Method, device, medium and product for identifying rock mass cracks in front of tunnel face

Abstract

The invention discloses a method, device, medium and product for identifying rock cracks in front of a tunnel face, and relates to the technical field of rock crack identification. The method comprises the following steps: training the feature extraction model with a data set containing drilling parameters of a plurality of adjacent borehole pairs to obtain a trained encoder, wherein a target loss function used in the training is determined by specific energy features, predicted specific energy and label specific energy corresponding to each borehole in the adjacent borehole pairs; inputting tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters and the target specific energy of a target tunnel construction operation into a trained rock crack identification model including the trained encoder to obtain the rock crack type in front of the target tunnel face, wherein the extracted drill rod geological feature response is taken into account in the rock crack identification model, so that the recognition accuracy of the rock crack type is higher.

Description

Technical Field

[0001] The invention relates to the technical field of rock mass crack identification, and in particular to a method, device, medium and product for identifying rock mass cracks in front of a tunnel face. Background Art

[0002] With the rapid development of my country's transportation infrastructure construction, tunnel engineering, as an important component, is constantly developing in the direction of longer, larger and deeper to meet the needs of extending the transportation network to complex geological areas such as mountainous areas. In the construction of mountain tunnel engineering, drilling and blasting, tunnel boring machine (TBM) and shield method are widely used. Mechanized construction methods such as TBM and shield method are highly efficient. Due to the strong flexibility and low economic cost of drilling and blasting construction, it is one of the mainstream methods for mountain tunnel construction, that is, drilling and blasting is still the most widely used tunnel construction method. With the upgrading of tunnel construction industry and the process of digitalization and intelligence, a large amount of data is generated in the construction of drilling and blasting tunnels, and geological information is the core of tunnel engineering construction. Among them, the data generated by the drilling rig in the tunnel blasting operation is closely related to the information of the rock mass in front of the face. The interpretation of this information can realize the refined identification of the rock mass in front of the face. On this basis, the blasting quality of the tunnel and the tunnel support structure can be optimized to improve the quality of blasting, reduce the waste of resources and extension of construction period caused by over-excavation and under-excavation, and ensure the stability of the tunnel support-surrounding rock structure system.

[0003] Traditional advanced geological prediction technologies ahead of the face include geological radar (electromagnetic waves), TSP (elastic waves), advanced drilling, and other technologies. Among the common advanced geological prediction methods in construction, non-destructive testing methods are easily affected by the environmental conditions of the project site, resulting in inaccurate test results. In addition, the use of TSP equipment requires the drilling of excitation holes, and the use of blasting to generate elastic waves requires the interruption of construction, which causes interference with the construction period. In addition, this type of testing equipment has problems with test accuracy. The accuracy of identifying small-scale rock cracks is poor, and it can only identify large-scale geologically unfavorable bodies. Using advanced geological drilling for advanced geological prediction is currently the most intuitive means of advanced geological prediction, but the drilling time is long, and only local drilling can be performed on the face. The cost of testing is high and normal construction needs to be interrupted.

[0004] It is of great value to identify the geology ahead of the face based on the drilling parameters of the rock drilling rig in the tunnel. On the one hand, it makes full use of the data of the tunnel construction stage at that time, and there is no need to block the construction to carry out advanced geological prediction on site; on the other hand, the traditional advanced geological prediction is not effective for the rock mass about 5m ahead of the face, while the drilling data can obtain the information of the rock mass to be excavated ahead of the face. Based on this, the artificial intelligence model and the drilling parameters of the rock drilling rig in the tunnel can be used to identify the geology ahead of the face.

[0005] CN201811547236.1 discloses a method and system for calculating rock strength using drilling parameters. The method collects parameter data of the rock drill during drilling and combines it with the rock compressive strength test results to establish a mapping relationship between the drilling parameters and the rock compressive strength. The specific steps include: preprocessing the drilling parameters; applying the data dimensionality reduction method to analyze the parameter contribution rate, and determining the main characteristic parameters by ordered weighted average operation; and establishing a rock compressive strength calculation model using the compressive strength test results and characteristic parameters.

[0006] CN202211472310.4 discloses a method and system for inversion of rock and soil parameters based on drilling test. The main steps of the method include: obtaining the drilling monitoring data during the drilling process and performing standardized processing; obtaining rock and soil parameters, including compressive strength, construction grade, etc.; establishing a mapping relationship data set between standardized drilling data and rock and soil parameters; according to the data set, inverting the drilling data to obtain the corresponding rock and soil parameters. The method establishes a mapping relationship between different drilling data and rock and soil parameters through regression analysis, BP neural network and support vector machine, so as to realize parameter inversion. Correspondingly, an inversion system is also proposed, including four units: data acquisition, parameter acquisition, data processing and data inversion.

