Tunnel soft rock large deformation evaluation method based on multi-source parameter analysis

Abstract

The invention relates to the technical field of tunnel engineering, and discloses a tunnel soft rock large deformation assessment method based on multi-source parameter analysis, and the method comprises the steps: firstly constructing an assessment model based on a grey clustering theory, and carrying out the preliminary risk classification of tunnel surrounding rock; after a tunnel face is revealed through tunnel excavation, apparent characteristic data are quantitatively collected, a comprehensive score is calculated, and an apparent large deformation level is determined; meanwhile, a BP neural network model optimized through a particle swarm optimization algorithm is utilized, and the cumulative deformation of the surrounding rock is predicted based on multi-source input parameters; calculating a strength-stress ratio by combining ground stress and rock mass strength data actually measured on site, verifying and correcting the preliminary grade, and determining a final comprehensive large deformation grade; and finally, mapping the grades into support parameters, and executing dynamic adjustment based on monitoring data. According to the method, through the fusion of theoretical calculation, field quantitative scoring and intelligent prediction, accurate judgment and closed-loop control of large deformation of the soft rock are realized, and the construction risk is effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering technology, specifically to a method for evaluating large deformation in soft rock tunnels based on multi-source parameter analysis. Background Technology

[0002] As transportation infrastructure construction extends into deeper and more complex geological areas, soft rock tunnel projects are becoming increasingly common. Under the combined effects of high ground stress and weak surrounding rock, tunnel excavation often triggers large deformation disasters in the surrounding rock, leading to cross-sectional shrinkage, distortion and failure of the initial support structure, and even collapse, severely impacting construction safety and project progress. Therefore, accurately assessing the risk level of large deformation in the surrounding rock and determining a reasonable allowance for deformation is crucial for the construction control of soft rock tunnels.

[0003] Early large deformation assessments primarily relied on geological survey data from the design phase, employing empirical formulas based on rock strength or geological structure for qualitative analysis. However, due to the discreteness of survey data and the concealed nature of the deep geological environment within tunnels, theoretical predictions from the design phase often fail to accurately reflect the true surrounding rock condition revealed after tunnel excavation, resulting in a disconnect between design parameters and actual on-site geological conditions. While real-time identification can be achieved by observing the tunnel face characteristics at the construction site, existing on-site assessments largely depend on the subjective experience of geological technicians, lacking unified, visualized, and quantitative scoring standards. This leads to discrepancies in the descriptions and judgments of the same geological phenomena by different technicians, making it difficult to guarantee the objectivity and stability of large deformation level assessment results.

[0004] Furthermore, in controlling construction for large deformations in soft rock, the determination of the allowable deformation is usually based on fixed recommended values ​​given in relevant specifications. This static approach ignores the differences in the rheological properties of surrounding rock under different lithological combinations and geostress environments. In actual construction, when the actual deformation of the surrounding rock exceeds the fixed allowable amount, it can lead to the initial support encroaching on the clearance of the secondary lining, forcing the project to be demolished and replaced with a new arch. Conversely, when the allowable amount is much larger than the actual deformation, it results in excessive over-excavation and waste of backfill concrete. Currently, there is a lack of an effective means to quantitatively calculate the cumulative deformation under specific working conditions by combining multi-source data from the field.

[0005] Meanwhile, existing monitoring and feedback mechanisms are often only used for post-event alarms after deformation exceeds the limit. They lack a closed-loop process for dynamically correcting the initial assessment level based on measured ground stress, rock mass strength, and real-time deformation rate. This results in the establishment of support parameters not matching the actual mechanical response of the surrounding rock on site, making it difficult to implement effective dynamic graded control before a disaster occurs. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis. This method solves the problems of low assessment accuracy due to relying solely on geological theory or field experience for qualitative judgment, mismatch between the fixed value of the reserved deformation amount and the actual rheological properties of the surrounding rock, and the lack of a closed-loop control mechanism for dynamically correcting the risk level based on measured stress and strength data.

[0007] To achieve the above objectives, this invention provides a method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis. This method includes the following steps:

[0008] A theoretical assessment model for large deformation of soft rock was constructed, and a preliminary risk classification of the tunnel surrounding rock was carried out based on the geological parameters obtained in the design stage to obtain a preliminary theoretical level.

[0009] After the tunnel face is exposed during excavation, surface feature data of the tunnel face are collected, and a comprehensive score is calculated using a quantitative scoring method. The apparent large deformation level of the surrounding rock at the tunnel face is determined based on the numerical range of the comprehensive score.

[0010] Multi-source parameters from the tunnel excavation section are collected to construct an input dataset. A BP neural network model optimized by particle swarm optimization is used to calculate and output the predicted cumulative deformation of the surrounding rock.

[0011] Based on the data from the on-site in-situ stress test and rock point load test, the rock mass strength-stress ratio is calculated, and combined with the preliminary theoretical level, the apparent large deformation level, and the predicted cumulative deformation of the surrounding rock, the final comprehensive large deformation level for guiding construction is determined.

[0012] Based on the comprehensive large deformation level, corresponding construction control measures are implemented, and the comprehensive large deformation level is dynamically adjusted according to the monitoring data during the construction process.

[0013] When constructing a theoretical evaluation model for large deformation in soft rock, the following evaluation indicators were selected: uniaxial saturated compressive strength of the rock, maximum initial ground stress, groundwater development, tunnel span, tunnel depth, the angle between the main structural plane and the tunnel axis, the dip angle of the main structural plane, and support stiffness. A hierarchical structure of the evaluation indicators was established using fuzzy interactive structural modeling and network analysis, and the weights of each indicator were calculated through consistency checks. Combining grey clustering evaluation theory, a whitening weight function was constructed to describe the membership of the evaluation indicators to different risk levels. The measured indicator data were substituted into the whitening weight function to calculate the membership degree, and the clustering coefficients for each risk level were calculated based on the weights. Finally, the preliminary theoretical level was determined according to the principle of maximum membership.

