Method and device for rapidly identifying tunnel rockburst site

Abstract

The invention relates to the field of road tunnel construction, in particular to a tunnel rockburst site rapid identification method and device. According to the method, two analysis methods of qualitative analysis and quantitative analysis are fused, the rockburst grade is determined on site through rapid qualitative analysis based on artificial intelligence, the experience of geological personnel is solidified into a computer model by applying a qualitative judgment model learning algorithm and an informatization means, the accuracy and usability of on-site qualitative criteria are improved, and the working efficiency is improved. Primary technicians or on-site constructors can share knowledge of geological personnel with rich experience to carry out on-site primary judgment, construction measures are taken in time, and on-site personnel are guided to be evacuated or disposed in time. And then quantitatively analyzing and verifying the deformation grade, carrying out a ground stress test and a rock test on rockburst of medium or above according to an initial judgment result, quantitatively determining the rockburst grade on site, and taking a reinforcing measure, thereby providing a basic support for tunnel stability evaluation and deformation control.

Description

Technical Field

[0001] The present invention relates to the field of road tunnel construction, and in particular to a method and device for quickly identifying rock bursts on site in a tunnel. Background Art

[0002] Tunnel rock burst refers to the rock burst phenomenon caused by the internal stress of the rock or rock soil exceeding its bearing capacity during the construction of an underground tunnel. Common rock burst phenomena include earthquakes, mine collapses, tunnel rock bursts, cave collapses, excavation collapses, etc., which cause the rock mass around the tunnel to become unstable, the rock layer to be damaged, cracked or peeled off, releasing a large amount of energy, and instantly producing a high-pressure explosion. Tunnel rock bursts will not only cause casualties and property losses, but also affect the construction progress and quality of the tunnel. For such phenomena, safety measures should be taken, such as pre-treatment of the rock layer, adoption of appropriate support methods, and enhanced monitoring to prevent and reduce the hazards of tunnel rock bursts. The existing tunnel rock burst confirmation standard is to judge rock bursts in accordance with the "Regulations on Unfavorable Geological Surveys for Railway Engineering" (TB10027-2022). See the table below for details:

[0003]

[0004] Currently, the standards for rockburst classification are clearly defined in the regulations. By comparing the classification table, the corresponding rockburst grade can be determined. However, in practice, obtaining and clarifying the various characteristics mentioned in the classification standards has always been a relatively slow process. On-site stress testing is required, and rock must be extracted and returned to the laboratory for rock testing. This process is heavily dependent on the experience of geologists. Furthermore, rockbursts involve a large number of parameters, a large amount of on-site data, and a high intensity of manual analysis. Therefore, it is currently very difficult to quickly confirm rockburst classification on-site. There is an urgent need for a rockburst identification method that can serve the field, is fast, and combines qualitative and quantitative analysis to greatly help on-site geologists and construction personnel improve efficiency and work accuracy. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned deficiencies in the prior art and to provide a method for rapid on-site identification of tunnel rockbursts.

[0006] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0007] A method for rapid on-site identification of tunnel rockbursts comprises the following steps:

[0008] S1: Acquire data to be analyzed, preprocess the data to be analyzed, and output a rockburst area data set;

[0009] S2: Import the rockburst area dataset into the pre-built qualitative identification model to generate a qualitatively identified rockburst grade; the rockburst grade is divided into four grades: mild, moderate, strong, and extremely strong; when the qualitatively identified rockburst grade is mild or moderate, proceed to S4; otherwise, proceed to S3;

[0010] S3: Perform preliminary quantitative judgment on the rockburst area dataset;

[0011] When the rockburst level determined by preliminary quantitative analysis is mild or moderate, proceed to S4;

[0012] When the rockburst level determined by the preliminary quantitative judgment is strong or extremely strong, a secondary quantitative judgment is performed to generate a secondary quantitative judgment rockburst level;

[0013] S4: Generate the rockburst grade of the data to be analyzed based on the results output by S2 and S3, and feedback the corresponding support measures;

[0014] The qualitative identification model adopts a one-dimensional convolutional neural network.

[0015] As a preferred embodiment of the present invention, S1 comprises the following steps:

[0016] S11: Acquire data to be analyzed; the data to be analyzed includes data from the survey period and data from the construction period; the data dimensions of the data to be analyzed are: lithology, rock structure, fracture opening, fracture filling, blasthole operation, acoustic and visual characteristics, and rockburst level; the data format of the data values ​​of the data to be analyzed is the corresponding category classification text;

[0017] S12: Converting the data to be analyzed into a structured data set; the structured data set includes input variables and output variables; the input variables are data on lithology, rock structure, fracture opening, fracture filling, blasthole construction, and acoustic and auditory characteristics; and the output variables are data on rockburst grade.

