A method for advanced pre-support and dynamic support decision of underground cavern in adverse geological section

Abstract

This invention discloses a method for advance support and dynamic support decision-making in adverse geological sections of underground caverns, relating to the field of tunnel construction. This method collects surrounding rock deformation data by setting up monitoring sections, using this as the basis for inversion. It employs forward and inverse analysis combined with optimization algorithms to automatically obtain surrounding rock mechanical parameters that match the actual on-site conditions, and updates the finite element model to conduct advance predictions for subsequent excavation cycles. The prediction results are then quantitatively compared with preset control standards. Dynamic adjustments to the support scheme are initiated only when an early warning is triggered. Differential measures are taken according to the actual exposed surrounding rock conditions, such as increasing the system support strength, initiating or strengthening advance support, or appropriately reducing the support strength. Simultaneously, the method operates on a rolling basis with each excavation cycle, allowing the support parameters to be continuously optimized as the surrounding rock information becomes clearer. This systematically improves the safety, economy, and scientific decision-making of construction in adverse geological sections without significantly increasing on-site workload.

Description

Technical Field

[0001] This invention relates to the field of tunnel construction, specifically to a method for advance support and dynamic support decision-making in underground caverns with adverse geological conditions. Background Technology

[0002] When underground cavern engineering traverses fault fracture zones, weak interlayers, and other adverse geological formations, the surrounding rock has poor self-stabilizing capacity, making it prone to large deformations or even collapses after excavation and unloading, resulting in significant construction safety risks. In existing technologies, the formulation of advanced support schemes mainly relies on geological data from the survey and design phase. However, the actual surrounding rock conditions revealed during excavation often deviate from prior predictions, leading to a mismatch between the pre-determined support scheme and on-site requirements. Furthermore, advanced support and systematic support are typically two relatively independent processes in current practices. The former is determined and implemented before excavation, while the latter is constructed according to the design after excavation, lacking a dynamic coordination mechanism based on real-time geological information. The application of surrounding rock classification methods suffers from a large range of levels; the transition states between levels III and IV, and between levels IV and V, are difficult to accurately correspond to specific support parameter combinations. Adopting a conservative strategy of lowering the level by one level increases engineering costs, and for special geological hazards such as rock bursts and large deformations in soft rock, simple classification methods are insufficient to provide targeted support decisions.

[0003] On the other hand, while numerical analysis methods can provide quantitative mechanical judgments on surrounding rock stability, they have significant bottlenecks in construction applications. Each parameter adjustment or scheme change requires manual model reconstruction, calculation submission, and result extraction—a cumbersome and time-consuming process that struggles to provide timely support for on-site decision-making during the tight schedule of construction in adverse geological sections. Existing research has made valuable explorations in automating the inversion of surrounding rock mechanical parameters and intelligent prediction of support parameters, but these remain at the level of single-technology verification and have not yet formed a systematic method that organically integrates monitoring data acquisition, parameter inversion, advanced prediction, and dynamic decision-making. Therefore, it is necessary to establish a method that can continuously operate throughout the excavation process and dynamically adjust the support scheme based on real-time monitoring data. Summary of the Invention

[0004] The purpose of this invention is to provide a method for advance support and dynamic support decision-making in adverse geological sections of underground caverns. It aims to solve the technical problems in the prior art, such as the disconnect between advance support schemes and dynamic support decisions, the lag in obtaining surrounding rock mechanical parameters, the lack of quantitative basis for support decisions, and the inability to automate the numerical modeling and parameter inversion process. This will enable rapid, accurate, and dynamic decision-making on support schemes during the construction of adverse geological sections of underground caverns.

[0005] This invention is achieved through the following technical solution:

[0006] A method for decision-making on advanced pre-support and dynamic support in adverse geological sections of underground caverns includes the following steps:

[0007] Step 1: During construction in areas with unfavorable geological conditions, monitoring sections are set up along the tunnel cross-section to collect surrounding rock deformation monitoring data, including multi-point displacement inside the surrounding rock and tunnel cross-section convergence displacement.

[0008] Step 2: Based on the surrounding rock deformation monitoring data as the inversion basis information, a surrounding rock mechanical parameter optimization inversion model is established using the forward and inverse analysis method. The optimization algorithm drives the inversion iterative calculation to automatically obtain surrounding rock mechanical parameters that match the actual surrounding rock condition on site.

[0009] Step 3: Based on the acquired surrounding rock mechanical parameters, update the finite element forward modeling analysis model, carry out advance prediction of the surrounding rock mechanical response for at least one subsequent excavation cycle, and output the predicted values ​​of the mechanical state of the surrounding rock and support structure after the completion of the subsequent excavation steps.

[0010] Step 4: Compare the obtained advanced prediction results with the preset control standards to determine the applicability of the existing support scheme; when the prediction results trigger the early warning conditions, make dynamic decisions and adjustments to the advanced pre-support scheme and system support parameters for subsequent excavation based on the actual exposure of the surrounding rock and the advanced prediction results.

