Tunnel surrounding rock quality probabilistic evaluation method integrating multi-source heterogeneous geological exploration information

Abstract

The application discloses a tunnel surrounding rock quality probabilistic evaluation method for calculating multi-source heterogeneous geological exploration information, relates to the technical field of tunnel and underground engineering construction geological evaluation, and comprises the following steps: collecting multi-source geological prediction parameters and measured BQ values of a detection section, and constructing a sample data set through preprocessing and spatial matching; a nonlinear mapping model and a probability distribution model from multi-index parameters and their fusion features to BQ values are established based on the data set to randomly generate virtual input feature samples; the samples are input into the nonlinear mapping model to obtain BQ prediction values; and reliability probability analysis is performed in combination with a surrounding rock grade function to obtain probabilistic evaluation results of each surrounding rock grade. The application can comprehensively utilize multi-index information to represent different characteristics of surrounding rock, realize probability grading and risk early warning of surrounding rock quality before construction under the condition of limited samples, reduce construction safety hazards such as insufficient support and collapse, and improve the safety of tunnel construction.

Description

Technical Field

[0001] This invention relates to the field of geological evaluation technology for tunnel and underground engineering construction, and in particular to a probabilistic evaluation method for the quality of surrounding rock in tunnels based on multi-source heterogeneous geological exploration information. Background Technology

[0002] In tunnel construction, accurate prediction of basic rock quality (BQ) indicators is crucial for ensuring project safety and controlling costs. Overly optimistic or conservative predictions of BQ significantly increase construction safety risks and project costs, directly impacting the life-cycle benefits of tunnel projects. Currently, rock mass parameters are primarily obtained through advanced geological forecasting methods such as seismic wave analysis and ground-penetrating radar. However, different detection methods reflect different geological characteristics, such as wave velocity structure, fracture development, and water-bearing anomalies. A single indicator can only characterize local information about the surrounding rock and is easily affected by noise interference and human interpretation, making it difficult to accurately reflect the quality of the surrounding rock. While machine learning prediction models have improved prediction accuracy to some extent in recent years, most remain limited to deterministic predictions and have relatively simple input variables. They fail to effectively integrate multiple advanced geological detection indicators and their complementary information, making it difficult to fully characterize the multi-factor coupling characteristics of surrounding rock quality and limiting the stability and generalization ability of the models under complex geological conditions. Especially when tunnels and underground engineering projects pass through fault fracture zones, water-rich areas and other adverse geological sections, if the prediction results lack comprehensive utilization of multi-source geological information and uncertainty assessment, it is easy to misjudge the surrounding rock grade, which in turn leads to insufficient or excessive support. This makes it difficult to provide a reliable basis for construction risk warning and support parameter optimization, affecting project safety, progress and cost control. Summary of the Invention

[0003] This invention provides an intelligent-probabilistic evaluation method for tunnel surrounding rock quality based on finite sample statistics, which overcomes the technical problems of insufficient utilization of multi-source information, single deterministic prediction results and inability to assess their uncertainty, as well as poor model reliability under small sample conditions in the prediction of surrounding rock quality in existing technologies.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows: A probabilistic evaluation method for tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information includes: S1: Collect multi-source geological prediction parameters and measured BQ values ​​for each detection section and preprocess them. Spatially match the multi-source geological prediction parameters and measured BQ values ​​for each detection section to obtain a sample dataset for training. S2: Based on the sample dataset, establish a nonlinear mapping model from multi-source feature parameters to BQ values; S3: Construct a probability distribution model based on the sample dataset, randomly generate virtual input feature samples according to the probability distribution model, input the virtual input feature samples into the nonlinear mapping model, and obtain the corresponding BQ prediction value; S4: Construct a rock mass grade function, construct a joint probability density function based on a probability distribution model, and perform a reliability probability analysis of the BQ prediction value based on the rock mass grade function and the joint probability density function to obtain the reliability probability of the rock mass grade, which serves as the evaluation result of the tunnel rock mass quality.

[0005] Furthermore, based on the aforementioned sample dataset, a nonlinear mapping model from multi-source feature parameters to BQ values ​​is established, including: S21. Construct an initial nonlinear mapping model between multi-source geological prediction parameters and BQ values; S22. Construct a loss function to measure the error between the model's predicted value and the actual BQ value; S23. Minimize the loss function, train the initial nonlinear mapping model based on K-fold cross-validation and the sample dataset, obtain the optimal parameters of the initial nonlinear mapping model, and then obtain the final nonlinear mapping model.

