Optimized construction method for horizontal circumferential drilling and blasting parameter model of spherical crown dome

Abstract

The invention discloses a spherical crown dome horizontal circumference drilling and blasting parameter model optimization construction method, and the method comprises the steps: collecting geological, crustal stress, hydrological and structural plane data, building a three-dimensional geological information model, carrying out the working condition partitioning based on a surrounding rock quality evaluation index, and constructing a parameter probability distribution model and an uncertainty boundary of each partition. Therefore, precise space description and parameter variability quantification of the spherical crown dome construction area are realized. The method comprises the following steps: establishing a three-dimensional numerical model of a cavern group, integrating blasting power, seepage and stress multi-physical field effects by adopting a finite element-discrete element coupling method, introducing a surrounding rock deterioration constitutive model considering cumulative damage and unloading aging, calibrating key parameters through an indoor test and field trial explosion, and establishing a three-dimensional model of the cavern group. And the numerical model can truly reflect the mechanical property degradation track of the surrounding rock under repeated blasting disturbance.

Description

Technical Field

[0001] This application relates to the field of civil engineering technology, and in particular to a method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome. Background Technology

[0002] The blasting excavation technology for underground cavern complexes in large-scale water conservancy projects is crucial to ensuring project safety and quality. Especially in the construction of ultra-large span underground powerhouses in deep-buried, high-stress environments, the dome structure, as the main load-bearing system of the main powerhouse's arch, faces multiple challenges during excavation, including complex spatial geometry, variable stress states of the surrounding rock, and extremely high precision requirements. The horizontal circular drill-and-blast method, as the mainstream technology for dome excavation, achieves surface shaping through layer-by-layer, circle-by-circle controlled blasting. The scientific configuration of its drilling and blasting parameters directly affects the degree of surrounding rock disturbance, the accuracy of over- and under-excavation control, and the reliability of subsequent support structures, representing a core technical bottleneck restricting the construction level of ultra-large span underground caverns.

[0003] The fundamental flaw in current drilling and blasting parameter optimization technology for dome arches lies in its neglect of the spatiotemporal evolution characteristics throughout the construction process. Existing static optimization models simplify the complex dynamic construction process into a single working condition analysis, failing to capture the trajectory of mechanical property degradation of the surrounding rock under continuous unloading and repeated blasting disturbances. Furthermore, the lack of organic coupling between the real-time monitoring system and the parameter optimization model results in a significant waste of information resources, as a large amount of field monitoring data is used only for post-construction verification rather than dynamic control. The lack of understanding of the parameter degradation patterns caused by differences in the stress paths of the surrounding rock at different excavation stages (initial excavation at the arch foot, widening of the waist, and closure of the arch crown) leads to frequent local failures when a unified parameter scheme is applied throughout the entire process. Summary of the Invention

[0004] The main objective of this application is to provide a method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this application provides the following technical solution: A method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome, the specific steps of which are as follows: S1. Collect geological, geostress, hydrological and structural surface data, establish a three-dimensional geological information model, divide the working conditions based on the surrounding rock quality evaluation index, and construct the probability distribution model of parameters and uncertainty boundary of each zone. S2. Establish a three-dimensional numerical model of the cavern group, adopt the finite element-discrete element coupling method, integrate the blasting dynamics, seepage and stress multi-physics field effects, introduce a constitutive model of surrounding rock deterioration considering cumulative damage and unloading aging, and calibrate key parameters through indoor tests and field test blasts. S3. Using the perforation layout, charge structure and detonation network as design variables, a multi-objective optimization mathematical model is established, which includes minimizing over-excavation and under-excavation, vibration control, forming quality optimization and cost constraints. A surrogate model and an improved genetic algorithm are used for global optimization to obtain the Pareto optimal solution set and form a partitioned baseline parameter scheme library. S4. Deploy laser scanning, vibration velocity sensing, displacement monitoring, acoustic emission and microseismic monitoring systems, establish a virtual model isomorphic to physical engineering, and realize real-time acquisition, transmission and feature extraction of monitoring data; S5. Establish a closed-loop mechanism of prediction-comparison-correction-re-optimization, adopt Bayesian update and Kalman filter to correct the surrounding rock parameters, use constrained optimization algorithm to perform small-step fast optimization, generate dynamic parameter adjustment scheme, and realize the coordinated optimization of drilling and blasting and support. S6. Conduct static and dynamic scheme comparison tests to evaluate the differences in forming quality, vibration control and advance efficiency, and use machine learning algorithms to adaptively calibrate vibration velocity, displacement and over- and under-excavation control thresholds. S7. Establish a knowledge graph linking geological zoning, parameter configuration, construction effects, and risk response; develop a case-based reasoning decision support system; generate emergency response templates for extreme working conditions; and compile standardized technical procedures.

[0006] Preferably, step S1 is performed in the following manner: S1.1 Collect multi-source geological survey data of the excavation area of ​​the dome, including borehole columnar section data, geological sketch logging data, and geophysical survey line data. Obtain physical and mechanical parameters of the rock mass through indoor rock mechanics tests, including uniaxial compressive strength, elastic modulus, Poisson's ratio, internal friction angle, cohesion, and density. Conduct in-situ stress testing using hydraulic fracturing and stress relief methods to obtain the three-dimensional principal stress field distribution. Measure the groundwater level, rock mass permeability coefficient, and unit inflow rate. Map the geometric parameters of the structural surfaces, including strike, dip, dip angle, spacing, ductility, opening, infill type, and roughness coefficient. Integrate the multi-source data based on three-dimensional geological modeling software to establish a three-dimensional geological information model. S1.2. Based on the RMR rock mass quality scoring system or the Norwegian Q system, the surrounding rock quality of each spatial unit of the dome is quantitatively evaluated. The evaluation indicators include rock strength, integrity index RQD, structural surface conditions, groundwater status, and in-situ stress ratio. Combining the geometric characteristics and stress state differences of the dome, the construction conditions are divided into zones, including the high stress zone at the arch foot, the transition zone at the arch waist, the unloading zone at the arch crown, the risk zone of the fracture zone, and the water-rich and weak zone. Rock mass parameter samples are collected for each zone. Statistical analysis methods are used to construct parameter probability distribution models. The mean, standard deviation, and confidence interval of the parameters are calculated through Monte Carlo simulation or Bayesian inference to form a prior probability knowledge base for the zone parameters.

[0007] Preferably, step S2 is performed as follows: S2.1 Based on the three-dimensional geological information model in step S1, construct an integrated numerical calculation model of the cavern group and the dome. The model space range is 3 to 5 times the characteristic size of the caverns beyond the excavation boundary. The finite element method is used to describe the elastoplastic behavior of the continuous medium, and the discrete element method is used to describe the discontinuous deformation controlled by the structural surface. A finite element-discrete element coupled framework is established. A 0.5-meter to 1-meter fine mesh is used at the stress concentration points of the arch foot and arch crown, and a 3-meter to 5-meter coarse mesh is used in the far field. Boundary conditions are set and the initial stress state is established. The multi-physics field coupled control equations of blasting dynamics, seepage, and stress are integrated. Rock mass parameters and structural surface parameters are assigned according to spatial partitions. S2.2 In the numerical model of step S2.1, the blasting cyclic cumulative damage constitutive model and the unloading aging creep constitutive model are introduced. The damage constitutive model establishes the correlation equation between damage variables and blasting vibration parameters. The creep constitutive model uses a combination of viscoelastic and plastic elements to establish the stress-strain-time correlation equation. The micro parameters are calibrated through indoor tests such as uniaxial compression, triaxial compression, cyclic loading and unloading, Hopkinson bar impact, and multi-stage creep. Field test blasts are carried out to deploy vibration velocity sensors and displacement monitoring networks. The inversion analysis method is used to iteratively optimize the macro parameters such as damping ratio, blasting load parameters, and structural surface mechanical parameters with the measured data as the objective function, and completes multi-scale parameter transformation and model verification.

[0008] Preferably, step S3 is performed as follows: S3.1 Establish a multi-objective optimization mathematical model for drilling and blasting parameters. Define design variables including borehole layout parameters, charge structure parameters, and detonation network parameters. Borehole parameters include ring spacing of 1.5 m to 3.0 m, hole spacing of 0.8 m to 2.0 m, hole diameter of 40 mm to 90 mm, and inclination angle of 60 degrees to 90 degrees. Charge parameters include single-hole charge amount, segmented charge position ratio of 0.3 to 0.7, and plugging length of 20% to 40% of hole depth. Detonation parameters include detonation sequence and delay time of 25 ms to 100 ms. Establish a multi-objective function set F(X). ={f1(X),f2(X),f3(X),f4(X)}, including minimizing over-excavation and under-excavation volume, minimizing peak vibration velocity, minimizing the standard deviation of profile flatness, and minimizing unit advance cost. Constraints are set, including safety constraints, geometric constraints, and process constraints. Based on the working condition zoning results in step S1.2, differentiated variable value boundaries are set to form the feasible region Ω_k of the zoning, where k is the working condition zoning number. The numerical model from step S2 is called, with input parameter combination X and output performance response value Y. A parameter-performance mapping database D={(X)} is established. i ,Y i )}; S3.2. Using Latin hypercube sampling, an initial sample point set is generated within the feasible region. The number of samples N = 10d to 20d, where d is the dimension of the design variable. For each sample point Xi, the numerical model from step S2 is called to calculate the objective function value Yi = F(Xi), forming a training dataset Dtrain. A surrogate model is constructed using a radial basis function neural network or a Gaussian process regression model. The parameters are fitted using the training dataset. The radial basis function neural network prediction formula is as follows: = Σwi·φ(||X-Xi||); The radial basis function is φ(r) = exp(-r² / (2σ²)), wi is the weight coefficient, and sigma is the kernel function bandwidth parameter. The prediction formula for the Gaussian process regression model is as follows: = μ(X) + k(X,Xtrain)·[K(Xtrain,Xtrain) + σn²·I] - ¹·(Ytrain - μ(Xtrain)); Where k is the covariance function, K is the covariance matrix, σn² is the observation noise variance, I is the identity matrix, and μ is the mean function; k-fold cross-validation was used to evaluate the prediction accuracy of the surrogate model, and the coefficient of determination and root mean square error were calculated. The formula for calculating the coefficient of determination is: R² = 1 - SSres / SStot = 1 - Σ(Yval,i - val,i)² / Σ(Yval,i - val)²; Where SSres is the residual sum of squares and SStot is the total sum of squares. val is the mean of the true values ​​in the validation set; The formula for calculating the root mean square error is:

[0009] Where nval is the number of validation samples, and the accuracy thresholds are set to R²≥0.90 and RMSE≤10%. When the accuracy does not meet the threshold, adaptive sampling is used to supplement training samples and update the surrogate model. Based on the surrogate model, the improved non-dominated sorting genetic algorithm NSGA-Ⅲ is used for multi-objective optimization. The population size is set to 100 to 200, the crossover probability is 0.8 to 0.9, the mutation probability is 0.05 to 0.1, and the maximum number of generations is 200 to 500. The population is initialized, the objective function value of each individual is calculated, and selection, crossover, and mutation genetic operations are performed. The parent and offspring populations are merged to form a mixed population. Non-dominated sorting is used to divide the mixed population into multiple non-dominated layers. The crowding distance is calculated for individuals within the same non-dominated layer. The formula for calculating the crowding distance is as follows: CD(Xi) = Σk[fk(Xi+1) - fk(Xi-1)] / (fk,max - fk,min); Where fk is the kth objective function value, fk,max and fk,min are the maximum and minimum values ​​of the objective function in the current population, respectively, and Xi+1 and Xi-1 are the neighboring individuals of the i-th individual after sorting; Individuals are selected to form the next generation of the population using an elite retention strategy that prioritizes non-dominated hierarchies and those with greater crowding distances. This process is iterated until the convergence criterion is met or the maximum number of generations is reached. The first non-dominated hierarchy is extracted as the Pareto optimal solution set P*. Based on the working condition partitioning in step S1.2, surrogate model construction and multi-objective optimization processes are performed for the high-stress stable zone at the arch foot, the stress transition zone at the arch waist, the unloading sensitive zone at the arch crown, the fault fracture zone risk zone, and the water-rich soft interlayer zone, respectively. The Pareto optimal solution set Pk corresponding to each working condition partition is obtained. A partition baseline parameter scheme library Library={Pk} is established, indexed by the working condition type number k, and stores the set of optimization parameters, including perforation layout parameters, charge structure parameters, detonation network parameters, and expected performance indicators.

[0010] Preferably, step S4 is performed as follows: S4.1. Based on the working condition zoning results in step S1.2, formulate a monitoring network scheme, deploy a three-dimensional laser scanner with a measurement accuracy of ±2 mm and a scanning cycle of 1 to 2 times per cycle, deploy triaxial velocity sensors with a measurement point spacing of 5 to 10 meters and a sampling frequency of not less than 5000 Hz, deploy multi-point displacement gauges with a measurement point spacing of 3 to 5 meters and a monitoring accuracy of 0.01 mm, deploy acoustic emission sensors with a spacing of 10 to 15 meters and a working frequency of 50 kHz to 400 kHz, deploy micro-vibration sensors with a spacing of 30 to 50 meters and a positioning accuracy of not less than 5 meters, establish a data acquisition and transmission system with an end-to-end delay of no more than 1 second, use a time-series database for storage and establish an index; S4.2. Based on the numerical model in step S2, construct a digital twin virtual model containing geometric, physical, and behavioral layers. Use the iterative nearest point algorithm or normal distribution transformation algorithm to register point clouds and update geometric boundaries. Establish a parameter inversion objective function and use gradient descent, genetic algorithm, or particle swarm optimization algorithm to solve for the optimal parameters of rock mass elastic modulus, Poisson's ratio, cohesion, and internal friction angle to update the physical layer. Update the behavioral layer according to the actual borehole location, charge amount, detonation sequence, and delay time. Preprocess and extract features from the monitoring data, including over- and under-excavation volume, half-hole ratio, profile standard deviation, peak velocity, dominant frequency, vibration duration, cumulative displacement, displacement rate, acoustic emission event rate, acoustic emission b-value, microseismic magnitude, source coordinates, and apparent volumetric strain. Use a timestamp alignment method to establish a multi-source data fusion mechanism. Set the model to be updated after each excavation cycle is completed. Establish a three-dimensional visualization platform to render the virtual model and monitoring data distribution cloud map in real time.

[0011] Preferably, step S5 is performed as follows: S5.1 Establish a prediction-comparison-correction closed-loop mechanism. Before each construction cycle, the blasting effect is predicted based on the digital twin model in step S4.2, outputting the predicted over- and under-excavation volume, peak vibration velocity, displacement field, and damage zone. After construction, the measured data is obtained from step S4.1, and the root mean square error of profile deviation RMSEcontour, relative vibration velocity error εPPV, and displacement error RMSEdisp are calculated. The state vector of surrounding rock parameters θ={E,ν,c,φ,K,σt,ρ} is defined. According to the prior distribution P(θ) in step S1.2, the likelihood function P(Dobs|θ) is established. The Metropolis-Hastings algorithm is used for MCMC sampling, or the extended Kalman filter EKF is used to recursively update and calculate the posterior mean θupdated, or the unscented Kalman filter UKF is used to update the parameters through sigma point propagation. θupdated is input into the physical model layer in step S4.2 to complete the correction. Three parameter update mechanisms are established: error triggering, periodic triggering, and state triggering. S5.2 Establish an online dynamic re-optimization module. Based on the updated parameters in step S5.1 and the engineering state characteristics in step S4.2, define design variables including hole mesh parameters, charge parameters, and detonation parameters. Establish multi-objective functions including minimizing over- and under-excavation, vibration control, forming quality, and cost constraints. Set safety constraints, geometric constraints, process constraints, and schedule constraints. Using the baseline parameters in step S3.2 as the initial solution, within a disturbance range of ±10% to 20%, use sequential quadratic programming, interior point method, particle swarm optimization, and differential evolution algorithm for optimization. Call the surrogate model in step S3.2 to accelerate evaluation, output optimized parameters to generate construction instructions, determine the surrounding rock stability based on the damage characteristics in step S4.2, and trigger support coordination adjustment when the acoustic emission event rate or micro-vibration frequency exceeds the threshold. Expand design variables to include drilling and blasting parameters and support parameters including anchor length, spacing, and concrete thickness. Determine the support timing and strength based on the damage depth and surrounding rock grade, establish an optimization history database, and use machine learning to train a fast decision-making model.

[0012] Preferably, step S6 is performed as follows: S6.1 Select typical working conditions and set the test section length for 3 to 5 excavation cycles. Implement the static optimization scheme (using step S3.2 to keep the baseline parameters fixed) and the dynamic adaptive optimization scheme (using step S5 to adjust the parameters in real time via a closed-loop mechanism). Collect construction effect data synchronously through the monitoring network in step S4.1. Establish forming quality evaluation indicators, including over-excavation and under-excavation rate, profile deviation standard deviation, half-hole rate, and forming compliance rate. Establish vibration control evaluation indicators, including peak vibration velocity compliance rate, vibration energy, and dominant frequency distribution. Establish advance efficiency evaluation indicators, including cycle time, monthly advance speed, and explosive consumption. Statistically compare the indicators of the two schemes using t-test and Mann-Whitney U test. Use the analytic hierarchy process (AHP) with weighted coefficients of forming quality 0.4, vibration control 0.3, and advance efficiency 0.3 for weighted comprehensive scoring and cost-benefit analysis. S6.2. Based on the experimental data from step S6.1 and the historical data from steps S4 and S5, a vibration velocity threshold calibration mechanism is established. Support Vector Regression (SVR) is used to build a prediction model, and Random Forest is used to build a vibration velocity-damage correlation model. The calibration threshold is determined through Bayesian optimization. A displacement threshold calibration mechanism is established. Long Short-Term Memory (LSTM) network is used to build a time-series prediction model, and Logistic Regression is used to build a displacement-instability classification model. The threshold is determined through ROC curves to keep the instability risk within 5%. An over- or under-excavation threshold calibration mechanism is established. Gaussian Process Regression (GPR) is used to build a prediction model, and K-means clustering is used to identify over- or under-excavation patterns. Pareto optimal search is used to balance control and efficiency. A dynamic threshold update mechanism is established, updating every 10 to 20 cycles. Q-learning or Deep Q-Network (DQN) is used to optimize the decision rules. The state space is defined as surrounding rock parameters and monitoring characteristics, the action space as parameter adjustment amplitude, and the reward function as performance improvement and cost weighting. The calibration threshold and decision rules are updated to step S5.

[0013] Preferably, step S7 is performed as follows: S7.1 Integrate the data from steps S1 to S6 to construct a four-dimensional relational knowledge graph. Define the ontology model, including an entity type layer containing geological zoning, parameter configuration, construction effect, and risk response entities; an attribute type layer containing numerical, categorical, temporal, and spatial attributes; and a relationship type layer containing association, causal, and triggering relationships. Use Apriori and FP-Growth algorithms to extract association rules, setting a minimum support of 0.3 and a minimum confidence of 0.7. Use structural equation modeling (SEM) to establish causal relationships and temporal association analysis to establish triggering relationships. Use the Neo4j graph database for storage. Establish a case-based reasoning (CBR) system, including modules for case representation, retrieval, reuse, correction, and retention. Case retrieval uses weighted similarity to return Top-K similar cases, with K equal to 3 to 5. Case reuse uses parameter interpolation or rule correction methods. Establish a decision confidence assessment mechanism with a threshold of 0.6. S7.2. Based on the knowledge graph in step S7.1 and the emergency plan database in step S5.2, extreme working condition types are sorted out, and a two-level classification system is defined. The first-level classification includes geological disaster type, construction accident type, equipment failure type, and environmental factor type. Geological disaster type is further subdivided into sudden water inrush, fault fracture zone, rock burst, and gas outburst. Emergency response templates are established for each type of extreme working condition. The template structure includes six modules: working condition identification characteristics, risk level determination, emergency response process, technical measures plan, resource allocation plan, and restoration construction conditions. The risk level is divided according to the water inrush volume: general water inrush of 50 to 100 cubic meters per hour, large water inrush of 100 to 500 cubic meters per hour, and major water inrush of more than 500 cubic meters per hour. Technical measures include grouting and water plugging, dewatering and drainage, and advanced pre-reinforcement. An emergency response decision tree is established to automatically match emergency templates based on monitoring data. An emergency drill simulation system is established based on the digital twin model in step S4.2.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. By constructing a data-driven closed-loop system, the problems of time-varying surrounding rock conditions and parameter adaptability were solved, realizing the transformation from static design to dynamic control throughout the entire process. This effectively addressed the problem of insufficient capture of the mechanical property degradation trajectory of surrounding rock under continuous unloading and repeated blasting disturbances. At the same time, it solved the problems of disconnection between monitoring data and parameter optimization models, as well as the local failure of drilling and blasting parameters caused by differences in the stress path of surrounding rock at different excavation stages.