[0011] Step 5: After implementing the adjusted support plan, continue excavation and enter the next cycle. Repeat steps one through four until the construction of the unfavorable geological section is completed.

[0012] This scheme uses measured deformation data during construction as input for the inversion of surrounding rock mechanical parameters. It automatically acquires surrounding rock mechanical parameters that match the actual field conditions using forward and inverse analysis combined with a highly efficient optimization algorithm. This overcomes the problems of long testing cycles, poor sampling representativeness, and the disconnect between traditional indoor testing parameters and the actual surrounding rock conditions during construction. Based on the inverted parameters, the finite element forward analysis model is updated, and advance predictions for subsequent excavation cycles are conducted. This ensures that adjustments to the support scheme are based on quantitative mechanical predictions rather than empirical judgments. Furthermore, by quantitatively comparing the advance prediction results with preset control standards, adjustments to the support scheme are only initiated when warning conditions are triggered, avoiding over-adjustment. The method addresses both insufficient and inadequate support, and allows for differentiated measures based on the actual exposure of the surrounding rock, such as increasing the strength of the system support, initiating or strengthening pre-support, and appropriately reducing the support strength. This achieves a synergistic match between pre-support and system support in time and space. Furthermore, the method operates on a rolling basis, with each round of decision-making adjustments followed by the next cycle of monitoring, inversion, prediction, and decision-making. This ensures that support parameters are continuously optimized as the surrounding rock information becomes clearer during excavation, adapting to the temporal development of surrounding rock deformation. Consequently, without significantly increasing on-site workload, the method systematically improves the safety, economy, and scientific decision-making of construction in adverse geological sections.

[0013] Furthermore, to ensure timely capture of the initiation, acceleration, and spatial transmission characteristics of surrounding rock deformation caused by excavation in unfavorable geological sections, in step one, the spacing between the monitoring sections in Class II to III surrounding rock is no more than 50m, and the spacing in Class IV to V surrounding rock and unfavorable geological sections is 20 to 30m; for each monitoring section, multiple displacement gauges are arranged radially along the tunnel at the arch crown, arch shoulder, arch waist, and arch foot, and each multi-point displacement gauge has no less than 3 measuring points at different depths.

[0014] Furthermore, in order to reduce the adverse disturbance of random errors in displacement monitoring data on the inversion results, in step two, the objective function of the rock mechanics parameter optimization inversion model is the root mean square error between the measured displacement and the calculated displacement, and the inversion parameters are at least one of deformation parameters, strength parameters, or a combination of deformation parameters and strength parameters.

[0015] Furthermore, to further ensure the reliability of stress field, displacement field, and plastic zone calculations in subsequent advanced predictions, the objective function is specifically as follows:

[0016] ;

[0017] Where N is the total number of displacement measuring points, and M is the number of displacement components at each measuring point. The calculated displacement value is the one corresponding to the inversion parameter vector X. The measured displacement value is used. When the objective function value is less than a predetermined threshold, the inversion is considered to have converged, and the current parameter is the optimal estimate.

[0018] Furthermore, to shorten the analysis cycle and meet the timeliness requirements of on-site decision-making, the forward modeling analysis in step two adopts the finite element method, and the calculation files are automatically generated and updated in the following way:

[0019] The finite element calculation file is divided into several data blocks according to its structured features;

[0020] The information in each data block is grouped and sequentially encoded;

[0021] Pseudocode is set in the calculation file to save the dynamically changing inversion parameter information separately in an external data file and update it automatically, so that successive calculations corresponding to different inversion parameters are completed based on the same calculation file.

[0022] Furthermore, to avoid over-reliance on a single indicator that may lead to missed or incorrect judgments, in step four, the preset control criteria include at least one of the following: the predicted displacement of key parts exceeds the set percentage threshold of the designed reserved deformation, the predicted stress of the support structure exceeds the design allowable value, and the expansion range of the plastic zone of the surrounding rock exceeds the pre-support area; when any criterion is triggered, dynamic decision-making adjustment of the support scheme is performed.

[0023] Furthermore, to cover different working conditions ranging from good to extremely poor surrounding rock, the dynamic decision adjustment includes:

[0024] When the actual exposure of the surrounding rock is worse than expected, but the expansion of the plastic zone is still controllable, the strength of the system support should be increased.

[0025] When the actual exposure of the surrounding rock deteriorates significantly and advanced predictions indicate that the plastic zone will expand beyond the pre-supported area or the support structure will be overloaded, advanced pre-support measures should be initiated or strengthened.

[0026] When the actual exposure of the surrounding rock is better than expected, the system support strength should be reduced while meeting the minimum requirements of the design specifications.

