A method and system for classifying tunnel surrounding rock

Abstract

The application discloses a kind of grading methods for tunnel surrounding rock, comprising: collecting surrounding rock characteristic data and corresponding surrounding rock grade;Constructing preset fusion model, and according to surrounding rock characteristic data and corresponding surrounding rock grade, preset fusion model is trained, and surrounding rock intelligent grading model is formed, preset fusion model includes feature recognition part and target learning part, feature recognition part has the feature learning network respectively established for different types of measurement data;Surrounding rock intelligent grading model is used to carry out grading evaluation on target surrounding rock.The application mines the internal connection between multiple measurements and surrounding rock grading targets, ensures the data foundation, constructs an intelligent grading model with good flexibility and wide applicability, and can effectively realize feature learning and target learning of surrounding rock intelligent grading task.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel surrounding rock evaluation, and in particular to a grading method and system for tunnel surrounding rock. BACKGROUND

[0002] Tunnel surrounding rock grading is the most direct and commonly used method for quantitatively evaluating the quality and stability of surrounding rock, and the grading result directly determines the tunnel excavation method, support parameters and the like, and is one of the very key indicators in rock tunnel design and construction.

[0003] Due to the fact that detailed geological data of the tunnel excavation site cannot be obtained at the survey stage, the surrounding rock grade determined according to the survey often does not match the actual situation. In order to maximize the stability of the surrounding rock, the surrounding rock grading of the exposed tunnel face needs to be performed. However, the traditional surrounding rock grading technology has some defects: (1) limited by the sample quantity and experience classification, each grading method is usually only applicable to a specific type of surrounding rock; (2) different grading methods consider different factors, and are not comprehensive and complete, and the obtained surrounding rock grading conclusions are inconsistent; (3) the input parameters are mostly qualitative descriptions, and the qualitative description or quantitative conversion of each parameter has strong uncertainty, and the implementation of surrounding rock grading is heavily dependent on the experience of the implementer.

[0004] In addition, the prior art also includes a scheme of introducing machine learning technology into the surrounding rock grading work to improve the automation degree of surrounding rock grading. However, with the development of tunnel construction towards deep-buried super-long tunnels, more and more complex geological conditions are faced, and the commonly used surrounding rock grading methods have the problems of low intelligence, imperfect classification standard system, insufficient utilization of multi-source geological information, not fine consideration of adverse geological conditions, and limited precision and efficiency in practical application. SUMMARY

[0005] With the rapid development of big data and artificial intelligence technology, the present application aims to provide an intelligent surrounding rock grading method based on fusion of multi-source heterogeneous geological information.

[0006] To solve the above technical problems, the present application provides a grading method for tunnel surrounding rock, comprising: collecting surrounding rock characteristic data and corresponding surrounding rock grades; constructing a preset fusion model and training the preset fusion model according to the surrounding rock characteristic data and corresponding surrounding rock grades to form a surrounding rock intelligent grading model, wherein the preset fusion model comprises a feature recognition part and a target learning part, and the feature recognition part has a feature learning network established for different types of measurement data; and using the surrounding rock intelligent grading model to perform grading evaluation on target surrounding rock.

[0007] Preferably, in the step of collecting the surrounding rock characteristic data and the corresponding surrounding rock grade, the following is included: collecting the surrounding rock characteristic data of each sample and labeling the corresponding surrounding rock grade label, thereby forming a surrounding rock characteristic data and surrounding rock grade dataset, wherein the surrounding rock characteristic data includes multiple surrounding rock characteristic parameters of different sources, different formats and different scales.

[0008] Preferably, the surrounding rock characteristic data includes but is not limited to rock mineral property characteristics, discontinuity development characteristics, geological condition backgrounds and engineering construction characteristics.

