Tunnel surrounding rock characteristic characterization and grading method, device and system and storage medium

Abstract

The invention discloses a tunnel surrounding rock characteristic characterization and grading method and device, a system and a storage medium, a multi-modal heterogeneous database is constructed by fusing while-drilling parameters, geological radar advanced geological forecast images, geological exploration texts and construction data, and through depth feature extraction and cross-modal alignment of multi-source data, the tunnel surrounding rock characteristic characterization and grading method is constructed. And the surrounding rock strength, the integrity coefficient and the grading result are synchronously inverted, so that intelligent grading and comprehensive evaluation are performed on the front surrounding rock more comprehensively, the grading effect of the surrounding rock is improved, the construction quality of a project is improved, the construction safety is guaranteed, and the intelligent construction level of a tunnel is improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of tunnel exploration, and particularly relates to a tunnel surrounding rock characteristic representation and grading method and device, system and storage medium. BACKGROUND

[0002] Tunnel is an important part of high-speed traffic line construction, but due to the concealment and limitation of its operation environment, construction faces great challenges. In order to ensure the safety of tunnel construction period and operation period, it is necessary to accurately judge the tunnel surrounding rock conditions and give corresponding support measures. Tunnel surrounding rock evaluation is an important link in tunnel construction, and surrounding rock grade division and quality evaluation can provide basis for support design and blasting parameter adjustment. The traditional surrounding rock evaluation seriously relies on field operation personnel, and has low precision and poor efficiency. With the rapid development of computer field and the rise of big data and artificial intelligence, the intelligent evaluation of surrounding rock has also developed rapidly, and a large number of new technologies have been introduced, and the data sources are also more diverse, such as drilling parameters, digital images and three-dimensional laser scanning point clouds, which greatly improve the accuracy and work efficiency of surrounding rock quality evaluation and provide support for dynamic design of tunnel construction.

[0003] The current intelligent grading method of tunnel surrounding rock has limitations: the data source is highly single, the existing technology only relies on drilling parameters or wave velocity information, or lacks geological text description, so that the key features such as joint development and weathering degree of rock mass are not fully utilized. Such single modal data is difficult to fully represent the mechanical behavior of surrounding rock, which restricts the accuracy of grading. The output dimension of the existing model is insufficient: only the surrounding rock grading label is output, and the mechanical parameter information such as rock mass strength and integrity coefficient cannot be provided, which reduces the result reliability. In addition, the existing technology has highly limited functions: the existing model only serves the single task of grading, cannot simultaneously support the research and demand of multi-task such as surrounding rock strength and integrity coefficient, and causes waste of data value. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a tunnel surrounding rock characteristic representation and grading method and device, system and storage medium.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0006] A tunnel surrounding rock characteristic representation and grading method, comprising:

[0007] obtaining drilling data of rock mass in front of the working face, geological radar advanced geological prediction map, geological exploration text and construction data;

[0008] obtaining tunnel surrounding rock geological fusion features of the working face scale according to the drilling data and the geological radar advanced geological prediction map;

[0009] According to the geological exploration text and the construction data, tunnel surrounding rock geological fusion characteristics of engineering scale are obtained.

[0010] The tunnel surrounding rock geological fusion characteristics of the engineering scale are input into the surrounding rock characteristic intelligent identification and grading model based on the cross attention mechanism to obtain the surrounding rock characteristic representation and the surrounding rock grading.

[0011] Preferably, obtaining the tunnel surrounding rock geological fusion characteristics of the engineering scale comprises:

[0012] The while-drilling data are input into the rock mass feature main data feature extraction model 1 to obtain the spatiotemporal feature data of the rock mass;

[0013] The geological radar advanced geological prediction map is input into the rock mass feature main data feature extraction model 2 to obtain the spatial feature of the adverse geology of the surrounding rock in front of the working face;

[0014] According to the spatiotemporal feature data of the rock mass and the spatial feature of the adverse geology of the surrounding rock in front of the working face, the tunnel surrounding rock geological fusion characteristics of the engineering scale are obtained.