[0015] 2. By collecting geological, geostress, hydrological, and structural surface data, a three-dimensional geological information model is established. Based on the surrounding rock quality evaluation index, working condition zones are divided, and probability distribution models and uncertainty boundaries of parameters for each zone are constructed, thereby achieving accurate spatial description and quantification of parameter variability in the construction area of ​​the dome. A three-dimensional numerical model of the cavern group is established, and the finite element-discrete element coupling method is used to integrate blasting dynamics, seepage, and stress multi-physics field effects. At the same time, a constitutive model of surrounding rock deterioration considering cumulative damage and unloading aging is introduced. Through indoor tests and field test blasts, key parameters are calibrated, enabling the numerical model to truly reflect the mechanical property degradation trajectory of the surrounding rock under repeated blasting disturbances. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the method steps of this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0018] The terms "first," "second," and "third" in this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" 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. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0019] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] Example 1: Please refer to Figure 1 A method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome, the specific steps of which are as follows: S1. Collect geological, geostress, hydrological and structural surface data, establish a three-dimensional geological information model, divide the working conditions based on the surrounding rock quality evaluation index, and construct the probability distribution model of parameters and uncertainty boundary of each zone. S2. Establish a three-dimensional numerical model of the cavern group, adopt the finite element-discrete element coupling method, integrate the blasting dynamics, seepage and stress multi-physics field effects, introduce a constitutive model of surrounding rock deterioration considering cumulative damage and unloading aging, and calibrate key parameters through indoor tests and field test blasts. S3. Using the perforation layout, charge structure and detonation network as design variables, a multi-objective optimization mathematical model is established, which includes minimizing over-excavation and under-excavation, vibration control, forming quality optimization and cost constraints. A surrogate model and an improved genetic algorithm are used for global optimization to obtain the Pareto optimal solution set and form a partitioned baseline parameter scheme library. S4. Deploy laser scanning, vibration velocity sensing, displacement monitoring, acoustic emission and microseismic monitoring systems, establish a virtual model isomorphic to physical engineering, and realize real-time acquisition, transmission and feature extraction of monitoring data; S5. Establish a closed-loop mechanism of prediction-comparison-correction-re-optimization, adopt Bayesian update and Kalman filter to correct the surrounding rock parameters, use constrained optimization algorithm to perform small-step fast optimization, generate dynamic parameter adjustment scheme, and realize the coordinated optimization of drilling and blasting and support. S6. Conduct static and dynamic scheme comparison tests to evaluate the differences in forming quality, vibration control and advance efficiency, and use machine learning algorithms to adaptively calibrate vibration velocity, displacement and over- and under-excavation control thresholds. S7. Establish a knowledge graph linking geological zoning, parameter configuration, construction effects, and risk response; develop a case-based reasoning decision support system; generate emergency response templates for extreme working conditions; and compile standardized technical procedures.

[0021] In this embodiment, the problem of time-varying surrounding rock conditions and parameter adaptability is solved by constructing a data-driven closed-loop system, realizing the transformation from static design to dynamic control throughout the entire process. This effectively addresses the problem of insufficient capture of the mechanical property degradation trajectory of surrounding rock under continuous unloading and repeated blasting disturbances. At the same time, it solves the problem of disconnect between monitoring data and parameter optimization models, as well as the problem of local failure of drilling and blasting parameters caused by differences in the stress path of surrounding rock at different excavation stages.

[0022] By collecting geological, geostress, hydrological, and structural surface data, a three-dimensional geological information model was established. Based on the surrounding rock quality evaluation index, working condition zones were divided, and probability distribution models and uncertainty boundaries of parameters for each zone were constructed, thereby achieving accurate spatial description and quantification of parameter variability in the construction area of ​​the dome. A three-dimensional numerical model of the cavern group was established, and the finite element-discrete element coupling method was used to integrate blasting dynamics, seepage, and stress multi-physics field effects. At the same time, a constitutive model of surrounding rock deterioration considering cumulative damage and unloading aging was introduced. Through indoor tests and field test blasts, key parameters were calibrated, enabling the numerical model to truly reflect the trajectory of mechanical property degradation of the surrounding rock under repeated blasting disturbances.

[0023] Based on this, a multi-objective optimization mathematical model was established, using the perforation layout, charge structure, and detonation network as design variables. This model incorporates minimizing over- and under-excavation, vibration control, forming quality optimization, and cost constraints. A surrogate model and an improved genetic algorithm were used for global optimization to obtain the Pareto optimal solution set and form a regional baseline parameter scheme library. This provides targeted initial parameter configurations for different working conditions. Simultaneously, laser scanning, vibration velocity sensing, displacement monitoring, acoustic emission, and microseismic monitoring systems were deployed. A virtual model isomorphic to the physical engineering was established to achieve real-time acquisition, transmission, and feature extraction of monitoring data, ensuring that changes in the surrounding rock condition during construction can be detected promptly.

[0024] A closed-loop mechanism of prediction-comparison-correction-re-optimization is established. Bayesian updating or Kalman filtering is used to correct surrounding rock parameters, and constrained optimization algorithms are employed for small-step rapid optimization to generate dynamically adjusted parameter schemes, achieving coordinated optimization of drilling and blasting with support. Specifically, static and dynamic scheme comparison tests are conducted to evaluate differences in forming quality, vibration control, and advance efficiency. Machine learning algorithms are used to adaptively calibrate vibration velocity, displacement, and over- and under-excavation control thresholds, thereby improving the accuracy and adaptability of parameter adjustments.

[0025] By establishing a knowledge graph linking geological zoning, parameter configuration, construction effects, and risk response, a case-based reasoning decision support system was developed. This system generates emergency response templates for extreme conditions and compiles standardized technical procedures, systematically storing and solidifying construction experience and optimization results. The entire technical solution, through multi-stage collaborative operation, deeply integrates geological data, numerical simulation, real-time monitoring, and dynamic optimization. This effectively addresses the time-varying nature of surrounding rock conditions and construction uncertainties, ensuring dynamic adaptation between drilling and blasting parameters and the actual response of the surrounding rock. This solves problems such as insufficient capture of the degradation trajectory of surrounding rock mechanical properties, disconnect between monitoring data and parameter optimization models, and localized failure of drilling and blasting parameters at different excavation stages.

[0026] Three-dimensional geological information models are a digital representation of integrated multi-source geological data. They can be implemented through geostatistical methods, spatial interpolation algorithms, or rule-based modeling techniques, such as Kriging interpolation and inverse distance weighted interpolation. Their main purpose is to achieve an accurate description of the spatial distribution characteristics of complex geological conditions.

[0027] Working condition zoning can be achieved through cluster analysis methods, such as K-means clustering and hierarchical clustering. The purpose is to divide the construction area according to the differences in surrounding rock characteristics in order to achieve differentiated parameter configuration.

[0028] The finite element-discrete element coupled method is a numerical calculation method used to simulate the interaction between continuous and discontinuous media. It can be solved by explicit or implicit time integration algorithms, such as the central difference method and the Newmark method. It is mainly used to handle complex mechanical behavior simulation problems.

[0029] Constitutive models of surrounding rock deterioration due to cumulative damage and unloading aging can be established through empirical formulas or theoretical derivations, such as damage evolution equations based on Weibull distribution or creep models based on viscoelastic theory.

[0030] The design variables for multi-objective optimization mathematical models can be generated using experimental design methods to create initial sample point sets, such as full factorial designs and partial factorial designs. The aim is to provide diverse parameter combinations for the optimization process. Surrogate models can be constructed using various machine learning algorithms, such as support vector machines and artificial neural networks, primarily to improve the computational efficiency of the optimization process.

[0031] The deployment of the monitoring system can achieve data transmission through wired or wireless communication technologies, such as fiber optic communication and Zigbee wireless sensor networks, with the aim of ensuring the real-time performance and reliability of the monitoring data. Parameter updates in the prediction-comparison-correction-reoptimization closed-loop mechanism can be implemented using various filtering algorithms, such as extended Kalman filtering and unscented Kalman filtering, primarily used to reflect the dynamic changes in the surrounding rock condition.

[0032] Example 2: Please refer to Figure 1 The specific method for step S1 is as follows: S1.1 Collect multi-source geological survey data of the excavation area of ​​the dome, including borehole columnar section data, geological sketch logging data, and geophysical survey line data. Obtain physical and mechanical parameters of the rock mass through indoor rock mechanics tests, including uniaxial compressive strength, elastic modulus, Poisson's ratio, internal friction angle, cohesion, and density. Conduct in-situ stress testing using hydraulic fracturing and stress relief methods to obtain the three-dimensional principal stress field distribution. Measure the groundwater level, rock mass permeability coefficient, and unit inflow rate. Map the geometric parameters of the structural surfaces, including strike, dip, dip angle, spacing, ductility, opening, infill type, and roughness coefficient. Integrate the multi-source data based on three-dimensional geological modeling software to establish a three-dimensional geological information model. S1.2. Based on the RMR rock mass quality scoring system or the Norwegian Q system, the surrounding rock quality of each spatial unit of the dome is quantitatively evaluated. The evaluation indicators include rock strength, integrity index RQD, structural surface conditions, groundwater status, and in-situ stress ratio. Combining the geometric characteristics and stress state differences of the dome, the construction conditions are divided into zones, including the high stress zone at the arch foot, the transition zone at the arch waist, the unloading zone at the arch crown, the risk zone of the fracture zone, and the water-rich and weak zone. Rock mass parameter samples are collected for each zone. Statistical analysis methods are used to construct parameter probability distribution models. The mean, standard deviation, and confidence interval of the parameters are calculated through Monte Carlo simulation or Bayesian inference to form a prior probability knowledge base for the zone parameters.

[0033] In this embodiment: By collecting and integrating multi-source geological exploration data, a three-dimensional geological information model capable of accurately reflecting the spatial heterogeneity of the dome's curved surface structure was established, avoiding the limitations of a single data source. Based on this, physical and mechanical parameters of the rock mass were obtained through indoor rock mechanics tests, and combined with geostress distribution data obtained by hydraulic fracturing or stress relief methods, accurate quantification of the inherent properties of the surrounding rock was achieved. Simultaneously, by mapping the geometric parameters of the structural surfaces, the controlling effect of the structural surfaces on excavation stability was revealed, avoiding model distortion caused by neglecting structural surfaces. Furthermore, quantitative evaluation of the surrounding rock quality was conducted using the RMR or Norwegian Q system, and construction condition zoning was performed based on the geometric characteristics of the dome and differences in stress state, ensuring accurate matching between the zoning results and the actual construction process, solving the adaptability problem of the unified parameter scheme in areas of sudden stress change. Finally, by constructing a parameter probability distribution model and calculating the parameter mean, standard deviation, and confidence interval, the uncertainty boundary of the surrounding rock parameters was quantified, forming a prior probability knowledge base for zoning parameters, providing reliable support for subsequent dynamic optimization.

[0034] The above technical solution effectively solves the problem of local failure of drilling and blasting parameters caused by rough zoning and parameter uncertainty, realizes the refined characterization of geological conditions in the excavation area of ​​the spherical dome and the quantitative management of the uncertainty of surrounding rock parameters, and provides a scientific basis for subsequent optimization of drilling and blasting parameters.