[0009] Preferably, the rock mineral property characteristics include but are not limited to lithology, rock velocity, rock hardness type, rock wear resistance, uniaxial compressive strength, tensile strength, Young's modulus, triaxial stress, anisotropy, Poisson's ratio, porosity and permeability; the discontinuity development characteristics include but are not limited to discontinuity type and development parameters including width, spacing, number, occurrence, roughness, ductility, filler and weathering alteration degree; the geological condition background includes but is not limited to tectonic type, seismic intensity, groundwater, ground stress and ground temperature; the engineering construction characteristics include but are not limited to the angle between the tunnel axis and the main structural plane and the tunnel excavation method.

[0010] Preferably, the grading method further includes: preprocessing the surrounding rock characteristic data; dividing the preprocessed surrounding rock characteristic data and the corresponding surrounding rock grade into a training sample set, a validation sample set and a test sample set according to a preset data sample ratio, so as to train, cross-validate, and optimize and quality control evaluate the preset fusion model in the model training process by using each sample set.

[0011] Preferably, in the step of preprocessing the surrounding rock characteristic data, the following is included: encoding qualitative type data to complete quantitative conversion; resampling one-dimensional sequence data or two-dimensional image data respectively, so that the same dimensional data have the same data size; normalizing or standardizing quantitative data of the same dimension and different types; calculating the surrounding rock attribute of the measured data.

[0012] Preferably, for discrete numerical type training input data, a multilayer perceptron or a support vector machine is used to construct a corresponding feature learning network; for structured image or curve type training input data, a convolutional neural network is used to construct a corresponding feature learning network; for text type training input data, a recurrent neural network is used to construct a corresponding feature learning network.

[0013] Preferably, the target learning part adopts a fully connected neural network to construct, wherein a softmax function is applied to realize the classification and output of the surrounding rock grade.

[0014] Preferably, based on the advancement of different tunnels and different construction stages, the surrounding rock characteristic data and the surrounding rock grade data set are expanded, so as to update and apply the surrounding rock intelligent grading model by using the expanded surrounding rock characteristic data and the surrounding rock grade data set.

[0015] In another aspect, the embodiments of the present application also provide a grading system for tunnel surrounding rock, comprising: a data set collection module configured to collect surrounding rock characteristic data and corresponding surrounding rock grades; an intelligent grading model construction module configured to construct a preset fusion model and train the preset fusion model according to the surrounding rock characteristic data and the corresponding surrounding rock grades, so as to form a surrounding rock intelligent grading model, wherein the preset fusion model comprises a feature recognition part and a target learning part, and the feature recognition part has a feature learning network established for different types of measurement data; and a model application module configured to use the surrounding rock intelligent grading model to carry out grading evaluation on target surrounding rock.

[0016] Compared with the prior art, one or more embodiments in the above scheme can have the following advantages or beneficial effects:

[0017] The present application proposes a grading method and system for tunnel surrounding rock. The method and system are based on an intelligent grading scheme for surrounding rock proposed based on multi-source heterogeneous information that affects and describes the state of surrounding rock, and have the following effects:

[0018] (1) The present application fully considers two types of factors, i.e., cause and performance, which are directly related to the grading evaluation of surrounding rock. The cause type factors include macro-geological factors and micro-mechanical parameters that affect the state of surrounding rock; the performance type factors include various types of measurement and corresponding calculation data used to measure the state of surrounding rock. From the perspective of the target categories reflected by the considered factors, these factors are divided into four types of parameters, i.e., rock mineral properties, discontinuity development, geological condition background, and engineering construction, which have different sources, different formats, and different scales. In summary, the comprehensive consideration of multi-source heterogeneous data fully utilizes the advantages of big data and is suitable for mining more accurate internal relationships between various measurements and the grading target of surrounding rock.

[0019] (2) The present application constructs a hybrid deep neural network suitable for intelligent grading of surrounding rock. The network reasonably combines multiple network modules according to the types, formats, and other characteristics of input and output data, and shows good flexibility and wide applicability, effectively realizing feature learning and target learning of the intelligent grading task of surrounding rock.