[0027] Furthermore, to enhance the reliability and robustness of the final decision, it also includes establishing a machine learning-based intelligent decision-making model for support parameters as an auxiliary tool for decision adjustment in step four:

[0028] Establish a classification index system for support parameters;

[0029] Collect successful case data of completed support to establish a data sample library;

[0030] A multi-label, multi-output random forest algorithm is used to establish an intelligent decision-making model for support parameters;

[0031] The current surrounding rock classification index is input into the model to obtain the recommended combination of support parameters, which serves as a reference for dynamic decision-making and adjustment.

[0032] Furthermore, to further enhance the reliability and robustness of the final decision, the support parameter classification index system includes classification indicators for the survey and design stage and classification indicators for the construction stage; each sample in the data sample library contains a combination of classification indicators and its corresponding combination of support parameters, wherein the combination of support parameters includes at least one of the following: shotcrete thickness and layout range, system anchor parameters, steel mesh specifications, steel arch frame specifications, and advanced support parameters.

[0033] Furthermore, when pipe roof pre-support is adopted in adverse geological sections, the dynamic decision adjustment in step four also includes optimization of pipe roof support parameters: taking the maximum deflection of the pipe roof as the control index, sensitivity analysis is performed on the pipe roof diameter, circumferential spacing of the pipe roof, excavation advance and subgrade coefficient in front of the tunnel face, and the parameters with the highest sensitivity are adjusted first according to the sensitivity ranking, and the optimized values ​​are selected within the recommended parameter range. This enables engineering technicians to accurately identify the most critical influencing factors when facing the risk of pipe roof deflection exceeding the limit, avoiding decision delays and resource waste caused by trying different parameters one by one based on experience, thereby improving the efficiency of parameter optimization.

[0034] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0035] 1. This invention constructs a closed-loop control system that automates the entire process of "monitoring-inversion-prediction-decision", which solves the inherent defects of traditional methods, such as the separation of numerical modeling, parameter inversion and advanced prediction, and the reliance on manual coordination.

[0036] 2. This invention utilizes a combination of file segmentation, information grouping and sequential encoding techniques, as well as pseudocode techniques, to enable the successive finite element calculations corresponding to different inversion parameters to be automatically completed based on the same calculation file. This achieves fully automated operation of the entire process from monitoring data acquisition to mechanical parameter inversion and then to advanced prediction, thereby providing timely and efficient quantitative analysis support for on-site decision-making within the tight time window of construction in adverse geological sections.

[0037] 3. This invention uses measured deformation data during construction as the basic information for inverting the mechanical parameters of the surrounding rock, overcoming the problems of long testing cycles, poor sampling representativeness, and serious disconnect between exploration and design parameters and the actual surrounding rock condition during construction in traditional indoor rock mechanics tests.

[0038] 4. This invention quantitatively compares the advanced prediction results with the preset multi-dimensional control standards, and only initiates the adjustment of the support plan when the early warning conditions are triggered, realizing a fundamental shift from experience-based qualitative judgment to data-driven quantitative decision-making.

[0039] 5. Based on the actual exposure of the surrounding rock and the results of advanced prediction, the present invention also adopts differentiated measures in stages, such as increasing the strength of the system support, initiating or strengthening advanced pre-support, and appropriately reducing the support strength. This can effectively ensure the construction safety of adverse geological sections and avoid the waste of resources caused by excessive support, thus achieving a balance between safety and economy.

[0040] 6. This invention also establishes a classification index system for support parameters and collects successful case data of completed support, enabling the intelligent model to provide recommended combinations of multiple support parameters such as shotcrete, anchor bolts, steel mesh, steel arch frames, and advanced support within seconds based on the current surrounding rock conditions. These parameters are then cross-validated with the quantitative mechanical indicators predicted in advance, thereby effectively improving the reliability of decision-making and the operability of the project. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0042] Figure 1 This is a schematic diagram of the process of the present invention;

[0043] Figure 2 This is a flowchart illustrating the process of automatically generating and updating calculation files for forward modeling analysis in step two. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0045] Example

[0046] This embodiment provides a method for advance support and dynamic support decision-making in adverse geological sections of underground caverns. This method is applicable to the stability control of surrounding rock during construction of large underground cavern complexes such as pumped storage powerhouses, hydropower station underground powerhouses, and traffic tunnels when traversing adverse geological sections such as fault fracture zones, weak interlayers, and densely jointed fracture zones. Figure 1 As shown, it includes the following steps:

[0047] Step 1: During construction in areas with unfavorable geological conditions, monitoring sections are set up along the tunnel cross-section to collect surrounding rock deformation monitoring data, including multi-point displacement inside the surrounding rock and tunnel cross-section convergence displacement.

[0048] Specifically, before construction begins on sections with unfavorable geological conditions, the spatial distribution range of these sections is determined based on the survey and design data. Monitoring sections are then deployed along the tunnel axis within this range. For Class II to III surrounding rock, the spacing between monitoring sections is no greater than 50m; for Class IV to V surrounding rock and unfavorable geological sections, the spacing is increased to 20 to 30m. For example, when the tunnel section to be constructed is determined to be a fault fracture zone with Class IV surrounding rock, the spacing between monitoring sections is 25m.