[0035] Multi-source geological exploration data refers to a collection of data obtained from different sources that can reflect geological conditions. It may include borehole columnar section data, geological sketch logging data, and geophysical survey line data, etc., with the aim of comprehensively characterizing the geological characteristics of the dome region.

[0036] Rock mass physical and mechanical parameters are key indicators used to describe the mechanical behavior of rock masses, such as uniaxial compressive strength and elastic modulus. These parameters can be obtained through indoor rock mechanics tests to ensure their accuracy and reliability.

[0037] Hydraulic fracturing, or stress relief method, is a technique used to obtain the distribution of in-situ stress. It directly acquires the three-dimensional principal stress field distribution through field testing, thus providing basic data for subsequent analysis. Structural plane geometric parameters refer to parameters describing the morphology and distribution of structural planes within the rock mass, such as strike, dip, and dip angle. These parameters can be obtained through surveying techniques and, combined with the geometric characteristics of the dome, reveal the controlling role of structural planes in excavation stability.

[0038] The RMR rock mass quality rating system, or Norwegian Q system, is a standard system for the quantitative evaluation of surrounding rock quality. It achieves objective classification of surrounding rock conditions by comprehensively considering multiple dimensions such as rock strength, integrity index (RQD), and structural surface conditions.

[0039] Construction condition zoning refers to the division of different construction areas based on the geometric characteristics and stress state differences of the dome, such as the high stress zone at the arch foot and the unloading zone at the arch crown. Its purpose is to develop differentiated construction plans for the characteristics of different areas.

[0040] The parametric probability distribution model is a mathematical model used to quantify the uncertainty of surrounding rock parameters. It can be constructed through statistical analysis methods and combined with Monte Carlo simulation or Bayesian inference to estimate the mean, standard deviation and confidence interval of parameters, thereby providing probabilistic prior knowledge for subsequent optimization.

[0041] Example 3: Please refer to Figure 1 The specific method for step S2 is as follows: S2.1 Based on the three-dimensional geological information model in step S1, construct an integrated numerical calculation model of the cavern group and the dome. The model space range is 3 to 5 times the characteristic size of the caverns beyond the excavation boundary. The finite element method is used to describe the elastoplastic behavior of the continuous medium, and the discrete element method is used to describe the discontinuous deformation controlled by the structural surface. A finite element-discrete element coupled framework is established. A 0.5-meter to 1-meter fine mesh is used at the stress concentration points of the arch foot and arch crown, and a 3-meter to 5-meter coarse mesh is used in the far field. Boundary conditions are set and the initial stress state is established. The multi-physics field coupled control equations of blasting dynamics, seepage, and stress are integrated. Rock mass parameters and structural surface parameters are assigned according to spatial partitions. S2.2 In the numerical model of step S2.1, the blasting cyclic cumulative damage constitutive model and the unloading aging creep constitutive model are introduced. The damage constitutive model establishes the correlation equation between damage variables and blasting vibration parameters. The creep constitutive model uses a combination of viscoelastic and plastic elements to establish the stress-strain-time correlation equation. The micro parameters are calibrated through indoor tests such as uniaxial compression, triaxial compression, cyclic loading and unloading, Hopkinson bar impact, and multi-stage creep. Field test blasts are carried out to deploy vibration velocity sensors and displacement monitoring networks. The inversion analysis method is used to iteratively optimize the macro parameters such as damping ratio, blasting load parameters, and structural surface mechanical parameters with the measured data as the objective function, and completes multi-scale parameter transformation and model verification.

[0042] In this embodiment, a high-fidelity numerical model was constructed and dynamic constitutive relations were integrated to achieve accurate simulation of the surrounding rock response during the excavation of the dome. First, in the model construction stage, an integrated numerical calculation model of the cavern group and the dome was established based on a three-dimensional geological information model, ensuring that the model's geometric boundaries and rock mass parameters strictly matched the actual geological conditions. The model's spatial range was set to 3 to 5 times the characteristic dimensions of the caverns beyond the excavation boundary, avoiding boundary effect interference and controlling the computational scale. The adoption of a finite element-discrete element coupled framework enabled the coordinated solution of the continuous medium's elasto-plastic behavior and the discontinuous deformation of the structural surfaces. In particular, the design of using refined meshes at the stress concentration points of the arch foot and crown, while using coarser meshes in the far field, achieved a balance between high accuracy in key areas and overall computational efficiency.

[0043] By integrating the multi-physics field coupling control equations of blasting dynamics, seepage, and stress, and assigning rock mass parameters according to spatial partitions, the interaction mechanism of water-force-blasting under construction disturbance was fully characterized. Subsequently, in the constitutive model development stage, a blasting cycle cumulative damage constitutive model was introduced, and a correlation equation between damage variables and blasting vibration parameters was established, enabling the model to quantify the progressive damage effect of multiple blasting vibrations on the surrounding rock. At the same time, a combination of viscoelastic and plastic elements was used to construct an unloading-aged creep constitutive model, and a stress-strain-time correlation equation was established to realistically simulate the rheological properties and time-dependent deterioration process of the surrounding rock after excavation and unloading.

[0044] Microscopic parameters were calibrated through indoor tests such as uniaxial and triaxial compression. Combined with vibration velocity sensor and displacement monitoring network data from field blast tests, an inversion analysis method was employed to iteratively optimize macroscopic parameters using the measured response as the objective function. This achieved multi-scale parameter transformation from microscopic experiments to macroscopic field behavior. These technical features collectively address the problem of insufficient dynamic representation capabilities of numerical models, providing a reliable physical simulation foundation for subsequent drilling and blasting parameter optimization.

[0045] The integrated numerical model of the cavern complex and dome refers to a numerical model capable of simultaneously simulating the overall stress characteristics of the cavern complex and the fine-grained local response of the dome. It can be achieved through the coupling of the finite element method (FEM) and the discrete element method (DEM). The FEM is primarily used to describe the elastoplastic behavior of continuous media, while the DEM is used to handle discontinuous deformation problems controlled by structural surfaces. The purpose of this coupled framework is to consider both the mechanical properties of continuous and discontinuous media, thereby more comprehensively reflecting the actual behavior of the surrounding rock.

[0046] The blasting cycle cumulative damage constitutive model is a mathematical model used to describe the progressive failure process of surrounding rock under multiple blasting operations. It is achieved by establishing the relationship between damage variables and blasting vibration parameters. The introduction of this model aims to quantify the long-term impact of blasting vibration on the surrounding rock, thereby improving the model's adaptability to actual construction conditions. The unloading-aged creep constitutive model, on the other hand, establishes a stress-strain-time correlation equation through a combination of viscoelastic-plastic elements to simulate the time-dependent deterioration process of the surrounding rock after excavation unloading. The introduction of this model helps to more accurately predict the long-term stability of the surrounding rock under construction disturbances.

[0047] Inversion analysis is a technique for optimizing model parameters based on measured data. It can calibrate macroscopic parameters such as damping ratio, blast load parameters, and structural surface mechanical parameters through iterative optimization of the objective function. The advantage of this method lies in its ability to organically combine laboratory test results with field monitoring data, thereby completing multi-scale parameter transformation from micro to macro, ensuring a high degree of consistency between model predictions and actual engineering responses.

[0048] Example 4: Please refer to Figure 1 The specific method for step S3 is as follows: S3.1 Establish a multi-objective optimization mathematical model for drilling and blasting parameters. Define design variables including borehole layout parameters, charge structure parameters, and detonation network parameters. Borehole parameters include ring spacing of 1.5 m to 3.0 m, hole spacing of 0.8 m to 2.0 m, hole diameter of 40 mm to 90 mm, and inclination angle of 60 degrees to 90 degrees. Charge parameters include single-hole charge amount, segmented charge position ratio of 0.3 to 0.7, and plugging length of 20% to 40% of hole depth. Detonation parameters include detonation sequence and delay time of 25 ms to 100 ms. Establish a multi-objective function set F(X). ={f1(X),f2(X),f3(X),f4(X)}, including minimizing over-excavation and under-excavation volume, minimizing peak vibration velocity, minimizing the standard deviation of profile flatness, and minimizing unit advance cost. Constraints are set, including safety constraints, geometric constraints, and process constraints. Based on the working condition zoning results in step S1.2, differentiated variable value boundaries are set to form the feasible region Ω_k of the zoning, where k is the working condition zoning number. The numerical model from step S2 is called, with input parameter combination X and output performance response value Y. A parameter-performance mapping database D={(X)} is established. i ,Y i )}; S3.2. Using Latin hypercube sampling, an initial sample point set is generated within the feasible region. The number of samples N = 10d to 20d, where d is the dimension of the design variable. For each sample point Xi, the numerical model from step S2 is called to calculate the objective function value Yi = F(Xi), forming a training dataset Dtrain. A surrogate model is constructed using a radial basis function neural network or a Gaussian process regression model. The parameters are fitted using the training dataset. The radial basis function neural network prediction formula is as follows: = Σwi·φ(||X-Xi||); The radial basis function is φ(r) = exp(-r² / (2σ²)), wi is the weight coefficient, and sigma is the kernel function bandwidth parameter. The prediction formula for the Gaussian process regression model is as follows: = μ(X) + k(X,Xtrain)·[K(Xtrain,Xtrain) + σn²·I] - ¹·(Ytrain - μ(Xtrain)); Where k is the covariance function, K is the covariance matrix, σn² is the observation noise variance, I is the identity matrix, and μ is the mean function; k-fold cross-validation was used to evaluate the prediction accuracy of the surrogate model, and the coefficient of determination and root mean square error were calculated. The formula for calculating the coefficient of determination is: R² = 1 - SSres / SStot = 1 - Σ(Yval,i - val,i)² / Σ(Yval,i - val)²; Where SSres is the residual sum of squares and SStot is the total sum of squares. val is the mean of the true values ​​in the validation set; The formula for calculating the root mean square error is:

[0049] Where nval is the number of validation samples, and the accuracy thresholds are set to R²≥0.90 and RMSE≤10%. When the accuracy does not meet the threshold, adaptive sampling is used to supplement training samples and update the surrogate model. Based on the surrogate model, the improved non-dominated sorting genetic algorithm NSGA-Ⅲ is used for multi-objective optimization. The population size is set to 100 to 200, the crossover probability is 0.8 to 0.9, the mutation probability is 0.05 to 0.1, and the maximum number of generations is 200 to 500. The population is initialized, the objective function value of each individual is calculated, and selection, crossover, and mutation genetic operations are performed. The parent and offspring populations are merged to form a mixed population. Non-dominated sorting is used to divide the mixed population into multiple non-dominated layers. The crowding distance is calculated for individuals within the same non-dominated layer. The formula for calculating the crowding distance is as follows: CD(Xi) = Σk[fk(Xi+1) - fk(Xi-1)] / (fk,max - fk,min); Where fk is the kth objective function value, fk,max and fk,min are the maximum and minimum values ​​of the objective function in the current population, respectively, and Xi+1 and Xi-1 are the neighboring individuals of the i-th individual after sorting; Individuals are selected to form the next generation of the population using an elite retention strategy that prioritizes non-dominated hierarchies and those with greater crowding distances. This process is iterated until the convergence criterion is met or the maximum number of generations is reached. The first non-dominated hierarchy is extracted as the Pareto optimal solution set P*. Based on the working condition partitioning in step S1.2, surrogate model construction and multi-objective optimization processes are performed for the high-stress stable zone at the arch foot, the stress transition zone at the arch waist, the unloading sensitive zone at the arch crown, the fault fracture zone risk zone, and the water-rich soft interlayer zone, respectively. The Pareto optimal solution set Pk corresponding to each working condition partition is obtained. A partition baseline parameter scheme library Library={Pk} is established, indexed by the working condition type number k, and stores the set of optimization parameters, including perforation layout parameters, charge structure parameters, detonation network parameters, and expected performance indicators.