[0015] Preferably, obtaining the tunnel surrounding rock geological fusion characteristics of the engineering scale comprises:

[0016] The construction data are input into the rock mass auxiliary feature extraction model 1 to obtain the engineering scale geostress feature;

[0017] The geological exploration data are input into the rock mass auxiliary feature extraction model 2 based on the geological text to obtain the engineering scale surrounding rock feature;

[0018] According to the engineering scale geostress feature and the engineering scale surrounding rock feature, the tunnel surrounding rock geological fusion characteristics of the engineering scale are obtained.

[0019] The application further provides a tunnel surrounding rock characteristic representation and grading device, which comprises:

[0020] The first processing module is used for acquiring the while-drilling data, the geological radar advanced geological prediction map, the geological exploration text and the construction data of the rock mass in front of the working face;

[0021] The second processing module is used for obtaining the tunnel surrounding rock geological fusion characteristics of the engineering scale according to the while-drilling data and the geological radar advanced geological prediction map;

[0022] The third processing module is used for obtaining the tunnel surrounding rock geological fusion characteristics of the engineering scale according to the geological exploration text and the construction data;

[0023] The fourth processing module is used to input the geological fusion characteristics of the tunnel surrounding rock at the working face scale and the geological fusion characteristics of the tunnel surrounding rock at the engineering scale into the intelligent identification and classification model of surrounding rock characteristics based on the cross-attention mechanism, so as to obtain the characterization and classification of surrounding rock characteristics.

[0024] Preferably, the second processing module includes:

[0025] The first processing unit is used to input the drilling data into the main data feature extraction model 1 of the rock mass to obtain the spatiotemporal feature data of the rock mass;

[0026] The second processing unit is used to input the geological radar advanced geological prediction map into the rock mass feature extraction model 2 to obtain the spatial characteristics of the adverse geology of the surrounding rock in front of the tunnel face.

[0027] The third processing unit is used to obtain the geological fusion characteristics of the tunnel surrounding rock at the tunnel face scale based on the spatiotemporal characteristic data of the rock mass and the spatial characteristics of the adverse geology of the surrounding rock in front of the tunnel face.

[0028] Preferably, the third processing module includes:

[0029] The fourth processing unit is used to input the construction data into the rock mass auxiliary feature extraction model 1 to obtain the geostress characteristics at the engineering scale.

[0030] The fifth processing unit is used to input geological exploration data into the rock mass auxiliary feature extraction model 2 based on geological text to obtain the surrounding rock features at the engineering scale.

[0031] The sixth processing unit is used to obtain the geological fusion characteristics of the tunnel surrounding rock at the engineering scale based on the engineering-scale geostress characteristics and the engineering-scale surrounding rock characteristics.

[0032] The present invention also provides a tunnel surrounding rock characteristic characterization and classification system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a tunnel surrounding rock characteristic characterization and classification method when executed by the processor.

[0033] The present invention also provides a storage medium storing a computer program, which executes a method for characterizing and classifying the characteristics of tunnel surrounding rock when running.

[0034] This invention integrates drilling parameters, ground-penetrating radar advanced geological prediction images, geological exploration texts, and construction data to construct a multimodal heterogeneous database. By extracting deep features from multi-source data and aligning them across modes, it synchronously inverts the surrounding rock strength, integrity coefficient, and classification results. This enables a more comprehensive intelligent classification and integrated evaluation of the surrounding rock ahead, improving the classification effect of the surrounding rock, helping to improve the construction quality of the project, ensuring construction safety, and enhancing the level of intelligent tunnel construction. Attached Figure Description

[0035] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0036] Figure 1 This is a flowchart of the tunnel surrounding rock characteristic characterization and classification method according to an embodiment of the present invention;

[0037] Figure 2 The training loss for the model;

[0038] Figure 3 To verify and test the rock mass quality. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0041] Example 1:

[0042] like Figure 1 As shown, this embodiment of the invention provides a method for characterizing and classifying the properties of tunnel surrounding rock, including:

[0043] Step S1: Obtain the main data and auxiliary data. The main data are the drilling data of the rock mass in front of the tunnel face and the advanced geological prediction map of the ground-penetrating radar. The auxiliary data are geological exploration text and construction data.

