A refined advance prediction method for tunnel surrounding rock grade based on drilling parameters

Abstract

The present invention relates to the technical field of tunnel surrounding rock, and more specifically, to a method for fine-grained advance prediction of tunnel surrounding rock levels based on drilling parameters. The method comprises the following steps: Step 1: Using a drilling rig to collect drilling parameters in a continuous cycle; Step 2: Extracting statistical features of the drilling parameters; Step 3: Using a long short-term memory network model to train a tunnel surrounding rock level advance prediction model based on drilling parameters; Step 4: Using a three-dimensional spatial unit partitioning method for surrounding rock, the entire cycle is horizontally divided into multiple blocks to perform fine-grained advance prediction of surrounding rock levels. The present invention can continuously collect drilling parameters during the excavation cycle, ensuring the continuity of advance prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel surrounding rock, and in particular to a method for advanced prediction of tunnel surrounding rock levels based on drilling parameters. Background Art

[0002] Advanced geological prediction during tunnel construction is a key technology for effectively understanding the geological conditions ahead of the tunnel face. Since the 1970s, research on the theory, technology, and engineering practice of advanced geological prediction during tunnel construction has focused on this field, resulting in the development of numerous advanced geological prediction technologies. Advanced geological prediction for railway tunnels can be categorized by prediction type: geological survey, advanced drilling (advanced horizontal drilling, deepened blastholes), geophysical exploration (seismic wave reflection, electromagnetic wave reflection, transient electromagnetic, and induced polarization), and advanced pilot pit prediction. Based on prediction distance, it can be categorized as long-distance prediction (forecast lengths over 100 meters), medium-to-long-distance prediction (forecast lengths between 30 and 100 meters), and short-distance prediction (forecast lengths under 30 meters).

[0003] Different types of advanced geological prediction techniques have distinct physical foundations. A single advanced geological prediction technique can only determine a specific physical property of a geological body and, in turn, infer its geological attributes. However, the accuracy of such a single technique is not entirely reliable, and different methods also have varying predictive effects on different geological defects. This has prompted the development of comprehensive advanced geological prediction techniques, and numerous researchers have conducted extensive research on this topic. Some researchers have proposed a comprehensive advanced geological prediction method for double-shield TBM construction based on multi-source information, including ground geological analysis, tunnel face surrounding rock observation, tunneling parameters and rock slag analysis, hammer-induced 3D seismic analysis, 3D resistivity, and advance drilling. Other researchers have proposed a comprehensive advanced geological prediction method that uses geological radar as the primary tool for advanced prediction, combined with the TGP206 seismic reflection method and engineering geological surveys. This method has been successfully applied to the prediction of fault fracture zones and water-rich zones in actual engineering projects. Some researchers have proposed a comprehensive advance geological prediction technology based on tunnel face cataloging and geological radar, which has been successfully implemented in a long, deep tunnel project. Overall, advanced geological prediction is effective in determining the geological conditions ahead of the tunnel face during tunnel construction. However, due to cost and time constraints, it is currently not fully implemented during tunnel construction.

[0004] Compared to surrounding rock classification during the design phase, surrounding rock classification during tunnel construction focuses more on re-evaluating the surrounding rock grade based on the conditions at the excavation face. This is crucial and objective. Furthermore, surrounding rock classification during tunnel construction directly guides construction and provides feedback for design, making it a crucial step in tunnel construction. To improve the accuracy and reliability of surrounding rock classification during tunnel construction, numerous researchers have conducted extensive research on this topic.

[0005] With the increasing mechanization of tunnel construction, drilling rigs are increasingly being used for drill-and-blast tunneling operations. The drilling parameters collected by these rigs are also being used for surrounding rock classification. Based on this, the three-dimensional spatial characteristics of drilling parameters are utilized to achieve three-dimensional refined surrounding rock classification within the cyclic footage range of the rig's blasthole drilling parameters. However, both overall surrounding rock classification based on drilling parameters and three-dimensional refined surrounding rock classification only focus on the cycle in which the excavation face is currently located. Research on identifying surrounding rock levels in unexcavated cycles has not yet been conducted, resulting in a relatively short range of geological conditions ahead of the face that can be assessed.