[0014] When determining the apparent large deformation level, the following factors are selected: the angle between the main structural plane and the tunnel axis, the dip angle of the main structural plane, the thickness of the rock strata, the integrity of the rock mass, the degree of interlayer bonding, the degree of influence of geological structures, and the state of groundwater exposure. The scores for each factor are determined based on field monitoring data. Each score is then weighted and summed with its corresponding weighting coefficient to obtain a comprehensive score characterizing the tunnel face condition. This comprehensive score is compared with a preset grading interval to determine the apparent large deformation level. This step also includes a correction judgment for special working conditions: when the dip angle of the rock strata at the tunnel face is detected to be less than 30 degrees, and the maximum initial ground stress is greater than 10 MPa or the tunnel depth exceeds 300 meters, if the rock strata thickness is simultaneously detected to be less than 10 cm or the rock mass integrity index is less than 0.3, the judgment based on the comprehensive score is stopped, and the apparent large deformation level is directly determined to be extremely severe large deformation.

[0015] When using a BP neural network model optimized by particle swarm optimization (PSO) for calculations, water content, construction method type, support conditions, tunnel depth, dip angle of main structural surfaces, rock layer thickness, and lateral pressure coefficient are selected as input parameters. These input parameters are subjected to minimax regularization to map them to a specific numerical range. The weights and biases of the BP neural network are used as particle positions in the PSO algorithm, and the prediction error is used as the fitness function. The global optimal weights and biases are searched by iteratively updating particle velocity and position. The optimal weights and biases are assigned to the BP neural network and trained using gradient descent. The normalized input parameters are then imported into the trained model, outputting the predicted cumulative deformation of the surrounding rock. This model employs a sliding window-based update mechanism: after completing the construction and monitoring of a standard excavation cycle segment, the measured geological parameters and stable cumulative deformation of that segment are stored as new sample pairs in the training set. The earliest old samples are removed chronologically to maintain a constant total sample size and activate the retraining process.

[0016] When determining the final comprehensive large deformation level for guiding construction, the maximum initial in-situ stress at the measuring points was obtained using the hydraulic fracturing method. Rock mass strength was then calculated through point load tests, and the ratio of rock mass strength to the maximum initial in-situ stress was used as the rock mass strength-stress ratio. Criteria for judging slight, moderate, severe, and extremely severe large deformations corresponding to different numerical ranges of the rock mass strength-stress ratio were established. If the level determined based on the strength-stress ratio differed from the preliminary theoretical level, the theoretical boundary was corrected using the measured strength-stress ratio results. Finally, the preliminary theoretical level, the apparent large deformation level, and the level corresponding to the predicted cumulative deformation of the surrounding rock were converted into numerical quantities. These converted numerical quantities were then weighted, summed, and rounded to obtain the comprehensive large deformation level. If the measured deformation rate of the tunnel exceeded the rapid deformation threshold or the cumulative deformation reached the extremely severe standard, it was directly judged as an extremely severe large deformation.

[0017] When implementing construction control measures, the comprehensive large deformation level is mapped to specific reserved deformation values ​​and support parameters. When an extremely severe large deformation is determined, the reserved deformation value is taken as the product of the predicted cumulative deformation of the surrounding rock and the safety reserve coefficient, wherein the safety reserve coefficient ranges from 1.15 to 1.20. The dynamic adjustment mechanism includes: increasing the inspection frequency when the monitoring data reaches the first-level warning threshold; forcibly stopping work and implementing radial grouting when the second-level warning threshold is reached; if the deformation rate still exceeds the convergence criterion after the reinforcement support is applied, an upgrade warning is automatically triggered, and the comprehensive large deformation level is upgraded by one level.

[0018] This invention provides a method for evaluating large deformation in soft rock tunnels based on multi-source parameter analysis. It has the following beneficial effects:

[0019] 1. This invention combines a theoretical evaluation model based on design parameters with a quantitative scoring method based on tunnel face characteristics to construct a multi-stage evaluation system. The theoretical risk level of the surrounding rock is determined through fuzzy interactive structural modeling and grey clustering algorithms, and quantitative scoring is performed using seven apparent factors measured in the field. This complementary approach of theoretical calculation and field measurement eliminates the limitations of traditional single-engineering analogy methods, reduces evaluation bias caused by differences in human experience, and improves the objectivity of determining the large deformation level of soft rock.

[0020] 2. This invention utilizes a BP neural network model optimized by particle swarm optimization and a sliding window update mechanism for the sample library to achieve quantitative calculation of the cumulative deformation of the surrounding rock. For extremely severe large deformation conditions, the technical solution uses model predictions combined with a safety reserve coefficient to determine the construction allowance deformation, replacing the commonly used fixed empirical values. This allows the allowance deformation parameter to adapt to the specific rheological characteristics of the surrounding rock in the current site, reducing encroachment and arch replacement due to insufficient allowance or over-excavation waste due to excessive allowance.

[0021] 3. This invention establishes a dynamic hierarchical control mechanism based on the measured verification of rock mass strength-stress ratio and feedback from monitoring data. This mechanism uses on-site hydraulic fracturing test data to correct the boundaries of theoretical evaluation results and defines two levels of early warning thresholds and mandatory work stoppage standards during the construction phase. This allows the comprehensive large deformation level to be dynamically adjusted according to the actual mechanical response of the surrounding rock, ensuring that the support strength matches the deformation pressure of the surrounding rock and reducing the risk of structural instability during soft rock construction. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the overall process of an embodiment of the present invention.

[0023] Figure 2 This is the overall flowchart of the FISM-ANP of the present invention;

[0024] Figure 3This is the error reduction and error change curve of the PSO-BP algorithm of the present invention;

[0025] Figure 4 This is the fitness value variation curve of the PSO-BP algorithm of the present invention;

[0026] Figure 5 This is a line graph comparing the predicted and actual values ​​of the present invention.