[0018] S13: Enumerate and encode the input variables of the structured data set and complete data normalization, perform one-hot encoding on the output variables of the structured data set, and generate a rockburst area data set after data normalization.

[0019] As a preferred embodiment of the present invention, S13 includes the following steps:

[0020] S131: Normalizing the input variables of the structured data set to form a two-dimensional array consisting of integer numbers with a size of 6*1;

[0021] S132: One-hot encoding is performed on the output variables of the structured data set. The encoding value converts the structured data into a one-hot encoding and outputs it as a two-dimensional array of size 1*n; wherein n=4, and the encoding includes mild rock burst, moderate rock burst, severe rock burst, and extremely severe rock burst;

[0022] S133: According to the number m of qualitative indicators, the input variable is a three-dimensional array of size m*6*1, denoted as X, and the output variable is a three-dimensional array of size m*1*n, denoted as Y, {X:Y}, which constitutes the rockburst area dataset after data standardization.

[0023] As a preferred embodiment of the present invention, the pre-construction of the qualitative identification model includes the following steps:

[0024] Obtaining rock burst area data of the rock burst area, and marking the rock burst level of the rock burst area data;

[0025] Preprocessing the marked rockburst area data to output a rockburst area data set; the preprocessing is the preprocessing in S1;

[0026] Generate a training test set based on the rock burst grade annotation of each rock burst area data in the rock burst area data set The training test set Including training set and test set; among them, is the rockburst area dataset, Mark the corresponding rock burst level;

[0027] The qualitative recognition model is a one-dimensional convolutional neural network, including three convolutional units connected in sequence, an average pooling layer, a fully connected layer and a softmax activation function; the convolutional unit includes a convolutional layer, a batch normalization layer and a relu activation function;

[0028] The qualitative identification model is trained using the training set, and after the model converges, it is verified using the test set. The qualitative identification model is output when the accuracy meets a preset threshold; otherwise, the qualitative identification model is pre-constructed again.

[0029] As a preferred embodiment of the present invention, S3 comprises the following steps:

[0030] S31: Preliminarily determining the rockburst level of the data to be analyzed based on the rockburst area dataset; the preliminary determination is to evaluate the data estimated based on the rockburst area dataset using a rock strength stress ratio method;

[0031] S32: If the rock burst level is medium or above, proceed to S33; otherwise, proceed to S4;

[0032] S33: Obtain the natural compressive strength and maximum initial principal stress of the rock mass in the area corresponding to the data to be analyzed, and use the rock mass strength-stress ratio method to determine the rockburst level of the data to be analyzed.

[0033] As a preferred embodiment of the present invention, S31 includes the following steps:

[0034] The estimated maximum initial principal stress of the rock mass, σ1max, is calculated based on the self-weight stress and the lateral pressure coefficient; the self-weight stress is estimated based on γh, and the lateral pressure coefficient is assigned based on the data during the survey period;

[0035] Estimate the rock strength and calculate the estimated value of the rock mass's natural compressive strength R1m based on the integrity of the surrounding rock;

[0036] Calculate the estimated value of rock mass strength stress ratio:

[0037] Estimated value of rock mass strength stress ratio = estimated value of rock mass natural compressive strength / estimated value of maximum initial principal stress = R1m / σ1max;

[0038] When the estimated value of the rock mass strength-stress ratio is greater than 0.25, the rock burst level is mild; when the estimated value of the rock mass strength-stress ratio is ∈(0.15, 0.25], the rock burst level is moderate; when the estimated value of the rock mass strength-stress ratio is ∈(0.05, 0.15], the rock burst level is strong; when the estimated value of the rock mass strength-stress ratio is ≤0.05, the rock burst level is extremely strong.