[0006] In summary, how to continuously predict the geological conditions of a certain length in front of the tunnel face during the tunnel construction phase has become the key to preparing for complex geological conditions in advance, and further research is needed. Summary of the Invention

[0007] The present invention provides a method for fine-grained advance prediction of tunnel surrounding rock levels based on drilling parameters, which can continuously collect drilling parameters of excavation cycles to ensure the continuity of advance prediction.

[0008] The method for advanced prediction of tunnel surrounding rock level based on drilling parameters according to the present invention comprises the following steps:

[0009] Step 1: Use a drilling rig to collect drilling parameters in a continuous cycle;

[0010] Step 2: Extracting statistical characteristics of drilling parameters;

[0011] Step 3: Use the long short-term memory network model to train an advanced prediction model for the tunnel surrounding rock level based on drilling parameters;

[0012] Step 4: Divide the entire cycle into multiple blocks horizontally through the surrounding rock three-dimensional space unit division method to perform refined advanced prediction of the surrounding rock level.

[0013] Preferably, in step 1, the drilling rig adopts an impact-rotation coupling mode for drilling. The drilling process includes four actions: propulsion, impact, rotation, and flushing. During this process, six drilling parameters, namely, feed speed, propulsion pressure, impact pressure, rotary pressure, water pressure, and water flow, as well as the corresponding drilling depth, interval time, and drilling coordinates, are collected. The feed speed and propulsion pressure correspond to the propulsion drilling action, the impact pressure corresponds to the impact drilling action, the rotary pressure corresponds to the rotary drilling action, and the water pressure and water flow correspond to the flushing drilling action.

[0014] Preferably, in step 1, during the drilling process, tunnel excavation is carried out in a cyclic manner, and 20,000 to 70,000 sets of drilling parameters are collected in each cycle. After each drilling cycle is completed, the drilling rig combines all sets of drilling parameters of each cycle in units of cycles to form a drilling parameter file as the drilling parameters of the cycle.

[0015] Preferably, in step 2, specifically:

[0016] The mean, standard deviation, three quartiles, and coefficient of variation were selected as the statistical features of drilling parameters for extraction;

[0017] The mean is the sum of all data in the data set divided by the number of data. The calculation formula is shown in formula (1):

[0018]

[0019] Where μ represents the mean value of the drilling parameters in the current cycle, x i Indicates the drilling parameters of each blasthole point in the current cycle, and n indicates the number of all blasthole points in the current cycle;

[0020] The standard deviation is the square root of the average of the sum of the squares of the differences between each data point and the mean in the data set. The standard deviation reflects the degree of dispersion of the data set. The larger the standard deviation, the more dispersed the data distribution is, and the smaller the standard deviation, the more concentrated the data distribution is. The calculation formula is shown in formula (2):

[0021]

[0022] Where σ represents the standard deviation of the drilling parameters of the current cycle, μ represents the mean value of the drilling parameters of the current cycle, and x i Indicates the drilling parameters of each blasthole point in the current cycle, and n indicates the number of all blasthole points in the current cycle;

[0023] Quartiles are the values ​​at the three dividing points after all the data in the data set are arranged from small to large, and then divided into four equal parts. The three quartiles are the first quartile Q1, the second quartile Q2, and the third quartile Q3. Among them, the first quartile Q1 is equal to the 25th percentile number after all the data in the data set are arranged from small to large; the second quartile Q2 is equal to the 50th percentile number after all the data in the data set are arranged from small to large; the third quartile Q3 is equal to the 75th percentile number after all the data in the data set are arranged from small to large.

[0024] The coefficient of variation is the ratio of the standard deviation to the mean. The coefficient of variation can eliminate the influence of the mean of the data set and only reflects the degree of dispersion of the data set relative to its mean. It is a relative indicator used to measure the degree of dispersion of the data set and is suitable for comparing the dispersion of different data sets. The calculation formula is shown in formula (3):

[0025]

[0026] Where c v represents the coefficient of variation of the drilling parameters of the current cycle, σ represents the standard deviation of the drilling parameters of the current cycle, and μ represents the mean of the drilling parameters of the current cycle;

[0027] Preferably, in the tunnel surrounding rock level advance prediction model based on drilling parameters in step 3, the drilling parameter statistical characteristics of four excavated cycles are added to the drilling parameter statistical characteristics of one drilled but not yet blasted cycle to form a set of cyclic sequence data with a step length of 5 to advance predict the surrounding rock level of the next unexcavated cycle.

