Classification method and system suitable for high ground stress area tunnel surrounding rock

Abstract

The application relates to the technical field of tunnel engineering, and discloses a grading method and system suitable for tunnel surrounding rock in a high ground stress area.The method adopts an operation door model based on a long short-term memory neural network (LSTM) and a BP neural network (BP), carries out weight processing and grade determination on seven grading indexes of the tunnel surrounding rock in the high ground stress area, and controls, evaluates and optimizes output values by using an output door, a feedback layer and a forgetting door.The method can comprehensively, accurately and quickly intelligently grade the tunnel surrounding rock in the high ground stress area, realizes that the surrounding rock grade evaluation result is in line with engineering practice, and builds a high ground stress area tunnel surrounding rock grading model with good regenerative learning and iterative optimization properties; and the application also discloses a grading system suitable for the tunnel surrounding rock in the high ground stress area, which comprises an input layer module, an operation door module, an output layer module, a feedback layer module and a forgetting door module, and is used for realizing the functions of the above method.

Description

Technical Field

[0001] This invention relates to the field of tunnel engineering technology, and specifically to a classification method and system for surrounding rock in tunnels in high ground stress zones. Background Technology

[0002] Accurate evaluation of the surrounding rock grade is a crucial issue in rock engineering during the construction of high-stress tunnels. Accurate classification results are essential for developing reasonable support schemes and ensuring safe construction.

[0003] Tunnel surrounding rock in high-stress zones refers to tunnel surrounding rock exhibiting characteristics such as high ground stress, rock bursts, and large deformations under high ground stress. The classification of tunnel surrounding rock in high-stress zones involves categorizing the surrounding rock into different grades based on its self-stability and support requirements, providing a basis for tunnel design and construction. The classification of tunnel surrounding rock in high-stress zones is a complex nonlinear problem, influenced by various factors such as rock strength, joint characteristics, groundwater conditions, and the magnitude and direction of ground stress.

[0004] Currently, commonly used surrounding rock classification methods at home and abroad include the RMR method, Q method, and BQ method. However, these methods have certain limitations and shortcomings in the classification of surrounding rock in tunnels in high ground stress areas. The mainstream surrounding rock classification method in my country, GB50218-2015 "Engineering Rock Mass Classification Standard" (i.e., BQ method), recommends that when evaluating the quality of surrounding rock in tunnel engineering, multiple surrounding rock classification methods should be used for comprehensive comparison and selection to finally determine the surrounding rock level. However, the problems existing in the actual use of the above surrounding rock classification methods are as follows: (1) They fail to fully consider the geological characteristics of high ground stress tunnels, such as high ground stress and rock bursts; (2) Some evaluation indicators and values ​​in the classification methods rely heavily on subjective determination and are closely related to the geological identification skills of on-site personnel, which leads to doubts about the reliability of the results; (3) The evaluation results of different surrounding rock classification methods are often inconsistent, which leads to blindness in the selection of actual judgment results.

[0005] In addition, although "artificial intelligence+" methods such as LIBSVM algorithm, TSP and PCA-Bayes method, DE-BP model, KNN method, and BP neural network algorithm have been initially applied in the field of surrounding rock classification methods, it should be noted that these methods are not suitable for the surrounding rock classification of tunnels in high ground stress areas, and a highly efficient and accurate classification method for surrounding rock that is fully suitable for the geological characteristics of high ground stress tunnels has not yet been formed. Summary of the Invention

[0006] The purpose of this invention is to provide a classification method and system for tunnel surrounding rock in high-stress areas, which solves the problems that existing classification methods for tunnel surrounding rock in high-stress areas cannot fully reflect the geological characteristics and stability of tunnel surrounding rock in high-stress areas, and that evaluation indicators and results lack objectivity and consistency.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A classification method applicable to tunnel surrounding rock in high ground stress zones, the method comprising:

[0009] The classification index of the surrounding rock of the tunnel in the high ground stress zone is selected to form the input layer data;

[0010] The input layer data is weighted and the surrounding rock grade is determined using arithmetic gates to form arithmetic gate data.