[0027] Figure 6 This is a scatter plot comparing the predicted and actual values ​​of the present invention. Detailed Implementation

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] See attached document Figure 1 This invention provides a method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis. This method integrates theoretical assessment models, on-site appearance characteristics, intelligent prediction algorithms, and measured verification data to achieve comprehensive identification and dynamic management of large deformation in soft rock.

[0030] Includes the following steps:

[0031] S100, construct a theoretical evaluation model for large deformation of soft rock, and conduct preliminary risk classification of the surrounding rock of the tunnel based on geological parameters obtained during the design stage or in advance prediction;

[0032] S200: After the tunnel face is exposed during excavation, quantitative scoring is performed based on the surface feature data of the tunnel face collected on site to determine the apparent large deformation level of the surrounding rock at the tunnel face.

[0033] S300 collects multi-source parameters of the tunnel excavation section, uses a BP neural network model optimized based on particle swarm optimization to calculate and output the predicted value of the maximum deformation of the surrounding rock.

[0034] S400, based on the in-situ ground stress test, rock point load test and monitoring measurement data, calculates the rock mass strength stress ratio, and determines the final large deformation level by combining the evaluation results of the aforementioned steps;

[0035] S500 involves implementing corresponding construction control measures based on the determined final large deformation level, and dynamically adjusting the level based on the structural response and monitoring data during construction.

[0036] In step S100, the uniaxial saturated compressive strength of rock, maximum initial ground stress, groundwater development, tunnel span, tunnel depth, the angle between the main structural plane and the tunnel axis, the dip angle of the main structural plane, and the support stiffness are selected as evaluation indicators. The fuzzy interactive structural modeling and network analysis method (FISM-ANP) is used to analyze the mutual influence relationships among the evaluation indicators and calculate the weight vector of each indicator. Combining grey clustering assessment theory, a whitening weight function is constructed and the clustering coefficient is calculated. Based on the principle of maximizing the clustering coefficient, the risk of large deformation in soft rock is divided into four preliminary theoretical levels: slight, moderate, severe, and extremely severe. This step provides a basic risk warning for subsequent construction.

[0037] In step S200, for the surrounding rock of the excavated face, metamorphic sandstone with high basic strength and sections that do not meet the characteristics of large deformation of soft rock are excluded. A scoring system is established, including seven factors: strike angle, stratum dip angle, stratum thickness, integrity, interlayer bonding, tectonic influence, and groundwater status. Corresponding scores are assigned to each factor based on the field measurements, and the total score is calculated according to the weight of each factor. The apparent large deformation level of the surrounding rock of the face is determined based on the range of the total score (40-60 points, 60-75 points, 75-90 points, 90-100 points). For gently dipping strata with a dip angle of less than 30 degrees and a ground stress greater than 10 MPa, if there are extremely thin strata or severely fractured structures, the level is directly determined according to specific rules.

[0038] In step S300, multi-source input parameters of the current excavation section are obtained, including water content, construction method type, support conditions, tunnel depth, structural surface dip angle, rock layer thickness, and lateral pressure coefficient. These input parameters are then subjected to Min-Max normalization, mapping the data to the [-1,1] interval. A BP neural network model is constructed, and the initial weights and biases of the BP neural network are iteratively optimized globally using the Particle Swarm Optimization (PSO) algorithm until convergence is satisfied. The processed input parameters are imported into the trained PSO-BP neural network model, outputting the predicted value of the maximum deformation of the surrounding rock, which serves as a quantitative auxiliary classification basis.

[0039] In step S400, in-situ hydraulic fracturing stress testing and rock point load testing are conducted to obtain the maximum principal stress value and rock mass compressive strength value, and the rock mass strength-stress ratio is calculated. A comprehensive identification standard including the strength-stress ratio, relative deformation, and surrounding rock failure characteristics is established. When the rock mass strength-stress ratio falls within the intervals of [0.25, 0.5], [0.15, 0.25], [0.05, 0.15], and [0, 0.05], it is successively identified as a slight, moderate, severe, and extremely severe deformation level. If on-site stress testing conditions are lacking, the deformation rate index obtained from monitoring measurements is used to identify the severe deformation level according to the set rate threshold interval. At the same time, the apparent level from step S200 and the predicted deformation from step S300 are combined to comprehensively determine the final severe deformation level guiding construction.

[0040] In step S500, based on the determined final large deformation level, the corresponding reserved deformation amount is set, excavation method is selected, support parameters are implemented, and monitoring and measurement frequency is set. During construction, the integrity and deformation rate of the initial support structure are continuously monitored. When it is observed that the initial support exhibits continuous length exceeding the set threshold for peeling, cracking, or steel frame twisting, or when the cumulative deformation exceeds the set proportion of the control benchmark value when the invert arch forms a ring, upgrade management is implemented, increasing the deformation level and strengthening support measures. When the lithology and integrity indicators of the working face and the surrounding rock in front continuously improve for a length exceeding the set limit, and the monitoring and measurement data are normal, downgrade management is implemented, reducing the large deformation level or canceling the large deformation measures.

[0041] See attached document Figure 2 In step S100, a theoretical assessment model for large deformation of soft rock is constructed. Based on geological parameters obtained during the design phase or in advance prediction, a preliminary risk classification of the tunnel surrounding rock is performed. This includes the following sub-steps:

[0042] S101, Determine the index system for assessing large deformation in soft rock. Based on the formation mechanism and influencing factors of large deformation in soft rock, eight assessment indicators are selected: uniaxial saturated compressive strength R of the rock. c Maximum initial ground stress σ max Groundwater development, tunnel span, tunnel depth, angle between the main structural planes and the tunnel axis, dip angle of the main structural planes, and support stiffness K. s For the qualitative indicators, pre-set quantitative evaluation standards.

[0043] For example, the development of groundwater can be assigned values ​​of 1, 2, 3, and 4 according to whether it is dry, wet, dripping, or gushing.

[0044] The support stiffness is converted into an equivalent elastic modulus value based on the design parameters.