[0006] Furthermore, an initial nonlinear mapping model between multi-source geological prediction parameters and BQ values ​​is constructed, including: The initial nonlinear mapping model between multi-source geological prediction parameters and BQ values ​​is constructed as shown in Equation (1). (1) in, For the initial nonlinear mapping model, = , This represents a set of multi-source advanced geological prediction parameters. express , , , and The original eigenvector matrix; This represents the BQ predicted value corresponding to the multi-source geological forecast parameters. This represents the weighting coefficient of each multi-source geological prediction parameter. Indicates the first The linear coefficients of each feature; Indicates the transverse wave velocity. Indicates the longitudinal wave velocity. Represents resistivity. Represents Poisson's ratio. Indicates drilling speed. Indicates the number of features; This represents the weight coefficient for each feature; , express , and The vector matrix, Reflecting the relative relationship between the velocities of longitudinal and transverse waves, It reflects the coordinated changes in resistivity and drilling speed. It reflects the modulation effect of Poisson's ratio on the propagation characteristics of longitudinal waves; and control The intensity of participation, It is a constant; Indicates feature splicing; This represents the original eigenvector matrix. With weighting coefficient Modulated auxiliary feature vector The new joint feature vector is formed by concatenating the features.

[0007] Furthermore, the loss function is constructed, including: The initial nonlinear mapping model is rewritten as a linear model, as shown in formula (2). (2) in, To augment the parameter vector, ; The loss function is constructed based on the linear model, as shown in formula (3). (3) in, The value of the loss function. This represents the total number of multi-source geological forecast parameters.

[0008] Furthermore, a probability distribution model is constructed based on the sample dataset. Virtual input feature samples are randomly generated according to the probability distribution model, and these virtual input feature samples are input into a nonlinear mapping model to obtain the corresponding BQ prediction values, including: A probability distribution model is constructed based on the sample dataset, as shown in formula (4). (4) in, It follows a multivariate normal distribution. The sample mean vector The sample covariance matrix is ​​shown in equations (5) and (6). (5) (6) in, The sample mean vector The sample covariance matrix; Virtual input feature samples are randomly generated based on a probability distribution model; The virtual input feature samples are input into the nonlinear mapping model to obtain the corresponding set of BQ prediction values, as shown in formula (7). (7) in, Indicates the first One virtual input feature sample After input, the BQ prediction value is generated by the mapping model; This is a pre-trained nonlinear mapping model.

[0009] Furthermore, a function for assessing the surrounding rock grade is constructed, and a joint probability density function is built based on a probability distribution model, including: Construct the surrounding rock grade function, as shown in formula (8). (8) The surrounding rock grade function is normalized to obtain the normalized value, as shown in formula (9). (9) in, This is the lower limit of the BQ range within which the predicted BQ value falls in the surrounding rock grade. The surrounding rock grade, Each level corresponds to a set of preset BQ value ranges; The joint probability density function is constructed based on the probability distribution model, as shown in formula (10). (10) in, The determinant of the sample covariance matrix. It is the inverse of the sample covariance matrix.

[0010] Furthermore, based on the surrounding rock grade function and joint probability density function, a reliability probability analysis of the BQ prediction value is performed to obtain the reliability probability of the surrounding rock grade, including: Based on the surrounding rock grade function and the joint probability density function, a reliable probability calculation formula for the BQ prediction value belonging to grade G is constructed, as shown in formula (11). (11) in, The probability that the BQ predicted value belongs to the G category; express The probability of; Substituting the predicted BQ value into formula (9), the normalized value of the corresponding surrounding rock grade function is solved, and the selected values ​​are then selected. The virtual input feature samples are selected and substituted into formula (10). Formula (11) is used to solve the problem and obtain the reliable probability that the BQ prediction value corresponding to each virtual input feature sample belongs to the surrounding rock grade corresponding to the BQ interval.

[0011] Beneficial effects: This invention provides a probabilistic evaluation method for the quality of surrounding rock in tunnels by comprehensively calculating multi-source heterogeneous geological exploration information. By constructing a multi-index mapping model under limited sample size conditions and conducting probabilistic evaluation of the prediction results based on multi-parameter sampling statistics, it realizes the probabilistic and rapid identification of the surrounding rock grade and its uncertainty in actual tunnel engineering.