[0011] The output gate is used to control the output value of the computation gate data to obtain the surrounding rock classification result, which constitutes the output layer data.

[0012] The accuracy of the output layer data is evaluated to form the feedback layer data;

[0013] The feedback layer data is processed using a forget gate to obtain forget gate data;

[0014] The forget gate data and the next input layer data are input together into the operation gate to optimize the input layer data.

[0015] Furthermore, the classification indicators for the surrounding rock of the tunnel in the high ground stress zone include: rock mass quality index RQD, rock mass strength parameters, joint direction, rock mass integrity coefficient, groundwater seepage flow, ground stress, and rock burst, which are respectively denoted as c1, c2, c3, c4, c5, c6, and c7.

[0016] Furthermore, the computation gates include an LSTM input gate and a BP neural network computation gate. The LSTM input gate performs weight processing on the input layer data and the forget gate data, and the processing procedure is as follows:

[0017] i t =σ(Wi·[h t-1 ,x t ]+b i )

[0018] In the formula, i t W is the input value of the t-th unit of the input gate. i h represents the network layer weights for the input gate. t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b i σ is the bias term of the input gate, and σ is the sigmoid function.

[0019] Furthermore, the data processed by the LSTM input gate weights sequentially passes through the BP input layer, BP hidden layer, and BP output layer of the BP neural network computation gate;

[0020] The BP input layer consists of the coordinate values ​​of hierarchical indicators (c1, c2, c3, c4, c5, c6, c7), which can be represented as a 1*7 matrix.

[0021] The BP hidden layer consists of two n-dimensional matrices: the input layer-hidden layer and the hidden layer-output layer. The corresponding matrix operations are: input layer-hidden layer: H = X * W1 + b1, hidden layer-output layer: Y = H * W2 + b2.

[0022] The BP output layer is a 1*5 matrix mapped by two n-dimensional matrices, namely (z1, z2, z3, z4, z5), where z1 represents Class I surrounding rock, z2 represents Class II surrounding rock, z3 represents Class III surrounding rock, z4 represents Class IV surrounding rock, and z5 represents Class V surrounding rock.

[0023] Furthermore, an activation layer is added after the BP output layer. After activation operation, the binary calculation result is output, where 1 represents the surrounding rock of this level and 0 represents the surrounding rock of a different level.

[0024] Furthermore, the output gate is used to control the memory unit c. t For the output value h t The impact of this is addressed using the following formula:

[0025] ο t =σ(W ο ·[h t-1 ,x t ]+b ο )

[0026] h t =ο t e tanh(c t )

[0027] In the formula, W ο Here, σ represents the network layer weights for the output gate, σ is the sigmoid function, and h... t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b ο h is the bias term for the output gate. t Let ο be the hidden state of the t-th unit. t The output value of the output gate, c t The information input to the neural network computation gate at time t.

[0028] Furthermore, the feedback layer data is processed using a forgetting gate, specifically the impact of the accuracy evaluation of the tunnel surrounding rock grade obtained at the previous moment based on the above seven classification indicators on the accuracy of the tunnel surrounding rock grade determination at the current moment. This data is used as forgetting gate data, and the processing procedure is as follows:

[0029] f t=σ(W f ·[h t-1 x t ]+b f1 +b f2 )

[0030]

[0031]

[0032] In the formula, f t The output value of the forget gate, σ is the sigmoid function, and W is the output value of the forget gate. f h represents the network layer weights for the forget gate. t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b f1 and b f2 For the bias term of the forget gate, To calculate c t The middle value of W has no specific meaning. c For the calculation of the forget gate, the intermediate value weight, b c It is the network layer bias, f t The output value for the forget gate, c t Let t be the information input to the neural network computation gate at time t, ☉ be the Hadamard product operator, and σ be the sigmoid function.