[0020] (3) In practical applications, part of the data can be difficult to collect or measure. When only part of the multi-source heterogeneous information can be obtained, the modular assembly property of the network architecture designed by the present application can be simplified and adapted accordingly, ensuring the practicality of the network model in the application of intelligent grading of surrounding rock.

[0021] (4) The intelligent surrounding rock grading model provided by the application can perform adaptive incremental learning on the used hybrid deep neural network by using gradually accumulated measurement data for different tunnel projects and different construction stages, so that the accuracy and generalization ability of the intelligent surrounding rock grading model are continuously optimized.

[0022] Additional features and advantages of the application will be set forth in the description that follows, and in part will be apparent from the description, or can be learned by practice of the application. The objectives and other advantages of the application will be realized and attained by the structure particularly pointed out in the description and claims. BRIEF DESCRIPTION OF DRAWINGS

[0023] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings:

[0024] Figure 1 A step diagram of the grading method for tunnel surrounding rock of the embodiment of the present application.

[0025] Figure 2 A whole technical roadmap of the grading method for tunnel surrounding rock of the embodiment of the present application.

[0026] Figure 3 A realization principle diagram of the grading method for tunnel surrounding rock of the embodiment of the present application.

[0027] Figure 4 A frame structure schematic diagram of the grading system for tunnel surrounding rock of the embodiment of the present application. DETAILED DESCRIPTION

[0028] The embodiments of the present application will be described in detail below with reference to the accompanying drawings and embodiments, so that how the present application applies technical means to solve technical problems and achieves technical effects can be fully understood and implemented. It should be noted that, as long as there is no conflict, each embodiment in the present application and each feature in each embodiment can be combined with each other, and the formed technical solutions are all within the protection scope of the present application.

[0029] In addition, the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions. Moreover, although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown.

[0030] Tunnel surrounding rock classification is an effective means to quantitatively evaluate the quality and stability of surrounding rock, and effectively guide the design and construction of tunnel excavation and support. The commonly used surrounding rock classification methods have the problems of low intelligent degree, imperfect classification standard system, insufficient utilization of multi-source geological information, not fine consideration of adverse geological conditions, limited precision and efficiency of practical application, etc., and cannot be well applied to the increasingly complex geological conditions faced by deep-buried super-long tunnel construction.

[0031] To solve the problems in the above background art, the embodiments of the present application propose a method and system for classifying tunnel surrounding rock. The method and system are based on a big data intelligent driving method, and construct a hybrid deep neural network for multi-source heterogeneous information of different sources, different formats and different scales affecting or describing the surrounding rock state, so as to realize automatic learning of a surrounding rock intelligent classification model without human intervention in an end-to-end manner, and can be used to effectively realize intelligent fusion of multi-source geological information and intelligent evaluation of surrounding rock.

[0032] Figure 1 The steps of the method for classifying tunnel surrounding rock according to the embodiments of the present application are shown in the flowchart. Figure 1 As shown in the flowchart, the tunnel surrounding rock classification method according to the embodiments of the present application includes the following steps: step S110 collects surrounding rock feature data and corresponding surrounding rock grades; step S120 constructs a preset fusion model, and trains the preset fusion model according to the surrounding rock feature data and corresponding surrounding rock grades collected in step S110 to form a surrounding rock intelligent classification model, wherein the preset fusion model includes a feature recognition part and a target learning part connected in sequence, and the feature recognition part has a feature learning network established for different types of measurement data; step S130 uses the surrounding rock intelligent classification model constructed in step S120 to carry out classification and evaluation of the target surrounding rock.

[0033] Figure 2 The overall technical roadmap of the method for classifying tunnel surrounding rock according to the embodiments of the present application is shown in the flowchart. Figure 3 The implementation principle diagram of the method for classifying tunnel surrounding rock according to the embodiments of the present application is shown in the flowchart. Figure 1 and Figure 2 The specific process of the tunnel surrounding rock classification method according to the embodiments of the present application is described below.