[0007] The geological survey method is mainly to measure the physical properties, lithology, hydrogeological parameters and other data of the stratum below the drill bit during the drilling process, so as to achieve real-time monitoring and acquisition of underground geological conditions. At present, geological surveys in the tunnel design stage are large-scale surveys, and the distance between survey holes is large, which cannot effectively identify detailed rock cracks.

[0008] The above two existing technologies only consider the geological parameters of a single blasthole, ignoring the spatial correlation of geological features between different blastholes, resulting in low model accuracy and a small amount of labeled data in the drilling data center, which limits the model's overall understanding and prediction capabilities of the rock mass fracture types in front of the face. That is, artificial intelligence models based on data training require a large amount of labeled data. The method of marking drilling parameters on site includes borehole imaging technology, but this technology is time-consuming, so only a small amount of drilling data can be marked. Unlabeled drilling parameters cannot play a role in supervised learning algorithms, resulting in a waste of data, and the labeled data is small, so the recognition accuracy of the model obtained by training is poor. Therefore, making full use of unlabeled data to improve the accuracy of rock mass fracture identification in front of the tunnel face based on drilling data needs to be solved. Summary of the invention

[0009] The purpose of the present invention is to provide a method, device, medium and product for identifying rock cracks in front of a tunnel face. A feature extraction model can be trained by a target loss function determined by the specific energy characteristics of the boreholes in the adjacent borehole pair, the predicted specific energy and the label specific energy, thereby obtaining a trained encoder. The trained encoder can better extract geological characteristics related to the type of rock cracks in front of the tunnel face. The extracted drill rod geological characteristic response is taken into account in the rock crack identification model, thereby improving the accuracy of the trained rock crack identification model.

[0010] To achieve the above object, the present invention provides the following solution.

[0011] A method for identifying rock mass cracks in front of a tunnel face comprises the following steps.

[0012] A data set is acquired; the data set includes drilling parameters of multiple adjacent borehole pairs in a tunnel boring operation; the drilling parameters of each adjacent borehole pair include drilling parameters of two adjacent boreholes.

[0013] For each of the adjacent borehole pairs, the label specific energy of each borehole in the adjacent borehole pair is calculated according to the drilling parameters of each borehole in the adjacent borehole pair; the label specific energy of each borehole in the adjacent borehole pair is respectively input into the feature extraction model to obtain the specific energy feature and predicted specific energy corresponding to each borehole in the adjacent borehole pair; the feature extraction model includes an encoder and a decoder; the specific energy feature is the output of the encoder; the predicted specific energy is the output of the decoder; the drilling parameters are all drilling parameters except the drilling depth.

[0014] The feature extraction model is trained using a target loss function to obtain a trained feature extraction model, wherein the trained feature extraction model includes a trained encoder and a trained decoder; the target loss function is determined by the specific energy feature, predicted specific energy and label specific energy corresponding to each borehole in the adjacent borehole pair.

[0015] The tunnel geometry parameters, tunnel construction parameters, tunnel support parameters and drilling parameters of the target borehole of the target tunnel construction operation are obtained; the tunnel geometry parameters include burial depth, tunnel width, tunnel height and tunnel area; the tunnel construction parameters include construction footage; the tunnel support parameters include support stiffness; the drilling parameters include drilling speed, drilling depth, rotation speed, impact pressure, jacking pressure, rotation pressure, water pressure and water flow.

[0016] A target specific energy is calculated based on the drilling parameters of the target borehole.

[0017] The tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters of the target drilling and the target specific energy of the target tunnel construction operation are input into the trained rock fracture identification model to obtain the rock fracture type in front of the target tunnel heading; the trained rock fracture identification model is constructed by the trained encoder.

[0018] A computer device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned method for identifying rock mass fractures in front of a tunnel face.

[0019] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for identifying rock mass fractures in front of a tunnel face.

[0020] A computer program product includes a computer program, which implements the steps of the above-mentioned method for identifying rock mass fractures in front of a tunnel face when executed by a processor.

[0021] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects: the present invention provides a method, device, medium and product for identifying rock cracks in front of a tunnel face. First, the feature extraction model is trained by a data set containing the drilling parameters of multiple adjacent borehole pairs to obtain a trained encoder. The target loss function used in the training is determined by the specific energy characteristics, predicted specific energy and label specific energy corresponding to each borehole in the adjacent borehole pairs; the tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters and target specific energy of the target tunnel construction operation are input into the trained rock crack identification model including the trained encoder to obtain the rock crack type in front of the target tunnel face. The present invention trains the feature extraction model by unlabeled data (data set), and the target loss function of the feature extraction model considers the specific energy characteristics, predicted specific energy and label specific energy of the boreholes in the adjacent borehole pairs, so that the trained encoder can better extract the geological characteristics related to the rock crack type in front of the face. The rock crack identification model considers the extracted drill rod geological feature response, so that the recognition accuracy of the trained rock crack identification model is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 A schematic flow chart of a method for identifying rock mass cracks in front of a tunnel face provided in Example 1 of the present invention.