[0028] As a preferred method, the training process of the tunnel surrounding rock grade advance prediction model based on drilling parameters is as follows:

[0029] 3.1) Dataset division;

[0030] The dataset is divided into a training set and a test set in a ratio of 4:1. The training set is used to train the model, and the test set is used to evaluate the model. The dataset consists of samples, and a sample consists of two parts: sample features and sample labels. The sample features are passed to the model as model input to obtain the model output, and then the model output is compared with the sample label to achieve the purpose of training or testing the model.

[0031] A sample consists of six consecutive cycles arranged in the order of excavation. The statistical characteristics of drilling parameters of the first five cycles constitute the sample characteristics, and the surrounding rock level of the last cycle constitutes the sample label.

[0032] 3.2) SMOTE oversampling of training set;

[0033] SMOTE oversampling creates new samples by interpolating between minority class samples;

[0034] 3.3) Training set standardization;

[0035] The training set was standardized using Z-score normalization. The mean of each eigenvalue was subtracted and divided by its standard deviation, so that the mean of the processed data was 0 and the standard deviation was 1, while maintaining the distribution of the original data. This eliminates the influence of different dimensions on model training. At the same time, to maintain the consistency of data processing and the fairness of model evaluation, the test set was standardized using the same Z-score normalization parameters as the training set.

[0036] 3.4) K-fold cross validation training model;

[0037] K is set to 5. First, a 5-fold cross-validation is performed to divide the training set into 5 subsets. One subset is selected as the cross-validation validation set, and the other 4 subsets are used as cross-validation training sets. Then, the model is trained, and the model evaluation index of the trained model on the cross-validation validation set is obtained. Finally, the above steps are repeated until each subset has served as the cross-validation validation set once, and the model evaluation index of the model trained 5 times on the cross-validation validation set is obtained. The optimal model is selected as the advanced prediction model for surrounding rock grade.

[0038] Preferably, in step 4, the surrounding rock three-dimensional space unit body division method is:

[0039] Based on the non-uniform characteristics of the loop, the entire loop is divided into multiple blocks in the horizontal direction. During the horizontal block division process, the tunnel centerline is used as a vertical reference line and offset to the left or right until the left and right boundaries of the tunnel outline are reached. The horizontal reference line near the bottom of the rail is offset up or down until the upper and lower boundaries of the tunnel outline are reached. Finally, the entire loop is divided into multiple blocks in the horizontal direction.

[0040] According to the above-mentioned horizontal block division rule, the drilling parameters of each cycle collected by the drilling rig are synchronously divided, and finally a plurality of blocks and drilling parameters within the range of each block are obtained for each cycle.

[0041] Preferably, in step 4, the four most recently excavated cycles and one drilled but not yet blasted cycle are selected and arranged in order of excavation, and the statistical features of the drilling parameters of the blocks at the same position in these five cycles are selected to form sequence data at the cycle block level, which are input into the above-trained rock mass level advance prediction model to obtain the rock mass levels of multiple blocks of the next unexcavated cycle, thereby realizing refined advance prediction of the rock mass level.

[0042] The present invention can provide a reference for judging geological conditions at a greater distance from the tunnel face. At the same time, since the drilling rig serves as a fixed excavation and drilling operation equipment, it can continuously collect drilling parameters of the excavation cycle without adding any working procedures, thereby ensuring the continuity of the advance prediction.

