismic energy index .

[0200] Calculate the error index using the following error formula. :

[0201] ;

[0202] in, , and Let be the weight coefficient, and satisfy... + + =1.

[0203] 3. Model parameter update:

[0204] If the error index If the value exceeds the set threshold, it indicates a large error, and an update needs to be initiated.

[0205] CNN-LSTM model update: adjusting the learning rate, loss function weights, etc.

[0206] PSO-NN inversion model update: Adjust PSO inertia weights, learning factors, etc.

[0207] 4. Closed-loop iteration:

[0208] After the model parameters are updated, the next construction cycle begins, repeating steps S2-S5 to achieve continuous closed-loop iteration of "perception-classification-decision-feedback-self-learning," making the system more and more accurate with long-term construction.

[0209] Although embodiments of the present invention have been described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present invention, and all such changes and alterations shall not depart from the protection scope of the present invention.

Claims

1. A method for classifying and deciding on surrounding rocks in high-energy geological environments based on digital twins and multi-source feedback, characterized in that, Includes the following steps: S1. Construct a digital twin of the tunnel and surrounding rock under a high-energy geological environment, wherein the digital twin integrates a three-dimensional geological model, tunnel geometric information and high-energy environmental field variables; S2. Real-time collection of geological and exploration data, construction disturbance data, and surrounding rock response data, which are then mapped to the digital twin after spatiotemporal alignment; S3. Preprocess and extract features from the collected multi-source data, and integrate them to form a feature vector; S4. Input the feature vector into the deep learning-inversion hybrid model, calculate the Dynamic Surrounding Rock Classification Index (DRCI), invert the key mechanical parameters of the surrounding rock, and update the digital twin. The deep learning-inversion hybrid model includes a CNN-LSTM deep learning model and a PSO-NN inversion model; The CNN-LSTM deep learning model takes feature vectors as input and outputs the Dynamic Rock Classification Index (DRCI) and rock category level corresponding to the current excavation cycle. Its training objective function is: ； Where N is the number of training samples; and These are the predicted value and the actual value of the k-th sample, respectively; and These represent the predicted and actual rock categories for the k-th sample, respectively. The cross-entropy loss function; These are the weighting coefficients; The objective function of the PSO-NN inversion model is: ； in, The objective function value; This is the set of constitutive parameters of the surrounding rock to be inverted. For cohesion, It is the internal friction angle. For deformation modulus, Poisson's ratio; Let be the measured response vector for the k-th operating condition; For parameters The predicted response vector is obtained through numerical simulation using a digital twin. The formula for calculating the Dynamic Surrounding Rock Classification Index (DRCI) is as follows: ； in, The surrounding rock stability index; The explosiveness index of the surrounding rock; It is a dynamic risk index; , and Let be the weight coefficient, and satisfy... ; ； Among them, U c U represents the convergence of the surrounding rock; c,lim This represents the limit of convergence allowed by the corresponding surrounding rock grade or specification; U v U represents the amount of settlement of the vault. v,lim The allowable ultimate settlement of the vault; N s For the internal forces of the support structure; N s,lim The allowable internal forces for the support structure design; α1, α2, and α3 are weighting coefficients, satisfying α1 + α2 + α3 = 1; ； Among them, F p F represents the average drilling thrust or equivalent drilling resistance. p,ref For reference thrust value; d 50 The median particle size of the slag; d 50,ref For target or reference particle size; v p V represents the peak value of the blast vibration velocity. p , ref The reference values ​​are for allowable or design vibrations; β1, β2, and β3 are weighting coefficients that satisfy β1 + β2 + β3 = 1. ； Among them, E m The cumulative energy of microseismic events per unit time or unit distance; E m,ref n is the reference energy threshold. m The frequency of microseismic events; n m , ref H is the reference frequency threshold. e These are high-energy geological environmental factors; γ1, γ2, and γ3 are weighting coefficients, satisfying γ1 + γ2 + γ3 = 1; ； Where, σ H ρ is the maximum principal stress; T is the surrounding rock temperature; p is the pore water pressure; D is the construction disturbance intensity index. , , , These are the corresponding reference values ​​for the maximum principal stress, surrounding rock temperature, pore water pressure, and construction disturbance intensity index, respectively; δ1, δ2, δ3, and δ4 are weighting coefficients, satisfying δ1+δ2+δ3+δ4=1; S5. Based on the Dynamic Surrounding Rock Classification Index (DRCI), determine the surrounding rock category and the corresponding constraint threshold. Based on the key mechanical parameters of the surrounding rock, quantify the correlation between drilling and blasting parameters and over- and under-excavation, surrounding rock damage, and blasting vibration through numerical simulation of digital twins. Construct and solve a multi-objective optimization model. Solve the multi-objective optimization model through a multi-objective optimization algorithm to obtain the optimal drilling and blasting scheme that adapts to the current surrounding rock condition. S6. After blasting is carried out according to the optimal drilling and blasting scheme, the actual construction feedback data is collected, and the error index is calculated by comparing it with the prediction results output by the digital twin through numerical simulation. The error index is then used to update the parameters of the deep learning-inversion hybrid model.