[0012] This invention, by comprehensively evaluating multiple advanced geological exploration indicators, changes the traditional method of relying on a single or few indicators for surrounding rock determination. It can give full play to the complementary advantages of each indicator in reflecting the surrounding rock structure, fractures, water content, etc., and more comprehensively and objectively characterize the quality of the surrounding rock. It improves the accuracy, stability and reliability of surrounding rock classification and judgment, and reduces misjudgments caused by the one-sidedness, fluctuation and interpretation bias of a single indicator. It is especially suitable for the identification of surrounding rock under complex and adverse geological conditions such as fault fracture zones and water-rich areas, and provides a more reliable basis for support design and construction safety control. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart of the method for probabilistic evaluation of tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information provided by the present invention; Figure 2 This is a BQ value distribution diagram according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the probability confidence interval for the surrounding rock grade according to an embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of 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, 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.

[0016] This embodiment provides a probabilistic evaluation method for tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information, such as... Figure 1 As shown, it includes: S1: Collect multi-source geological prediction parameters and measured BQ values ​​for each detection section and preprocess them. Spatially match the multi-source geological prediction parameters and measured BQ values ​​for each detection section to obtain a sample dataset for training. S2: Based on the sample dataset, establish a nonlinear mapping model from multi-source feature parameters to BQ values; S3: Construct a probability distribution model based on the sample dataset, randomly generate virtual input feature samples according to the probability distribution model, input the virtual input feature samples into the nonlinear mapping model, and obtain the corresponding BQ prediction value; S4: Construct a rock mass grade function, construct a joint probability density function based on a probability distribution model, and perform a reliability probability analysis of the BQ prediction value based on the rock mass grade function and the joint probability density function to obtain the reliability probability of the rock mass grade, which serves as the evaluation result of the tunnel rock mass quality.

[0017] Specifically, this invention discloses a probabilistic evaluation method for tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information. Addressing the problems of incomplete characterization and insufficient applicability in traditional tunnel surrounding rock prediction methods due to the single source of advance prediction parameters, limited index selection, and small number of effective samples, this invention establishes a mapping relationship between "multi-source advance prediction parameters and surrounding rock quality index BQ value." Multiple physical parameters are obtained using various advance prediction methods such as TSP, TEM, and drilling, and a nonlinear mapping relationship between "multi-source advance prediction parameters and surrounding rock basic quality index BQ value" is established based on these parameters. This invention achieves a more comprehensive and accurate characterization of surrounding rock quality by establishing a mapping relationship based on multiple indices. Furthermore, probabilistic sampling evaluation is introduced based on this mapping, randomly generating samples that conform to the probability according to the probability distribution of the input parameters, thereby generating a probabilistic distribution index of surrounding rock quality grades, achieving a leap from traditional deterministic grading to probabilistic quantitative assessment. This invention can output the probability of occurrence of different surrounding rock grades, providing a quantitative decision-making basis for dynamic design and construction risk management in tunnel engineering.

[0018] In a specific embodiment, the following scheme is used to collect and preprocess multi-source geological prediction parameters and measured BQ values ​​for each detection section, and then spatially match the multi-source geological prediction parameters and measured BQ values ​​for each detection section to obtain the sample dataset for training: The system systematically collects multi-source characteristic parameters of the rock mass ahead of the tunnel face using various advanced geological prediction methods, including the TSP seismic wave method and drilling. The collected parameters include: P-wave velocity. (Range 2.7-3.5 km / s), shear wave velocity (Range 5.0-5.8 km / s), dynamic Poisson's ratio (Range 0.14-0.34), resistivity (range 170-270) The Z-score standardization process eliminates the differences in the dimensions of each feature, forming a standardized feature vector. ; The specific steps for collecting wave velocity and Poisson's ratio indices include: Tunnel Seismic Prediction (TSP) seismic wave reflection method was used to collect rock mass wave velocity indicators. This method belongs to multi-wave, multi-component, high-resolution seismic reflection technology. Source boreholes and receiver boreholes were precisely arranged in a linear array on the tunnel sidewall (left or right side). The source boreholes were configured with 24 shot points, spaced 1.5-2.0 meters apart; the receiver boreholes were equipped with two three-component high-sensitivity seismic detectors, spaced 10-20 meters apart. All boreholes were drilled to a depth of 1.5-2.0 meters to ensure good coupling with the rock mass.