[0024] Figure 2 A schematic diagram of the implementation flow of the rock mass fracture identification method provided in Example 1 of the present invention.

[0025] Figure 3 A schematic diagram of the structure and training process of the feature extraction model provided in Example 1 of the present invention.

[0026] Figure 4 This is a schematic diagram of the rock fracture identification model structure provided in Example 1 of the present invention.

[0027] Figure 5 A schematic diagram of drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to normal rock fracture types provided in Example 1 of the present invention.

[0028] Figure 6 A schematic diagram of the drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to the intrusive rock fracture type provided in Example 1 of the present invention.

[0029] Figure 7 A schematic diagram of drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to a single fracture or hollow rock fracture type provided in Example 1 of the present invention.

[0030] Figure 8 A schematic diagram of drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to the fracture types of a rock mass with multiple fractures densely distributed provided in Example 1 of the present invention.

[0031] Fig. 9 This is a schematic diagram of the accuracy comparison results of the three models provided in Example 1 of the present invention.

[0032] Fig.10 A schematic diagram of the accuracy of the training set used in the trained rock fracture identification model provided in Example 1 of the present invention.

[0033] Fig.11 A schematic diagram of the accuracy of the test set used for the trained rock fracture identification model provided in Example 1 of the present invention.

[0034] Fig.12 This is a diagram of the internal structure of the computer device provided by the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] The purpose of the present invention is to provide a method, device, medium and product for identifying rock cracks in front of a tunnel face, aiming to train a feature extraction model through unlabeled data (data set), the target loss function of the feature extraction model takes into account the specific energy characteristics, predicted specific energy and label specific energy of the boreholes in the adjacent borehole pairs, so that the trained encoder can better extract the geological characteristics of the rock crack types in front of the face, and the rock crack recognition model takes into account the extracted drill rod geological characteristic response, so that the trained rock crack recognition model constructed by the trained encoder has a higher recognition accuracy; and multi-source data (tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters and target specific energy) are used to identify rock cracks in front of the face, with a higher accuracy.

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0038] Example 1.

[0039] like Figure 1 As shown, a method for identifying rock cracks in front of a tunnel face in this embodiment includes the following steps.

[0040] S1: Acquire a data set; the data set includes drilling parameters of multiple adjacent borehole pairs in a tunnel boring operation; the drilling parameters of each adjacent borehole pair include drilling parameters of two adjacent boreholes.

[0041] Drilling data: various geological, engineering and operational data acquired and recorded in real time during the drilling process. These data can help engineers and technicians understand the situation of the formation below the wellhead, optimize drilling operations, ensure the stability of the wellbore, and take necessary measures to deal with potential problems. Drilling data are usually collected through various sensors and measuring devices, which can be installed in places such as drill pipes, drill bits, and drilling fluid treatment systems. The drilling data in the fully intelligent rock drilling rig generally include drilling speed, drilling depth, impact pressure, rotary pressure, rotary speed, water pressure, and water flow. The fully intelligent rock drilling rig is an intelligent rock drilling equipment that uses the drill and blast method for tunnel and underground engineering construction. It uses a computer intelligent control system for precise calculation and high-precision control of the three drill arms to automatically drill rocks in an orderly manner according to the pre-designed hole layout, effectively reducing the construction difficulty of the operator and improving the efficiency of tunnel construction.

[0042] S2: For each of the adjacent borehole pairs, the label specific energy of each borehole in the adjacent borehole pair is calculated according to the drilling parameters of each borehole in the adjacent borehole pair; the label specific energy of each borehole in the adjacent borehole pair is respectively input into the feature extraction model to obtain the specific energy feature and predicted specific energy corresponding to each borehole in the adjacent borehole pair; the feature extraction model includes an encoder and a decoder; the specific energy feature is the output of the encoder; the predicted specific energy is the output of the decoder; the drilling parameters are all drilling parameters except the drilling depth.

[0043] S3: Use the target loss function to train the feature extraction model to obtain a trained feature extraction model, wherein the trained feature extraction model includes a trained encoder and a trained decoder; the target loss function is determined by the specific energy feature, predicted specific energy and label specific energy corresponding to each borehole in the adjacent borehole pair.

[0044] S4: Acquire the tunnel geometry parameters, tunnel construction parameters, tunnel support parameters and drilling parameters of the target borehole of the target tunnel construction operation; the tunnel geometry parameters include burial depth, tunnel width, tunnel height and tunnel area; the tunnel construction parameters include construction footage; the tunnel support parameters include support stiffness; the drilling parameters include drilling speed, drilling depth, rotation speed, impact pressure, jacking pressure, rotation pressure, water pressure and water flow.

[0045] S5: Calculating a target specific energy according to the drilling parameters of the target borehole.