[0043] The present invention obtains a refined advance prediction result of surrounding rock level through a refined advance prediction method of tunnel surrounding rock level based on drilling parameters, which is effective. It can grasp some surrounding rock level conditions in front of the tunnel face in advance, pre-evaluate potential construction risks, adjust excavation footage and blasting parameters, ensure construction safety and efficiency, and provide suggestions for timely adjustment of on-site construction plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a flow chart of a method for advanced prediction of tunnel surrounding rock level based on drilling parameters in an embodiment;

[0045] Figure 2 Schematic diagram of blasthole arrangement on the tunnel face in the embodiment;

[0046] Figure 3 Schematic diagram of single blasthole drilling parameter acquisition in the embodiment;

[0047] Figure 4 Schematic diagram of the corresponding relationship between hardness, integrity and basic classification of surrounding rock in the embodiment;

[0048] Figure 5 Schematic diagram of sample types and quantity details in the embodiment;

[0049] Figure 6 Schematic diagram of drilling parameter statistics characteristics in the embodiment;

[0050] Figure 7 Schematic diagram of the advanced prediction of surrounding rock level in the embodiment;

[0051] Figure 8 This is a schematic diagram of the data set structure in the embodiment;

[0052] Figure 9 Schematic diagram of the sample structure in the embodiment;

[0053] Figure 10 This is a schematic diagram of the confusion matrix of the test set results in the embodiment;

[0054] Figure 11 This is a schematic diagram of horizontal blocks in the embodiment;

[0055] Figure 12 Schematic diagram of cyclic block advance prediction in the embodiment;

[0056] Figure 13Schematic diagram of the refined advance prediction results of the surrounding rock level of cycle DK3+431.4~DK3+434.4 in the embodiment;

[0057] Figure 14 Schematic diagram of the tunnel face with cycles DK3+431.4 to DK3+434.4 in the embodiment. DETAILED DESCRIPTION

[0058] In order to further understand the content of the present invention, the present invention is described in detail with reference to the accompanying drawings and embodiments. It should be understood that the embodiments are merely for explaining the present invention and are not intended to limit the present invention.

[0059] Example

[0060] like Figure 1 As shown, this embodiment provides a method for advanced prediction of tunnel surrounding rock level based on drilling parameters, which includes the following steps:

[0061] Step 1: Use a drilling rig to collect drilling parameters in a continuous cycle;

[0062] Step 2: extracting statistical features of drilling parameters;

[0063] Step 3: Use the long short-term memory network model to train an advanced prediction model for the tunnel surrounding rock level based on drilling parameters;

[0064] Step 4: Divide the entire cycle into multiple blocks horizontally through the surrounding rock three-dimensional space unit division method to perform refined advanced prediction of the surrounding rock level.

[0065] Drilling parameter collection

[0066] In step 1, the drilling rig uses an impact-rotation coupling method to drill holes. The drilling process includes four actions: propulsion, impact, rotation, and flushing. During this process, a total of six drilling parameters, including feed speed, propulsion pressure, impact pressure, rotation pressure, water pressure, and water flow, as well as the corresponding drilling depth, interval time, and drilling coordinates, are collected. Among them, the feed speed and propulsion pressure correspond to the propulsion drilling action, the impact pressure corresponds to the impact drilling action, the rotary pressure corresponds to the rotary drilling action, and the water pressure and water flow correspond to the flushing drilling action.

[0067] The definitions and acquisition methods of various drilling parameters are shown in Table 1.

[0068] Table 1 Drilling parameter definitions and acquisition methods

[0069]

[0070]

[0071] In step 1, during the drilling and blasting tunnel boring process, the tunnel excavation is carried out in a cyclic manner. The excavation footage of each cycle is about 2 to 5 meters (related to the excavation method, surrounding rock grade, etc.). Each cycle of excavation is carried out in a drilling and blasting manner. In each cycle, 200 to 300 blast holes are arranged at a blast hole spacing of 50 to 70 cm. The blast hole arrangement on the tunnel face is as follows: Figure 2 shown.

[0072] The acquisition method of each drilling parameter of each blasthole is to record at equal depth intervals, with an acquisition depth interval of 0.02m. That is, in each blasthole, one blasthole point is set every 2cm of drilling, and one set of drilling parameters is collected, such as Figure 3 As shown, approximately 100 to 250 sets of drilling parameters are collected for each blasthole, and approximately 20,000 to 70,000 sets of drilling parameters are collected for each cycle. After each drilling cycle is completed, the drilling rig combines all the drilling parameters for each cycle into a single drilling parameter file, which serves as the drilling parameters for that cycle.