[0044] Step S2: Input the drilling data into the main data feature extraction model 1 of the rock mass to obtain the spatiotemporal feature data of the rock mass;

[0045] Step S3, input the geological radar advanced geological prediction map into the rock mass feature main data feature extraction model 2 to obtain the spatial features of the adverse geology of the surrounding rock in front of the working face;

[0046] Step S4, according to the spatiotemporal feature data of the rock mass and the spatial features of the adverse geology of the surrounding rock in front of the working face, the tunnel surrounding rock geology fusion features at the working face scale are obtained; wherein, the spatiotemporal feature data of the rock mass and the spatial features of the adverse geology of the surrounding rock in front of the working face are fused by using the DNN model 1 to obtain the tunnel surrounding rock geology fusion features at the working face scale;

[0047] Step S5, input the construction data into the rock mass auxiliary feature extraction model 1 to obtain the engineering scale ground stress features;

[0048] Step S6, input the geological survey data into the rock mass auxiliary feature extraction model 2 based on geological text to obtain the engineering scale surrounding rock features;

[0049] Step S7, according to the engineering scale ground stress features and the engineering scale surrounding rock features, the engineering scale tunnel surrounding rock geology fusion features are obtained; the engineering scale ground stress features and the engineering scale surrounding rock features are fused by using the DNN model 2 to obtain the engineering scale tunnel surrounding rock geology fusion features;

[0050] Step S8, input the tunnel surrounding rock geology fusion features at the working face scale and the tunnel surrounding rock geology fusion features at the engineering scale into the surrounding rock property intelligent identification and classification model based on the cross attention mechanism to obtain the surrounding rock property representation and the surrounding rock classification; wherein, the surrounding rock intelligent identification and classification model includes: a prediction model 1, a prediction model 2 and a prediction model 3.

[0051] As an embodiment of the present application, in step S1, various geological, engineering and operational data are acquired and recorded in real time during the drilling process. These data can help engineers and technicians understand the conditions of the formation below the wellhead, optimize drilling operations, ensure the stability of the wellbore, and take necessary measures to deal with potential problems. While drilling data is usually collected through various sensors and measuring devices, which can be installed in the drill pipe, drill bit and drilling fluid handling system, etc. The while drilling data generally includes drilling speed, drilling depth, impact pressure, rotation pressure, rotation speed, water pressure, water flow, specific energy, etc. Field data is collected through engineering construction, rock drilling jumbo and geological radar, including while-drilling parameters of the rock drill (drilling speed, drilling depth, rotation speed, impact pressure, jacking pressure, rotation pressure, water pressure, water flow, specific energy), tunnel construction data (construction footage, support stiffness, tunnel area, tunnel height, tunnel width, tunnel burial depth), geological radar data (radar image), geological survey text data (geological text).

[0052] As an embodiment of the embodiment of the application, in step S2, the while-drilling data is obtained by the while-drilling acquisition technology, and the obtained while-drilling data includes drilling depth, drilling speed, impact pressure, jacking pressure, rotation pressure, rotation speed, water flow, water pressure, specific energy, etc. These data are input into the rock mass space-time feature main data feature extraction model 1 to obtain the space-time feature data of the rock mass.

[0053] The input while-drilling data (spatial sequence type) of the model 1 is the surrounding rock characteristics (strength, integrity and surrounding rock classification) related to the high-dimensional geological space-time characteristics of the working face; wherein the while-drilling data is selected from 7 drill holes, the length is selected as 2m, and each 0.02m is recorded once, a total of 100 drill hole data acquisition points, 9 parameters of drilling speed, drilling depth, rotation speed, impact pressure, jacking pressure, rotation pressure, water pressure, water flow and specific energy are obtained respectively, and a data form of 7*9*100 is formed as input.