[0046] S6: Input the tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters of the target drilling and the target specific energy of the target tunnel construction operation into the trained rock fracture identification model to obtain the rock fracture type in front of the target tunnel heading; the trained rock fracture identification model is constructed by the trained encoder.

[0047] like Figure 2 As shown in the figure, the process of obtaining the trained rock fracture recognition model is divided into two parts. The first part is to establish a geological feature extraction model for drill pipe while drilling data based on unsupervised learning, and the second part is to integrate the geological feature extraction model and multi-source data to establish a rock fracture recognition model in front of the face. Unsupervised learning: The data will not be specifically identified (unlabeled data), and the learning model is to infer some inherent structure of the data. Supervised learning: Use labeled data sets to train algorithms in order to classify data or accurately predict results.

[0048] (1) Field data are collected through engineering construction and rock drilling rigs, including tunnel geometry parameters (burial depth, tunnel width, tunnel height, tunnel area), tunnel construction parameters (construction footage, construction methods), tunnel support parameters (support stiffness, support composition, etc.), and rock drill parameters while drilling (drilling speed, drilling depth, rotation speed, impact pressure, jacking pressure, rotation pressure, water pressure and water flow). (2) Data are preprocessed, including (data cleaning, data noise reduction, data extraction, etc.). Field data are marked through borehole imaging, and the rock mass fracture types of field data are divided into four types of rock mass discontinuities: normal rock mass, local intrusion rock mass, single fracture or cavity, and multiple fractures with dense distribution. After regularization, the data is divided into labeled data and unlabeled data according to whether the data has a label. (3) The unlabeled data is input into the first part of the feature extraction model for training, and the specific energy feature calculation model (i.e., the trained encoder) is obtained by training the model with unlabeled data. (4) The marked results are put into the rock fracture identification model training in the second part. The feature extraction of MSE data in the rock fracture identification model adopts the specific energy feature calculation model obtained by the first part of the training. The marked data is used to train the second part of the model to obtain the rock fracture identification model. (5) After the trained rock fracture identification model, the data obtained by the data collection system can be directly input into the trained rock fracture identification model after preprocessing and regularization to determine the rock fracture type corresponding to the downhole data input, and then identify and locate the rock fracture in front of the face according to the borehole and data position corresponding to the downhole data.

[0049] Part I: The geological feature extraction method of the drill pipe while drilling data geological feature extraction model (i.e., feature extraction model) based on unsupervised learning mainly includes a loss function that considers the similar geological responses of adjacent drill pipe while drilling parameters and an unsupervised encoding and decoding feature extraction model. The structure and training process of the feature extraction model are as follows: Figure 3 shown.

[0050] 1. Figure 3The MSE (specific energy, which is the energy required for mechanical drilling per cubic meter of rock) curve on the left is calculated from the input drilling data. 2. The feature extraction model consists of two modules, the encoder and the decoder. The encoder is used to extract the specific energy feature, and the decoder is used to reconstruct the feature to obtain the initial data. If the extracted feature can contain the feature of the original data, then after passing through the decoder, it can be re-fitted with the initial data. 3. The calculated MSE curve encoder outputs the extracted feature to Denseblock3. 4. The extracted feature passes through the decoder to obtain the reconstructed MSE curve. The encoder and decoder data outputs are symmetrical. The more highly refined the extracted features are, the higher the matching degree between the input MSE and the output MSE. The model training process is to continuously correct the extracted features (the output result of Denseblock3) so that the output and input results match. Among them, the training process needs to use the loss function to determine the error between the input and output MSE curves to correct the features extracted by the model. The Loss function mentioned in the training is (Loss-feature) to guide the above model to extract the features of interest.

[0051] The structure of the feature extraction model includes an encoder and a decoder. The encoder includes a first dense block (Dense Block1), a first transition layer (TransitionLayer 1), a second dense block (Dense Block2), a second transition layer (Transition Layer2), and a third dense block (Dense Block3); the decoder includes a third dense block (Dense Block3), a third transition layer (TransitionLayer3), a fourth dense block (Dense Block4), a fourth transition layer (Transition Layer4), and a fifth dense block (Dense Block 5). Among them, the number of layers of the first dense block, the second dense block, and the third dense block is 2, and the growth rate is 3; the number of layers of the fourth dense block and the fifth dense block is 2, and the growth rate is 3; the structure of the first transition layer and the second transition layer is: a batch normalization layer (BN), a ReLU activation function, a convolution layer (Conv), and an average pooling layer (AvgPool) connected in sequence; the structure of the third transition layer and the fourth transition layer is: a batch normalization layer (BN), a ReLU activation function, a convolution layer (Conv), and an upsample (Upsample) connected in sequence.

[0052] It should be noted that the Dense blackcok in this embodiment is a Densenet network model.