[0049] On each monitoring section, multiple displacement gauges are arranged radially along the tunnel at the arch crown, arch shoulder, arch waist, and arch foot. Each multi-point displacement gauge has three measuring points at different depths, located in the shallow (0.5–1.0 m from the tunnel wall), middle (1.5–2.5 m from the tunnel wall), and deep (3.0–5.0 m from the tunnel wall) parts of the surrounding rock, respectively, to obtain displacement gradient information from shallow to deep within the surrounding rock. Simultaneously, convergence monitoring points are arranged on each monitoring section to measure the convergence displacement of the tunnel section.

[0050] During excavation in areas with unfavorable geological conditions, displacement data at each monitoring point was collected every 2–4 hours to obtain time-series data on multi-point displacement within the surrounding rock and the convergence displacement of the tunnel cross-section. For tunnel sections with high deformation rates (e.g., daily deformation exceeding 1 mm / d), the collection frequency was increased to once every hour. Displacement-time curves at each measuring point were recorded, the deformation rate of the surrounding rock was calculated, and the stage characteristics of the surrounding rock deformation were identified, with a focus on capturing the starting time point and peak deformation rate of the rapid deformation stage.

[0051] Step 2: Based on the surrounding rock deformation monitoring data as the inversion basis information, a surrounding rock mechanical parameter optimization inversion model is established using the forward and inverse analysis method. The optimization algorithm drives the inversion iterative calculation to automatically obtain surrounding rock mechanical parameters that match the actual surrounding rock condition on site.

[0052] Specifically, the measured deformation data collected in step one is used as the basic information for inversion, and the forward and inverse analysis methods are used to establish an optimized inversion model for the mechanical parameters of the surrounding rock.

[0053] First, determine the surrounding rock mechanical parameters that need to be inverted. Based on the lithological characteristics of the adverse geological section, select the combination of inversion parameters. For example, for weak interlayer sections where deformation is the primary control, select inversion deformation parameters (deformation modulus and Poisson's ratio); for fault fracture zones where stability is the primary control, select inversion strength parameters (cohesion and internal friction angle); for tunnel sections with complex surrounding rock conditions, select a combined inversion of deformation and strength parameters. The range of inversion parameter values ​​is determined based on exploration and design data and engineering experience. For example, the range for deformation modulus is set to 1.0–30.0 GPa, the range for cohesion is set to 0.05–2.0 MPa, and the range for internal friction angle is set to 20°–50°.

[0054] Secondly, a forward analysis model is established. A general-purpose finite element software (such as ABAQUS) is used as the forward analysis solver. Based on the actual engineering situation, a finite element calculation model reflecting the tunnel geometry, excavation sequence, support scheme, and monitoring point layout is established. This model includes node definition data blocks, element definition data blocks, material grouping data blocks, boundary node set data blocks, support information data blocks, excavation element set data blocks, material assignment data blocks, boundary condition data blocks, and analysis step data blocks.

[0055] Then, an optimization inversion objective function is constructed. The objective function is defined as the root mean square error between the measured displacement and the calculated displacement:

[0056] ;

[0057] Where N is the total number of displacement measuring points, and M is the number of displacement components at each measuring point. The calculated displacement value is the one corresponding to the inversion parameter vector X. The measured displacement value is used. When the objective function value is less than a predetermined threshold (e.g., 0.01), the inversion is considered to have converged, and the current parameter is the optimal estimate.

[0058] Next, the asynchronous particle swarm optimization (APSO) algorithm is used to drive the inversion iterative computation. Specifically, the search behavior of each particle is treated as an independent process. Once any particle finds that its fitness is better than the optimal value in the current shared information, it immediately updates the global shared information. That is, assuming a particle swarm X = { in an m-dimensional search space, n particles constitute a particle swarm X = { , ... , , ... , }, where the position of the a-th particle is =[ , , ... , ] T Its speed is = [ , , ... , [T], whose individual extreme value is = [ , , ... , ] T The global extremum is = [ , , ... , ] T Then the update of particle position and velocity follows the formula:

[0059]

[0060] Where a = 1, 2, ..., n; d = 1, 2, ..., m; k is the particle update generation. The inertial weight decreases linearly with increasing algebraic degree; The acceleration constant is 2.0; rand() is a random number uniformly distributed between (0,1).

[0061] During the inversion iteration process, file segmentation, information grouping, and sequential encoding techniques are employed to automatically generate and update finite element calculation files. Specifically, the finite element calculation file is divided into several data blocks based on structured features; the information within each data block is grouped and sequentially encoded; pseudocode is set in the inversion INP file to store dynamically changing inversion parameter information in a separate external data file, which is then dynamically and automatically updated through programming. For example, for a data block containing excavation element sets, if N steps of excavation are performed during the analysis, the excavation element set information is divided into N groups, encoded as 1, 2, ..., N according to the actual excavation sequence. Successive calculations corresponding to different inversion parameters are all performed based on the same calculation file, and the pseudocode calls this automatically updated data file during the analysis process.