[0034] In step S110, the surrounding rock feature data of each sample is collected and the corresponding surrounding rock grade label is labeled, so as to form a surrounding rock feature data and surrounding rock grade data set. The surrounding rock feature data includes multiple surrounding rock feature parameters of different sources, different formats and different scales. That is, the data set contains multiple samples, each sample is a combination of surrounding rock features and corresponding surrounding rock grade. It can be understood that each sample corresponds to the surrounding rock at a specific mileage in a specific tunnel. For each sample, multiple surrounding rock feature parameters of different sources, different formats and different scales need to be obtained first, and a surrounding rock grade label is labeled.

[0035] For the training process of supervised learning, a complete data set containing (training) input data and target label, i.e. surrounding rock feature data and surrounding rock grade data set, needs to be prepared in the data collection stage. The preparation of input surrounding rock feature parameters includes but is not limited to the following four categories: parameters related to the description of rock mineral properties (rock mineral property features), parameters related to the development of discontinuities (discontinuity development features), parameters related to the geological conditions of surrounding rock (geological condition background), and parameters related to engineering construction (engineering construction features), which are provided in various formats such as binary files, images, texts, numerical values, types, etc.

[0036] Further, the rock mineral property features are obtained by field measurement, laboratory test, etc. Specifically, the rock mineral property features include but are not limited to: lithology, rock velocity, rock hardness type, rock wear resistance, uniaxial compressive strength, tensile strength, Young's modulus, triaxial stress, anisotropy, Poisson's ratio, porosity and permeability, etc.

[0037] Further, the discontinuity development features can be obtained based on the face image, three-dimensional laser scanning, etc. Specifically, the discontinuity development features include but are not limited to: discontinuity type and development parameters. The discontinuity type includes but is not limited to: fault, bedding, joint, fracture and microcrack, etc.; the development parameters include but are not limited to: width, spacing, number, occurrence, roughness, ductility, filler and weathering alteration degree, etc.

[0038] Further, the geological condition background includes but is not limited to: tectonic type, seismic intensity, groundwater, ground stress, ground temperature, etc. The engineering construction features include but are not limited to: the angle between the tunnel axis and the main structural plane, the tunnel excavation method, etc.

[0039] Further, after obtaining multiple surrounding rock feature parameters of different formats, step S110 also labels the corresponding surrounding rock grade label for each data (sample) in each surrounding rock feature parameter. In labeling the surrounding rock grade target label for each data sample parameter, the existing surrounding rock classification method or expert evaluation method can be applied.

[0040] After the collection of surrounding rock characteristic data and the surrounding rock level labeling are completed, the surrounding rock intelligent grading model is established in step S120.

[0041] Due to the multi-source nature of the surrounding rock characteristic data, the formats of the characteristic data from different sources are diverse. Therefore, before entering step S120 to construct the grading model, the embodiment of the present application needs to first preprocess all the surrounding rock characteristic data collected in step S110, and then divide the preprocessed surrounding rock characteristic data and the corresponding surrounding rock level into a training sample set, a validation sample set and a test sample set according to a preset data sample ratio, so as to train, cross-validate, and optimize and quality control evaluate the preset fusion model in step S120 using each sample set.

[0042] Further, in the process of preprocessing the collected surrounding rock characteristic data, the following steps are included: encoding qualitative data to complete quantitative conversion; resampling one-dimensional sequence data or two-dimensional image data so that data of the same dimension have the same data size; normalizing or standardizing data of different unit levels; and calculating the surrounding rock attributes of the measured data.