[0019] In each source borehole, micro-explosives are used to precisely generate seismic waves, with the explosive amount controlled within the range of 20-50 grams. When the seismic waves encounter interfaces with different wave impedances during propagation through the rock mass, reflected and transmitted signals are generated. The reflected seismic signals are received by a high-sensitivity three-component geophone, which fully records P-wave, S-wave, and other multi-wavefield information. The P-wave and S-wave velocities can be obtained through processing with TSPwin software. By generating seismic waves on the tunnel sidewalls and receiving signals reflected from the geological interface in front of the tunnel, the P-wave velocity of the rock mass is calculated by processing these signals. and transverse wave velocity The Poisson ratio is obtained through calculation, as shown in formula (12). (12) TSP data may produce unreliable wave velocity values ​​in areas with unfavorable geological conditions such as faults and fracture zones. These anomalies need to be identified and addressed in conjunction with geological reports, and then removed to ensure data reliability.

[0020] The specific steps for collecting resistivity data include: Ground-penetrating radar is used to conduct surveys along the tunnel face or tunnel wall in the form of survey lines. The transmitting antenna transmits high-frequency electromagnetic pulses with a center frequency of f, and the receiving antenna records the reflected wave signals. The acquired raw radar waveforms are preprocessed, including DC drift removal, gain restoration, and bandpass filtering. From the preprocessed radar waveforms, the amplitude A of the echo at a specific reflection interface is determined by measuring the exponential decay law of the echo amplitude A with the propagation depth z, as shown in formula (13). (13) in, The initial amplitude, Because electromagnetic waves have long propagation paths, This is the electromagnetic wave amplitude attenuation coefficient; By collecting electromagnetic wave two-way travel time and amplitude data using ground-penetrating radar, the propagation velocity and attenuation coefficient can be calculated. Substituting these values ​​into the formulas, the resistivity of the medium can be indirectly obtained. The specific formulas are shown in (14)-(17). (14) (15) (16) (17) in, The electrical conductivity of the rock mass (S / m); The depth of the reflector This represents the two-way travel time of the corresponding reflector in the radar waveform; The radar operating frequency (Hz) The magnetic permeability of non-magnetic rock mass ; Angular frequency of electromagnetic waves (rad / s); The dielectric constant of the rock mass is denoted by F / m. The speed of electromagnetic wave propagation (m / s); The specific steps for collecting drilling speed data include: The drilling speed index is obtained in real time using an advanced horizontal drilling system. By arranging horizontal boreholes at the tunnel face, digital drilling monitoring technology is used to record the advance speed of the drill rod (range 30-150 cm / min) throughout the entire process. This drilling speed parameter directly reflects the drillability and mechanical strength characteristics of the rock mass. The drilling speed increases significantly when encountering soft and fractured rock layers, while it slows down relatively when encountering hard and intact rock masses. Simultaneously collect the field-measured BQ values ​​of the corresponding sections to construct a training dataset containing n samples. The number of training samples .

[0021] This scheme constructs a training dataset of size 30. This dataset comes from mechanical tests conducted in the laboratory. The uniaxial saturated compressive strength (Rc) of the rock was accurately determined by loading standard rock core specimens with a pressure testing machine. The rock mass integrity coefficient (Kv) was determined by referring to a table using the rock mass volume joint number (Jv) measured in the field. The obtained Rc and Kv were then used as core input parameters and substituted into the BQ calculation formula specified in the national standard "Engineering Rock Mass Classification Standard" to obtain the measured BQ value. Using various advanced geological prediction methods such as TSP seismic wave method, ground-penetrating radar, and drilling method, the system systematically collected multi-source characteristic parameters of the rock mass in front of the tunnel face. The dataset is shown in Table 1. Table 1 Dataset

[0022] .