[0054] The model 1 is a hybrid neural network structure, which combines the advantages of LSTM, CNN and DNN to process time series data. The dimension of the input data is [5, 9, 20], which respectively represents that the batch size is 5, the sequence length is 20, and the feature dimension is 9. The model 1 extracts features through three parallel paths, and finally splices and fuses the features.

[0055] 1. LSTM part

[0056] A 3-layer LSTM network is used to process time series data:

[0057] h t ,c t =LSTM(x t ,h t-1 ,c t-1 )

[0058] Input: x t (input vector at time step t, dimension 20)

[0059] Output: h t (hidden state, dimension d h ), c t (cell state, dimension d h )

[0060] d h represents the dimension of the LSTM hidden layer

[0061] Final output dimension: [5, d h ](take the output of the last time step)

[0062] 2. CNN part

[0063] Process data with 1D CNN:

[0064] y i = ReLU(W*x i:i+k-1 +b)

[0065] where y_i is the value of the output feature map of the convolutional layer at position i, representing the response of the input data at this position after convolution and activation; ReLU(·) is the activation function, defined as max(0, z), used to introduce nonlinearity and enhance the expressive power of the model; W is the weight vector of the convolution kernel (filter), with a length of k, the size of the convolution kernel, which is a learnable parameter, used to extract local features in the input; * is the convolution operation, representing the dot product (element-wise multiplication and summation) of the convolution kernel W and the local segment x_{i:i+k-1} of the input; x_{i:i+k-1} is a subsequence of the input sequence starting from index i with a length of k, which is the input data window covered by the current convolution operation; b is the bias term, a learnable scalar parameter, added to the convolution result to improve the fitting ability of the model.

[0066] 3. Feature combination section

[0067] Feature 1: Post-processing after adding LSTM and CNN results

[0068] Feature1 = LayerNorm(ReLU(W d1 (LSTM out + CNN out )+ b d1 )

[0069] where LayerNorm(·) is the layer normalization operation, which normalizes the output of ReLU, stabilizes the training process, and improves the convergence and generalization ability of the model; LSTM out , CNN out are the results of LSTM and CNN outputs, respectively, LSTM corresponds to h t , and CNN corresponds to y i ; W d1 is a learnable weight matrix used to process the fused features, acting on the results of LSTM out + CNN out to achieve feature transformation; b d1 is the bias vector corresponding to W d1 , which is a learnable parameter used to adjust the output of linear transformation.

[0070] Feature 2: Separate processing of LSTM results

[0071] Feature2 = W d2 LSTMout +b d2

[0072] where W d2 is the weight matrix for processing the LSTM individual output, is a learnable parameter; b d2 is the bias vector corresponding to W d2 .

[0073] Feature 3: CNN result individual processing

[0074] Feature3 = W d3 CNN out +b d3

[0075] where W d3 is the weight matrix for processing the LSTM individual output, is a learnable parameter; B d3 is the bias vector corresponding to W d3 .

[0076] 4. Feature fusion

[0077] Three feature splicing fusion:

[0078] FinalFeature = [Feature1; Feature2; Feature3]

[0079] DNN part

[0080] The model contains 4 layers of DNN, and each layer structure is:

[0081] h L = ReLU(W L h L-1 +b L ), L = 1, 2, 3

[0082] h4 = W4h3 + b4

[0083] where h L is the output of the Lth hidden layer, h0 is the input to be processed data, such as FinalFeature; W L and b L are the weight and bias matrix of the Lth layer, respectively.

[0084] The main data extraction model of the rock mass space-time feature based on the while-drilling data is obtained through the following steps, including: obtaining a training set, the training set including while-drilling data such as drilling depth, drilling speed, impact pressure, jacking pressure, rotary pressure, rotary speed, water flow, water pressure, specific energy, etc.; inputting the while-drilling data into the main data extraction model of the space-time feature of the surrounding rock, and constructing a mapping relationship between the while-drilling data and the space-time feature of the surrounding rock characteristics; and outputting the space-time feature data of the rock mass based on the training model.