[0033] A grading system suitable for tunnel surrounding rock in high ground stress areas includes: an input layer module, an operation gate module, a feedback layer module, an output layer module, and a forget gate module;

[0034] The input layer module is used to select the grading index of the surrounding rock of the tunnel in the high ground stress zone to form the input layer data;

[0035] The computation gate module is used to perform weight processing and surrounding rock grade determination on the input layer data using computation gates to form computation gate data;

[0036] The output layer module is used to control the output value of the operation gate data using an output gate to obtain the surrounding rock classification result, thus forming the output layer data.

[0037] The feedback layer module is used to evaluate the accuracy of the output layer data and form feedback layer data.

[0038] The forget gate module is used to process the feedback layer data using a forget gate to form forget gate data. Then, the forget gate data and the next input layer data are input together into the operation gate module to optimize the input layer data.

[0039] Electronic equipment suitable for classifying the surrounding rock of tunnels in high-stress zones includes:

[0040] At least one processor; and at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, which, when invoked, enable the processor to perform the hierarchical method as described in any one of claims 1 to 6.

[0041] A storage medium suitable for classifying the surrounding rock of tunnels in high-stress areas, the storage medium comprising a stored program that, when executed by a processor, implements the aforementioned classification method.

[0042] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0043] This invention provides a classification method for tunnel surrounding rock in high-stress areas. Taking into account the geological characteristics of the surrounding rock in high-stress areas, it delves into seven targeted classification indicators, including high stress and rock bursts. A tunnel surrounding rock classification system based on LSTM-BP neural network (LSTM-BP) and its iterative optimization function are constructed. This classification method can perform comprehensive, accurate, and rapid intelligent classification of tunnel surrounding rock data in high-stress areas, ensuring that the surrounding rock grade evaluation results align with engineering realities. It also constructs a high-stress area tunnel surrounding rock classification model with good regenerative learning and iterative optimization capabilities.

[0044] Of course, implementing the various technical solutions of this invention does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0045] 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 only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained from these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating the composition and method of the LSTM-BP tunnel surrounding rock classification system in high ground stress zone according to an embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram of the physical structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0048] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of this application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to this application are not shown or described in the specification. This is to avoid obscuring the core parts of this application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0049] Furthermore, the features, operations, or characteristics described in the specification can be combined in any suitable manner to form various embodiments. At the same time, the steps or actions in the method description can be rearranged or adjusted in a manner obvious to those skilled in the art. Therefore, the various orders in the specification and drawings are only for the clear description of a particular embodiment and do not imply a necessary order, unless otherwise stated that a particular order must be followed.

[0050] The component numbers used in this document, such as "S1" and "S2," are merely for distinguishing the described objects and have no sequential or technical meaning. The terms "connection" and "linkage" used in this application, unless otherwise specified, include both direct and indirect connections (linkages).

[0051] Example 1:

[0052] S1. Select the classification index for the surrounding rock of tunnels in high ground stress areas.

[0053] The selection criteria are as follows:

[0054] The selection of grading index parameters for the integrated surrounding rock grading method for tunnels in high-stress zones must reflect both the accuracy of surrounding rock grading and the unique geological characteristics of tunnels in high-stress zones. Based on research into various grading methods and considering the characteristics of tunnels in high-stress zones, this invention ultimately selected seven grading parameters: Rock Mass Quality Determination (RQD), Rock Mass Strength Parameter, Rock Mass Integrity Coefficient (Kv), Joint Direction, Groundwater Seepage, Ground Stress, and Rockburst. Among these, RQD, Rock Mass Strength Parameter, Rock Mass Integrity Coefficient (Kv), Joint Direction, and Groundwater Seepage primarily reference the RMR and BQ grading methods, while Ground Stress and Rockburst Parameter mainly refer to the "Code for Geological Investigation of Hydropower Engineering" (GB 50287-2016) and a rockburst classification table based on Chinese engineering experience.

[0055] Among all the grading indicators, the rock mass quality index RQD, rock integrity coefficient (Kv), groundwater seepage, geostress, and rockburst are directly taken from the calculated values ​​under their respective standards.