[0023] In a specific embodiment, the scheme for establishing a nonlinear mapping model from multi-source feature parameters to BQ values ​​based on the sample dataset is as follows: S21. Construct an initial nonlinear mapping model between multi-source geological prediction parameters and BQ values: The initial nonlinear mapping model between multi-source geological prediction parameters and BQ values ​​is constructed as shown in formula (18). (18) in, For the initial nonlinear mapping model, = , This represents a set of multi-source advanced geological prediction parameters. express , , , and The original eigenvector matrix; This represents the BQ predicted value corresponding to the multi-source geological forecast parameters. This represents the weighting coefficient of each multi-source geological prediction parameter. Indicates the first The linear coefficients of each feature; Indicates the transverse wave velocity. Indicates the longitudinal wave velocity. Represents resistivity. Represents Poisson's ratio. Indicates drilling speed. Indicates the number of features; This represents the weight coefficient for each feature; , express , and The vector matrix, Reflecting the relative relationship between the velocities of longitudinal and transverse waves, It reflects the coordinated changes in resistivity and drilling speed. It reflects the modulation effect of Poisson's ratio on the propagation characteristics of longitudinal waves; and control The intensity of participation, It is a constant; Indicates feature splicing; This represents the original eigenvector matrix. With weighting coefficient Modulated auxiliary feature vector The new joint feature vector formed by concatenating features; The initial nonlinear mapping model is rewritten as a linear model, as shown in equation (19). (19) in, Represents a linear model. To augment the parameter vector, ; To simplify the derivation, the offset k is transformed into the coefficients of the "virtual feature," and the following matrix vector is constructed:

[0024] S22. Construct a loss function based on the linear model to measure the error between the model's predicted value and the actual BQ value. The loss function is shown in formula (20). (20) in, The value of the loss function. This represents the total number of multi-source geological forecast parameters; S23. Minimize the loss function, train the initial nonlinear mapping model based on K-fold cross-validation and the sample dataset, obtain the optimal parameters of the initial nonlinear mapping model, and then obtain the final nonlinear mapping model. Specifically, this scheme employs a 10-fold cross-validation method, randomly dividing the total sample dataset into 10 subsets. In each model training iteration, one subset is selected sequentially as the validation set, while the remaining nine subsets are used as the training set. This process is repeated 10 times until all subsets have been used as validation sets. Finally, the parameter with the smallest error across the 10 training iterations is selected as the optimal parameter. This method makes full use of limited sample data and ensures the stability and reliability of model evaluation results through iterative validation, effectively avoiding overfitting or underfitting problems caused by the randomness of data partitioning. The model is evaluated using the mean squared error and the coefficient of determination, as shown in equations (21) and (22). (twenty one) (twenty two) in, For the first For each sample's predicted value, the smaller the MSE, the better the fit. In this scheme, the MSE value is compared with... If any value reaches the set threshold requirement, the model will be retrained until the set threshold requirement is reached, and the training will be completed. In this scheme, by adding a column of all 1s as "virtual features" before the original detection parameter matrix, the original model is uniformly transformed into a pure linear mapping form, so that the offset k is integrated into the weights corresponding to the virtual features. This avoids the complex process of solving the weights and offsets separately, and improves the rigor and simplicity of the model derivation.

[0025] Augmented parameter vector It is a unified set of parameters that integrates the original weight coefficients and offsets. The purpose of solving this vector is to provide a unified parameter optimization object for the loss function, so that subsequent iterations to optimize the weights and offsets do not need to be performed separately, thereby improving the efficiency and stability of model training.

[0026] In a specific embodiment, the scheme of constructing a probability distribution model based on the sample dataset, randomly generating virtual input feature samples according to the probability distribution model, and inputting the virtual input feature samples into a nonlinear mapping model to obtain the corresponding BQ prediction value is as follows: A probability distribution model is constructed based on the sample dataset, as shown in formula (23). (twenty three) in, It follows a multivariate normal distribution. The sample mean vector The sample covariance matrix is ​​shown in equations (24) and (25). (twenty four) (25) in, The sample mean vector The sample covariance matrix; 10,000 sets of input feature samples that conform to actual geological laws are randomly generated based on the probability distribution model; The virtual input feature samples are input into the nonlinear mapping model to obtain the corresponding set of BQ prediction values, as shown in formula (26). (26) in, Indicates the first A virtual sample After input, the BQ prediction value is generated by the mapping model; This is a pre-trained nonlinear mapping model.