[0085] For the while-drilling data, the feature extraction model 1 is established by the CNN model and the LSTM model, and the space-time feature of the data about the surrounding rock characteristics in the while-drilling data is extracted.

[0086] As an embodiment of the present application, in step S3, the geological radar image data of the surrounding rock in front of the tunnel face is obtained through radar scanning technology, and the processed 256*512 image data is obtained through image preprocessing, including cropping, image enhancement and normalization operations. The processed geological radar image data is input into the ResNet model, and the spatial feature of the adverse geology of the surrounding rock in front of the tunnel face is obtained through the feature extraction of the ResNet model.

[0087] The rock mass ResNet feature extraction model based on the tunnel face image is obtained through the following steps, including: obtaining image training set data; inputting the image data into the rock mass ResNet feature extraction model, and constructing a mapping relationship between the image data and the surrounding rock structure feature (adverse geology); and outputting the spatial feature of the adverse geology of the surrounding rock in front of the tunnel face based on the training model.

[0088] As an embodiment of the present application, in step S5, the construction data of the tunnel is obtained through historical construction records, and the construction data includes tunnel construction footage, support stiffness, tunnel area, tunnel height, tunnel width, and tunnel burial depth. These data are input into the rock mass auxiliary feature extraction model 1 to obtain the engineering scale in-situ stress feature. The construction data is the construction data of the first 40m before the historical excavation, and is recorded every 0.5m, a total of 80 data, each of which involves construction footage, support stiffness, tunnel area, tunnel height, tunnel width, and tunnel burial depth, forming a 6*80 data form.

[0089] The surrounding rock auxiliary feature extraction model 1 based on the construction data is obtained through the following steps, including: obtaining a training set, the training set including construction data such as tunnel construction footage, support stiffness, tunnel area, tunnel height, tunnel width, and tunnel burial depth; inputting the construction data into the rock mass auxiliary feature extraction model 1, and constructing a mapping relationship between the construction data and the surrounding rock auxiliary data; and outputting the engineering scale in-situ stress feature based on the training model.

[0090] For the construction data, the DNN model is used to establish the mapping relationship between the rock mass auxiliary features and the construction data, and the engineering scale stress features can be obtained through the model.

[0091] As an embodiment of the embodiment of the application, in step S6, geological survey data of the tunnel is obtained through geological survey, the geological survey data includes rock mass category, weathering degree, fracture development condition, groundwater condition, rock layer strike and inclination angle and the like, and the data is input into the rock mass auxiliary feature extraction model 2 (Bert text model) based on geological text to obtain engineering scale surrounding rock features.

[0092] The input data of the rock mass auxiliary feature extraction model 2 is the geological survey data which is text data, and the extracted data features are engineering scale surrounding rock features (rock mass type, weathering degree, groundwater and the like)

[0093] · The bottom layer EN marked Embedding module

[0094] · The middle layer Trm marked Transformer module

[0095] · The uppermost layer TN marked pre-fine-tuning module

[0096] BERT is a bidirectional language representation model based on the Transformer architecture. Unlike traditional unidirectional language models, BERT adopts Masked Language Model (MLM) and Next Sentence Prediction (NSP) strategies for unsupervised pre-training, so as to model the semantic representation of words on both sides of the context.

[0097] Let the text input sequence be X after word embedding and position encoding text The text encoder processes the text similarly to the image:

[0098]

[0099] H' = LayerNorm(X text +A')

[0100] Z text = LayerNorm(H' + FFN(H'))

[0101] Where, W Q′ ,W K′ ,W V′ is the self-attention of the text encoder.

[0102] After multi-layer stacking, the global semantic vector Z text can be further extracted.

[0103] The rock mass auxiliary feature extraction model 2 based on geological survey data is trained by the following steps, including: obtaining a training set, the training set including rock mass categories, weathering degrees, fracture development conditions, groundwater conditions, and stratum strike inclination angles and other survey data; inputting the geological survey data into the surrounding rock auxiliary feature extraction model 2 to construct a mapping relationship between the survey data and the rock mass auxiliary data; and outputting the engineering scale surrounding rock features based on the training model.