[0039] As a preferred embodiment of the present invention, S33 includes the following steps:

[0040] The directions and magnitudes of the three-dimensional ground stresses are obtained through ground stress testing, and the actual value of the maximum initial principal stress of the rock mass σ2max is obtained;

[0041] The actual value of the rock mass's natural compressive strength R2m is calculated through uniaxial compressive strength tests and the integrity of the rock mass;

[0042] Calculate the actual value of rock mass strength stress ratio:

[0043] Actual value of rock mass strength stress ratio = actual value of rock mass natural compressive strength / actual value of maximum initial principal stress = R2m / σ2max;

[0044] When the actual value of the rock mass strength stress ratio is greater than 0.25, the rock burst level is mild; when the actual value of the rock mass strength stress ratio is ∈(0.15, 0.25], the rock burst level is moderate; when the actual value of the rock mass strength stress ratio is ∈(0.05, 0.15], the rock burst level is severe; when the actual value of the rock mass strength stress ratio is ≤0.05, the rock burst level is extremely severe.

[0045] As a preferred embodiment of the present invention, S4 comprises the following steps:

[0046] Generate the rockburst level of the data to be analyzed according to the results output by S2 and S3;

[0047] When the rockburst level is medium or above, the rockburst confirmation length in front of the tunnel face is determined by the seismic wave reflection method, and the corresponding preset support measures are fed back as a reminder.

[0048] A device for rapid on-site identification of tunnel rockbursts comprises at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any one of the methods described above.

[0049] Compared with the prior art, the present invention has the following beneficial effects:

[0050] The present invention combines qualitative and quantitative analysis methods, using artificial intelligence-based rapid qualitative analysis to determine the rockburst grade on-site. It applies qualitative identification model learning algorithms and information technology to solidify the experience of geologists into computer models, improving the accuracy and usability of on-site qualitative judgment criteria. This allows junior technicians or on-site construction personnel to share the knowledge of experienced geologists to make preliminary on-site judgments, take timely construction measures, and guide on-site personnel to evacuate or dispose of the rocks in a timely manner. Quantitative analysis is then performed to verify the deformation grade. Based on the preliminary judgment results, ground stress tests and rock tests are conducted for rockbursts of medium or higher levels. The rockburst grade is quantitatively determined on-site and reinforcement measures are taken, providing basic support for tunnel stability evaluation and deformation control. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 This is a flow chart of a method for rapid on-site identification of tunnel rockburst according to Example 1 of the present invention;

[0052] Figure 2 This is a diagram of a qualitative identification network structure in a method for rapid on-site identification of a tunnel rockburst according to Example 2 of the present invention;

[0053] Figure 3 This is a schematic structural diagram of a tunnel rockburst on-site rapid identification device according to Example 3 of the present invention, which utilizes the tunnel rockburst on-site rapid identification method according to Example 1. DETAILED DESCRIPTION

[0054] The present invention will be further described in detail below in conjunction with test examples and specific embodiments. However, this should not be understood as limiting the scope of the present invention to the following embodiments, and all technologies implemented based on the present invention fall within the scope of the present invention.

[0055] Example 1

[0056] like Figure 1 As shown, a method for rapid on-site identification of tunnel rockburst includes the following steps:

[0057] S1: Acquire data to be analyzed, pre-process the data to be analyzed, and output a rockburst area data set.

[0058] S2: Import the rockburst area dataset into the pre-built qualitative identification model to generate a qualitatively identified rockburst level; the rockburst level is divided into four levels: mild, moderate, strong, and extremely strong; when the generated qualitatively identified rockburst level is mild or moderate, enter S4, otherwise enter S3.

[0059] S3: Perform preliminary quantitative judgment on the rockburst area dataset;

[0060] When the rockburst level determined by preliminary quantitative analysis is mild or moderate, proceed to S4;

[0061] When the rockburst level determined by the preliminary quantitative judgment is strong or extremely strong, a secondary quantitative judgment is performed to generate a rockburst level determined by the secondary quantitative judgment.

[0062] S4: Generate the rockburst grade of the data to be analyzed based on the results output by S2 and S3, and feedback the corresponding support measures.

[0063] The qualitative recognition model uses a one-dimensional convolutional neural network (1D-CNN) as the basic framework for feature extraction. Through fine network structure adjustment and large-scale data training, a neural network model with significant recognition effect is finally obtained.

[0064] Example 2

[0065] This embodiment is a specific implementation of the method for rapid on-site identification of tunnel rockbursts described in Example 1, comprising the following steps:

[0066] S1: Acquire data to be analyzed, pre-process the data to be analyzed, and output a rockburst area data set.

[0067] S11: Acquire data to be analyzed; the data to be analyzed includes survey period data and construction period data.