[0056] Specifically,

[0057] The rock integrity coefficient (Kv) is calculated using a formula; groundwater seepage is taken as the water inflow per 10m of tunnel length; in-situ stress is taken as the maximum principal stress σ1 of the surrounding rock; rockburst is taken as the maximum circumferential stress σ of the chamber. θ uniaxial compressive strength σ of rock c The ratio. Rock mass strength parameters and joint orientation are scored using the RMR method.

[0058] The main grading index parameters are described in detail below:

[0059] (1) Rock mass strength parameters

[0060] Rock compressive strength is one of the fundamental mechanical properties of rock, significantly influencing its classification and quality. For a long time, uniaxial compressive strength has been used as a rock classification standard and to evaluate rock quality and stability. Point load testing can also be used to determine the uniaxial strength of rock, with lower requirements for rock fabrication specifications. This is a rapid, economical, and effective method for in-situ determination of rock mass strength, widely used in rock mass quality grading. The strength of rock blocks under point load can be calculated according to the 1985 revised "Recommended Method for Determining Point Load Strength" by the International Society for Rock Mechanics. Table 1 shows the correspondence between the point load strength index and the uniaxial compressive strength of rock mass strength parameters in the RMR surrounding rock grading method. Since the strength of the tunnel rock mass is relatively low, uniaxial compressive strength is selected as the input value for the rock mass strength parameter.

[0061] Table 1 Rock mass strength parameters using RMR classification method

[0062]

[0063] (2) Rock integrity coefficient (Kv)

[0064] The integrity of a rock mass primarily refers to the degree of cutting by structural planes, the size of unit blocks, and the bonding state between blocks. Therefore, the integrity of a rock mass is related to the geometric characteristics and properties of structural planes, determined by their density, number of groups, orientation, extension, opening, roughness, undulation, filling condition, and the properties of the infill material. The rock integrity coefficient reflects the structural characteristics of the rock mass. In surrounding rock classification methods, the rock mass integrity coefficient is one of the commonly used and important parameters. In engineering construction, the rock mass integrity coefficient is relatively easy to obtain and can be expressed as the square of the ratio of the P-wave velocity of the rock mass to the P-wave velocity of the rock block. Alternatively, the number of joints per unit volume (Jv) can also reflect the integrity of the rock mass.

[0065] As shown in Table 2.

[0066]

[0067] In the formula: K v V is the rock mass integrity coefficient; pm V represents the longitudinal wave velocity of the rock mass. pr The velocity is the longitudinal wave velocity of the rock block.

[0068] Based on the different values ​​of the rock mass integrity coefficient, the surrounding rock can be divided into 5 levels, as shown in Table 2.

[0069] Table 2 Rock Mass Integrity Coefficient Table

[0070]

[0071] (3) Joint direction

[0072] In tunnel surrounding rock classification methods, the orientation of structural joints and the tunnel axis are also factors influencing the surrounding rock grade. This influencing factor is addressed in commonly used classification methods. Based on the evaluation relationship between the orientation of structural joints and the tunnel axis, and referring to the RMR classification method, this paper determines the values ​​of the joint direction parameters according to Table 3.

[0073] Table 3 Joint Strike Parameters

[0074]

[0075] (4) Groundwater seepage

[0076] The development of groundwater reflects the conditions of the rock mass's environment. Because groundwater can soften the surrounding rock, erode the infill materials on the rock's structural surfaces, and reduce the effectiveness of the rock support, thus negatively impacting the project, groundwater is considered one of the main factors affecting the stability of the surrounding rock. The groundwater seepage input parameters are based on the water inflow rate per 10m of tunnel length, as shown in Table 4.

[0077] Table 4. Groundwater Condition Classification Table (RMR)

[0078]

[0079] (5) Geostress

[0080] In-situ stress is stress existing within the Earth's crust. It mainly consists of two parts: gravity caused by the weight of the overlying rock and tectonic stress transmitted from the surrounding landmass. During tunnel excavation, not only does the self-weight stress of the surrounding rock change, but tectonic stress also occurs. Compared to low-altitude areas, tunnels in high-stress zones have a higher distribution of in-situ stress in the surrounding rock, even posing a risk of rockburst, significantly impacting the safety of the rock mass. Therefore, it is essential to consider in-situ stress factors in the classification of surrounding rock for tunnels in high-stress zones. To ensure convenience and speed in classification, the maximum principal stress σ1 of the surrounding rock is used as one of the classification parameters. The stress parameters are shown in Table 5.