[0104] As an embodiment of the present application, in step S8, the rock mass main data and auxiliary data features are input into the cross attention mechanism model, the auxiliary data mainly assisting the main data to further extract data features. For the three tasks set, the data features are respectively input into the respective single networks of the tasks for respective data application of the tasks, and the single prediction module models of the three tasks are prediction model 1, prediction model 2 and prediction model 3 respectively, and the prediction model is a model composed of DNN, an activation function and a skip connection structure, which respectively outputs three results: the integrity, the strength and the surrounding rock classification of the rock mass.

[0105] The input data of the cross attention mechanism model is the tunnel surrounding rock geological fusion features at the face scale (fusing the face geological features and the front adverse geological features) and the tunnel surrounding rock geological fusion features at the engineering scale (fusing the ground stress and the geological features such as lithology, integrity and groundwater), and the output data is the multi-scale cross-fused tunnel surrounding rock geological features.

[0106] In the cross attention, there are two sequences x1 and x2. For each element in sequence x1, the attention weight of it to all elements in sequence x2 is calculated. These weights reflect which elements in sequence x2 should be paid more attention to by the element in sequence x1. This mechanism enables the model to learn which features should be focused on when one sequence pays attention to another sequence, thereby improving the representation ability of the fused features.

[0107] Let Q=x1W Q and K=V=x2W K Then the calculation of cross attention is as follows:

[0108]

[0109] where d k is the column number of Q, K matrix, i.e. the dimension of the vector.

[0110] The surrounding rock intelligent identification and grading model based on multi-source data fusion is trained by the following steps, including main rock mass data and auxiliary data; inputting the multi-source data features into the surrounding rock characteristic intelligent identification and grading model to construct the mapping relationship between the multi-source data and the surrounding rock characteristics and grading; and outputting the surrounding rock characteristic representation (rock strength, rock mass integrity) and the surrounding rock grading based on the trained model.

[0111] The input of the prediction model 1 is the multi-scale cross-fusion tunnel surrounding rock geological characteristics, and the output is the integrity parameter of the surrounding rock; the input of the prediction model 2 is the multi-scale cross-fusion tunnel surrounding rock geological characteristics, and the output is the strength parameter of the surrounding rock; the input of the prediction model 3 is the multi-scale cross-fusion tunnel surrounding rock geological characteristics, and the output is the surrounding rock grading; the difference between the three models lies in the different numbers of layers of FC-Tanh structure, and the remaining model structures are as shown in the figure, and the right FC-Tanh BatchNorm is a jump connection structure.

[0112] The input of the surrounding rock intelligent identification and grading model based on multi-source data fusion is the fused data features, and the output is set to the integrity parameter, the strength parameter and the surrounding rock grading according to the requirements.

[0113] For the main rock mass data and auxiliary data, the mapping relationship between the extracted data features and the surrounding rock characteristics and grading is established by using the prediction models 1-3, and the corresponding rock mass characteristics and grading results can be obtained through the classification of the feature data.

[0114] The present application performs multi-modal fusion on the while-drilling data collected in the tunneling process, radar images, geological exploration texts and historical construction data, uses a multi-task learning method, realizes multi-modal intelligent fine grading of the tunnel surrounding rock and output of the surrounding rock characteristics in the intermediate process. Through the method, the various data obtained in the tunnel construction are fully utilized, the multi-source data are effectively processed and fused, the shortcomings of incomplete input samples in the model are improved, the accuracy of the surrounding rock grading is effectively improved; meanwhile, through the identification and prediction of the combined multiple models, not only the surrounding rock grading result is obtained, but also the strength information and integrity information of the surrounding rock are output, the diversity of the output information is improved, the comprehensiveness and accuracy of the surrounding rock characteristic and grade identification are improved, the roughness of the overall identification and the subjectivity of the human identification are avoided, which is conducive to formulating more accurate construction schemes, reducing construction risks, improving construction quality and assisting intelligent tunnel construction.