[0045] S102, a method combining fuzzy interactive structural modeling (FISM) and analogous network analysis (ANP) is used to determine the weights of each evaluation indicator. An evaluation indicator set including the above eight indicators is established. A fuzzy adjacency matrix is ​​constructed based on the mutual influence relationships between the indicators, using expert scoring or on-site engineering experience.

[0046] The elements in the matrix represent the membership degree of the direct influence between indicators, with values ​​ranging from 0 to 1.

[0047] S103. Calculate the reachability matrix and divide the hierarchy based on the fuzzy adjacency matrix. Using Boolean matrix operations, add the fuzzy adjacency matrix to the identity matrix, and then perform continuous self-multiplication on the resulting matrix until the matrix state no longer changes, thus obtaining the reachability matrix. Based on the reachability matrix, determine the reachable set and antecedent set for each indicator. By identifying the intersection elements of the reachable set and the antecedent set, divide the indicator system into a hierarchical structure of different levels from top to bottom.

[0048] S104, construct the ANP network structure model and calculate the weight vector. Based on the hierarchical structure and feedback relationships between indicators determined in S103, construct the network analysis structure diagram. Construct the judgment matrix through pairwise comparisons. Before calculating the eigenvectors, to ensure the self-consistency of the expert scoring logic, the judgment matrix must be checked for consistency. Define the consistency ratio CR, and its calculation formula is:

[0049]

[0050] In the formula, CI is the consistency index, and λ max To determine the largest eigenvalue of a matrix, k is the order of the matrix, and RI is the average random consistency index (obtained from a table based on the matrix order, e.g., when k=3).

[0051] (RI = 0.52). When the calculated CR < 0.1, the judgment matrix is ​​deemed to have passed the consistency test; otherwise, the relative importance scores of the indicators need to be readjusted.

[0052] After the test is passed, eigenvectors are calculated to form an unweighted hypermatrix, and then the limiting hypermatrix is ​​obtained through exponentiation to determine the global weight w of each index. j .

[0053] S105. Construct a grey clustering assessment model. Divide the risk level of large deformation in soft rock into four grey categories: slight (Level I), moderate (Level II), severe (Level III), and extremely severe (Level IV). Determine the whitening weight function for each indicator with respect to each grey category. The whitening weight function describes the degree to which an indicator value belongs to a certain risk level, and adopts an ascending semi-trapezoidal, descending semi-trapezoidal, or trapezoidal distribution form. Set a quantitative grading threshold, using the rock saturated compressive strength R... cFor example, we set corresponding risk thresholds for each level. When the measured value falls into a certain range, we calculate its membership value as a slight, moderate, strong, or extremely strong risk using a whitening weight function.

[0054] S106, calculate the grey clustering coefficient for each evaluation object. For each evaluation object, obtain its measured index value x. ij Substituting these values ​​into the whitening weight function constructed in step S105, we obtain the membership values ​​of each indicator to different risk levels. Then, we correlate the membership values ​​of each indicator with their corresponding weights w. j By performing a weighted summation, we obtain the clustering coefficient σ for the assessed object to belong to the four risk levels. k .

[0055] S107, Determine the preliminary risk level. Compare the four gray clustering coefficients calculated in step S106. Based on the principle of maximum membership, select the risk level corresponding to the clustering coefficient with the largest value as the preliminary theoretical level of large deformation of soft rock for the assessment object.

[0056] In step S200, after the tunnel face is exposed during excavation, a quantitative score is performed based on the surface feature data of the tunnel face collected on-site to determine the apparent large deformation level of the surrounding rock at the tunnel face. This specifically includes the following sub-steps:

[0057] S201, conduct preliminary screening of the suitability of the surrounding rock at the tunnel face. Before quantitative scoring, conduct on-site rapid testing of the rock mineral composition and basic mechanical properties of the lithology revealed by the current excavation. If the revealed rock mass is intact metamorphic sandstone, granite, or a hard rock group with a uniaxial saturated compressive strength greater than a preset threshold (e.g., 30 MPa) that has not been severely damaged by tectonic activity, it is determined that this section does not have the material basis for large deformation of soft rock, the current large deformation assessment process is terminated, and conventional hard rock tunnel construction proceeds. If the revealed rock mass is a soft rock group such as phyllite, carbonaceous shale, slate, or mudstone, or a rock mass that, although the original rock strength is high, is extremely fragmented by tectonic activity into mylonitic or pulverized form, proceed to the subsequent quantitative scoring step.

[0058] S202, establish a seven-factor scoring system for the surrounding rock at the tunnel face. Seven apparent characteristics affecting the stability of the surrounding rock are selected as scoring factors: angle α1 between the strike of the main structural plane and the tunnel axis, dip angle α2 of the main structural plane, layer thickness α3, rock mass integrity α4, interlayer bonding degree α5, degree of influence of geological structure α6, and groundwater exposure state α7. The total baseline score of the seven-factor scoring system is set at 100 points, and each factor is assigned an upper limit of score according to its weight in the stability of the surrounding rock. In this embodiment, the full scores of α1 to α7 are set as follows: 15 points, 10 points, 15 points, 20 points, 15 points, 15 points, and 10 points, respectively.

[0059] S203, Collect apparent characteristic data and calculate the total score. Specific indicators for each factor are obtained using a geological compass, steel tape measure, and on-site observation methods, and corresponding scores are assigned according to a pre-set scoring standard table. To eliminate the arbitrariness of human judgment, this embodiment has developed a standardized "Appearance Characteristic Scoring Table for Large Deformation of Soft Rock at the Working Face," as shown in Table 1. On-site technicians determine the specific score based on the measured data falling within the specific characteristic intervals in Table 1 using linear interpolation.

[0060] Table 1: Scoring Table for Apparent Characteristics of Large Deformation in Soft Rock at the Working Face

[0061]

[0062]

[0063]

[0064] To comprehensively consider the cumulative effect of various influencing factors, the total score Q of the apparent characteristics of the surrounding rock at the tunnel face is calculated using a weighted summation model, and the calculation formula is as follows:

[0065]

[0066] In the formula: W is the raw score obtained from column 4 of the i-th factor comparison table; i The values ​​in column 2 of Table 1 represent the weighting coefficients; Sα1, Sα2, Sα3, Sα4, Sα5, Sα6, and Sα7 represent the original scores of the seven factors: strike angle, dip angle, layer thickness, integrity, cohesion, tectonic influence, and groundwater, respectively; 0.34, 0.22, ..., 0.03 are the weighting coefficients for each factor.