[0027] In a specific embodiment, a rock mass grade function is constructed, a joint probability density function is constructed based on a probability distribution model, and a reliability probability analysis of the rock mass grade is performed on the BQ prediction value based on the rock mass grade function and the joint probability density function to obtain the reliability probability of the rock mass grade. The scheme for using this as the evaluation result of the tunnel surrounding rock quality is as follows: Construct the surrounding rock grade function, as shown in formula (27). (27) The surrounding rock grade function is normalized to obtain the normalized value, as shown in formula (28). (28) in, This is the lower limit of the BQ range within which the predicted BQ value falls in the surrounding rock grade. The surrounding rock grade, Each level corresponds to a preset range of BQ values ​​(e.g., level 1). The BQ value of the surrounding rock ranges from 350 to 450. Z is used to describe the state of the rock mass, when When, it indicates that the structure is in a reliable state; when When, it indicates that the structure is in a limit state, when When this occurs, it indicates that the structure is in a state of failure; The joint probability density function is constructed based on the probability distribution model, as shown in formula (29). (29) in, The determinant of the sample covariance matrix. It is the inverse of the sample covariance matrix; Based on the rock mass grade function and joint probability density function, a reliability probability analysis of the BQ prediction value is performed to obtain the rock mass grade reliability probability, including: Based on the surrounding rock grade function and the joint probability density function, a reliable probability calculation formula for the BQ prediction value belonging to grade G is constructed, as shown in formula (30). (30) in, The probability that the BQ predicted value belongs to the G category; express The probability of; Substituting the predicted BQ value into formula (28), the normalized value of the corresponding surrounding rock grade function is solved, and the selected values ​​are then selected. The virtual input feature samples are selected and substituted into formula (29) and solved by formula (30) to obtain the reliable probability that the BQ prediction value corresponding to each virtual input feature sample belongs to the surrounding rock grade corresponding to the BQ interval it is in. Statistical analysis was performed on the obtained BQ prediction values. Statistical methods were used to calculate the 95% confidence intervals for each probability level to determine the distribution of BQ prediction values ​​within each surrounding rock grade. The results are as follows: Figure 2 As shown in the figure, most BQ values ​​are within the range of... grade; The reliability of the probability estimate is assessed as shown in Equation (31). (31) The half-width of the confidence interval defines the range of fluctuation for the probability estimate; 1.96 is the critical value for the standard normal distribution, this specific value corresponds to a 95% confidence level. Indicates the total number of samples; Judge the calculated The value, if Then the probability distribution model needs to be improved to identify and remove abnormal samples. This means we have a 95% confidence level that the true probability value falls within the calculated interval. We output the rock mass grade with the highest final confidence interval, thus obtaining the final rock mass grade confidence interval as shown below. Figure 3 As shown, after a comprehensive reliability analysis based on the prediction model and sampling statistics, the probability distribution of the tunnel surrounding rock grade was finally quantified, with the predicted probability of Grade III surrounding rock reaching as high as 84.3%. This result indicates that, under current engineering geological conditions, the probability of the surrounding rock quality being classified as Grade III is significantly higher, providing clear and reliable data support for determining tunnel support design parameters and optimizing construction schemes. This high-probability result not only verifies the accuracy of the model's predictions but also significantly enhances confidence in engineering decisions, effectively avoiding technical risks that may arise from misjudgments of the surrounding rock grade.

[0028] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A probabilistic evaluation method for tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information, characterized in that, include: S1: Collect multi-source geological prediction parameters and measured BQ values ​​for each detection section and preprocess them. Spatially match the multi-source geological prediction parameters and measured BQ values ​​for each detection section to obtain a sample dataset for training. S2: Based on the sample dataset, establish a nonlinear mapping model from multi-source feature parameters to BQ values; S3: Construct a probability distribution model based on the sample dataset, randomly generate virtual input feature samples according to the probability distribution model, input the virtual input feature samples into the nonlinear mapping model, and obtain the corresponding BQ prediction value; S4: Construct a rock mass grade function, construct a joint probability density function based on a probability distribution model, and perform a reliability probability analysis of the BQ prediction value based on the rock mass grade function and the joint probability density function to obtain the reliability probability of the rock mass grade, which serves as the evaluation result of the tunnel rock mass quality.

2. The method for probabilistic evaluation of tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information as described in claim 1, characterized in that, Based on the aforementioned sample dataset, a nonlinear mapping model from multi-source feature parameters to BQ values ​​is established, including: S21. Construct an initial nonlinear mapping model between multi-source geological prediction parameters and BQ values; S22. Construct a loss function to measure the error between the model's predicted value and the actual BQ value; S23. Minimize the loss function, train the initial nonlinear mapping model based on K-fold cross-validation and the sample dataset, obtain the optimal parameters of the initial nonlinear mapping model, and then obtain the final nonlinear mapping model.