[0062] The automated implementation process of the entire inversion is as follows: After acquiring the finite element calculation model and optimization parameter information, the system automatically generates inversion calculation files and prediction calculation files. An initial particle swarm is formed using a random method. For each particle's position, an external data file storing inversion parameter information is automatically generated and dynamically updated for use by the inversion calculation file. The finite element solver is invoked to complete the stress and deformation calculation of the surrounding rock and support structure system. The displacement calculation values ​​of measuring points are extracted from the calculation results, and the particle fitness is calculated. The particle position and velocity are updated according to the asynchronous particle swarm optimization algorithm. The iteration is repeated until the optimal particle that meets the predetermined minimum threshold or the maximum number of iterations is reached. After the inversion converges, the current globally optimal particle position is the optimal estimate of the mechanical parameters that matches the actual surrounding rock condition on site. This parameter is then used for the advanced prediction of the surrounding rock mechanical response in step three.

[0063] Step 3: Based on the acquired surrounding rock mechanical parameters, update the finite element forward modeling analysis model, carry out advance prediction of the surrounding rock mechanical response for at least one subsequent excavation cycle, and output the predicted values ​​of the mechanical state of the surrounding rock and support structure after the completion of the subsequent excavation steps.

[0064] Specifically, after completing the parameter inversion and obtaining the optimal surrounding rock mechanical parameters in step two, the following operations are performed based on these parameters:

[0065] First, using the forward analysis model established in step two, a forward analysis is carried out based on the optimal surrounding rock mechanical parameters obtained from the inversion, to calculate the surrounding rock stress field and displacement field after the current excavation step is completed.

[0066] Then, based on the forward analysis, the predicted INP file is automatically run for restart analysis. The predicted INP file is not a complete calculation data file, but rather a restart calculation data file defined on the basis of the inverted INP file. Its structure consists of restart information and analysis step data blocks. Through restart analysis, the stress field and displacement field of the surrounding rock of the cavern, the stress state of the support structure (including anchor bolt axial force, shotcrete layer stress, steel arch stress, etc.), and the expansion range and trend of the plastic zone of the surrounding rock are calculated after the completion of the next 1-2 excavation cycles.

[0067] Finally, the following key quantitative indicators are output: the predicted displacement increment and cumulative displacement of key parts of the cavern (arch crown, arch shoulder, arch waist, and arch foot) after the completion of subsequent excavation steps; the predicted stress state of each component of the support structure; and the predicted range of expansion of the plastic zone of the surrounding rock.

[0068] Step 4: Compare the obtained advanced prediction results with the preset control standards to determine the applicability of the existing support scheme; when the prediction results trigger the early warning conditions, make dynamic decisions and adjustments to the advanced pre-support scheme and system support parameters for subsequent excavation based on the actual exposure of the surrounding rock and the advanced prediction results.

[0069] Specifically, the advanced prediction results obtained in step three are compared with the preset control standards to determine the applicability of the existing support scheme.

[0070] The preset control criteria include the following three items: (1) the predicted displacement of key parts exceeds 70% of the design allowable deformation; (2) the predicted stress of the support structure exceeds the design allowable value; and (3) the expansion range of the plastic zone of the surrounding rock exceeds the pre-supported area. When any criterion is triggered, dynamic decision-making adjustment of the support scheme is performed.

[0071] Dynamic decision adjustments are implemented according to the following hierarchical rules:

[0072] When the actual exposed condition of the surrounding rock is worse than expected, but the expansion of the plastic zone is still controllable, the strength of the system support should be increased. Specific measures include: reducing the spacing between anchor bolts to 0.7 to 0.8 times the original design (e.g., the original design spacing between anchor bolts was 1.2m × 1.2m, adjusted to 0.9m × 0.9m); lengthening the anchor bolts by 1.0 to 1.5m based on the original design (e.g., the original design anchor bolt length was 3.0m, adjusted to 4.0 to 4.5m); increasing the thickness of shotcrete by 3 to 5cm based on the original design (e.g., the original design shotcrete thickness was 15cm, adjusted to 18 to 20cm); upgrading the steel arch frame type or increasing the spacing between steel arch frames (e.g., replacing I16 I-beams with I18 I-beams, or increasing the spacing from 100cm to 80cm).

[0073] When the actual exposure of the surrounding rock deteriorates significantly and advanced predictions indicate that the plastic zone will expand beyond the pre-support area or the support structure will be overloaded, advanced pre-support measures should be initiated or strengthened. Specific measures include: installing additional advanced anchor bolts in front of the tunnel face (anchor bolt length 4.5–6.0 m, spacing 30–50 cm, outward insertion angle 5°–15°); increasing the density of advanced small-diameter pipe grouting (small-diameter pipe length 3.0–5.0 m, spacing 20–40 cm, grouting pressure 0.5–2.0 MPa); increasing the density of pipe roof arrangement (reducing the circumferential spacing of pipe roofs from 40 cm to 30 cm) or expanding the range of pipe roof arrangement (expanding the 120° range of the arch to 150°); and increasing the range of curtain grouting.