[0050] In this embodiment: By establishing a multi-objective optimization mathematical model, the complex drilling and blasting parameter optimization problem is transformed into a mathematical programming problem, providing a theoretical foundation for subsequent optimization. Based on this, Latin hypercube sampling is used to generate an initial sample point set, requiring only a finite number of value model calculations to construct the parameter-performance mapping database, significantly reducing the computational burden. By constructing a surrogate model, kernel functions or covariance functions are used to capture the nonlinear relationship between parameters and performance, simplifying the expression of complex numerical models and significantly accelerating the evaluation of the objective function while ensuring accuracy. A k-fold cross-validation mechanism is used to rigorously verify the reliability of the surrogate model. When the accuracy does not meet the requirements, adaptive sampling is triggered to ensure that the surrogate model continuously approximates the real numerical model. Based on the surrogate model, an improved non-dominated sorting genetic algorithm is used for multi-objective optimization. Efficient elite retention is achieved through non-dominated sorting and crowding distance calculation, accurately extracting the Pareto optimal solution set. Optimization processes are executed separately for different working condition zones. The surrogate model and optimization process are customized according to the geological zone characteristics, solving the failure problem of a unified parameter scheme in different working conditions. Finally, a partitioned baseline parameter scheme library was established to provide a reliable initial solution for subsequent dynamic adjustments, effectively bridging the efficiency gap between numerical simulation and engineering practice.

[0051] A surrogate model is a mathematical model used to replace complex numerical models for rapid prediction. It can be implemented using machine learning algorithms such as support vector machines and artificial neural networks, with the aim of reducing the computational cost of numerical simulation and improving optimization efficiency.

[0052] Latin hypercube sampling is an efficient experimental design method. It is a stratified random sampling technique that dynamically determines the number of samples according to the dimensions of the design variables. Its purpose is to ensure that the sample points are evenly distributed in the design space and avoid the redundant calculations of traditional grid sampling.

[0053] The non-dominated sorting genetic algorithm is a multi-objective optimization algorithm based on the idea of ​​evolution. It can use various encoding methods such as binary encoding and real number encoding. Its purpose is to optimize multiple conflicting objective functions at the same time and maintain population diversity.

[0054] Example 5: Please refer to Figure 1 The specific method for step S4 is as follows: S4.1. Based on the working condition zoning results in step S1.2, formulate a monitoring network scheme, deploy a three-dimensional laser scanner with a measurement accuracy of ±2 mm and a scanning cycle of 1 to 2 times per cycle, deploy triaxial velocity sensors with a measurement point spacing of 5 to 10 meters and a sampling frequency of not less than 5000 Hz, deploy multi-point displacement gauges with a measurement point spacing of 3 to 5 meters and a monitoring accuracy of 0.01 mm, deploy acoustic emission sensors with a spacing of 10 to 15 meters and a working frequency of 50 kHz to 400 kHz, deploy micro-vibration sensors with a spacing of 30 to 50 meters and a positioning accuracy of not less than 5 meters, establish a data acquisition and transmission system with an end-to-end delay of no more than 1 second, use a time-series database for storage and establish an index; S4.2. Based on the numerical model in step S2, construct a digital twin virtual model containing geometric, physical, and behavioral layers. Use the iterative nearest point algorithm or normal distribution transformation algorithm to register point clouds and update geometric boundaries. Establish a parameter inversion objective function and use gradient descent, genetic algorithm, or particle swarm optimization algorithm to solve for the optimal parameters of rock mass elastic modulus, Poisson's ratio, cohesion, and internal friction angle to update the physical layer. Update the behavioral layer according to the actual borehole location, charge amount, detonation sequence, and delay time. Preprocess and extract features from the monitoring data, including over- and under-excavation volume, half-hole ratio, profile standard deviation, peak velocity, dominant frequency, vibration duration, cumulative displacement, displacement rate, acoustic emission event rate, acoustic emission b-value, microseismic magnitude, source coordinates, and apparent volumetric strain. Use a timestamp alignment method to establish a multi-source data fusion mechanism. Set the model to be updated after each excavation cycle is completed. Establish a three-dimensional visualization platform to render the virtual model and monitoring data distribution cloud map in real time.

[0055] In this embodiment, the core problem of the disconnect between static optimization and dynamic construction is solved by constructing a real-time collaborative system of a multi-source monitoring network and a digital twin virtual model. The monitoring network scheme is formulated based on the results of working condition zoning, which can rationally allocate monitoring resources and avoid monitoring blind spots. The 3D laser scanner, with a measurement accuracy of ±2 mm and a scanning cycle of 1 to 2 times per cycle, achieves high-frequency and high-precision capture of the excavation contour, ensuring the timeliness of geometric data. The three-phase vibration velocity sensor, with a measuring point spacing of 5 to 10 meters and a sampling frequency of no less than 5000 Hz, comprehensively records the spatial distribution and temporal evolution characteristics of blasting vibration. The multi-point displacement gauge, combined with acoustic emission and microseismic sensors, forms a multi-scale deformation monitoring network, enabling the effective identification of minute displacements and internal rock damage signals. The end-to-end delay of the data acquisition and transmission system is controlled within 1 second, ensuring the real-time flow of monitoring data from acquisition to processing. The introduction of a time-series database supports efficient time-series querying and historical data backtracking, facilitating the tracking of the evolution trend of surrounding rock parameters.

[0056] The digital twin virtual model achieves a virtual isomorphic mapping of physical engineering through a layered design of geometric, physical, and behavioral layers. The geometric layer uses an iterative nearest-point algorithm to register point clouds and update geometric boundaries, ensuring the virtual model's geometric accuracy is dynamically maintained as construction progresses. The physical layer solves for optimal parameters such as the rock mass's elastic modulus through parameter inversion of the objective function, enabling the virtual model's physical properties to reflect the surrounding rock's deterioration state in real time. The behavioral layer is updated based on actual borehole locations, charge quantities, and other construction details, improving the model's simulation fidelity for the drilling and blasting process. Monitoring data undergoes preprocessing and feature extraction to extract key indicators such as over- and under-excavation volume and acoustic emission b-value, providing quantitative input for performance evaluation. A timestamp alignment method integrates multi-source data to form a unified state view. The model is updated after each excavation cycle to ensure the virtual model continuously reflects the latest construction status. A 3D visualization platform renders the virtual model and monitoring data distribution cloud map in real time, assisting construction personnel in quickly identifying high-risk areas.

[0057] The above technical solution solves the problem that static optimization models cannot effectively integrate on-site monitoring data for dynamic parameter adjustment, significantly improves the control accuracy of over-excavation and under-excavation, reduces the risk of excessive vibration, and improves the stability of forming quality.

[0058] The monitoring network scheme refers to the technical means of configuring monitoring density in a targeted manner based on differences in geological risks. It can be achieved by adopting a differentiated deployment strategy, with the aim of ensuring high-precision data coverage of key parts such as the high-stress zone at the arch foot and the fracture zone.

[0059] A 3D laser scanner is a device used to capture the excavation contour at high frequency and with high precision. It can be achieved by setting a specific scanning cycle, with the aim of providing reliable geometric basis for over-excavation and under-excavation analysis.

[0060] A triaxial velocity sensor is a device used to comprehensively capture the characteristics of blasting vibrations. It can be implemented using specific spacing and high-frequency sampling settings to avoid missing vibration peaks caused by low-frequency sampling.

[0061] Multi-point displacement gauges are monitoring devices used to identify minute displacement changes. They can be implemented using high-precision measurement technology and are intended to effectively identify the progressive failure process of surrounding rock.

[0062] The data acquisition and transmission system refers to the technical architecture that ensures the real-time flow of monitoring data. It can be implemented using low-latency communication protocols to avoid delays caused by accumulated latency.

[0063] A digital twin virtual model is a computational framework that realizes a virtual isomorphic mapping of physical engineering. It can be implemented through a multi-layered structure design, with the aim of supporting real-time decision-making based on the current surrounding rock conditions.

[0064] Example 6: Please refer to Figure 1 The specific method for step S5 is as follows: S5.1 Establish a prediction-comparison-correction closed-loop mechanism. Before each construction cycle, the blasting effect is predicted based on the digital twin model in step S4.2, outputting the predicted over- and under-excavation volume, peak vibration velocity, displacement field, and damage zone. After construction, the measured data is obtained from step S4.1, and the root mean square error of profile deviation RMSEcontour, relative vibration velocity error εPPV, and displacement error RMSEdisp are calculated. The state vector of surrounding rock parameters θ={E,ν,c,φ,K,σt,ρ} is defined. According to the prior distribution P(θ) in step S1.2, the likelihood function P(Dobs|θ) is established. The Metropolis-Hastings algorithm is used for MCMC sampling, or the extended Kalman filter EKF is used to recursively update and calculate the posterior mean θupdated, or the unscented Kalman filter UKF is used to update the parameters through sigma point propagation. θupdated is input into the physical model layer in step S4.2 to complete the correction. Three parameter update mechanisms are established: error triggering, periodic triggering, and state triggering. S5.2 Establish an online dynamic re-optimization module. Based on the updated parameters in step S5.1 and the engineering state characteristics in step S4.2, define design variables including hole mesh parameters, charge parameters, and detonation parameters. Establish multi-objective functions including minimizing over- and under-excavation, vibration control, forming quality, and cost constraints. Set safety constraints, geometric constraints, process constraints, and schedule constraints. Using the baseline parameters in step S3.2 as the initial solution, within a disturbance range of ±10% to 20%, use sequential quadratic programming, interior point method, particle swarm optimization, and differential evolution algorithm for optimization. Call the surrogate model in step S3.2 to accelerate evaluation, output optimized parameters to generate construction instructions, determine the surrounding rock stability based on the damage characteristics in step S4.2, and trigger support coordination adjustment when the acoustic emission event rate or micro-vibration frequency exceeds the threshold. Expand design variables to include drilling and blasting parameters and support parameters including anchor length, spacing, and concrete thickness. Determine the support timing and strength based on the damage depth and surrounding rock grade, establish an optimization history database, and use machine learning to train a fast decision-making model.