[0081] Table 5 Stress Parameter Input Table

[0082]

[0083] (6)Rockburst

[0084] During the excavation of tunnels in high-stress areas, due to the high in-situ stress in the rock mass, the strain energy accumulated in the free-floating rock mass may be released suddenly and violently, leading to explosion-like damage to the surrounding rock. Severe rockbursts can damage tunnels, destroy machinery and equipment, and even cause casualties. In high-stress areas, the occurrence of rockbursts has a significant impact on the stability of the surrounding rock and the type of support structure. Therefore, this paper considers rockburst risk as one of the influencing parameters of the surrounding rock grade of tunnels in high-stress areas. To enable the parameter to be recognized by the BP neural network system, the comprehensive evaluation result of rockburst is converted into a dimensionless value between 0 and 1, as shown in Table 6.

[0085] Table 6 Rockburst Parameter Input Table

[0086]

[0087] S2. Establish a classification system for surrounding rock suitable for high geostress zones.

[0088] Based on the geological characteristics of high ground stress and rockburst in tunnels located in high ground stress zones, factors influencing the surrounding rock grade are divided into basic factors and additional factors, forming an integrated rapid classification system for surrounding rock. The classification indices for the basic factors are derived from the RMR and BQ methods, while ground stress and rockburst are selected as classification indices for the additional factors, primarily reflecting the geological characteristics of the tunnel's environment in high ground stress zones. The hierarchical classification of tunnels in high ground stress zones according to this method is shown in Table 7.

[0089] Table 7 Classification System of Surrounding Rock of Tunnels in High Geostress Zones

[0090]

[0091] S3. Construct a classification method and system suitable for tunnel surrounding rock in high-stress areas.

[0092] This grading method is based on the LSTM-BP neural network (LSTM-BP) tunnel surrounding rock grading system. The grading system structure and grading method process are as follows: Figure 1 As shown, the hierarchical system includes, in sequence: an input layer module, an LSTM forget gate module, an LSTM-BP operation gate module, an output layer module, and a feedback layer module; among which, the LSTM-BP operation gate module includes an LSTM input gate module and a BP neural network computation gate module.

[0093] The grading method is summarized as follows: Rock mass quality index (RQD), rock mass strength parameters, joint direction, rock mass integrity coefficient, groundwater seepage flow, geostress, and rockburst (c1, c2, c3, c4, c5, c6, c7) are taken from the surrounding rock in high-stress areas and fed into the entire grading system via the input layer module. The LSTM forget gate module optimizes the grading indices from the input layer module. The results from both the input layer module and the LSTM forget gate module are fed into the LSTM-BP computation gate module. The grading results processed by the LSTM-BP computation gate module are transmitted to the output layer module. Engineers provide an accuracy evaluation of the grading results in the feedback layer module, which is then fed back to the LSTM forget gate module as a reference for the next grading process to continuously improve the accuracy of the grading results.

[0094] Specifically, the functional modules and specific grading steps involved in the above grading method are as follows:

[0095] S301, Input Layer Module: Selects the grading indicators for tunnel surrounding rock in high-stress areas to form the input layer data. S2 has already determined the grading indicators for the grading system of tunnel surrounding rock in high-stress areas, which are: c1, c2, c3, c4, c5, c6, and c7, respectively representing: rock mass quality index RQD, rock mass strength parameters, joint direction, rock mass integrity coefficient, groundwater seepage flow, ground stress, and rock burst.