[0074] When the actual exposed surrounding rock conditions are better than expected, the system support strength should be reduced while still meeting the minimum requirements of the design specifications. Specific measures include: increasing the spacing between anchor bolts (e.g., from 1.0m × 1.0m to 1.2m × 1.2m); reducing the thickness of shotcrete (e.g., from 15cm to 10cm); and reducing the type of steel arch or increasing the spacing between steel arches (e.g., replacing I18 I-beams with I16 I-beams, or increasing the spacing from 80cm to 100cm). However, the adjusted support parameters must not be lower than the minimum requirements of the design specifications for the corresponding surrounding rock category.

[0075] For specific decision-making adjustments regarding rockburst hazards, the maximum tangential stress σθ of the tunnel wall is calculated based on the ground stress monitoring data obtained in step one. This, combined with the surrounding rock mechanical parameters obtained through inversion in step two, is used to comprehensively determine the rockburst risk level. Under a slight rockburst risk, conventional support is maintained and a flexible protective net is added. Under a moderate rockburst risk, stress relief holes (50–100 mm diameter, 3–5 m depth, 1–2 m spacing) are drilled at the tunnel face and nearby tunnel walls, the system anchor bolts are densified, and a steel mesh is added. Under a severe rockburst risk, in addition to densified anchor bolts and a steel mesh, a layer of steel fiber shotcrete (5–8 cm thick) and a steel arch frame are added. If necessary, advanced stress relief holes are drilled in front of the tunnel face.

[0076] After the pre-support scheme is adjusted, steps one to three are repeated to verify the applicability of the adjusted support scheme until the pre-prediction results meet all control criteria.

[0077] Step 5: After implementing the adjusted support plan, continue excavation and enter the next cycle. Repeat steps one through four until the construction of the unfavorable geological section is completed.

[0078] Specifically, after implementing the adjusted support scheme, excavation continues, entering the next excavation cycle. After each excavation cycle is completed, steps one through four are repeated, namely, re-collecting monitoring data, performing automatic inversion of surrounding rock mechanical parameters, conducting advance prediction of subsequent excavation cycles, and making dynamic decision adjustments, until all construction of the unfavorable geological section is completed.

[0079] In each cycle, the actual combination of support parameters and its corresponding surrounding rock classification index are added to the data sample library to achieve dynamic expansion of the data sample library and adaptive updating of the model.

[0080] In this embodiment, the method further includes establishing a machine learning-based intelligent decision-making model for support parameters as an auxiliary tool for decision adjustment in step four. Specifically, this includes:

[0081] A classification index system for support parameters was established. Based on the different information available during the exploration and design phase and the construction phase, two separate classification index systems were established: the exploration and design phase classification indexes include five indicators: rock uniaxial saturated compressive strength Rc, rock mass integrity coefficient Kv, groundwater influence index W, structural plane influence index S, and in-situ stress influence index C; the construction phase classification indexes include seven indicators: rock hardness, rock mass integrity, structural plane interlocking degree, rock mass structure type, groundwater condition, weathering condition, and in-situ stress state.

[0082] Collect successful case data on completed support procedures and establish a data sample library. Each data sample contains a combination of classification indicators and its corresponding combination of support parameters. The combination of support parameters includes the thickness and layout range of shotcrete (e.g., "15~18cm / B" indicates a thickness of 15~18cm and a layout range of half a ring), the number and spacing of system anchors and their layout range (e.g., "20~22 anchors / 0.8m / B" indicates 20~22 anchors per ring, a spacing of 0.8m, and a layout range of half a ring), the specifications and spacing of steel mesh and their layout range (e.g., "Ф6.5@25 / B" indicates a diameter of 6.5mm, a spacing of 25cm, and a layout range of half a ring), the specifications and spacing of steel arch frames and their layout range (e.g., "I16@80 / B" indicates I16 I-beams, a spacing of 80cm, and a layout range of half a ring), and the type, length, and layout range of advanced support (e.g., "Ф42 / 450 / 120°" indicates a diameter of 42mm, a length of 450cm, and a layout range of 120° in the arch).

[0083] A multi-label, multi-output random forest (MOC-RF) algorithm is used to establish an intelligent decision-making model for support parameters. A multi-output strategy is introduced to improve the random forest algorithm, training a classifier for each output index to form a set of classifiers for multiple objective variables. An improved particle swarm optimization (IPSO) algorithm is used to optimize the hyperparameters of the MOC-RF algorithm. The IPSO algorithm introduces a chaotic strategy to initialize the particle population, adaptively updates the learning factors and inertia weight coefficients, and improves population diversity based on a natural selection strategy. The trained IPSO-MOC-RF model is integrated into the decision-making process in step four. When a rapid assessment of the support scheme adjustment direction is needed, the current surrounding rock classification index is input into the model to obtain recommended support parameter combinations as a decision reference. When a new combination outside the sample library appears, it is reviewed according to the judgment rules in step four and the design specifications.