[0065] In this embodiment, a closed-loop mechanism of prediction-comparison-correction-re-optimization is constructed to achieve dynamic control of drilling and blasting parameters driven by monitoring data. In each construction cycle, a digital twin model is first used to predict the blasting effect, ensuring that the prediction starting point is consistent with the current surrounding rock conditions. Then, the prediction deviation is quantified using measured data to accurately locate the model's inaccurate points, and the surrounding rock parameters are updated by combining prior distribution and real-time monitoring information. Based on this, a multi-trigger mechanism ensures the timeliness of parameter correction, avoiding optimization failure due to update lag.

[0066] The dynamic re-optimization module redefines the optimization problem based on updated surrounding rock parameters and engineering condition characteristics, tightly coupling design variables and constraints with the latest surrounding rock state. By using baseline parameters as the initial solution and optimizing within a small range of disturbances, performance evaluation is accelerated using a pre-trained surrogate model, thus ensuring real-time optimization. Furthermore, surrounding rock stability is determined by acoustic emission event rate and microseismic frequency, and drilling and blasting parameters are optimized in conjunction with support elements to achieve dynamic matching between excavation and support. Finally, by establishing an optimization history database and training a fast decision-making model, iterative experience is accumulated to improve subsequent adjustment efficiency, forming a continuous learning capability throughout the construction process.

[0067] The above technical solution effectively solves the problem that monitoring data is only used for post-event verification rather than dynamic control, realizes adaptive and precise control of drilling and blasting construction, suppresses over-excavation and under-excavation fluctuations and vibration effects, and ensures the stability of the surrounding rock and construction safety.

[0068] The prediction-comparison-correction closed-loop mechanism is a dynamic control method that continuously corrects and optimizes construction parameters through real-time monitoring data feedback. It can accurately predict blasting effects by constructing a digital twin model and perform deviation analysis based on monitoring data after actual construction, thereby ensuring that the model always remains consistent with the current surrounding rock conditions. The purpose of introducing this mechanism is to solve the problem in existing technologies where monitoring data is only used for post-event verification, enabling drilling and blasting parameters to dynamically respond to the evolution of the surrounding rock condition.

[0069] The surrounding rock parameter state vector θ={E,ν,c,φ,K,σt,ρ} is a set of key parameters describing the mechanical properties of the surrounding rock, specifically including parameters such as elastic modulus E, Poisson's ratio ν, cohesion c, internal friction angle φ, bulk modulus K, tensile strength σt, and density ρ. These parameters can be dynamically corrected through Bayesian updates or Kalman filtering methods to reflect the deterioration trajectory of the surrounding rock during construction.

[0070] The online dynamic re-optimization module is an optimization tool based on real-time parameter updates. Its core function is to redefine the optimization problem according to the latest surrounding rock conditions and accelerate performance evaluation through a surrogate model. The purpose of this module is to eliminate the operational adaptability defects of static optimization schemes while ensuring the real-time nature and efficiency of the optimization process.

[0071] Example 7: Please refer to Figure 1 The specific method for step S6 is as follows: S6.1 Select typical working conditions and set the test section length for 3 to 5 excavation cycles. Implement the static optimization scheme (using step S3.2 to keep the baseline parameters fixed) and the dynamic adaptive optimization scheme (using step S5 to adjust the parameters in real time via a closed-loop mechanism). Collect construction effect data synchronously through the monitoring network in step S4.1. Establish forming quality evaluation indicators, including over-excavation and under-excavation rate, profile deviation standard deviation, half-hole rate, and forming compliance rate. Establish vibration control evaluation indicators, including peak vibration velocity compliance rate, vibration energy, and dominant frequency distribution. Establish advance efficiency evaluation indicators, including cycle time, monthly advance speed, and explosive consumption. Statistically compare the indicators of the two schemes using t-test and Mann-Whitney U test. Use the analytic hierarchy process (AHP) with weighted coefficients of forming quality 0.4, vibration control 0.3, and advance efficiency 0.3 for weighted comprehensive scoring and cost-benefit analysis. S6.2. Based on the experimental data from step S6.1 and the historical data from steps S4 and S5, a vibration velocity threshold calibration mechanism is established. Support Vector Regression (SVR) is used to build a prediction model, and Random Forest is used to build a vibration velocity-damage correlation model. The calibration threshold is determined through Bayesian optimization. A displacement threshold calibration mechanism is established. Long Short-Term Memory (LSTM) network is used to build a time-series prediction model, and Logistic Regression is used to build a displacement-instability classification model. The threshold is determined through ROC curves to keep the instability risk within 5%. An over- or under-excavation threshold calibration mechanism is established. Gaussian Process Regression (GPR) is used to build a prediction model, and K-means clustering is used to identify over- or under-excavation patterns. Pareto optimal search is used to balance control and efficiency. A dynamic threshold update mechanism is established, updating every 10 to 20 cycles. Q-learning or Deep Q-Network (DQN) is used to optimize the decision rules. The state space is defined as surrounding rock parameters and monitoring characteristics, the action space as parameter adjustment amplitude, and the reward function as performance improvement and cost weighting. The calibration threshold and decision rules are updated to step S5.

[0072] In this embodiment, a comparative verification and data-driven threshold calibration system is constructed to address the insufficient adaptability caused by static control thresholds during dynamic optimization. First, test sections are set up in typical working condition zones, enabling precise positioning of the applicable boundaries for parameter optimization based on the differences in surrounding rock characteristics across different geological zones. Static optimization and dynamic adaptive optimization schemes are implemented separately, and by comparing the effects of fixed parameters versus real-time parameter adjustments, the dynamic evolution of surrounding rock mechanical behavior with the excavation process is revealed. Simultaneous collection of construction data through a monitoring network achieves a closed-loop correlation between multi-source monitoring information and the optimization process, providing an empirical basis for threshold calibration. Secondly, a vibration velocity threshold calibration mechanism was established based on experimental and historical data. A prediction model was constructed using Support Vector Regression (SVR) to capture the nonlinear relationship between vibration velocity and surrounding rock damage. Combined with random forest analysis of the vibration velocity-damage correlation model, the calibration threshold was dynamically determined through Bayesian optimization, ensuring that the threshold setting is closely related to the actual damage evolution. A displacement threshold calibration mechanism was also established, utilizing a Long Short-Term Memory (LSTM) network to process time-series monitoring data and accurately predict displacement development trends. A displacement-instability classification model was constructed using logistic regression, and the threshold was optimized through ROC curves to control instability risk, ensuring dynamic matching between the safety boundary and the surrounding rock condition. Finally, an over-excavation / under-excavation threshold calibration mechanism was established, employing Gaussian over-excavation... This approach uses GPR (Geometric Regression-Based Particle Size) to predict over- and under-excavation patterns, combined with K-means clustering to identify typical over- and under-excavation characteristics under different geological conditions. A Pareto optimal search achieves a balance between control accuracy and construction efficiency. A dynamic threshold update mechanism is established, automatically iterating the threshold every 10 to 20 cycles to adapt to the cumulative effect of surrounding rock deterioration. Q-learning or Deep Q-Network (DQN) is employed to optimize decision rules. The state space is defined based on surrounding rock parameters and monitoring characteristics, with the parameter adjustment range forming the action space. Intelligent decision-making is driven by a performance improvement and cost-weighted reward function, feeding the calibration results back to the dynamic optimization stage, forming a co-evolutionary closed loop of threshold calibration and parameter optimization. Overall, the effectiveness of the optimization mechanism is verified through empirical comparison, and machine learning is used to achieve adaptive threshold calibration, enabling the control threshold to respond to the spatiotemporal evolution characteristics of the surrounding rock, significantly improving the accuracy and robustness of dynamic optimization.

[0073] A test section refers to a representative area selected during actual construction. Its length is usually set to 3 to 5 excavation cycles, with the aim of verifying the applicability of different optimization schemes.

[0074] Static optimization schemes are construction methods that maintain a fixed baseline based on pre-set parameters, while dynamic adaptive optimization schemes adjust parameters in real time through a closed-loop mechanism to adapt to changes in surrounding rock conditions.

[0075] The monitoring network is used to synchronously collect construction effect data, with the aim of comprehensively evaluating key performance indicators during the construction process. Finishing quality evaluation indicators include over-excavation / under-excavation rate and profile deviation standard deviation, which are introduced to quantify construction precision.

[0076] Vibration control evaluation indicators include peak velocity compliance rate and other aspects, with the aim of ensuring the safety of blasting vibration.

[0077] The evaluation indicators for advance efficiency focus on cycle time, monthly advance rate, etc., with the goal of improving construction efficiency. Statistical comparisons use t-tests or Mann-Whitney U tests to scientifically distinguish significant differences between different schemes.

[0078] The Analytic Hierarchy Process (AHP) is used to set weight coefficients and perform comprehensive scoring, with the aim of achieving quantitative decision-making on the results of multi-objective optimization.

[0079] Example 8: Please refer to Figure 1 The specific method for step S7 is as follows: S7.1 Integrate the data from steps S1 to S6 to construct a four-dimensional relational knowledge graph. Define the ontology model, including an entity type layer containing geological zoning, parameter configuration, construction effect, and risk response entities; an attribute type layer containing numerical, categorical, temporal, and spatial attributes; and a relationship type layer containing association, causal, and triggering relationships. Use Apriori and FP-Growth algorithms to extract association rules, setting a minimum support of 0.3 and a minimum confidence of 0.7. Use structural equation modeling (SEM) to establish causal relationships and temporal association analysis to establish triggering relationships. Use the Neo4j graph database for storage. Establish a case-based reasoning (CBR) system, including modules for case representation, retrieval, reuse, correction, and retention. Case retrieval uses weighted similarity to return Top-K similar cases, with K equal to 3 to 5. Case reuse uses parameter interpolation or rule correction methods. Establish a decision confidence assessment mechanism with a threshold of 0.6. S7.2. Based on the knowledge graph in step S7.1 and the emergency plan database in step S5.2, extreme working condition types are sorted out, and a two-level classification system is defined. The first-level classification includes geological disaster type, construction accident type, equipment failure type, and environmental factor type. Geological disaster type is further subdivided into sudden water inrush, fault fracture zone, rock burst, and gas outburst. Emergency response templates are established for each type of extreme working condition. The template structure includes six modules: working condition identification characteristics, risk level determination, emergency response process, technical measures plan, resource allocation plan, and restoration construction conditions. The risk level is divided according to the water inrush volume: general water inrush of 50 to 100 cubic meters per hour, large water inrush of 100 to 500 cubic meters per hour, and major water inrush of more than 500 cubic meters per hour. Technical measures include grouting and water plugging, dewatering and drainage, and advanced pre-reinforcement. An emergency response decision tree is established to automatically match emergency templates based on monitoring data. An emergency drill simulation system is established based on the digital twin model in step S4.2.