[0043] Specifically, since data preprocessing is a necessary step for machine learning training sample preparation, it ensures that each type of data input into the deep learning model has the same data size and numerical range. For some qualitative inputs commonly used in surrounding rock grading description, they need to be encoded (for example, One-Hot encoding) first for quantitative conversion to facilitate computer processing. There may be a large difference in the numerical magnitude between different quantitative inputs, so it is necessary to normalize or standardize the quantitative data of the same dimension and different types to eliminate the influence of the difference in the numerical magnitude of data from different sources on the training accuracy of the model. When inputting two-dimensional images or one-dimensional sequences (such as curves), resampling processing needs to be performed on each dimension of different characteristic data respectively to ensure that data of the same dimension have the same data size. Based on the measured data, specific attribute calculations can be performed to extract effective surrounding rock characteristics. Next, the preprocessed surrounding rock characteristic data samples are randomly divided into training samples, validation samples and test samples according to a preset ratio (for example, 70%, 20% and 10%), which are used for model training, validation and quality control evaluation, respectively.

[0044] Next, enter step S120 to construct a suitable preset fusion model according to the types and formats of the inputs and outputs in the prepared data set and train the model.

[0045] In order to simultaneously consider the input of multiple different types of data, the preset fusion model is configured as two sequentially connected parts, a feature recognition part and a target learning part. The feature recognition part applies a designated network model to different types of input to perform feature learning, and then fuses the features learned from all input data together, so as to link the target label of the surrounding rock level through a proper target learning network, and realize target learning.

[0046] Specifically, in the feature recognition part, the embodiment of the present application constructs a corresponding feature learning network for each data type (data format). When constructing the corresponding feature learning network for each data type, for discrete numerical type training input data, a multi-layer perceptron or support vector machine network can be applied to construct a feature learning network for this type of input; for structured image or curve type training input data, a convolutional neural network can be applied to construct a feature learning network for this type of input; for text type training input data, a recurrent neural network can be applied to construct a feature learning network for this type of input.

[0047] In addition, the target learning part adopts an algorithm such as a fully connected neural network to construct a multi-level classification and output of the surrounding rock level. Among them, the softmax function is applied as the activation function of the fully connected neural network to realize the classification and output of the surrounding rock level.

[0048] In addition, the preset fusion model described in the embodiment of the present application further includes a feature fusion part between the feature recognition part and the target learning part. The feature fusion part is a feature conversion processing model that fuses different features together. On the basis of independent feature learning, the feature fusion part connects all learned features to form a fused feature set. The features output by each feature learning network in the feature recognition part are essentially arrays. When connecting different groups of feature data (a feature learning network outputs a group of feature arrays), according to the format type of the input data, the feature arrays output by different feature learning networks can be connected horizontally (row expansion) or vertically (column expansion) according to the horizontal direction (feature direction) or vertical direction (channel direction) of the matrix, to form a feature set matrix.

[0049] Specifically, in the process of forming the feature set matrix, the feature fusion unit will learn the feature learning network for the feature containing discrete data in the input data, and concatenate each group of arrays in the feature direction according to the feature array output by the feature learning network; for the feature learning network of one-dimensional or two-dimensional structured data, the feature array output by the current feature learning network is first connected in parallel in the channel direction, and then the connected array is matrix converted to form an array with the same dimension as the matrix that has been connected in series, so that the array formed after the current conversion is connected in series with the matrix connected in series, so that the feature set matrix is formed.

[0050] That is, in the embodiment of the application, a plurality of feature learning networks constructed for different input data types are connected in parallel to form a feature recognition unit, and then the feature recognition unit is connected in series with a feature fusion unit and a target learning unit to form a preset fusion model.

[0051] Then, based on the collected (preprocessed) surrounding rock feature data and the training sample set, the validation sample set and the test sample set corresponding to the surrounding rock level data set, the preset fusion model based on multi-source information is trained, validated and tested, so as to form a surrounding rock intelligent grading model.

[0052] According to the specific characteristics of the training data and the network structure of the preset fusion model, a suitable loss function is selected as the objective function of the network model. Based on the constructed surrounding rock intelligent grading hybrid deep neural network (preset fusion model), the prepared training samples, validation samples and test samples are used to carry out model training, parameter debugging, model validation and quality control evaluation, and the optimal solution is carried out, and cross-validation, hyperparameter optimization and quality control evaluation are carried out. Thus, the optimal training of the model is completed, and a trained high-precision and strong-generalization-capability surrounding rock intelligent grading model is obtained.