[0053] The features extracted by this type of autoencoder network model (feature extraction model) during feature extraction training have no clear direction. In order to make it clear that the autoencoder network model focuses on geological fracture-related features when extracting features, the extracted loss function is used to determine the directionality of the extracted features and constrain the feature extraction model.

[0054] This embodiment takes into account the similarity of the rock mass fracture response of two groups of adjacent drill rods, and constructs a data set for training the unsupervised encoding and decoding feature extraction model in the form of a pair of adjacent drill rods, that is, the unlabeled drilling data obtained on site is combined into pairs of data pairs according to the adjacent holes in the tunnel drilling operation as a set of training data, that is, the data set obtained by S1 includes the drilling parameters of multiple adjacent borehole pairs in the tunnel drilling operation; the drilling parameters of each adjacent borehole pair include the drilling parameters of two adjacent boreholes. The MSE curve is established with the drilling depth as the horizontal coordinate and the specific energy calculated from the drilling speed, rotation speed, impact pressure, jacking pressure, rotation pressure, water pressure and water flow corresponding to each drilling depth as the vertical coordinate, as the input of the feature extraction model.

[0055] The variable of the drilling parameters used to extract the features in this part is MSE, that is, the input of the feature extraction model is MSE. The rock drill will have different responses under different working parameters, such as different drilling speeds and rotation speeds (i.e., slewing speeds). When selecting such parameters, the subjective changes caused by mechanical operation are relatively strong, which is not conducive to analyzing the features related to rock mass cracks. Therefore, in order to avoid the above problems, this embodiment uses the specific energy MSE calculated from the drilling parameters of the rock drill (including other drilling parameters except drilling depth) to reflect the response of the rock drill parameters to the rock mass during the rock drilling process. The specific energy MSE follows the calculation rule shown in formula (1).

[0056]

[0057] in, is the propulsion energy, P1 is the propulsion pressure (jacking pressure); A1 is the cross-sectional area of ​​the guide beam propulsion cylinder, and the cylinder diameter is 70 mm; D is the drilling diameter, 45 mm; is the impact energy, η tr Collision energy transfer efficiency; P2 punching pressure; A2 impact piston cylinder area 0.0003m 2 ; L m Impact piston stroke 50mm; H z Impact frequency, in units of 3800bpm; is the rotational energy, RPM is the drill bit speed; TOB is the torque; TOB = KP3; P3 is the rotary pressure; K is the torque coefficient of the hydraulic rotary motor, which is 4; ROP is the drilling speed; is water energy, Pw is the water pressure; Q w is the water flow rate; λ is the impact energy efficiency; A is the drilling area; d b is the drill bit diameter, the value is 45mm; d n is the nozzle diameter, the value is 5mm; n is the number of nozzles, the value is 4; L is the jet constant velocity core length, which is 4d n ; D1 is the distance between the nozzle and the bottom of the well, the value is 1d n , α is the jet diffusion angle, and its value is 3°.

[0058] Based on the above formula (1), the label specific energy of each borehole in each adjacent borehole pair in the data set can be calculated. After each input of the label specific energy of each borehole in a group of adjacent boreholes, the loss function value is calculated by the specific energy features, predicted specific energy and label specific energy corresponding to each borehole in all the adjacent borehole pairs currently input to train the feature extraction model. Loss function: An important concept in machine learning and deep learning, used to measure the gap between the model prediction value and the actual observation value. It is a key part of the optimization algorithm, helping the model to automatically adjust parameters to minimize the prediction error. In the process of training a machine learning model, the model gradually optimizes its prediction ability by continuously adjusting internal parameters. The loss function provides a numerical indicator for measuring the performance of the model on the training data. The goal of the model is to make the prediction result closer to the actual observation value by minimizing the value of the loss function. In this embodiment, the target loss function follows the calculation rule shown in formula (2).

[0059]

[0060] Wherein, Loss is the target loss function; m is the total number of adjacent borehole pairs currently input (the total number of training inputs this time), one of the boreholes in each adjacent borehole pair is recorded as the first borehole (also recorded as drill rod 1), and the other borehole is recorded as the second borehole (also recorded as drill rod 2); x i，1 is the label specific energy of the first borehole in the i-th group of adjacent borehole pairs; is the predicted specific energy of the first borehole in the i-th group of adjacent boreholes; f i，1 is the specific energy characteristic of the first borehole in the i-th group of adjacent boreholes; f i，2 is the specific energy characteristic of the second borehole in the i-th group of adjacent borehole pairs; x i，2 is the label specific energy of the second borehole in the i-th group of adjacent borehole pairs; is the predicted specific energy of the second borehole in the i-th group of adjacent borehole pairs.

[0061] The above objective constraint function is used to train the unsupervised codec feature extraction model. When the model is trained to the set Loss threshold, the training is stopped and the encoder part of the trained unsupervised codec feature extraction model is taken out as the feature extractor (i.e., the trained encoder Encoder).