[0067] The total score Q calculated using this formula is the sole quantitative basis for determining the apparent large deformation level in subsequent S204.

[0068] S204, the apparent large deformation level is determined based on the comprehensive score. A mapping relationship is established between the comprehensive score Q and the large deformation level of soft rock. Four classification intervals are defined:

[0069] When 90≤Q≤100, it is judged as a minor large deformation (Level I);

[0070] When 75≤Q<90, it is judged as medium-large deformation (Level II);

[0071] When 60≤Q<75, it is judged as a severe large deformation (Level III);

[0072] When Q < 60, it is judged as extremely severe large deformation (Level IV).

[0073] The lower the score, the more unfavorable the development of the surrounding rock structure, the worse its integrity, the more severe the influence of water and tectonics, and the higher the risk of large deformation.

[0074] S205 involves a grade correction for special gently dipping rock strata conditions. Based on the preliminary assessment in step S204, it checks for the existence of specific extremely hazardous conditions. When the measured rock strata dip angle is less than 30°, and the on-site in-situ stress test shows a maximum principal stress exceeding 10 MPa or a burial depth exceeding 300 m, if it is accompanied by an extremely thin layered structure with a rock strata thickness of less than 10 cm, or if the rock mass integrity index Kv calculated by wave velocity testing is <0.3, then the apparent large deformation grade of this segment is directly determined to be extremely severe (Level IV) without relying on the score result of step S204. This correction step is used to compensate for the assessment bias of the linear cumulative scoring method under nonlinear instability modes by identifying the specific hazardous mode of high stress + gently dipping angle + thin soft rock.

[0075] In step S300, a backpropagation (BP) neural network model optimized by the particle swarm optimization (PSO) algorithm is established. Using the geological parameters and monitoring data obtained in the aforementioned evaluation stage, the deformation of the tunnel surrounding rock is nonlinearly predicted. This specifically includes the following sub-steps:

[0076] S301, construct the input and output sample sets for the prediction model and perform normalization processing. Collect the geological parameters of the surrounding rock in the excavated section and the corresponding historical deformation monitoring data as the original database. Given the significant differences in the dimensions of different input factors such as rock strength (MPa level), tunnel depth (hundreds of meters level), and layer thickness (centimeter level), directly inputting them into the neural network would lead to convergence difficulties. Therefore, the Min-Max regularization method is used to map all input data to the [-1,1] interval. For any input feature x, its normalized value x normalize The calculation formula is:

[0077]

[0078] Alternatively, the standard formula mapping to the [0,1] interval can be used:

[0079]

[0080] In the formula, x min and x max These are the minimum and maximum values ​​of the feature in the sample set, respectively.

[0081] S302, determine the topology of the BP neural network. The number of input layer nodes is determined based on the number of input variables, and the number of output layer nodes is determined based on the output variables (i.e., cumulative deformation). The number of hidden layer nodes is set using a trial-and-error method; that is, within a range determined by empirical formulas, the number of nodes is gradually increased or decreased during pre-training, and the network convergence error under different node numbers is compared. The node number corresponding to the minimum error is selected as the final structure. The hidden layer transfer function is set to a hyperbolic tangent sigmoid function, and the output layer transfer function is set to a linear function.

[0082] S303, Initialize Particle Swarm Optimization (PSO) Algorithm Parameters. Construct a particle encoding mapping mechanism, arranging the connection weights and neuron thresholds across all layers of the BP neural network in a predetermined order to form a high-dimensional vector. This vector represents the position of a particle in the PSO algorithm. The particle dimension is equal to the total number of all weights and thresholds in the network. Set the population size, maximum number of iterations, and learning factor, and initialize the random positions and random velocities of each particle.

[0083] See attached document Figure 4 S304 defines and calculates the particle's fitness function. The particle's position vector is decoded and restored to the specific weights and threshold parameters of a backpropagation (BP) neural network. Forward propagation is performed on the network with these parameters using the training set data to obtain the predicted deformation value. The mean square error (MSE) between the predicted deformation value and the actual monitored deformation value is calculated, and this MSE is directly used as the particle's fitness value.

[0084] S305 executes particle velocity and position updates. In each iteration, the particle adjusts its velocity and position based on its own historical best position (individual extreme) and the historical best position of the entire population (global extreme). The specific adjustment logic is as follows: the current velocity is multiplied by the inertia weight, plus the individual cognitive component (the vector pointing from the current position to the individual extreme), plus the social cognitive component (the vector pointing from the current position to the global extreme), and the weighted sum of these three factors constitutes the new velocity; the current position plus the new velocity yields the updated position. During the update process, if the particle's position or velocity exceeds the preset boundary limit, it is forcibly pulled back to the boundary value.

[0085] See attached document Figure 3 In step S306, the optimal weights are obtained and the BP network is trained. When the number of iterations reaches the upper limit or the fitness value meets the convergence accuracy, PSO optimization is stopped. The position vector corresponding to the globally optimal particle is extracted and assigned to the BP neural network as the initial weights and thresholds for optimization. Based on this, the Levenberg-Marquardt algorithm is used to continue backpropagation training of the network, and gradient descent is used to fine-tune the weights and thresholds until the overall network error reaches the preset standard.

[0086] See attached document Figure 5 - Appendix Figure 6 S307, Output the prediction result and denormalize it. The prediction result output by the network is a normalized value. A linear transformation opposite to step S301 needs to be performed. Using the maximum and minimum values ​​of the output variables in the training set, the prediction result is restored to a millimeter-level deformation with actual physical dimensions. The final predicted cumulative deformation is compared with a preset deformation management level threshold. If the predicted value exceeds the set threshold (e.g., 80% of the reserved deformation), an early warning signal is triggered.