[0068] Furthermore, survey data collection focuses on rockburst phenomena, establishing a survey dataset and laying the foundation for later automated algorithms. Specifically, this involves extensively collecting geological reports, drill hole histograms, hydrological reports, and other data from the survey period, extracting data based on rock quality and groundwater development. Because survey data is primarily regional, geographic coordinates are used as unique identifiers during the collection phase.

[0069] Furthermore, construction-phase data collection utilizes various on-site reports, reports, monitoring, and experimental data to extract parameters such as rock layer thickness, rock type (soft rock, hard rock), rock structure, blasthole drilling progress, blasthole footage, acoustic and auditory characteristics, and surrounding rock failure characteristics at the tunnel face. Each parameter is then classified. Because construction-phase data is primarily point-based, such as rock layer thickness at a specific mileage, the mileage at that point is used as a unique identifier during the collection phase.

[0070] Furthermore, the data dimensions of the data to be analyzed are: lithology, rock structure, fracture opening, fracture filling condition, blasthole operation condition, acoustic and auditory characteristics, and rockburst level; the data format of the data value of the data to be analyzed is the corresponding category classification text.

[0071] S12: Converting the data to be analyzed into a structured data set; the structured data set includes input variables and output variables; the input variables are data on lithology, rock structure, fracture opening, fracture filling, blasthole construction, and acoustic and auditory characteristics; and the output variables are data on rockburst grade.

[0072] This includes cleaning and processing the text, using regular expressions to extract predefined strings from the original text, and then matching them to specified enumeration values.

[0073] For example, the dimension of sound and auditory features predefines the following strings: occasional blasting sounds, rock tearing sounds, muffled sounds of rock tearing sounds, and muffled thunder sounds of rock tearing sounds. These predefined strings are extracted from the various descriptive texts collected and matched to enumeration values.

[0074] S13: Enumerate and normalize the input variables of the structured dataset, and perform one-hot encoding on the output variables of the structured dataset to generate a standardized rockburst area dataset. One-hot encoding ensures that each output category has a unique, non-zero vector representation, making the differences between different categories clearer and significantly improving model classification results.

[0075] S131: Normalizing the input variables of the structured data set to form a two-dimensional array consisting of integer numbers with a size of 6*1;

[0076] This step can eliminate the dimensional differences of different indicators and make all indicators at the same order of magnitude, which makes it easier for the gradient descent optimization algorithm to converge in subsequent training, accelerates the model training process and improves model robustness.

[0077] S132: One-hot encoding is performed on the output variables of the structured data set. The encoding value converts the structured data into a one-hot encoding and outputs it as a two-dimensional array of size 1*n; wherein n=4, and the encoding includes mild rock burst, moderate rock burst, strong rock burst, and extremely strong rock burst.

[0078] S133: According to the number of qualitative indicators m, the input variable is a three-dimensional array of size m*6*1, denoted as X, and the output variable is a three-dimensional array of size m*1*n, denoted as Y, {X:Y}, which constitutes the rockburst area dataset X after data standardization.

[0079] S2: Import the rockburst area dataset into a pre-built qualitative identification model to generate a qualitatively identified rockburst grade. Because rockburst data has numerous parameters and a large volume of data, neural networks offer better learning results. The classification goal is to construct a multi-class neural network model, with each class corresponding to a rockburst grade. The model is categorized into four levels: mild, moderate, severe, and extremely severe. If the rockburst grade is mild or moderate, the model proceeds to S4; otherwise, the model proceeds to S3.

[0080] The main purpose of introducing the qualitative identification model in this embodiment is that on-site construction personnel do not necessarily have extensive geological knowledge. The occurrence of rock burst is based on the m and n values ​​in the rock burst area dataset X. The placeholder input by the design network is [None, [m, n]].

[0081] Furthermore, the pre-construction of the qualitative identification model includes the following steps:

[0082] 1) Obtain rock burst area data of the rock burst area, and label the rock burst level of the rock burst area data.

[0083] Data is filtered based on rockburst areas. The database stores rockburst location information in the form of regional coordinates (Region) and point coordinates (Point). Simply spatially intersecting the location information with the query area yields the rockburst region dataset X. Based on the criteria, each rockburst region data set is manually labeled with a rockburst level Y. The manual criteria are as follows:

[0084]

[0085] 2) Preprocessing the labeled rockburst area data to output a rockburst area data set; the preprocessing is the preprocessing in S1.