[0053] After the construction of the surrounding rock intelligent grading model is completed, step S130 is entered to put the constructed surrounding rock intelligent grading model into application.

[0054] In step S130, the trained surrounding rock intelligent grading model is applied, and the surrounding rock feature measurement data preprocessed in the same way after data type classification is input into the surrounding rock intelligent grading model, and the target surrounding rock is graded and evaluated, so that the surrounding rock level information output by the intelligent grading model is used to guide the tunnel construction (which can include design before construction and design change during construction, such as guiding the design and change of tunnel excavation method (full-face, three-step method, etc.) and the design and change of support parameters, etc.).

[0055] In addition, in order to improve the applicability and accuracy of the intelligent surrounding rock grading model, the tunnel surrounding rock grading method provided by the embodiment of the present application also needs to continuously update the intelligent surrounding rock grading model. Based on the advancement of different tunnels and different construction stages, the surrounding rock feature data and the surrounding rock level data set are expanded, so as to update and apply the intelligent surrounding rock grading model by using the expanded surrounding rock feature data and the surrounding rock level data set.

[0056] Specifically, with the continuous accumulation of a large number of data samples related to surrounding rock grading in the implementation process of a large number of tunnel projects, in order to improve the accuracy or accuracy of surrounding rock grading, incremental learning is carried out at different tunnels and different construction stages, and the intelligent surrounding rock grading model is continuously updated and improved.

[0057] On the other hand, based on the above-mentioned tunnel surrounding rock grading method, the embodiment of the present application also provides a grading system for tunnel surrounding rock. Figure 4 The frame structure of the grading system for tunnel surrounding rock of the embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the tunnel surrounding rock grading system provided by the embodiment of the present application comprises a data collection module 41, an intelligent grading model construction module 42 and a model application module 43. Figure 4

[0058] Specifically, the data set collection module 41 is implemented according to the above-mentioned step S110 and is configured to collect surrounding rock feature data and corresponding surrounding rock levels; the intelligent grading model construction module 42 is implemented according to the above-mentioned step S120 and is configured to construct a preset fusion model, and train the preset fusion model according to the surrounding rock feature data and the corresponding surrounding rock levels, to form an intelligent surrounding rock grading model, wherein the preset fusion model comprises a feature recognition part and a target learning part, and the feature recognition part has a feature learning network established for different types of measurement data; the model application module 43 is implemented according to the above-mentioned step S130 and is configured to use the intelligent surrounding rock grading model to carry out grading evaluation on the target surrounding rock.

[0059] ​The application discloses a kind of grading method and system for tunnel surrounding rock.The intelligent grading method of surrounding rock based on multi-source heterogeneous information fusion proposed in the method and system first comprehensively considers four categories of data, such as rock mineral properties, discontinuous body development, geological condition background and engineering construction, which are directly related to surrounding rock classification, fully utilizes the advantages of big data, and ensures the data foundation for mining the internal relationship between various measurements and surrounding rock classification goals.Secondly, a hybrid deep neural network model is constructed for intelligent classification of surrounding rock.According to the characteristics of input and output data types, formats, etc., multiple network modules are reasonably combined, showing good flexibility and wide applicability, and can effectively realize feature learning and target learning of surrounding rock intelligent classification task.Thirdly, considering that part of the input data cannot be obtained in practical application, the modular assembly of the network architecture designed in the application can be simplified and adapted accordingly, ensuring the practicability of the network model in the application of surrounding rock intelligent classification.In addition, the intelligent classification method of surrounding rock proposed in the application applies the incremental learning of the hybrid deep neural network used by the gradually accumulated data, and continuously optimizes the precision and generalization ability of the surrounding rock intelligent classification model.