[0062] Part 2: Before S4, it also includes training the rock fracture identification model to obtain a trained rock fracture identification model, specifically including: tunnel construction generates multi-source information, and the parameter categories related to the support-surrounding rock effect include: tunnel geometric parameters (burial depth, tunnel width, tunnel height, tunnel area), tunnel construction parameters (construction footage, construction method), tunnel support parameters (support stiffness, support composition, etc.). These parameters are continuously distributed along the longitudinal direction of the tunnel and will have a potential impact on the next stage of construction. Therefore, this embodiment uses an LSTM network to extract the impact of the constructed area on the subsequent construction area.

[0063] The process of obtaining the label data set: For the tunnel depth, which is a parameter known in the geological survey stage, the selected length is 60m in front of the face and 80m behind the face, and one data is collected every 1m, so the data format of the depth is 1×140. The remaining parameters distributed along the longitudinal direction of the tunnel, such as support stiffness, tunnel width, tunnel height, tunnel area, and footage, are selected in the range of 80m behind the face, and one data is collected every 1m, so the data format is 5×80. For the drilling parameters (data other than depth), the DNN block is used to extract data features. The recording distance of the drilling parameters is recorded every 2cm, 40cm each time, and the data format is 7×20. The above data are labeled according to the rock fracture types determined by borehole imaging corresponding to the 40cm drilling data, that is, according to each drilling, a set of training data (including tunnel geometry parameters, tunnel construction parameters, tunnel support parameters and drilling parameters) is corresponding, and the label of each set of training data adopts the rock fracture type determined by each borehole imaging. The rock fracture types include normal rock, intrusive rock, single fracture or cavity, and multiple fractures with dense distribution. The rock fracture recognition model is trained through a database with rock fracture types.

[0064] The structure of the rock mass fracture identification model constructed in this embodiment is as follows: Figure 4 As shown, the rock mass fracture recognition model includes a feature extraction module, a flattening layer (Flatten) and a recognition module connected in sequence; the feature extraction module includes a first feature extraction unit, a second feature extraction unit, a third feature extraction unit and a fourth feature extraction unit. Among them, the first feature extraction unit and the second feature extraction unit are LSTM models (i.e. Figure 4In the identification of rock fracture types in front of the tunnel face, two LSTM blocks are used to extract the accumulated features in the longitudinal direction of the tunnel.

[0065] The third feature extraction unit and the recognition module may be a DNN model (i.e. Figure 4 The fourth feature extraction unit is the trained encoder obtained in the first part. In addition, the MSE is obtained by calculating the drilling parameters, and the encoder obtained by the first part of the training ( Figure 4 The features extracted by LSTM block 1 (first feature extraction unit), LSTM block 2 (second feature extraction unit), DNN block (third feature extraction unit) and Encoder (fourth feature extraction unit) are fused and flattened, and then the next step is to classify the rock discontinuity type in the DNN network (recognition module).

[0066] After the above process, a trained rock fracture identification model can be obtained.

[0067] The first and second components above form a rock crack identification method ahead of the tunnel face based on the geological feature extraction of drilling parameters of the drilling rig. The penetration rate (ROP), impact pressure, thrust pressure and rotation pressure of the four types of rock cracks are as follows: Figure 5-Figure 8 As shown, Figure 5 The drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to the normal rock fracture type are: Figure 6 The drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to the fracture type of the intruding rock mass are: Figure 7 The drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to the single fracture or hollow rock fracture type are: Figure 8 The drilling speed, impact pressure, thrust pressure and rotation pressure corresponding to the fracture types of rock masses with multiple densely distributed fractures.

[0068] This embodiment also establishes three models to compare the rock mass fracture type identification effects, and the comparison data are shown in Table 1.

[0069] Table 1 Comparative data.

[0070] name Model database Model 1 Part 1 + Part 2 Upsampling balanced database Model 2 DNN Upsampling balanced database Model 3 Part 1 + Part 2 Original database

[0071] In Table 1, the first part + the second part refers to the trained rock fracture recognition model including the encoder trained in the first part. In order to avoid the training prediction results caused by too little labeled data of one type, upsampling is used to balance the labeled data set to obtain an upsampled balanced database. The upsampled balanced database is a database obtained by expanding the labeled data set. The original database is the labeled data set. The upsampled balanced database is divided into a training set and a test set. The training set is used to test the models in Table 1. The accuracy comparison results of the above three models are shown in the figure. Fig. 9 The trained rock mass fracture recognition model in this embodiment uses the accuracy of the training set and the test set as shown in Fig.10 and Fig.11 shown. Fig.10 and Fig.11 1, 2, 3 and 4 represent the sample data of rock mass fracture types in the training set or test set, namely normal rock mass, intrusive rock mass, single fracture or cavity, and dense distribution of multiple fractures.