[0086] 3) Generate a training test set based on the rock burst level annotation of each rock burst area data in the rock burst area dataset (This embodiment is split according to the ratio of 8:2); the training test set Including training set and test set; among them, is the rockburst area dataset, The corresponding rock burst level is marked.

[0087] 4) The qualitative identification model is trained using the training set. After the model converges, it is verified using the test set. The qualitative identification model is output when the accuracy meets a preset threshold; otherwise, the qualitative identification model is pre-constructed again.

[0088] Furthermore, the qualitative identification model is a one-dimensional convolutional neural network, including three convolutional units connected in sequence, an average pooling layer, a fully connected layer and a softmax activation function; the convolutional unit includes a convolutional layer, a batch normalization layer and a relu activation function. Figure 2 As shown, after each layer of convolution, a batch normalization improvement is performed to alleviate the problem of gradient vanishing or gradient exploding caused by a large amount of training. Finally, it is passed to the next layer of convolution after activation by relu. After the three layers of convolution, an average pooling operation and full connection are performed, and finally activated by softmax. The goal of using three layers of convolution and gradually increasing its filters is to extract richer features by increasing the complexity of the model to overcome the problems that the present invention may face, such as small amount of data and sparse output variables. The possible problems of gradient vanishing and gradient exploding are eliminated as much as possible through batch normalization.

[0089] Furthermore, the qualitative judgment is saved as an h5 file for easy calling during regression prediction.

[0090] S3: Perform preliminary quantitative judgment on the rockburst area dataset;

[0091] When the rockburst level determined by preliminary quantitative analysis is mild or moderate, proceed to S4;

[0092] When the rockburst level determined by the preliminary quantitative judgment is strong or extremely strong, a secondary quantitative judgment is performed to generate a rockburst level determined by the secondary quantitative judgment.

[0093] S31: Preliminarily judging the rockburst level of the data to be analyzed based on the rockburst area dataset; the preliminary judgment is to evaluate the data estimated based on the rockburst area dataset using a rock strength stress ratio method.

[0094] The estimated value of the maximum initial principal stress of the rock mass, σ1max, is calculated based on the self-weight stress and the lateral pressure coefficient; among them, the self-weight stress is estimated based on γh, and the lateral pressure coefficient is assigned based on the data during the survey period, which is generally 1.1 to 1.2, and can reach 1.3 to 1.4 in some areas.

[0095] Estimate the rock strength and calculate the estimated natural compressive strength R1m of the rock mass based on the integrity of the surrounding rock. The rock strength can be determined by hammering, pinching, or performing rebound tests on rocks of different lithologies at the tunnel face.

[0096] Calculate the estimated value of rock mass strength stress ratio:

[0097] Estimated rock mass strength stress ratio = estimated rock mass natural compressive strength / estimated maximum initial principal stress = R1m / σ1max. The criteria are shown in the following table:

[0098]

[0099] S32: If the rock burst level is medium or above, proceed to S33; otherwise, proceed to S4;

[0100] S33: Obtain the natural compressive strength and maximum initial principal stress of the rock mass in the area corresponding to the data to be analyzed, and use the rock mass strength-stress ratio method to determine the rockburst level of the data to be analyzed.

[0101] In-situ stress testing determines the direction and magnitude of three-dimensional in-situ stresses and the actual value of the maximum initial principal stress in the rock mass, σ2max. For example, in areas that do not interfere with tunnel construction, such as crane tunnels, escape tunnels, and cross passages, directional drilling can be performed around the tunnels at intervals of 500 to 1000 meters to conduct in-situ stress tests, including the stress relief method. The length of the maximum deformation ahead of the tunnel face can be determined based on changes in parameters such as TSP longitudinal waves, density, Poisson's ratio, and dynamic Young's modulus, with a single confirmed length of 10 to 20 meters.

[0102] The actual value of the rock mass's natural compressive strength, R2m, is calculated based on uniaxial compressive strength tests and the rock mass integrity. For example, uniaxial compressive strength tests are performed on rock blocks of different lithologies (at least three specimens of each type) to obtain the rock's uniaxial compressive strength. This is then combined with the rock mass integrity to convert the rock mass's natural compressive strength (Rm). Alternatively, rebound tests are performed on rocks of different lithologies at the tunnel face to convert the rock's natural compressive strength. This is then combined with the rock mass integrity to convert the rock mass's natural compressive strength (R2m).