[0073] Since the water pressure and water flow record the water outlet pressure and water outlet flow of the drilling rig water pump and are not related to the quality of the surrounding rock, the drilling parameters used in this embodiment mainly refer to the four drilling parameters of feed speed, thrust pressure, impact pressure, and rotary pressure.

[0074] Surrounding rock level collection

[0075] In actual production, the surrounding rock grade is used in design based on cycles. The entire rock mass of each cycle is considered as a whole, and the rock mass quality is comprehensively judged based on the whole to determine the surrounding rock grade.

[0076] In this embodiment, the corresponding surrounding rock grade of the drilling parameter cycle is mainly determined by the surrounding rock classification method in the "Code for Design of Railway Tunnel" (TB 10003-2016) through the geological sketch of the tunnel face carried out by the geological engineer. After the blasting and slag removal of the previous cycle, the lithology, weathering degree, hardness and integrity of the surrounding rock are determined by observing the exposed tunnel face of the current cycle, and then the basic surrounding rock grade is determined. The basic surrounding rock grade is based on the following: Figure 4 As shown, the groundwater discharge status, initial ground stress status and main structural surface attitude status are determined through on-site measurement and other means, and then the surrounding rock classification is corrected.

[0077] Number of samples collected

[0078] According to statistics, a total of 2670 cycles of drilling parameters and their corresponding surrounding rock levels were collected, including four surrounding rock levels: II, III, IV and V. The sample types and quantity details are as follows: Figure 5 shown.

[0079] Drilling parameter statistical feature extraction

[0080] As mentioned above, about 20,000 to 70,000 sets of drilling parameters are collected in each cycle. Since the surrounding rock level is used in actual production in cycles, it is necessary to perform mathematical statistics on the drilling parameters of a large number of blasthole points in each cycle and extract the drilling parameters at the cycle level. Specifically:

[0081] The mean, standard deviation, three quartiles and coefficient of variation are selected as the statistical features of drilling parameters for extraction, such as Figure 6 As shown in the figure, 6 statistical features can be extracted from each original feature of the drilling parameter, and finally 24 statistical features are obtained in each cycle.

[0082] The mean is the sum of all data in the data set divided by the number of data. The calculation formula is shown in formula (1):

[0083]

[0084] Where μ represents the mean value of the drilling parameters in the current cycle, x i It represents the drilling parameters of each blasthole point in the current cycle, and n represents the number of all blasthole points in the current cycle.

[0085] The standard deviation is the square root of the average of the sum of the squares of the differences between each data point and the mean in the data set. The standard deviation reflects the degree of dispersion of the data set. The larger the standard deviation, the more dispersed the data distribution is, and the smaller the standard deviation, the more concentrated the data distribution is. The calculation formula is shown in formula (2):

[0086]

[0087] Where σ represents the standard deviation of the drilling parameters of the current cycle, μ represents the mean value of the drilling parameters of the current cycle, and x i It represents the drilling parameters of each blasthole point in the current cycle, and n represents the number of all blasthole points in the current cycle.

[0088] Quartiles are the values ​​at the three dividing points after all the data in the data set are arranged from small to large and divided into four equal parts; these three quartiles are the first quartile Q1, the second quartile Q2 and the third quartile Q3, among which the first quartile Q1 is also called the "lower quartile" or "smaller quartile", which is equal to the 25th percentile number after all the data in the data set are arranged from small to large; the second quartile Q2 is also called the "median", which is equal to the 50th percentile number after all the data in the data set are arranged from small to large; the third quartile Q3 is also called the "upper quartile" or "larger quartile", which is equal to the 75th percentile number after all the data in the data set are arranged from small to large.