[0096] S302, LSTM Forget Gate Module: This module uses a forget gate to process the feedback layer data, serving as the forget gate data. The forget gate is used to determine the information c of the memory unit from the previous time step. t-1 For the current memory unit c t The degree of influence, that is, the impact of the accuracy evaluation of the tunnel surrounding rock grade obtained based on the above seven classification indicators at the previous moment on the accuracy of the current surrounding rock grade determination, is processed by the following formula:

[0097] ft =σ(W f ·[h t-1 x t ]+b f1 +b f2 )

[0098]

[0099]

[0100] In the formula, f t The output value of the forget gate, σ is the sigmoid function, and W is the output value of the forget gate. f h represents the network layer weights for the forget gate. t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b f1 and b f2 For the bias term of the forget gate, To calculate c t The middle value of W has no specific meaning. c For the calculation of the forget gate, the intermediate value weight, b c It is the network layer bias, f t The output value for the forget gate, c t Let t be the information input to the neural network computation gate at time t, ☉ be the Hadamard product operator, and σ be the sigmoid function.

[0101] The sigmoid function, also known as the logistic activation function, is commonly used in logistic regression. It compresses a real number to the range of 0 to 1, making it useful for binary classification. When the ultimate goal is to predict probabilities, it can be applied to the output layer. Its most significant characteristic is its ability to transform large negative numbers towards 0 and large positive numbers towards 1. The sigmoid function is mathematically represented as:

[0102]

[0103] S303, LSTM-BP Operation Gate Module: The LSTM-BP operation gate module consists of two sub-modules: the LSTM input gate module and the BP neural network operation gate module. The main function of the LSTM input gate module is to weight the current input rock grading index based on the accuracy of the tunnel rock grading level obtained from the previous time step using the seven grading indicators. The main function of the BP neural network operation gate module is to receive the weighted rock grading index and determine the rock grading level.

[0104] Specifically:

[0105] S3031, LSTM Input Gate Module: The LSTM input gate module controls the number of current input data flowing into the memory unit, stored in c. The processing procedure is as follows:

[0106] i t =σ(Wi·[h t-1 ,x t ]+b i )

[0107] In the formula, i t W is the input value of the t-th unit of the input gate. i h represents the network layer weights for the input gate. t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b i σ is the bias term of the input gate, and σ is the sigmoid function.

[0108] S3031, BP Neural Network Calculation Gate Module: The surrounding rock grade determination parameters optimized by the LSTM input gate module are input into the input layer of the BP neural network. The input indicators are c1, c2, c3, c4, c5, c6, and c7, respectively, resulting in a dataset of (c1, c2, c3, c4, c5, c6, c7). A three-layer neural network is then constructed, with the following structure:

[0109] ①BP Input Layer: The BP input layer is the coordinate values ​​(c1, c2, c3, c4, c5, c6, c7) representing the surrounding rock information. This is an array containing seven data points, which can also be regarded as a 1*7 matrix.

[0110] ②BP Hidden Layer: The BP hidden layer represents an algorithm from the surrounding rock information (c1, c2, c3, c4, c5, c6, c7) to the output result. Essentially, it consists of two n-dimensional matrices: the input layer-hidden layer and the hidden layer-output layer, corresponding to two matrix operations:

[0111] Input layer - hidden layer: H = X * W1 + b1

[0112] Hidden layer - Output layer: Y = H * W² + b²

[0113] ③BP Output Layer: The output layer is controlled as a 1*5 matrix using two n-dimensional matrices from the hidden layer, i.e., (z1, z2, z3, z4, z5), where z1 represents Class I surrounding rock, z2 represents Class II surrounding rock, z3 represents Class III surrounding rock, z4 represents Class IV surrounding rock, and z5 represents Class V surrounding rock. The output matrix result might be (0, -4, 5, 8, -7), which obviously cannot intuitively display the specific surrounding rock level. Therefore, an activation layer is added to the hidden layer, and after activation operations, the binary calculation result is output as (1, 0, 0, 0, 0), where 1 represents the current level of surrounding rock and 0 represents a different level.