[0103] Calculate the actual value of rock mass strength stress ratio:

[0104] The actual value of rock mass strength stress ratio = the actual value of rock mass natural compressive strength / the actual value of maximum initial principal stress = R2m / σ2max. The judgment criteria are shown in the following table:

[0105]

[0106] S4: Generate the rockburst grade of the data to be analyzed based on the results output by S2 and S3, and feedback the corresponding support measures.

[0107] Among them, when the output result of S2 conflicts with the output result of S3, the output result of S3 shall prevail.

[0108] Furthermore, when the rockburst level is medium or above, the rockburst length ahead of the tunnel face is determined using seismic wave reflection, and corresponding pre-set support measures are provided as a reminder. Specifically, the rockburst confirmation standards in the "Regulations on Unfavorable Geological Surveys for Railway Engineering" (TB10027-2022) are implemented. Based on the deformation and damage characteristics of the surrounding rock support, support measures are reinforced as necessary. If the rockburst cannot be effectively controlled after reinforcement, the rockburst level should be increased, and the next level of support measures should be implemented, and the support should be replaced if necessary.

[0109] This qualitative and quantitative rockburst classification method has been applied hundreds of times in the construction of multiple railway tunnels. Field applications have demonstrated that the method has an accuracy rate exceeding 80%, saving significant manual effort and improving the efficiency and accuracy of on-site geologists. It holds great promise for future application.

[0110] Example 3

[0111] like Figure 3 As shown, a device for rapid on-site identification of tunnel rockbursts includes at least one processor, a memory communicatively connected to the at least one processor, and at least one input / output interface communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform the method for rapid on-site identification of tunnel rockbursts described in the aforementioned embodiments. The input / output interface may include a display, a keyboard, a mouse, and a USB interface for inputting and outputting data.

[0112] Those skilled in the art will understand that all or part of the steps of implementing the above-mentioned method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above-mentioned method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, and other media that can store program codes.

[0113] When the above-mentioned integrated unit of the present invention is implemented in the form of a software functional unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0114] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for rapid on-site identification of tunnel rockburst, characterized in that: The following steps are involved: S1: Acquire data to be analyzed, preprocess the data to be analyzed, and output a rockburst area data set; S2: Import the rockburst area dataset into the pre-built qualitative identification model to generate a qualitatively identified rockburst grade; the rockburst grade is divided into four grades: mild, moderate, strong, and extremely strong; when the qualitatively identified rockburst grade is mild or moderate, proceed to S4; otherwise, proceed to S3; S3: Perform preliminary quantitative judgment on the rockburst area dataset; When the rockburst level determined by preliminary quantitative analysis is mild or moderate, proceed to S4; When the rockburst level determined by the preliminary quantitative judgment is strong or extremely strong, a secondary quantitative judgment is performed to generate a secondary quantitative judgment rockburst level; S4: Generate the rockburst grade of the data to be analyzed based on the results output by S2 and S3, and feedback the corresponding support measures; The qualitative identification model adopts a one-dimensional convolutional neural network.

2. A method for rapid on-site identification of tunnel rockburst according to claim 1, characterized in that: Said S1 comprises the following steps: S11: Acquire data to be analyzed; the data to be analyzed includes data from the survey period and data from the construction period; the data dimensions of the data to be analyzed are: lithology, rock structure, fracture opening, fracture filling, blasthole operation, acoustic and visual characteristics, and rockburst level; the data format of the data values ​​of the data to be analyzed is the corresponding category classification text; S12: Converting the data to be analyzed into a structured data set; the structured data set includes input variables and output variables; the input variables are data on lithology, rock structure, fracture opening, fracture filling, blasthole construction, and acoustic and auditory characteristics; and the output variables are data on rockburst grade. S13: Enumerate and encode the input variables of the structured data set and complete data normalization, perform one-hot encoding on the output variables of the structured data set, and generate a rockburst area data set after data normalization.

3. A method for rapid on-site identification of tunnel rockburst according to claim 2, characterized in that: The S13 comprises the following steps: S131: Normalizing the input variables of the structured data set to form a two-dimensional array consisting of integer numbers with a size of 6*1; S132: One-hot encoding is performed on the output variables of the structured data set. The encoding value converts the structured data into a one-hot encoding and outputs it as a two-dimensional array of size 1*n; wherein n=4, and the encoding includes mild rock burst, moderate rock burst, severe rock burst, and extremely severe rock burst; S133: According to the number m of qualitative indicators, the input variable is a three-dimensional array of size m*6*1, denoted as X, and the output variable is a three-dimensional array of size m*1*n, denoted as Y, {X:Y}, which constitutes the rockburst area dataset after data standardization.