[0087] In step S400, a multi-source data fusion and verification mechanism is established to integrate the theoretical risk level, apparent large deformation level, and AI prediction results obtained from the previous steps, and to correct for deviations using field measurement data, ultimately determining the comprehensive large deformation level of the surrounding rock. This includes the following sub-steps:

[0088] S401 involves the field measurement verification of key parameters related to ground stress and large deformation. After tunnel excavation, a typical cross-section is selected for refined ground stress testing. The maximum initial ground stress σ at the measuring points is obtained using either the hydraulic fracturing method or the hollow inclusion stress relief method. max Simultaneously, point load tests were conducted on the rock mass to calculate the rock mass strength R. m (Or use uniaxial compressive strength). Calculate the rock mass strength-stress ratio R. m / σ max In conjunction with the deformation characteristics of the surrounding rock revealed on site, the measured grade was verified according to the "Classification Criteria for Large Deformation of Soft Rock" shown in Table 2.

[0089] Table 2. Criteria for classifying large deformations in soft rock (based on strength-stress ratio verification)

[0090]

[0091]

[0092]

[0093] If there is a deviation between the measured and calculated strength-stress ratio and the theoretical assessment level, the boundary conditions of the theoretical model shall be corrected according to the strength-stress ratio criterion in Table 2. In particular, when the measured local stress value is between 8 and 10 MPa and the stress ratio meets the large deformation condition, considering the rheological effect of soft rock, the large deformation level can be reduced by one level according to the standard in Table 2 (e.g., from strong to moderate) to avoid over-support.

[0094] S402. Conduct real-time monitoring and inversion analysis of surrounding rock deformation. Using a total station or 3D laser scanner, monitoring points are deployed to acquire convergent deformation data around the tunnel. Based on the changes in the observed side length at different times, the relative deformation around the tunnel and the deformation rate per unit time are calculated. The measured cumulative deformation is compared with the predicted value output by the neural network in the previous steps. If the absolute error between the two exceeds a preset allowable range (e.g., 10 mm), it indicates that the current geological conditions have changed unexpectedly. In this case, the measured geological parameters and corresponding deformation data of this segment are used as new training samples and incorporated into the original neural network sample database for incremental training of the model, thereby improving the accuracy of predictions for subsequent segments.

[0095] S403, Construct a comprehensive grading model integrating multi-source information. To address potential inconsistencies in results from single-method assessments, a weighted voting logic is used to determine the final major deformation level. First, a grading quantification mapping relationship is established, assigning values ​​1, 2, 3, and 4 to Level I (Slight), Level II (Moderate), Level III (Strong), and Level IV (Extremely Strong), respectively. The theoretical assessment level, apparent identification level, and AI prediction level are then converted into corresponding numerical values.

[0096] A veto mechanism based on measured data is implemented: It checks whether the measured deformation rate exceeds the abrupt deformation threshold (e.g., 10 mm / d) or whether the cumulative deformation has reached Level IV. If either condition is met, no weighted calculation is performed, and the overall large deformation level is directly determined to be Level IV (extremely severe).

[0097] If the veto mechanism is not triggered, the weighted sum of the grade values ​​from the three sources is used to calculate the comprehensive large deformation index. The weighting principle is as follows: the apparent judgment of the field measurement data is given the highest weight (e.g., 0.4), and theoretical assessment and AI prediction are given auxiliary weights (e.g., 0.3 each). The calculation result is rounded to the nearest integer, and the rounded value is restored to the corresponding Roman numeral grade (I-IV) as the final comprehensive large deformation grade.

[0098] S404, Implement a graded support dynamic adjustment strategy. Based on the comprehensive large deformation level determined in step S403, match the corresponding specific support parameters:

[0099] When the classification is Level I, the original design parameters are maintained, and only routine monitoring is performed.

[0100] When the classification is Level II, reinforced support is implemented, including reducing the system anchor spacing (e.g., from 1.2m to 1.0m) and increasing the thickness of the shotcrete.

[0101] When the classification is Level III, in addition to Level II, a steel frame or grid frame is added, with the support spacing controlled between 0.8m and 1.0m, and advanced small-diameter pipe grouting is carried out, covering a 120-degree range of the arch, in order to pre-reinforce the surrounding rock in front.

[0102] When the risk level is determined to be Level IV, the extremely high risk response plan is activated, the excavation method is adjusted to the reserved core soil method or the double-side wall pilot tunnel method, long pipe roof advanced support is constructed in conjunction with advanced curtain grouting, and a retractable steel frame is installed to adapt to the large deformation release of the surrounding rock and prevent the rigid support from breaking.

[0103] S405 establishes closed-loop control through continuous feedback of deformation monitoring data. After implementing the appropriate level of support measures, the deformation rate in step S402 is continuously monitored. A convergence criterion is set (e.g., a deformation rate less than 0.2 mm / d for three consecutive days). If, after implementing reinforced support, the deformation rate during the continuous monitoring period still exceeds the convergence criterion, or if cracking and spalling of the shotcrete occur, the current comprehensive level assessment is determined to be too low. An upgrade warning is automatically triggered, directly raising the comprehensive level by one level, and the support reinforcement measures in step S404 are re-implemented according to the higher-level standard, even including radial grouting backfilling, until the surrounding rock deformation tends to stabilize.

[0104] In step S500, a dynamic hierarchical management system covering the entire lifecycle is established. The aforementioned assessment and grading results are transformed into specific construction control parameters. Through continuous monitoring data feedback, the system model parameters and on-site construction measures are subject to bidirectional correction and closed-loop control. This includes the following sub-steps:

[0105] S501 constructs a dynamically updated geological information database for large deformations in soft rock. Using tunnel mileage markers as index keys, it structurally stores the geological records exposed at the tunnel face, rock mechanical parameters obtained from experimental tests, convergence and in-situ stress data acquired through field monitoring, and large deformation rating results for each stage. A data variability verification logic is implemented. When the tunnel face advances to a new mileage segment, the geological parameters at that location are retrieved and compared with historical data from adjacent excavated sections. If the change in lithology or rock mass integrity index exceeds a preset variability threshold, the system will specially mark that segment in the database and output a signal prompting technicians to re-verify the input parameters to prevent model input distortion due to sudden geological changes.