[0114] S304, Output Layer Module: The output layer module uses output gates to control the output values ​​of the data from the LSTM-BP arithmetic gate module, that is, the output layer module controls the memory unit c. t For the output value h t The influence of these factors ultimately forms the output layer data, and the processing procedure is as follows:

[0115] ο t =σ(W ο ·[h t-1 ,x t ]+b ο )

[0116] h t =ο t e tanh(c t )

[0117] In the formula, W ο Here, σ represents the network layer weights for the output gate, σ is the sigmoid function, and h... t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b ο h is the bias term for the output gate. t Let ο be the hidden state of the t-th unit. t The output value of the output gate, c t The information input to the neural network computation gate at time t.

[0118] S4. Feedback Layer Module: The feedback layer module is used by engineers to quantitatively evaluate the accuracy of the surrounding rock grade results obtained from the LSTM-BP tunnel surrounding rock classification system based on actual working conditions, and to express the evaluation results in b... f2 The forget gate bias term is fed back to the LSTM forget gate module for accuracy optimization of the next evaluation result.

[0119] Example 2:

[0120] This embodiment relates to an electronic device suitable for classifying the surrounding rock of tunnels in high ground stress zones. Figure 2 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of the present invention. The electronic device may include: a processor 301, a communication interface 302, a memory 303, and a bus 304. The processor 301, communication interface 302, and memory 303 communicate with each other via the bus 304. The processor 301 can call a computer program stored in the memory 303 and executable on the processor 301 to perform the functions provided in Embodiment 1. For example, this includes: selecting grading indicators for the surrounding rock of tunnels in high-stress areas to form input layer data; using an arithmetic gate to perform weight processing and surrounding rock grade determination on the input layer data to form arithmetic gate data; using an output gate to control the output value of the arithmetic gate data to obtain the surrounding rock grading result to form output layer data; evaluating the accuracy of the output layer data to form feedback layer data; processing the feedback layer data using a forget gate to form forget gate data; and inputting the forget gate data and the next input layer data into the arithmetic gate to optimize the input layer data.