[0060] Based on the multi-source heterogeneous information affecting and describing the state of surrounding rock, the intelligent classification method and system of surrounding rock proposed in the application have the following effects:

[0061] (1) The application fully considers two types of factors, such as causes and manifestations, which are directly related to the classification and evaluation of surrounding rock.The cause factors include macro-geological factors and micro-mechanical parameters affecting the state of surrounding rock, and the manifestation factors include various measurements and corresponding calculation data for measuring the state of surrounding rock.From the perspective of the target categories reflected by the considered factors, these factors are divided into four types of parameters, such as rock mineral properties, discontinuous body development, geological condition background and engineering construction, which have different sources, formats and scales.In summary, the comprehensive consideration of multi-source heterogeneous data fully utilizes the advantages of big data, and is suitable for mining more accurate internal relationship between various measurements and surrounding rock classification goals.

[0062] (2) The application constructs a hybrid deep neural network suitable for intelligent classification of surrounding rock, which reasonably combines multiple network modules according to the characteristics of input and output data types, formats, etc., shows good flexibility and wide applicability, and effectively realizes feature learning and target learning of surrounding rock intelligent classification task.

[0063] (3) In practical application, part of the data may be difficult to collect or measure, and when only part of the multi-source heterogeneous information can be obtained, the modular assembly property of the network architecture designed in the application can be simplified and adapted accordingly, ensuring the practicability of the network model in the application of surrounding rock intelligent classification.

[0064] (4) The surrounding rock intelligent grading model provided by the application can perform adaptive incremental learning on the used hybrid deep neural network by using gradually accumulated measurement data, so that the precision and generalization ability of the surrounding rock intelligent grading model are continuously optimized for different tunnel projects and different construction stages.

[0065] The above merely describes preferred embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0066] It should be understood that the embodiments disclosed herein are not limited to the specific structures, processing steps or materials disclosed herein, but extend to equivalent alternatives of these features understood by those skilled in the relevant art. It should also be understood that the terminology used herein is for the purpose of describing specific embodiments only and is not intended to be limiting.

[0067] The phrase "one embodiment" or "an embodiment" as used throughout this description means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. Therefore, the appearances of the phrase "one embodiment" or "an embodiment" throughout the description are not necessarily all referring to the same embodiment.

[0068] Although the embodiments disclosed by the present application are as above, the content described is only the embodiments adopted for the purpose of facilitating the understanding of the present application, and is not intended to limit the present application. Any person skilled in the art of the present application can make any modification and change in the implementation form and details without departing from the spirit and scope of the present application, but the patent protection scope of the present application should be subject to the scope defined by the appended claims.

Claims

1. A method for the classification of tunnel surrounding rock, characterized in that, The method comprises the following steps: Collecting surrounding rock characteristic data and corresponding surrounding rock grades, wherein the surrounding rock characteristic data includes multiple surrounding rock characteristic parameters of different sources, different formats and different scales, and the categories of the multiple surrounding rock characteristic parameters include rock mineral property characteristics, discontinuous development characteristics, geological condition backgrounds and engineering construction characteristics; Constructing a preset fusion model and training the preset fusion model according to the surrounding rock characteristic data and corresponding surrounding rock grades to form a surrounding rock intelligent grading model; Using the surrounding rock intelligent grading model to carry out grading evaluation on target surrounding rock, wherein The preset fusion model is a feature recognition part formed by connecting multiple feature learning networks in parallel, and the feature recognition part is connected with a feature fusion part and a target learning part in sequence, The feature recognition part has feature learning networks established for different types of measurement data, wherein for discrete numerical type training input data, a multilayer perceptron or a support vector machine is used to construct a corresponding feature learning network; for structured image or curve type training input data, a convolutional neural network is used to construct a corresponding feature learning network; and for text type training input data, a recurrent neural network is used to construct a corresponding feature learning network; The feature fusion part is used to connect the feature arrays output from each feature learning network in the feature recognition part, wherein according to the format type of the input data, the feature arrays output by different feature learning networks are connected horizontally and / or vertically in the matrix horizontal direction representing the feature direction and the matrix vertical direction representing the channel direction to form the feature fusion part; The target learning part is used to realize multi-level classification and output of surrounding rock grades.