[0106] S502, determine the reserved deformation amount and construction step distance based on the comprehensive grading results. Map the comprehensive large deformation level output in step S403 to specific geometric dimension control parameters and construction procedure parameters.

[0107] The specific implementation logic is as follows:

[0108] For Level I (minor): the allowable deformation is set between 50mm and 100mm, the full-section method is used for construction, and the single excavation advance is controlled within 3.0m.

[0109] For Class II (Medium): The allowable deformation is set between 100mm and 150mm. The step method is adopted for construction, and the length of the upper step needs to be controlled at about 1 times the tunnel diameter.

[0110] For Level III (Strong): the allowable deformation is set between 150mm and 250mm, the short bench method is adopted, the excavation advance is limited to 1.0m to 1.5m, and it is stipulated that the distance between the initial support arch and the working face shall not exceed 20m, so as to close the support ring as soon as possible.

[0111] For Level IV (extremely severe): The logic for setting the allowable deformation amount is no longer based on fixed empirical values, but rather on dynamic calculations based on AI prediction results. Specifically, the predicted cumulative deformation amount output in step S300 is multiplied by a safety reserve factor (e.g., 1.15 to 1.20) to obtain the final allowable amount, and the lower limit of this allowable amount is not less than 300mm. At this time, the excavation advance is strictly controlled between 0.6m and 1.0m, and a close follow-up support procedure is implemented, excavating one steel frame and then supporting one steel frame.

[0112] S503 involves identifying and immediately correcting deviations during construction. During construction, daily measured arch settlement and surrounding convergence values ​​are collected and compared with two pre-defined control thresholds. The first-level warning threshold is set at 70% of the design allowable deformation, and the second-level warning threshold is set at 90% of the design allowable deformation.

[0113] When the monitoring data reaches the first-level warning threshold, the second-level response is activated, increasing the frequency of inspections of the appearance of the support structure, especially cracks on the surface of the shotcrete, and doubling the monitoring frequency.

[0114] When the monitoring data reaches the second-level warning threshold, a first-level response is initiated, a mandatory work stoppage order is executed, tunneling at the working face is halted, and personnel are evacuated. Subsequently, radial grouting reinforcement or additional long anchor cables are installed in the rapidly deforming sections until the deformation rate during the continuous monitoring period returns to a stable standard (e.g., less than 0.5 mm / d). Only then can the work stoppage order be lifted and tunneling resume.

[0115] S504, Implement adaptive incremental learning and parameter iteration of the prediction model. Establish a sliding window update mechanism for the sample library, and the neural network parameters in step S300 dynamically evolve with the construction progress. Each time a standard excavation cycle segment (e.g., 50m) is completed and monitored, the actual geological parameters exposed in that segment are used as input, and the final measured stable cumulative deformation is used as output to construct a new standard sample pair. The new samples are stored in the training set, and the earliest old sample is removed in chronological order, keeping the total number of samples constant while continuously updating the content. The updated sample set is used to activate the retraining process of the neural network, and the network weights are fine-tuned using gradient descent. Through this mechanism, the prediction model can continuously adapt to the gradual changes in geological conditions along the tunnel route, correcting errors caused by geological environment differences due to increased mileage.

[0116] S505 generates a visual guidance plan and archived records. The determined grading results, calculated reserved deformation values, matched support parameters, and triggered early warning information are displayed through a graphical user interface. An engineering report generation module is constructed, summarizing the excavation mileage for the day, surrounding rock grade identification results, measured deformation trend curves, and suggested construction parameters for the next day. This report data is synchronously stored on the server as the basis for completion and delivery documentation and subsequent comparative analysis of similar projects. The specific database construction instructions and data storage and retrieval interfaces are standard techniques in the field of computer data processing; those skilled in the art can implement them using general relational databases and programming interfaces, and will not be elaborated upon here.

Claims

1. A method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis, characterized in that, Includes the following steps: S100, construct a theoretical evaluation model for large deformation of soft rock, and conduct preliminary risk classification of the surrounding rock of the tunnel based on the geological parameters obtained in the design stage to obtain a preliminary theoretical level; S200: After the tunnel face is exposed during excavation, data on the appearance characteristics of the tunnel face are collected, and a comprehensive score is calculated using a quantitative scoring method. The apparent large deformation level of the surrounding rock at the tunnel face is determined based on the numerical range of the comprehensive score. S300 collects multi-source parameters from the tunnel excavation section to construct an input dataset, and uses a BP neural network model optimized by particle swarm optimization to calculate and output the predicted cumulative deformation of the surrounding rock. S400, based on the in-situ ground stress test and rock point load test data, calculates the rock mass strength stress ratio, and combines the preliminary theoretical level, the apparent large deformation level and the predicted cumulative deformation of the surrounding rock to determine the final comprehensive large deformation level to guide construction; S500, implement corresponding construction control measures based on the comprehensive large deformation level, and dynamically adjust the comprehensive large deformation level based on monitoring data during construction.

2. The method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that, S100 specifically includes: An evaluation index system was established, and the following were selected as evaluation indicators for constructing the theoretical evaluation model of large deformation of soft rock: uniaxial saturated compressive strength of rock, maximum initial ground stress, groundwater development, tunnel span, tunnel depth, angle between the main structural plane and the tunnel axis, dip angle of the main structural plane, and support stiffness. The hierarchical structure of the evaluation indicators is established using fuzzy interactive structure modeling and network analysis, and the weight vector of each evaluation indicator is calculated. Based on the grey clustering evaluation theory, a whitening weight function is constructed to describe the evaluation index belonging to different risk levels. The measured index data is substituted into the whitening weight function to calculate the membership degree, and the clustering coefficient of each risk level is calculated by combining the weight vector. Based on the principle of maximum membership, the risk level corresponding to the clustering coefficient with the largest value is selected as the preliminary theoretical level.