[0080] In this embodiment: the case representation module is responsible for converting historical construction data into a standardized format for easy subsequent retrieval and reuse; the case retrieval module returns Top-K similar cases through a weighted similarity algorithm to ensure that the matching results fit the characteristics of the current working conditions; the case reuse module uses parameter interpolation or rule correction methods to fine-tune the parameters of historical solutions according to the new working conditions, avoiding rote application. Simultaneously, the decision confidence assessment mechanism filters low-quality suggestions by setting thresholds to ensure the reliability of the inference results. The emergency response template is designed to transform fuzzy experience into standardized operating procedures. Its structure includes multiple modules such as working condition identification features, risk level determination, and emergency response procedures, covering the entire process from risk identification to solution generation. Taking sudden water inrush as an example, classifying risk levels by water inrush volume and clarifying technical measures can significantly improve the standardization of the response.

[0081] The introduction of an emergency response decision tree enables a dynamic response mechanism. It automatically matches emergency templates with real-time monitoring data, shortening the time lag between risk identification and solution generation. Combined with an emergency drill simulation system based on a digital twin model, it allows for the rehearsal of response plans in a virtual environment, verifying the feasibility of technical measures and avoiding trial-and-error risks in actual construction. Overall, this solution, by constructing a four-dimensional relational knowledge graph and an emergency response template system, transforms data from the entire construction process into reusable decision-making knowledge, solving the core problems of fragmented historical experience and slow response to extreme conditions, and forming a closed-loop decision support system from knowledge accumulation to emergency execution.

[0082] A four-dimensional relational knowledge graph is a tool for systematically representing multi-source heterogeneous data through a unified framework. It can be achieved by constructing a triple structure of entity-relationship-attribute, aiming to avoid data silos in traditional methods and thus improve the computability of the relationship between surrounding rock characteristics and construction activities.

[0083] An ontology model is a multi-layered form of knowledge organization, comprising entity type, attribute type, and relation type layers. Through structured classification and semantic definition, it enables elements such as geological zoning and parameter configuration to have clear logical relationships. For example, geological zoning in the entity type layer can be refined into specific categories such as high-stress zones at the arch foot and unloading zones at the arch crown, while the attribute type layer covers multiple dimensions of data description, including numerical, categorical, temporal, and spatial types.

[0084] The Apriori or FP-Growth algorithm is used to uncover hidden patterns in historical data. Its core principle lies in filtering reliable association rules by setting minimum support and confidence thresholds. These rules can reveal the potential relationship between parameter configurations and construction outcomes, providing a basis for optimization decisions.

[0085] Structural equation modeling is a method for quantifying causal relationships. It uses path analysis to quantify the impact path of changes in surrounding rock parameters on effects such as over-excavation, under-excavation, or vibration, thereby enhancing the interpretability of the prediction model.

[0086] Temporal correlation analysis focuses on capturing the temporal dependencies between key events and risks in the monitoring data sequence, with the aim of enabling proactive risk warning.

[0087] Neo4j graph database, as an efficient storage tool, leverages its graph traversal capabilities to support real-time queries of complex relationships, meeting the timeliness requirements of construction decision-making.

[0088] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0089] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome, characterized in that: The specific steps are as follows: S1. Collect geological, geostress, hydrological and structural surface data, establish a three-dimensional geological information model, divide the working conditions based on the surrounding rock quality evaluation index, and construct the probability distribution model of parameters and uncertainty boundary of each zone. S2. Establish a three-dimensional numerical model of the cavern group, adopt the finite element-discrete element coupling method, integrate the blasting dynamics, seepage and stress multi-physics field effects, introduce a constitutive model of surrounding rock deterioration considering cumulative damage and unloading aging, and calibrate key parameters through indoor tests and field test blasts. S3. Using the perforation layout, charge structure and detonation network as design variables, a multi-objective optimization mathematical model is established, which includes minimizing over-excavation and under-excavation, vibration control, forming quality optimization and cost constraints. A surrogate model and an improved genetic algorithm are used for global optimization to obtain the Pareto optimal solution set and form a partitioned baseline parameter scheme library. S4. Deploy laser scanning, vibration velocity sensing, displacement monitoring, acoustic emission and microseismic monitoring systems, establish a virtual model isomorphic to physical engineering, and realize real-time acquisition, transmission and feature extraction of monitoring data; S5. Establish a closed-loop mechanism of prediction-comparison-correction-re-optimization, adopt Bayesian update and Kalman filter to correct the surrounding rock parameters, use constrained optimization algorithm to perform small-step fast optimization, generate dynamic parameter adjustment scheme, and realize the coordinated optimization of drilling and blasting and support. S6. Conduct static and dynamic scheme comparison tests to evaluate the differences in forming quality, vibration control and advance efficiency, and use machine learning algorithms to adaptively calibrate vibration velocity, displacement and over- and under-excavation control thresholds. S7. Establish a knowledge graph linking geological zoning, parameter configuration, construction effects, and risk response; develop a case-based reasoning decision support system; generate emergency response templates for extreme working conditions; and compile standardized technical procedures.

2. The method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome according to claim 1, characterized in that, The specific method for step S1 is as follows: S1.1 Collect multi-source geological survey data of the excavation area of ​​the dome, including borehole columnar section data, geological sketch logging data, and geophysical survey line data. Obtain physical and mechanical parameters of the rock mass through indoor rock mechanics tests, including uniaxial compressive strength, elastic modulus, Poisson's ratio, internal friction angle, cohesion, and density. Conduct in-situ stress testing using hydraulic fracturing and stress relief methods to obtain the three-dimensional principal stress field distribution. Measure the groundwater level, rock mass permeability coefficient, and unit inflow rate. Map the geometric parameters of the structural surfaces, including strike, dip, dip angle, spacing, ductility, opening, infill type, and roughness coefficient. Integrate the multi-source data based on three-dimensional geological modeling software to establish a three-dimensional geological information model. S1.

2. Based on the RMR rock mass quality scoring system or the Norwegian Q system, the surrounding rock quality of each spatial unit of the dome is quantitatively evaluated. The evaluation indicators include rock strength, integrity index RQD, structural surface conditions, groundwater status, and in-situ stress ratio. Combining the geometric characteristics and stress state differences of the dome, the construction conditions are divided into zones, including the high stress zone at the arch foot, the transition zone at the arch waist, the unloading zone at the arch crown, the risk zone of the fracture zone, and the water-rich and weak zone. Rock mass parameter samples are collected for each zone. Statistical analysis methods are used to construct parameter probability distribution models. The mean, standard deviation, and confidence interval of the parameters are calculated through Monte Carlo simulation or Bayesian inference to form a prior probability knowledge base for the zone parameters.

3. The method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a dome-shaped arch as described in claim 2, characterized in that, The specific method for step S2 is as follows: S2.1 Based on the three-dimensional geological information model in step S1, construct an integrated numerical calculation model of the cavern group and the dome. The model space range is 3 to 5 times the characteristic size of the caverns beyond the excavation boundary. The finite element method is used to describe the elastoplastic behavior of the continuous medium, and the discrete element method is used to describe the discontinuous deformation controlled by the structural surface. A finite element-discrete element coupled framework is established. A 0.5-meter to 1-meter fine mesh is used at the stress concentration points of the arch foot and arch crown, and a 3-meter to 5-meter coarse mesh is used in the far field. Boundary conditions are set and the initial stress state is established. The multi-physics field coupled control equations of blasting dynamics, seepage, and stress are integrated. Rock mass parameters and structural surface parameters are assigned according to spatial partitions. S2.2 In the numerical model of step S2.1, the blasting cyclic cumulative damage constitutive model and the unloading aging creep constitutive model are introduced. The damage constitutive model establishes the correlation equation between damage variables and blasting vibration parameters. The creep constitutive model uses a combination of viscoelastic and plastic elements to establish the stress-strain-time correlation equation. The micro parameters are calibrated through indoor tests such as uniaxial compression, triaxial compression, cyclic loading and unloading, Hopkinson bar impact, and multi-stage creep. Field test blasts are carried out to deploy vibration velocity sensors and displacement monitoring networks. The inversion analysis method is used to iteratively optimize the macro parameters such as damping ratio, blasting load parameters, and structural surface mechanical parameters with the measured data as the objective function, and completes multi-scale parameter transformation and model verification.

4. The method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a dome-shaped arch as described in claim 3, characterized in that, The specific method for step S3 is as follows: S3.1 Establish a multi-objective optimization mathematical model for drilling and blasting parameters. Define design variables including borehole layout parameters, charge structure parameters, and detonation network parameters. Borehole parameters include ring spacing of 1.5 m to 3.0 m, hole spacing of 0.8 m to 2.0 m, hole diameter of 40 mm to 90 mm, and inclination angle of 60 degrees to 90 degrees. Charge parameters include single-hole charge amount, segmented charge position ratio of 0.3 to 0.7, and plugging length of 20% to 40% of hole depth. Detonation parameters include detonation sequence and delay time of 25 ms to 100 ms. Establish a multi-objective function set F(X). ={f1(X),f2(X),f3(X),f4(X)}, including minimizing over-excavation and under-excavation volume, minimizing peak vibration velocity, minimizing the standard deviation of profile flatness, and minimizing unit advance cost. Constraints are set, including safety constraints, geometric constraints, and process constraints. Based on the working condition zoning results in step S1.2, differentiated variable value boundaries are set to form the feasible region Ω_k of the zoning, where k is the working condition zoning number. The numerical model from step S2 is called, with input parameter combination X and output performance response value Y. A parameter-performance mapping database D={(X)} is established. i ,Y i )}; S3.