[0084] In this embodiment, when pipe roof pre-support is adopted in adverse geological sections, the dynamic decision adjustment in step four also includes optimization of pipe roof support parameters. Specifically, the maximum deflection of the pipe roof is used as the control index, and sensitivity analysis is performed on the pipe roof diameter, circumferential spacing of the pipe roof, excavation advance, and subgrade coefficient in front of the tunnel face. The sensitivity analysis uses a modified multi-factor sensitivity analysis method to calculate the degree of influence of changes in each parameter on the maximum deflection of the pipe roof. The sensitivity ranking from highest to lowest is: pipe roof diameter, circumferential spacing of the pipe roof, excavation advance, and subgrade coefficient in front of the tunnel face. The parameter with the highest sensitivity is adjusted first according to the sensitivity ranking. The pipe roof diameter is selected within the range of 70–150 mm, with 108 mm as the preferred value; the circumferential spacing of the pipe roof is selected within the range of 20–60 cm, with 40 cm as the preferred value; the equivalent thickness of the grouting reinforcement zone is selected within the range of 50–90 cm, with 70 cm as the preferred value; the equivalent elastic modulus of the grouting reinforcement zone is taken as 1–10 times the elastic modulus of the original surrounding rock.

[0085] When the increased excavation advance causes the pipe roof deflection to approach the control standard (e.g., 30mm), priority should be given to increasing the pipe roof diameter or increasing the circumferential spacing. When the pipe roof spacing is greater than 50cm, priority should be given to increasing the spacing rather than increasing the diameter, in order to ensure the formation of the micro-arch effect between pipe roofs and the integrity of the grouting reinforcement zone.

[0086] In summary, this invention continuously collects multi-point displacements within the surrounding rock and convergence displacements of the tunnel section at monitoring sections deployed during the construction phase. These deformation data essentially represent the comprehensive response of the surrounding rock to external excavation disturbances under the current stress state and material conditions. Using these measured deformation data as input, an optimized inversion model is established through forward and inverse analysis. An asynchronous particle swarm optimization algorithm drives the inversion iteration, treating the surrounding rock mechanical parameters as the solution space to be searched. The optimization objective is to minimize the deviation between the measured deformation and the forward-calculated deformation, automatically searching for the mechanical parameters that best match the actual state of the surrounding rock on site. These parameters are then used to update the finite element forward analysis model and to perform advance predictions for at least one subsequent excavation cycle. This calculates the stress field, displacement field, and plastic zone range of the surrounding rock and support structure after subsequent excavation, thereby elevating qualitative judgments based on engineering experience to quantitative predictions based on mechanical calculations.

[0087] The predicted results are compared with preset control standards (including displacement thresholds for key parts, allowable stress values ​​for support structures, and the expansion range of the plastic zone) to form the basis for determining whether to initiate adjustments. When the predicted results trigger an early warning, differentiated measures are taken in stages based on the correspondence between the actual exposed surrounding rock and the predicted results. These measures include increasing the strength of the system support, initiating or strengthening advanced pre-support, or appropriately reducing the support strength. This achieves a reasonable overlap between advanced pre-support and system support in time, and a coordinated matching between reinforcement in front of the tunnel face and constraint behind it in space. Simultaneously, a support parameter classification index system is established, and a multi-label, multi-output random forest algorithm is used to construct an intelligent decision-making model. This serves as an auxiliary reference to provide rapid recommended solutions based on successful engineering experience. For cases using pipe roof advanced pre-support, parameter sensitivity analysis is further conducted using the maximum deflection of the pipe roof as the control index. Based on the sensitivity ranking, parameters with the most significant impact on the control effect are adjusted first.

[0088] The above process operates on a rolling basis, with each excavation cycle as the unit. After each round of decision adjustments, excavation continues into the next cycle, and monitoring, inversion, prediction, and decision-making are repeated, allowing support parameters to be continuously optimized as surrounding rock information becomes clearer. Throughout the process, numerical modeling, parameter inversion, and advanced prediction are integrated into an automated pipeline through file segmentation, information grouping, and sequential coding techniques. This systematically improves the safety, economy, and scientific rigor of construction in adverse geological sections without significantly increasing on-site workload.