4. A method for rapid on-site identification of tunnel rockburst according to claim 1, characterized in that: The pre-construction of the qualitative identification model includes the following steps: Obtaining rock burst area data of the rock burst area, and marking the rock burst level of the rock burst area data; Preprocessing the marked rockburst area data to output a rockburst area data set; the preprocessing is the preprocessing in S1; Generate a training test set based on the rock burst grade annotation of each rock burst area data in the rock burst area data set The training test set Including training set and test set; among them, is the rockburst area dataset, Mark the corresponding rock burst level; The qualitative recognition model is a one-dimensional convolutional neural network, including three convolutional units connected in sequence, an average pooling layer, a fully connected layer and a softmax activation function; the convolutional unit includes a convolutional layer, a batch normalization layer and a relu activation function; The qualitative identification model is trained using the training set, and after the model converges, it is verified using the test set. The qualitative identification model is output when the accuracy meets a preset threshold; otherwise, the qualitative identification model is pre-constructed again.

5. The method for rapid on-site identification of tunnel rockburst according to claim 1, characterized in that: The S3 includes the following steps: S31: Preliminarily determining the rockburst level of the data to be analyzed based on the rockburst area dataset; the preliminary determination is to evaluate the data estimated based on the rockburst area dataset using a rock strength stress ratio method; S32: If the rock burst level is medium or above, proceed to S33; otherwise, proceed to S4; S33: Obtain the natural compressive strength and maximum initial principal stress of the rock mass in the area corresponding to the data to be analyzed, and use the rock mass strength-stress ratio method to determine the rockburst level of the data to be analyzed.

6. A method for rapid on-site identification of tunnel rockburst according to claim 5, characterized in that: The S31 includes the following steps: The estimated maximum initial principal stress of the rock mass, σ1max, is calculated based on the self-weight stress and the lateral pressure coefficient; the self-weight stress is estimated based on γh, and the lateral pressure coefficient is assigned based on the data during the survey period; Estimate the rock strength and calculate the estimated value of the rock mass's natural compressive strength R1m based on the integrity of the surrounding rock; Calculate the estimated value of rock mass strength stress ratio: Estimated value of rock mass strength stress ratio = estimated value of rock mass natural compressive strength / estimated value of maximum initial principal stress = R1m / σ1max; When the estimated value of the rock mass strength-stress ratio is greater than 0.25, the rock burst level is mild; when the estimated value of the rock mass strength-stress ratio is ∈(0.15, 0.25], the rock burst level is moderate; when the estimated value of the rock mass strength-stress ratio is ∈(0.05, 0.15], the rock burst level is strong; when the estimated value of the rock mass strength-stress ratio is ≤0.05, the rock burst level is extremely strong.

7. A method for rapid on-site identification of tunnel rockburst according to claim 5, characterized in that: The S33 includes the following steps: The directions and magnitudes of the three-dimensional ground stresses are obtained through ground stress testing, and the actual value of the maximum initial principal stress of the rock mass σ2max is obtained; The actual value of the rock mass's natural compressive strength R2m is calculated through uniaxial compressive strength tests and the integrity of the rock mass; Calculate the actual value of rock mass strength stress ratio: Actual value of rock mass strength stress ratio = actual value of rock mass natural compressive strength / actual value of maximum initial principal stress = R2m / σ2max; When the actual value of the rock mass strength stress ratio is greater than 0.25, the rock burst level is mild; when the actual value of the rock mass strength stress ratio is ∈(0.15, 0.25], the rock burst level is moderate; when the actual value of the rock mass strength stress ratio is ∈(0.05, 0.15], the rock burst level is severe; when the actual value of the rock mass strength stress ratio is ≤0.05, the rock burst level is extremely severe.

8. A method for rapid on-site identification of tunnel rockburst according to claim 1, characterized in that: The S4 comprises the following steps: Generate the rockburst level of the data to be analyzed according to the results output by S2 and S3; When the rockburst level is medium or above, the rockburst confirmation length in front of the tunnel face is determined by the seismic wave reflection method, and the corresponding preset support measures are fed back as a reminder.

9. A tunnel rockburst on-site rapid identification device, characterized in that: The invention comprises at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method according to any one of claims 1 to 8.

Images