2. Using Latin hypercube sampling, an initial sample point set is generated within the feasible region. The number of samples N = 10d to 20d, where d is the dimension of the design variable. For each sample point Xi, the numerical model from step S2 is called to calculate the objective function value Yi = F(Xi), forming a training dataset Dtrain. A surrogate model is constructed using a radial basis function neural network or a Gaussian process regression model. The parameters are fitted using the training dataset. The radial basis function neural network prediction formula is as follows: = Σwi·φ(||X-Xi||); The radial basis function is φ(r) = exp(-r² / (2σ²)), wi is the weight coefficient, and sigma is the kernel function bandwidth parameter. The prediction formula for the Gaussian process regression model is as follows: = μ(X) + k(X,Xtrain)·[K(Xtrain,Xtrain) + σn²·I] - ¹·(Ytrain - μ(Xtrain)); Where k is the covariance function, K is the covariance matrix, σn² is the observation noise variance, I is the identity matrix, and μ is the mean function; k-fold cross-validation was used to evaluate the prediction accuracy of the surrogate model, and the coefficient of determination and root mean square error were calculated. The formula for calculating the coefficient of determination is: R² = 1 - SSres / SStot = 1 - Σ(Yval,i - val,i)2 / Σ(Yval,i - val)2; Where SSres is the residual sum of squares and SStot is the total sum of squares. val is the mean of the true values ​​in the validation set; The formula for calculating the root mean square error is: Where nval is the number of validation samples, and the accuracy thresholds are set to R²≥0.90 and RMSE≤10%. When the accuracy does not meet the threshold, adaptive sampling is used to supplement training samples and update the surrogate model. Based on the surrogate model, the improved non-dominated sorting genetic algorithm NSGA-Ⅲ is used for multi-objective optimization. The population size is set to 100 to 200, the crossover probability is 0.8 to 0.9, the mutation probability is 0.05 to 0.1, and the maximum number of generations is 200 to 500. The population is initialized, the objective function value of each individual is calculated, and selection, crossover, and mutation genetic operations are performed. The parent and offspring populations are merged to form a mixed population. Non-dominated sorting is used to divide the mixed population into multiple non-dominated layers. The crowding distance is calculated for individuals within the same non-dominated layer. The formula for calculating the crowding distance is as follows: CD(Xi) = Σk[fk(Xi+1) - fk(Xi-1)] / (fk,max - fk,min); Where fk is the kth objective function value, fk,max and fk,min are the maximum and minimum values ​​of the objective function in the current population, respectively, and Xi+1 and Xi-1 are the neighboring individuals of the i-th individual after sorting; Individuals are selected to form the next generation of the population based on an elite retention strategy that prioritizes non-dominated hierarchies and those with greater crowding distances. This process is iterated until the convergence criterion is met or the maximum number of generations is reached. The first non-dominated hierarchy is extracted as the Pareto optimal solution set P*. According to the working condition partitioning in step S1.2, surrogate model construction and multi-objective optimization processes are performed for the high-stress stable zone at the arch foot, the stress transition zone at the arch waist, the unloading sensitive zone at the arch crown, the fault fracture zone risk zone, and the water-rich soft interlayer zone, respectively, to obtain the Pareto optimal solution set Pk corresponding to each working condition partition. A partition baseline parameter scheme library Library={Pk} is established, with the working condition type number k as the index, to store the set of optimization parameters, including perforation layout parameters, charge structure parameters, detonation network parameters, and expected performance indicators.

5. The method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome according to claim 4, characterized in that, The specific method for step S4 is as follows: S4.

1. Based on the working condition zoning results in step S1.2, formulate a monitoring network scheme, deploy a three-dimensional laser scanner with a measurement accuracy of ±2 mm and a scanning cycle of 1 to 2 times per cycle, deploy triaxial velocity sensors with a measurement point spacing of 5 to 10 meters and a sampling frequency of not less than 5000 Hz, deploy multi-point displacement gauges with a measurement point spacing of 3 to 5 meters and a monitoring accuracy of 0.01 mm, deploy acoustic emission sensors with a spacing of 10 to 15 meters and a working frequency of 50 kHz to 400 kHz, deploy micro-vibration sensors with a spacing of 30 to 50 meters and a positioning accuracy of not less than 5 meters, establish a data acquisition and transmission system with an end-to-end delay of no more than 1 second, use a time-series database for storage and establish an index; S4.

2. Based on the numerical model in step S2, construct a digital twin virtual model containing geometric, physical, and behavioral layers. Use the iterative nearest point algorithm or normal distribution transformation algorithm to register point clouds and update geometric boundaries. Establish a parameter inversion objective function and use gradient descent, genetic algorithm, or particle swarm optimization algorithm to solve for the optimal parameters of rock mass elastic modulus, Poisson's ratio, cohesion, and internal friction angle to update the physical layer. Update the behavioral layer according to the actual borehole location, charge amount, detonation sequence, and delay time. Preprocess and extract features from the monitoring data, including over- and under-excavation volume, half-hole ratio, profile standard deviation, peak velocity, dominant frequency, vibration duration, cumulative displacement, displacement rate, acoustic emission event rate, acoustic emission b-value, microseismic magnitude, source coordinates, and apparent volumetric strain. Use a timestamp alignment method to establish a multi-source data fusion mechanism. Set the model to be updated after each excavation cycle is completed. Establish a three-dimensional visualization platform to render the virtual model and monitoring data distribution cloud map in real time.

6. The method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome according to claim 5, characterized in that, The specific method for step S5 is as follows: S5.1 Establish a prediction-comparison-correction closed-loop mechanism. Before each construction cycle, the blasting effect is predicted based on the digital twin model in step S4.2, outputting the predicted over- and under-excavation volume, peak vibration velocity, displacement field, and damage zone. After construction, the measured data is obtained from step S4.1, and the root mean square error of profile deviation RMSEcontour, relative vibration velocity error εPPV, and displacement error RMSEdisp are calculated. The state vector of surrounding rock parameters θ={E,ν,c,φ,K,σt,ρ} is defined. According to the prior distribution P(θ) in step S1.2, the likelihood function P(Dobs|θ) is established. The Metropolis-Hastings algorithm is used for MCMC sampling, or the extended Kalman filter EKF is used to recursively update and calculate the posterior mean θupdated, or the unscented Kalman filter UKF is used to update the parameters through sigma point propagation. θupdated is input into the physical model layer in step S4.2 to complete the correction. Three parameter update mechanisms are established: error triggering, periodic triggering, and state triggering. S5.2 Establish an online dynamic re-optimization module. Based on the updated parameters in step S5.1 and the engineering state characteristics in step S4.2, define design variables including hole mesh parameters, charge parameters, and detonation parameters. Establish multi-objective functions including minimizing over- and under-excavation, vibration control, forming quality, and cost constraints. Set safety constraints, geometric constraints, process constraints, and schedule constraints. Using the baseline parameters in step S3.2 as the initial solution, within a disturbance range of ±10% to 20%, use sequential quadratic programming, interior point method, particle swarm optimization, and differential evolution algorithm for optimization. Call the surrogate model in step S3.2 to accelerate evaluation, output optimized parameters to generate construction instructions, determine the surrounding rock stability based on the damage characteristics in step S4.2, and trigger support coordination adjustment when the acoustic emission event rate or micro-vibration frequency exceeds the threshold. Expand design variables to include drilling and blasting parameters and support parameters including anchor length, spacing, and concrete thickness. Determine the support timing and strength based on the damage depth and surrounding rock grade, establish an optimization history database, and use machine learning to train a fast decision-making model.

7. The method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome according to claim 6, characterized in that, The specific method for step S6 is as follows: S6.1 Select typical working conditions and set the test section length for 3 to 5 excavation cycles. Implement the static optimization scheme (using step S3.2 to keep the baseline parameters fixed) and the dynamic adaptive optimization scheme (using step S5 to adjust the parameters in real time via a closed-loop mechanism). Collect construction effect data synchronously through the monitoring network in step S4.

1. Establish forming quality evaluation indicators, including over-excavation and under-excavation rate, profile deviation standard deviation, half-hole rate, and forming compliance rate. Establish vibration control evaluation indicators, including peak vibration velocity compliance rate, vibration energy, and dominant frequency distribution. Establish advance efficiency evaluation indicators, including cycle time, monthly advance speed, and explosive consumption. Statistically compare the indicators of the two schemes using t-test and Mann-Whitney U test. Use the analytic hierarchy process (AHP) with weighted coefficients of forming quality 0.4, vibration control 0.3, and advance efficiency 0.3 for weighted comprehensive scoring and cost-benefit analysis. S6.

2. Based on the experimental data from step S6.1 and the historical data from steps S4 and S5, a vibration velocity threshold calibration mechanism is established. Support Vector Regression (SVR) is used to build a prediction model, and Random Forest is used to build a vibration velocity-damage correlation model. The calibration threshold is determined through Bayesian optimization. A displacement threshold calibration mechanism is established. Long Short-Term Memory (LSTM) network is used to build a time-series prediction model, and Logistic Regression is used to build a displacement-instability classification model. The threshold is determined through ROC curves to keep the instability risk within 5%. An over- or under-excavation threshold calibration mechanism is established. Gaussian Process Regression (GPR) is used to build a prediction model, and K-means clustering is used to identify over- or under-excavation patterns. Pareto optimal search is used to balance control and efficiency. A dynamic threshold update mechanism is established, updating every 10 to 20 cycles. Q-learning or Deep Q-Network (DQN) is used to optimize the decision rules. The state space is defined as surrounding rock parameters and monitoring characteristics, the action space as parameter adjustment amplitude, and the reward function as performance improvement and cost weighting. The calibration threshold and decision rules are updated to step S5.

8. The method for optimizing and constructing a horizontal circumferential drilling and blasting parameter model for a spherical dome according to claim 7, characterized in that, The specific method for step S7 is as follows: S7.1 Integrate the data from steps S1 to S6 to construct a four-dimensional relational knowledge graph. Define the ontology model, including an entity type layer containing geological zoning, parameter configuration, construction effect, and risk response entities; an attribute type layer containing numerical, categorical, temporal, and spatial attributes; and a relationship type layer containing association, causal, and triggering relationships. Use Apriori and FP-Growth algorithms to extract association rules, setting a minimum support of 0.3 and a minimum confidence of 0.

7. Use structural equation modeling (SEM) to establish causal relationships and temporal association analysis to establish triggering relationships. Use the Neo4j graph database for storage. Establish a case-based reasoning (CBR) system, including modules for case representation, retrieval, reuse, correction, and retention. Case retrieval uses weighted similarity to return Top-K similar cases, with K equal to 3 to 5. Case reuse uses parameter interpolation or rule correction methods. Establish a decision confidence assessment mechanism with a threshold of 0.

6. S7.

2. Based on the knowledge graph in step S7.1 and the emergency plan database in step S5.2, extreme working condition types are sorted out, and a two-level classification system is defined. The first-level classification includes geological disaster type, construction accident type, equipment failure type, and environmental factor type. Geological disaster type is further subdivided into sudden water inrush, fault fracture zone, rock burst, and gas outburst. Emergency response templates are established for each type of extreme working condition. The template structure includes six modules: working condition identification characteristics, risk level determination, emergency response process, technical measures plan, resource allocation plan, and restoration construction conditions. The risk level is divided according to the water inrush volume: general water inrush of 50 to 100 cubic meters per hour, large water inrush of 100 to 500 cubic meters per hour, and major water inrush of more than 500 cubic meters per hour. Technical measures include grouting and water plugging, dewatering and drainage, and advanced pre-reinforcement. An emergency response decision tree is established to automatically match emergency templates based on monitoring data. An emergency drill simulation system is established based on the digital twin model in step S4.2.

Images