[0121] Furthermore, the logical instructions in the aforementioned memory 303 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention embodiment, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A classification method applicable to tunnel surrounding rock in high ground stress zones, characterized in that, The method includes: The classification index of the surrounding rock of the tunnel in the high ground stress zone is selected to form the input layer data; The input layer data is weighted and the surrounding rock grade is determined using arithmetic gates to form arithmetic gate data. The output gate is used to control the output value of the computation gate data to obtain the surrounding rock classification result, which constitutes the output layer data. The accuracy of the output layer data is evaluated to form the feedback layer data; The feedback layer data is processed using a forget gate to obtain forget gate data; The forget gate data and the next input layer data are input together into the arithmetic gate to optimize the input layer data; Based on the geological characteristics of high ground stress and rockburst in tunnels in high ground stress areas, the factors affecting the surrounding rock grade are divided into basic factors and additional factors, forming an integrated rapid classification system for surrounding rock. The classification index of the basic factors is taken from the RMR method and the BQ method, and ground stress and rockburst are selected as the classification index of the additional factors, which mainly reflect the geological characteristics of the tunnel occurrence environment in high ground stress areas. The classification indicators for the surrounding rock of the tunnel in the high ground stress zone include: rock mass quality index RQD, rock mass strength parameters, joint direction, rock mass integrity coefficient, groundwater seepage flow, ground stress, and rock burst, which are respectively denoted as c1, c2, c3, c4, c5, c6, and c7. The computation gates include an LSTM input gate and a BP neural network computation gate. The LSTM input gate performs weight processing on the input layer data and the forget gate data, and the processing procedure is as follows: In the formula, i t W is the input value of the t-th unit of the input gate. i h represents the network layer weights for the input gate. t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b i σ is the bias term of the input gate, and σ is the sigmoid function; The data processed by the LSTM input gate weights sequentially passes through the BP input layer, BP hidden layer, and BP output layer of the BP neural network computation gate; The BP input layer consists of the coordinate values ​​of hierarchical indicators (c1, c2, c3, c4, c5, c6, c7), which can be represented as a 1*7 matrix. The backpropagation (BP) hidden layer consists of two n-dimensional matrices: the input layer-hidden layer and the hidden layer-output layer. The corresponding matrix operations are: Input layer-hidden layer: Hidden layer - Output layer: ; The BP output layer is a 1*5 matrix mapped by two n-dimensional matrices, namely (z1, z2, z3, z4, z5), where z1 represents Class I surrounding rock, z2 represents Class II surrounding rock, z3 represents Class III surrounding rock, z4 represents Class IV surrounding rock, and z5 represents Class V surrounding rock. The memory unit c is controlled by the output gate. t For the output value h t The impact of this is addressed using the following formula: In the formula, W ο Here, σ represents the network layer weights for the output gates, σ is the sigmoid function, and h... t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b ο h is the bias term for the output gate. t Let ο be the hidden state of the t-th unit. t The output value of the output gate. t The information input to the neural network computation gate at time t; The feedback layer data is processed using a forgetting gate, specifically the impact of the accuracy evaluation of the tunnel surrounding rock grade obtained at the previous moment based on the above seven classification indicators on the accuracy of the tunnel surrounding rock grade determination at the current moment. This data is treated as forgetting gate data, and the processing procedure is as follows: In the formula, f t The output value of the forget gate, σ is the sigmoid function, and W is the output value of the forget gate. f h represents the network layer weights for the forget gate. t-1 Let x be the hidden state of the (t-1)th unit. t For the input of the t-th unit, b f1 and b f2 For the bias term of the forget gate, For calculation t The middle value of W has no specific meaning. c For the calculation of the forget gate, the intermediate value weight, b c It is the network layer bias, f t Output the value for the forget gate. t t represents the information input to the neural network computation gate at time t, ☉ is the Hadamard product operator, and σ is the sigmoid function; Among them, the rock mass strength parameters and joint direction are obtained according to the RMR method to obtain the corresponding score values; the rock mass quality index RQD, rock mass integrity coefficient, groundwater seepage flow, in-situ stress and rockburst are directly taken as the calculated values ​​under the corresponding standards. The groundwater seepage flow is the water inflow per 10m of tunnel length, the in-situ stress is the maximum principal stress σ1 of the surrounding rock, and the rockburst is the maximum circumferential stress σ of the chamber. θ uniaxial compressive strength σ of rock c The ratio; The feedback layer data is formed by evaluating the accuracy of the surrounding rock classification results given by the engineer in the feedback layer module, and is then input into the LSTM forget gate module as a reference for the next classification process, so that the LSTM forget gate module can undertake the optimization of the classification indicators of the input layer module.

2. The classification method for surrounding rock of tunnels in high-stress zones according to claim 1, characterized in that, An activation layer is added after the BP output layer. After activation operation, the binary calculation result is output, where 1 represents the surrounding rock of this level and 0 represents the surrounding rock of other levels.

3. The grading system for the grading method applicable to tunnel surrounding rock in high-stress zones according to claim 1 or 2, characterized in that, include: Input layer module, arithmetic gate module, feedback layer module, output layer module, and forget gate module; The input layer module is used to select the grading index of the surrounding rock of the tunnel in the high ground stress zone to form the input layer data; The computation gate module is used to perform weight processing and surrounding rock grade determination on the input layer data using computation gates to form computation gate data; The output layer module is used to control the output value of the operation gate data using an output gate to obtain the surrounding rock classification result, thus forming the output layer data. The feedback layer module is used to evaluate the accuracy of the output layer data and form feedback layer data. The forget gate module is used to process the feedback layer data using a forget gate to form forget gate data. Then, the forget gate data and the next input layer data are input together into the operation gate module to optimize the input layer data.

4. An electronic device suitable for classifying the surrounding rock of tunnels in high-stress zones, characterized in that, include: At least one processor; And at least one memory communicatively connected to the processor, wherein: the memory stores program instructions executable by the processor, and the processor invokes the program instructions to perform the hierarchical method as described in claim 1 or 2.

5. A storage medium suitable for classifying surrounding rock in tunnels in high-stress areas, characterized in that: The storage medium includes a stored program that, when executed by a processor, implements the hierarchical method as described in claim 1 or 2.

Images