[0089] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for advance support and dynamic support decision-making in unfavorable geological sections of underground caverns, characterized in that, Includes the following steps: Step 1: During construction in areas with unfavorable geological conditions, monitoring sections are set up along the tunnel cross-section to collect surrounding rock deformation monitoring data, including multi-point displacement inside the surrounding rock and tunnel cross-section convergence displacement. Step 2: Based on the surrounding rock deformation monitoring data as the inversion basis information, a surrounding rock mechanical parameter optimization inversion model is established using the forward and inverse analysis method. The optimization algorithm drives the inversion iterative calculation to automatically obtain surrounding rock mechanical parameters that match the actual surrounding rock condition on site. Step 3: Based on the acquired surrounding rock mechanical parameters, update the finite element forward modeling analysis model, carry out advance prediction of the surrounding rock mechanical response for at least one subsequent excavation cycle, and output the predicted values ​​of the mechanical state of the surrounding rock and support structure after the completion of the subsequent excavation steps. Step 4: Compare the obtained advanced prediction results with the preset control standards to determine the applicability of the existing support scheme; when the prediction results trigger the early warning conditions, make dynamic decisions and adjustments to the advanced pre-support scheme and system support parameters for subsequent excavation based on the actual exposure of the surrounding rock and the advanced prediction results. Step 5: After implementing the adjusted support plan, continue excavation and enter the next cycle. Repeat steps one through four until the construction of the unfavorable geological section is completed.

2. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 1, characterized in that, In step one, the spacing between the monitoring sections in Class II to III surrounding rock is no more than 50m, and the spacing in Class IV to V surrounding rock and unfavorable geological sections is 20 to 30m. For each monitoring section, multiple displacement gauges are arranged along the radial direction of the cavern at the arch crown, arch shoulder, arch waist, and arch foot. Each multi-point displacement gauge has no less than 3 measuring points at different depths.

3. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 1, characterized in that, In step two, the objective function of the rock mechanics parameter optimization inversion model is the root mean square error between the measured displacement and the calculated displacement, and the inversion parameters are at least one of deformation parameters, strength parameters, or a combination of deformation parameters and strength parameters.

4. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 3, characterized in that, The objective function is specifically: ； Where N is the total number of displacement measuring points, and M is the number of displacement components at each measuring point. The calculated displacement value is the one corresponding to the inversion parameter vector X. The measured displacement value is used. When the objective function value is less than a predetermined threshold, the inversion is considered to have converged, and the current parameter is the optimal estimate.

5. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 1, characterized in that, In step two, the forward modeling analysis employs the finite element method, and the calculation files are automatically generated and updated in the following manner: The finite element calculation file is divided into several data blocks according to its structured features; The information in each data block is grouped and sequentially encoded; Pseudocode is set in the calculation file to save the dynamically changing inversion parameter information separately in an external data file and update it automatically, so that successive calculations corresponding to different inversion parameters are completed based on the same calculation file.

6. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 1, characterized in that, In step four, the preset control criteria include at least one of the following: the predicted displacement of the key part exceeds the set percentage threshold of the design reserved deformation, the predicted stress of the support structure exceeds the design allowable value, and the expansion range of the plastic zone of the surrounding rock exceeds the pre-support area; when any criterion is triggered, dynamic decision adjustment of the support scheme is performed.

7. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 6, characterized in that, The dynamic decision adjustment includes: When the actual exposure of the surrounding rock is worse than expected, but the expansion of the plastic zone is still controllable, the strength of the system support should be increased. When the actual exposure of the surrounding rock deteriorates significantly and advanced predictions indicate that the plastic zone will expand beyond the pre-supported area or the support structure will be overloaded, advanced pre-support measures should be initiated or strengthened. When the actual exposure of the surrounding rock is better than expected, the system support strength should be reduced while meeting the minimum requirements of the design specifications.

8. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 1, characterized in that, This also includes establishing a machine learning-based intelligent decision-making model for support parameters as an auxiliary tool for decision adjustment in step four: Establish a classification index system for support parameters; Collect successful case data of completed support to establish a data sample library; A multi-label, multi-output random forest algorithm is used to establish an intelligent decision-making model for support parameters; The current surrounding rock classification index is input into the model to obtain the recommended combination of support parameters, which serves as a reference for dynamic decision-making and adjustment.

9. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 8, characterized in that, The support parameter classification index system includes classification indicators for the survey and design stage and classification indicators for the construction stage; each sample in the data sample library contains a combination of classification indicators and its corresponding support parameter combination, the support parameter combination including at least one of the following: shotcrete thickness and layout range, system anchor bolt parameters, steel mesh specifications, steel arch frame specifications, and advanced support parameters.

10. The method for advance support and dynamic support decision-making in adverse geological sections of underground caverns according to claim 1, characterized in that, When pipe roof pre-support is adopted in adverse geological sections, the dynamic decision adjustment in step four also includes optimization of pipe roof support parameters: taking the maximum deflection of the pipe roof as the control index, sensitivity analysis is performed on the pipe roof diameter, pipe roof circumferential spacing, excavation advance and subgrade coefficient in front of the tunnel face, and the parameters with the highest sensitivity are adjusted first according to the sensitivity ranking, and the optimized values ​​are selected within the recommended parameter range.

Images