[0072] After obtaining the trained rock mass fracture identification model, go to S4, that is, obtain the tunnel geometry parameters, tunnel construction parameters, tunnel support parameters and drilling parameters of the target drilling operation.

[0073] Before inputting the tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters of the target borehole and the target specific energy of the target tunnel construction operation into the trained rock fracture identification model, it also includes: preprocessing the tunnel geometry parameters, the tunnel construction parameters, the tunnel support parameters and the drilling parameters; the preprocessing includes data cleaning, data noise reduction and data extraction.

[0074] Then, based on formula (1), the target specific energy is calculated according to the drilling parameters of the target borehole. The tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters of the target borehole and the target specific energy of the target tunnel construction operation are input into the trained rock fracture identification model to obtain the rock fracture type in front of the target tunnel face.

[0075] The trained rock fracture recognition model includes a feature extraction module, a flattening layer and a recognition module connected in sequence; the feature extraction module includes a first feature extraction unit, a second feature extraction unit, a third feature extraction unit and a fourth feature extraction unit; the first feature extraction unit is used to extract features of the buried depth to obtain a first feature; the second feature extraction unit is used to extract features of support stiffness, construction footage, tunnel width, tunnel height and tunnel area to obtain a second feature; the third feature extraction unit is used to extract features of drilling parameters other than depth to obtain a third feature; the fourth feature extraction unit is the trained encoder, which is used to extract features of the target specific energy to obtain a specific energy feature; the flattening layer is used to fuse and flatten the first feature, the second feature, the third feature and the specific energy feature to obtain a fused flattened feature; the recognition module is used to identify the fused flattened feature to obtain the type of rock fracture in front of the target tunnel face.

[0076] In order to make full use of the large amount of data in the mechanized drilling and blasting construction, the present invention uses the drilling data of the drilling process of the rock drilling rig for blasting blasting operations, uses artificial intelligence methods to perform feature extraction unsupervised model training on unlabeled data, and uses labeled data and feature extraction modules to realize the identification of the crack information of the rock mass to be excavated in front of the tunnel face of the drilling and blasting method based on the drilling data. In order to make full use of the blasting hole data, an encoder-decoder model (feature extraction model) is introduced to extract key features related to rock mass cracks from adjacent blasting holes, and a rock mass crack identification model based on multi-source information is established. After the implementation of the method, the prediction model of different discontinuities is successfully obtained, and the correlation between adjacent blasting holes and a large amount of unlabeled data are fully utilized, which improves the recognition accuracy and reliability of blasting hole data. The geological feature extraction model of the drill rod drilling data based on unsupervised learning gives full play to the correlation between adjacent blasting holes. Through the rock mass crack identification model in front of the face, the accurate identification of rock mass cracks is realized, so that the unlabeled drilling data that has not been fully utilized in the past is better applied.

[0077] This embodiment proposes a rock mass fracture feature extraction model for drilling data trained by unsupervised learning with unlabeled data, and establishes a rock mass fracture identification fusion model based on multi-source data and the trained fracture feature extraction model, and trains the rock mass fracture identification fusion model in combination with labeled data. This method considers that the geology of adjacent boreholes in adjacent locations is similar, so the response of the drilling parameters in this area has similar characteristics, so similar rock mass fracture response characteristics between the drilling data of two boreholes can be extracted. First, the decoder extracts features from the original curve. Then, the encoder reverses the extracted features to the original curve. If the curve matching degree before and after decoding is high, it means that the extracted features can represent the overall features. On this basis, this embodiment proposes a loss function that considers the similar geological response of the drilling parameters of adjacent drill rods, which is used to constrain the unsupervised encoder and decoder to extract the geological features of the drilling parameters, so that it can extract the features related to rock mass fractures in the drilling parameters during the training process. The trained feature extraction model and multi-source heterogeneous information (multi-source data tunnel geometry parameters, support parameters, construction parameters and drilling parameters and their extracted MSE geological features) are jointly established to establish a rock mass fracture identification fusion model. By extracting the features of adjacent boreholes for prediction, better geological information can be obtained, which helps to improve the prediction effect. This method can obtain better geological information by extracting the features of adjacent boreholes and establishing a prediction model, solving the problems of existing methods.

[0078] Example 2.

[0079] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for identifying rock mass fractures in front of a tunnel face in Example 1.

[0080] Example 3.

[0081] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for identifying rock mass fractures in front of a tunnel face in Example 1.

[0082] Example 4.

[0083] A computer program product includes a computer program, which, when executed by a processor, implements the steps of the method for identifying rock mass fractures in front of a tunnel face in embodiment 1.

[0084] Example 5.

[0085] A computer device, which may be a database, may have an internal structure as shown in Fig.12As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the rock mass fracture identification method in front of the tunnel face in Example 1 is implemented.

[0086] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0087] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided by the present invention may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided by the present invention may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited thereto.