3. The method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that, The specific steps of S200 include: The seven apparent characteristic data, including the angle between the main structural plane and the tunnel axis, the dip angle of the main structural plane, the thickness of the rock strata, the integrity of the rock mass, the degree of interlayer bonding, the degree of influence of geological structure, and the state of groundwater exposure, are established as the scoring factors of the quantitative scoring method. Based on the on-site monitoring data and referring to the preset scoring standard table, each of the scoring factors is assigned a score; The scores of each of the scoring factors are weighted and summed with their corresponding weight coefficients to obtain the comprehensive score used to characterize the state of the tunnel face. The comprehensive score is compared with four preset grading intervals. When the comprehensive score falls into the first interval, the second interval, the third interval, and the fourth interval, the apparent large deformation level is determined to be slight large deformation, moderate large deformation, severe large deformation, and extremely severe large deformation, respectively.

4. The method for evaluating large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 3, characterized in that, Step S200 also includes a correction step for the gently dipping strata: If the dip angle of the rock strata at the tunnel face is less than 30 degrees and the maximum initial ground stress is greater than 10 MPa or the tunnel depth exceeds 300 meters, an additional judgment is executed. If the rock layer thickness is detected to be less than 10 cm or the rock mass integrity index is less than 0.3, then the judgment based on the comprehensive score will be stopped, and the apparent large deformation level will be directly judged as extremely severe large deformation.

5. The method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that, In step S300, the specific steps for calculation using the BP neural network model optimized by the particle swarm optimization algorithm include: Moisture content, construction method type, support conditions, tunnel depth, dip angle of main structural surfaces, rock layer thickness, and lateral pressure coefficient are selected as input parameters. The input parameters are mapped to a numerical range from negative 1 to positive 1 using the maximum-minimum regularization method; The weights and biases of the BP neural network are used as the particle positions in the particle swarm optimization algorithm, and the prediction error is used as the fitness function. By iteratively updating the particle velocity and position, the globally optimal weights and biases are searched. The optimal weights and biases are assigned to the BP neural network, and the model is trained using the gradient descent method. The normalized input parameters are then imported into the trained model to output the predicted cumulative deformation of the surrounding rock.

6. The method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that, The specific steps for calculating and verifying the rock mass strength-stress ratio in step S400 include: The maximum initial ground stress at the measuring point was obtained by hydraulic fracturing. The rock mass strength was converted by point load test, and the ratio of the rock mass strength to the maximum initial ground stress was calculated as the rock mass strength-stress ratio. When the rock mass strength-stress ratio is between 0.25 and 0.5, it is determined to be a slight large deformation; When the rock mass strength-stress ratio is between 0.15 and 0.25, it is determined to be a medium-to-large deformation. When the rock mass strength-stress ratio is between 0.05 and 0.15, it is determined to be a case of severe large deformation. When the rock mass strength-stress ratio is less than or equal to 0.05, it is determined to be an extremely strong large deformation; If the grade determined based on the rock mass strength-stress ratio is inconsistent with the preliminary theoretical grade, the boundary conditions of the theoretical model shall be corrected based on the grade determined based on the rock mass strength-stress ratio.

7. The method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that, In step S400, the specific steps for determining the final comprehensive large deformation level to guide construction, combining the preliminary theoretical level, the apparent large deformation level, and the predicted cumulative deformation of the surrounding rock, include: The preliminary theoretical level, the apparent large deformation level, and the level corresponding to the predicted cumulative deformation of the surrounding rock are respectively converted into numerical quantities. The three transformed numerical values ​​are weighted and summed to obtain a comprehensive index. The comprehensive index is then rounded down to restore the corresponding level, which is used as the comprehensive large deformation level. If the measured deformation rate of the tunnel is found to be greater than the threshold for rapid deformation, or if the measured cumulative deformation has reached the standard for extremely severe large deformation, then the weighted summation calculation will not be performed, and the comprehensive large deformation level will be directly determined as extremely severe large deformation.

8. The method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that, In step S500, the specific steps for implementing the corresponding construction control measures include: The comprehensive large deformation level is mapped to the reserved deformation amount and support parameters; When the comprehensive large deformation level is determined to be extremely strong large deformation, the value of the reserved deformation amount is no longer a fixed empirical value, but the product of the predicted cumulative deformation amount of the surrounding rock and the safety reserve coefficient is taken as the final reserved deformation amount. The safety reserve coefficient ranges from 1.15 to 1.

20.

9. The method for evaluating large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that: In step S500, the specific steps for dynamically adjusting the overall large deformation level include: Set a first-level warning threshold and a second-level warning threshold; When the monitoring data reaches the first-level warning threshold, the frequency of visual inspection of the support structure is increased and the monitoring frequency is doubled. When the monitoring data reaches the second-level warning threshold, a forced shutdown order is executed to stop the tunneling face and radial grouting reinforcement is carried out on the section with rapid deformation until the deformation rate drops back to the stable standard before tunneling resumes. If the deformation rate during the continuous monitoring period after the reinforcement is applied still exceeds the convergence criterion, an upgrade warning will be automatically triggered, and the comprehensive large deformation level will be directly increased by one level.

10. The method for assessing large deformation in soft rock tunnels based on multi-source parameter analysis according to claim 1, characterized in that, It also includes a model update step based on the newly added data: Establish a sliding window update mechanism for maintaining the training set, and perform the following operations based on the sliding window update mechanism: After the construction and monitoring of a standard excavation cycle are completed, the actual geological parameters exposed in the excavation cycle are used as input, and the measured stable cumulative deformation is used as output to construct a new sample pair. The new sample pairs are stored in the training set, and the oldest sample is removed in chronological order to keep the total number of samples in the training set constant. The retraining process of activating the BP neural network model using the updated training set.

Images