2. The grading method according to claim 1, wherein The rock mineral property characteristics include but are not limited to lithology, rock velocity, rock hardness type, rock abrasion resistance, uniaxial compressive strength, tensile strength, Young's modulus, triaxial stress, anisotropy, Poisson's ratio, porosity and permeability; The discontinuous development characteristics include but are not limited to discontinuous body types and development parameters including width, spacing, number, occurrence, roughness, ductility, filler and weathering alteration degree; The geological condition backgrounds include but are not limited to tectonic types, seismic intensity, groundwater, ground stress and ground temperature; The engineering construction characteristics include but are not limited to the angle between the tunnel axis and the main structure surface and the tunnel excavation method.

3. The ranking method of claim 1, wherein, The method further comprises the following steps: Preprocessing the surrounding rock characteristic data; Dividing the preprocessed surrounding rock characteristic data and corresponding surrounding rock grades into a training sample set, a verification sample set and a test sample set according to a preset data sample ratio, so as to train, cross-verify, and optimize and quality control evaluate the preset fusion model in the model training process by using each sample set.

4. The ranking method of claim 3, wherein, In the step of preprocessing the surrounding rock characteristic data, the following steps are included: encoding the qualitative type of data to complete quantitative transformation; resampling the one-dimensional sequence data or two-dimensional image data respectively, so that the same dimension data has the same data size; normalizing or standardizing the quantitative data of the same dimension and different types; calculating the surrounding rock attribute based on the measured data.

5. The ranking method of claim 1, wherein, The target learning part adopts a fully connected neural network to construct, and a softmax function is applied to realize the classification and output of the surrounding rock grade.

6. The grading method according to any one of claims 1-5, characterized in that, based on the advancement of different tunnels and different construction stages, the surrounding rock feature data and the surrounding rock grade data set are expanded, and the surrounding rock intelligent grading model is updated and applied by using the expanded surrounding rock feature data and surrounding rock grade data set.

7. A classification system for tunnel surrounding rock, characterized in that, It includes: a data set collection module configured to collect surrounding rock feature data and corresponding surrounding rock grades, wherein: collecting surrounding rock feature data of each sample and labeling the corresponding surrounding rock grade label, thereby forming a surrounding rock feature data and surrounding rock grade data set, the surrounding rock feature data includes multiple surrounding rock feature parameters of different sources, different formats and different scales, and the categories of the multiple surrounding rock feature parameters include rock mineral property characteristics, discontinuous development characteristics, geological condition background and engineering construction characteristics; an intelligent grading model construction module configured to construct a preset fusion model, and train the preset fusion model according to the surrounding rock feature data and the corresponding surrounding rock grades, to form a surrounding rock intelligent grading model; a model application module configured to use the surrounding rock intelligent grading model to carry out grading evaluation on the target surrounding rock, wherein, the preset fusion model is a feature recognition part formed by connecting in parallel multiple feature learning networks respectively constructed for different data types, and the feature recognition part is connected in series with a feature fusion part and a target learning part in turn, the feature recognition part has feature learning networks respectively established for different types of measured data, including: for discrete numerical type training input data, a multilayer perceptron or a support vector machine is used to construct the corresponding feature learning network; for structured image or curve type training input data, a convolutional neural network is used to construct the corresponding feature learning network; for text type training input data, a recurrent neural network is used to construct the corresponding feature learning network; the feature fusion part is used to connect the feature arrays output from each feature learning network in the feature recognition part, including: according to the format type of the input data, the feature arrays output by different feature learning networks are connected horizontally and / or vertically in the matrix horizontal direction representing the feature direction and the matrix vertical direction representing the channel direction, to form the feature fusion part; the target learning part is used to realize the multi-level classification and output of the surrounding rock grade.

Images