[0115] Figure 2 The four loss function change trends in the model training process are shown. It can be observed that the total loss value of the model and the loss value of each single task gradually decrease, which indicates that the training process of each sub-task has good synchronicity and cooperativity.

[0116] The task prediction results of the test set based on the model are as followsFigure 3 The closer the data points are to the black reference line with a slope of 1, the higher the prediction accuracy. Figure 3 The predicted values of P21 (rock mass integrity index, total length of all fissures (including joints, faults, fissures, etc.) center lines on the rock surface (such as outcrop surface, core surface or calculation plane) per unit area), uniaxial compressive strength (rock strength) and BQ rock mass quality classification index (surrounding rock classification index) in the verification set are shown.

[0117] Embodiment 2:

[0118] The embodiment of the present application also provides a tunnel surrounding rock characteristic characterization and classification device, comprising:

[0119] The first processing module is used for acquiring the while-drilling data of the rock mass in front of the working face, the advanced geological prediction map of the geological radar, the geological exploration text and the construction data;

[0120] The second processing module is used for obtaining the tunnel surrounding rock geological fusion characteristics of the working face scale according to the while-drilling data and the advanced geological prediction map of the geological radar;

[0121] The third processing module is used for obtaining the tunnel surrounding rock geological fusion characteristics of the engineering scale according to the geological exploration text and the construction data;

[0122] The fourth processing module is used for inputting the tunnel surrounding rock geological fusion characteristics of the working face scale and the tunnel surrounding rock geological fusion characteristics of the engineering scale into the surrounding rock characteristic intelligent identification and classification model based on the cross attention mechanism to obtain the surrounding rock characteristic characterization and the surrounding rock classification.

[0123] As an implementation manner of the embodiment of the present application, the second processing module comprises:

[0124] The first processing unit is used for inputting the while-drilling data into the rock mass characteristic main data feature extraction model 1 to obtain the spatio-temporal characteristic data of the rock mass;

[0125] The second processing unit is used for inputting the advanced geological prediction map of the geological radar into the rock mass characteristic main data feature extraction model 2 to obtain the spatial characteristics of the adverse geology of the surrounding rock in front of the working face;

[0126] The third processing unit is used for obtaining the tunnel surrounding rock geological fusion characteristics of the working face scale according to the spatio-temporal characteristic data of the rock mass and the spatial characteristics of the adverse geology of the surrounding rock in front of the working face.

[0127] As an implementation manner of the embodiment of the present application, the third processing module comprises:

[0128] The fourth processing unit is used for inputting the construction data into the rock mass auxiliary feature extraction model 1 to obtain the engineering scale geostress characteristics;

[0129] The fifth processing unit is configured to input the geological survey data into the geological text-based rock mass auxiliary feature extraction model 2 to obtain engineering scale surrounding rock features;

[0130] The sixth processing unit is configured to obtain engineering scale tunnel surrounding rock geological fusion features according to the engineering scale in-situ stress features and the engineering scale surrounding rock features.

[0131] Embodiment 3

[0132] The embodiment of the present application also provides a tunnel surrounding rock characteristic characterization and grading system, which comprises a memory and a processor, the memory has a computer program run by the processor stored thereon, and the computer program performs the tunnel surrounding rock characteristic characterization and grading method when run by the processor.

[0133] Embodiment 4

[0134] The embodiment of the present application also provides a storage medium, which has a computer program stored thereon, and the computer program performs the tunnel surrounding rock characteristic characterization and grading method when run.

[0135] The above-described embodiments only describe the preferred modes of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope of the present application defined by the claims.

Claims

1. A method for characterizing and classifying the properties of the surrounding rock of a tunnel, characterized by, The method comprises the following steps: obtaining the drilling data of the rock mass in front of the tunnel face, the geological radar advanced geological prediction map, the geological exploration text and the construction data; obtaining the tunnel surrounding rock geological fusion characteristics at the scale of the tunnel face according to the drilling data and the geological radar advanced geological prediction map; obtaining the tunnel surrounding rock geological fusion characteristics at the engineering scale according to the geological exploration text and the construction data; inputting the tunnel surrounding rock geological fusion characteristics at the scale of the tunnel face and the tunnel surrounding rock geological fusion characteristics at the engineering scale into the surrounding rock property intelligent identification and classification model based on the cross attention mechanism to obtain the surrounding rock property representation and the surrounding rock classification.