2. The high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback as described in claim 1, characterized in that, In step S1, the methods for constructing a tunnel-surrounding rock digital twin under high-energy geological conditions include: A three-dimensional geological model of the tunnel route was constructed based on the survey data; The designed tunnel geometry information is embedded into a three-dimensional geological model to form an initial digital twin integrating the tunnel and surrounding rock; High-energy environmental field variables are preset in the initial digital twin of the tunnel-surrounding rock integration, and the spatial distribution of high-energy characteristic regions is recorded to obtain the tunnel-surrounding rock digital twin.

3. The high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback as described in claim 1, characterized in that, In step S2, the geological and exploration data includes: geological images of the working face, information on anomalies identified by the advanced geological prediction system, and data on the size and shape of the slag blocks. The construction disturbance data includes drilling parameters such as drilling rig advance speed, thrust, and torque, as well as blasting parameters such as blasting hole spacing, hole depth, charge amount, and detonation network structure. The surrounding rock response data includes: surrounding rock displacement monitoring data, blasting vibration parameters, and magnitude, energy, and location coordinates of microseismic monitoring events.

4. The high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback as described in claim 1, characterized in that, In step S2, mapping the data to the digital twin after spatiotemporal alignment includes: During the data acquisition process, construction mileage station numbers, timestamps, and three-dimensional coordinates inside the tunnel are added to each type of data. Spatiotemporal alignment of multi-source data is achieved through time axis standardization and spatial coordinate matching technology. Subsequently, the aligned data is mapped to the digital twin in real time.

5. The high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback as described in claim 1, characterized in that, In step S3, the preprocessing includes: standardizing the time axis of data with different sampling frequencies and achieving time alignment through linear or spline interpolation; projecting the point monitoring data onto the tunnel axis and surrounding rock units and performing spatial interpolation through Kriging or inverse distance weighting methods.

6. The high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback as described in claim 1, characterized in that, In step S3, the feature extraction includes: extracting texture features, fracture orientation and block size distribution from image data; extracting average propulsion resistance and fluctuation coefficient from drilling parameters; and extracting energy spectrum, dominant frequency and duration from blasting vibration parameters and microseismic monitoring event data.

7. The high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback as described in claim 1, characterized in that, In step S5, the constraint thresholds include the minimum threshold for surrounding rock stability and the allowable threshold for blasting vibration; The objective function of the multi-objective optimization model is: ； The constraints include: Surrounding rock stability constraints: ; Vibration control constraints: ; Engineering constraints related to safe distances and propellant loading limits; in, Let be the drill and blast parameter vector. , The hole spacing is... For row spacing, For the depth of the hole, This refers to the amount of explosives loaded. This represents the minimum threshold for surrounding rock stability. Drilling and blasting parameters The maximum vibration velocity predicted by numerical simulation using a digital twin; This refers to the permissible threshold for blasting vibration. This could be due to over- or under-excavation or profile deviation. This is an indicator of the surrounding rock damage or fragmentation induced by blasting. This refers to the velocity or energy index of blasting vibration. Unit cycle cost.

8. The high-energy geological environment surrounding rock classification and decision-making method based on digital twin and multi-source feedback as described in claim 7, characterized in that, In step S6, the error index is calculated as follows: ； in, For error indicators; and The over- and under-excavation volumes are respectively measured and predicted by numerical simulation using a digital twin model; and The test results and numerical simulations using digital twin phantoms are used to predict surrounding rock damage indices, respectively. and These are respectively measured values ​​and numerical simulations using digital twin phantoms to predict microseismic energy or vibration energy; , and Let be the weight coefficient, and satisfy... + + =1.

Images