3. The method for probabilistic evaluation of tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information as described in claim 2, characterized in that, An initial nonlinear mapping model between multi-source geological prediction parameters and BQ values ​​is constructed, including: The initial nonlinear mapping model between multi-source geological prediction parameters and BQ values ​​is constructed as shown in Equation (1). （1） in, For the initial nonlinear mapping model, = , This represents a set of multi-source advanced geological prediction parameters. express , , , and The original eigenvector matrix; This represents the BQ predicted value corresponding to the multi-source geological forecast parameters. This represents the weighting coefficient of each multi-source geological prediction parameter. Indicates the first The linear coefficients of each feature; Indicates the transverse wave velocity. Indicates the longitudinal wave velocity. Represents resistivity. Represents Poisson's ratio. Indicates drilling speed. Indicates the number of features; This represents the weight coefficient for each feature; , express , and The vector matrix, Reflecting the relative relationship between the velocities of longitudinal and transverse waves, It reflects the coordinated changes in resistivity and drilling speed. It reflects the modulation effect of Poisson's ratio on the propagation characteristics of longitudinal waves; and For weighting coefficients, control The intensity of participation, It is a constant; Indicates feature splicing; This represents the original eigenvector matrix. With weighting coefficient Modulated auxiliary feature vector The new joint feature vector is formed by concatenating the features.

4. The method for probabilistic evaluation of tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information as described in claim 3, characterized in that, Constructing the loss function includes: The initial nonlinear mapping model is rewritten as a linear model, as shown in formula (2). （2） in, Represents a linear model. To augment the parameter vector, ; The loss function is constructed based on the linear model, as shown in formula (3). （3） in, The value of the loss function. This represents the total number of multi-source geological forecast parameters.

5. The method for probabilistic evaluation of tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information as described in claim 1, characterized in that, A probability distribution model is constructed based on the sample dataset. Virtual input feature samples are randomly generated according to the probability distribution model. These virtual input feature samples are then input into a nonlinear mapping model to obtain the corresponding BQ prediction values, including: A probability distribution model is constructed based on the sample dataset, as shown in formula (4). （4） in, It follows a multivariate normal distribution. The sample mean vector The sample covariance matrix is ​​shown in equations (5) and (6). （5） （6） in, The sample mean vector The sample covariance matrix; Virtual input feature samples are randomly generated based on a probability distribution model; The virtual input feature samples are input into the nonlinear mapping model to obtain the corresponding set of BQ prediction values, as shown in formula (7). （7） in, Indicates the first One virtual input feature sample After input, the BQ prediction value is generated by the mapping model; This is a pre-trained nonlinear mapping model.

6. The method for probabilistic evaluation of tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information as described in claim 1, characterized in that, Construct a function for the surrounding rock grade, and build a joint probability density function based on a probability distribution model, including: Construct the surrounding rock grade function, as shown in formula (8). （8） The surrounding rock grade function is normalized to obtain the normalized value, as shown in formula (9). （9） in, This is the lower limit of the BQ range within which the predicted BQ value falls in the surrounding rock grade. The surrounding rock grade, Each level corresponds to a set of preset BQ value ranges; The joint probability density function is constructed based on the probability distribution model, as shown in formula (10). （10） in, The determinant of the sample covariance matrix. It is the inverse of the sample covariance matrix.

7. The method for probabilistic evaluation of tunnel surrounding rock quality based on multi-source heterogeneous geological exploration information as described in claim 6, characterized in that, Based on the rock mass grade function and joint probability density function, a reliability probability analysis of the BQ prediction value is performed to obtain the rock mass grade reliability probability, including: Based on the surrounding rock grade function and the joint probability density function, a reliable probability calculation formula for the BQ prediction value belonging to grade G is constructed, as shown in formula (11). （11） in, The probability that the BQ predicted value belongs to the G category; express The probability of; Substituting the predicted BQ value into formula (9), the normalized value of the corresponding surrounding rock grade function is solved, and the selected values ​​are then selected. The virtual input feature samples are selected and substituted into formula (10). Formula (11) is used to solve the problem and obtain the reliable probability that the BQ prediction value corresponding to each virtual input feature sample belongs to the surrounding rock grade corresponding to the BQ interval.

Images