2. The method of characterizing and classifying tunnel surrounding rock properties according to claim 1, characterized in that, The tunnel surrounding rock geological fusion characteristics at the scale of the tunnel face include: inputting the drilling data into the rock mass feature main data feature extraction model 1 to obtain the spatiotemporal feature data of the rock mass; inputting the geological radar advanced geological prediction map into the rock mass feature main data feature extraction model 2 to obtain the spatial feature of the adverse geology of the surrounding rock in front of the tunnel face; obtaining the tunnel surrounding rock geological fusion characteristics at the scale of the tunnel face according to the spatiotemporal feature data of the rock mass and the spatial feature of the adverse geology of the surrounding rock in front of the tunnel face.

3. The method of tunnel surrounding rock characterization and classification according to claim 2, characterized in that, The tunnel surrounding rock geological fusion characteristics at the engineering scale include: inputting the construction data into the rock mass auxiliary feature extraction model 1 to obtain the engineering scale geostress feature; inputting the geological exploration data into the rock mass auxiliary feature extraction model 2 based on the geological text to obtain the engineering scale surrounding rock feature; obtaining the tunnel surrounding rock geological fusion characteristics at the engineering scale according to the engineering scale geostress feature and the engineering scale surrounding rock feature.

4. A device for characterizing and classifying the properties of the surrounding rock of a tunnel, characterized in that, The method comprises the following steps: a first processing module is configured to obtain the drilling data of the rock mass in front of the tunnel face, the geological radar advanced geological prediction map, the geological exploration text and the construction data; a second processing module is configured to obtain the tunnel surrounding rock geological fusion characteristics at the scale of the tunnel face according to the drilling data and the geological radar advanced geological prediction map; a third processing module is configured to obtain the tunnel surrounding rock geological fusion characteristics at the engineering scale according to the geological exploration text and the construction data; a fourth processing module is configured to input the tunnel surrounding rock geological fusion characteristics at the scale of the tunnel face and the tunnel surrounding rock geological fusion characteristics at the engineering scale into the surrounding rock property intelligent identification and classification model based on the cross attention mechanism to obtain the surrounding rock property representation and the surrounding rock classification.

5. The device for characterization and classification of tunnel surrounding rock mass properties as claimed in claim 4, characterized in that, The second processing module comprises: a first processing unit is configured to input the drilling data into the rock mass feature main data feature extraction model 1 to obtain the spatiotemporal feature data of the rock mass; a second processing unit is configured to input the geological radar advanced geological prediction map into the rock mass feature main data feature extraction model 2 to obtain the spatial feature of the adverse geology of the surrounding rock in front of the tunnel face; a third processing unit is configured to obtain the tunnel surrounding rock geological fusion characteristics at the scale of the tunnel face according to the spatiotemporal feature data of the rock mass and the spatial feature of the adverse geology of the surrounding rock in front of the tunnel face.

6. The device for characterization and classification of tunnel surrounding rock mass properties as claimed in claim 5, characterized in that, The third processing module comprises: a fourth processing unit is configured to input the construction data into the rock mass auxiliary feature extraction model 1 to obtain the engineering scale geostress feature; a fifth processing unit is configured to input the geological exploration data into the rock mass auxiliary feature extraction model 2 based on the geological text to obtain the engineering scale surrounding rock feature; A sixth processing unit is configured to obtain an engineering-scale tunnel surrounding rock geology fusion feature according to the engineering-scale stress feature and the engineering-scale surrounding rock feature.

7. A system for characterizing and classifying the properties of tunnel surrounding rock, characterized by The application further provides a computer readable storage medium having stored thereon a computer program. The application further provides a computer readable storage medium having stored thereon a computer program.

8. A storage medium, characterized by The application further provides a computer readable storage medium having stored thereon a computer program.

Images