[0088] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0089] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for identifying rock cracks in front of a tunnel face, characterized in that: include: Acquire a data set; the data set includes drilling parameters of multiple adjacent borehole pairs in a tunnel boring operation; the drilling parameters of each adjacent borehole pair include drilling parameters of two adjacent boreholes; the drilling parameters include drilling speed, drilling depth, rotation speed, impact pressure, jacking pressure, rotation pressure, water pressure and water flow; For each of the adjacent borehole pairs, the label specific energy of each borehole in the adjacent borehole pair is calculated according to the drilling parameters of each borehole in the adjacent borehole pair; the label specific energy of each borehole in the adjacent borehole pair is respectively input into the feature extraction model to obtain the specific energy feature and predicted specific energy corresponding to each borehole in the adjacent borehole pair; The feature extraction model includes an encoder and a decoder; the specific energy feature is the output of the encoder; the predicted specific energy is the output of the decoder; the drilling parameters are all drilling parameters except the drilling depth; The feature extraction model is trained using a target loss function to obtain a trained feature extraction model, wherein the trained feature extraction model includes a trained encoder and a trained decoder; the target loss function is determined by the specific energy feature, the predicted specific energy, and the label specific energy corresponding to each borehole in the adjacent borehole pair; Acquire tunnel geometry parameters, tunnel construction parameters, tunnel support parameters and drilling parameters of target boreholes of target tunnel construction operations; the tunnel geometry parameters include burial depth, tunnel width, tunnel height and tunnel area; the tunnel construction parameters include construction footage; the tunnel support parameters include support stiffness; Calculating a target specific energy according to a while-drilling parameter of the target borehole; Inputting the tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters of the target borehole and the target specific energy of the target tunnel construction operation into the trained rock fracture identification model to obtain the rock fracture type in front of the target tunnel face; the trained rock fracture identification model is constructed by the trained encoder; the trained rock fracture identification model includes a feature extraction module; The feature extraction module includes a first feature extraction unit, a second feature extraction unit, a third feature extraction unit and a fourth feature extraction unit; A first feature extraction unit is used to extract features of the buried depth to obtain a first feature; The second feature extraction unit is used to extract the features of the support stiffness, construction footage, tunnel width, tunnel height and tunnel area to obtain the second feature; The third feature extraction unit is used to extract features of drilling parameters other than depth to obtain a third feature; the fourth feature extraction unit is a trained encoder, which is used to extract features of target specific energy to obtain specific energy features.

2. A method for identifying rock cracks in front of a tunnel face according to claim 1, characterized in that: The objective loss function is as follows: Where Loss is the target loss function; m is the total number of adjacent borehole pairs currently input, one of the boreholes in each adjacent borehole pair is recorded as the first borehole, and the other borehole is recorded as the second borehole; x i，1 is the label specific energy of the first borehole in the i-th group of adjacent borehole pairs; is the predicted specific energy of the first borehole in the i-th group of adjacent boreholes; f i，1 is the specific energy characteristic of the first borehole in the i-th group of adjacent boreholes; f i，2 is the specific energy characteristic of the second borehole in the i-th group of adjacent borehole pairs; x i，2 is the label specific energy of the second borehole in the i-th group of adjacent borehole pairs; is the predicted specific energy of the second borehole in the i-th group of adjacent borehole pairs.

3. The method for identifying rock cracks in front of a tunnel face according to claim 1, characterized in that: Before inputting the tunnel geometry parameters, tunnel construction parameters, tunnel support parameters, drilling parameters of the target borehole and the target specific energy of the target tunnel construction operation into the trained rock mass fracture identification model, the method further includes: The tunnel geometric parameters, the tunnel construction parameters, the tunnel support parameters and the drilling parameters are preprocessed; the preprocessing includes data cleaning and data noise reduction.

4. The method for identifying rock mass cracks in front of a tunnel face according to claim 1, characterized in that: The trained rock mass fracture identification model also includes a flattening layer and an identification module; The flattening layer is used to fuse and flatten the first feature, the second feature, the third feature and the specific energy feature to obtain a fused flattened feature; The identification module is used to identify the fused flattened features to obtain the rock mass fracture type in front of the target tunnel face.

5. A method for identifying rock mass cracks in front of a tunnel face according to claim 4, characterized in that: The first feature extraction unit and the second feature extraction unit are LSTM models.

6. A method for identifying rock mass cracks in front of a tunnel face according to claim 4, characterized in that: The third feature extraction unit and the recognition module are DNN models.

7. The method for identifying rock mass cracks in front of a tunnel face according to claim 1, characterized in that: The rock mass fracture types include normal rock mass, intrusive rock mass, single fracture or cavity, and densely distributed multiple fractures.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of a method for identifying rock fractures in front of a tunnel face as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for identifying rock mass fractures in front of a tunnel face as described in any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of a method for identifying rock mass fractures in front of a tunnel face as described in any one of claims 1 to 7 are implemented.

Images