[0089] The coefficient of variation is the ratio of the standard deviation to the mean, also known as the coefficient of dispersion or standard deviation rate. The coefficient of variation can eliminate the influence of the mean of the data set and only reflects the degree of dispersion of the data set relative to its mean. It is a relative indicator used to measure the degree of dispersion of a data set and is therefore suitable for comparing the degree of dispersion of different data sets. The calculation formula is shown in formula (3):

[0090]

[0091] Where c v represents the coefficient of variation of the drilling parameters of the current cycle, σ represents the standard deviation of the drilling parameters of the current cycle, and μ represents the mean of the drilling parameters of the current cycle;

[0092] Advanced prediction model of tunnel surrounding rock grade based on drilling parameters

[0093] Long Short Term Memory (LSTM) is a special recurrent neural network (RNN) architecture that solves the gradient vanishing or gradient exploding problems encountered by traditional recurrent neural networks (RNN) when processing long sequence data. Since the Long Short Term Memory (LSTM) can capture and understand the complex long-term dependencies in sequence data, it can predict the future trend of sequence data and is often used for time series prediction. Therefore, this embodiment uses the Long Short Term Memory (LSTM) to train a rock mass level advance prediction model. Through the statistical features of the drilling parameters of four excavated cycles and the statistical features of the drilling parameters of a drilled but not yet blasted cycle, a group of cyclic sequence data with a step size of 5 is formed to advance the prediction of the rock mass level of the next unexcavated cycle, such as Figure 7 As shown in the figure, it is used for refined advance prediction of surrounding rock level.

[0094] 3.1) Dataset Division

[0095] The dataset is divided into a training set and a test set in a ratio of 4:1. The training set is used to train the model, and the test set is used to evaluate the model. The dataset is composed of samples, and a sample consists of two parts: sample features and sample labels. The sample features are passed to the model as model input to obtain the model output, and then the model output is compared with the sample label to achieve the purpose of training or testing the model. The dataset structure is as follows: Figure 8 shown.

[0096] Since the model input of the surrounding rock grade advance prediction model in this embodiment is a cyclic sequence data composed of the statistical characteristics of drilling parameters of four excavated cycles and one drilled but not yet blasted cycle, and the model output is the surrounding rock grade of the next unexcavated cycle, a sample in the data set of this embodiment includes six consecutive cycles arranged in the order of excavation, the statistical characteristics of drilling parameters of the first five cycles are taken to constitute the sample characteristics, and the surrounding rock grade of the last cycle is taken to constitute the sample label; the sample structure is shown in Figure 9.

[0097] Following the above method, the extracted drilling parameter statistical features were converted into standard training and testing samples to form the final dataset, and finally the dataset was partitioned. The details of the dataset partitioning are shown in Table 2.

[0098] Table 2 Dataset division details

[0099]

[0100] 3.2) SMOTE oversampling of training set

[0101] Because the number of samples from different surrounding rock classes is inconsistent during sample collection, the resulting training and data sets are unbalanced. When the number of samples in some categories far exceeds that of other categories, the model will overemphasize the categories with large data volumes during training and ignore the categories with small data volumes. This will cause the model to be more inclined to classify samples as belonging to the large data volume category, thereby reducing the model's accuracy and reliability. Therefore, the problem of sample imbalance needs to be addressed.

[0102] Oversampling and undersampling are two common methods for dealing with sample imbalance. Oversampling balances samples of different classes by increasing the number of minority class samples. Undersampling balances samples of different classes by reducing the number of majority class samples. Since the total number of samples collected in this embodiment is relatively small, oversampling is chosen for processing.

[0103] Conventional oversampling involves randomly selecting samples from the minority class and then replicating them to generate new samples, thereby increasing the number of minority class samples and achieving a relatively balanced number of samples for each class. However, directly replicating samples can lead to model overfitting, as the model will overemphasize these duplicated samples. SMOTE oversampling, on the other hand, creates new samples by interpolating between minority class samples; therefore, SMOTE oversampling generates synthetic samples of minority class samples, rather than direct replication. Therefore, SMOTE oversampling was ultimately chosen for this process. To maintain the independence of the test set and align with practical application scenarios, SMOTE oversampling was performed only on the training set. Table 3 shows the detailed data set partitioning after SMOTE oversampling.

[0104] Table 3 Details of the data set division after SMOTE oversampling

[0105]

[0106] 3.3) Training set standardization

[0107] Because the 24 drilling parameter statistical features that make up the sample characteristics have different magnitudes, their orders of magnitude vary. If features of different dimensions were directly used in model training, features with larger dimensions might dominate the model, while features with smaller dimensions would have little influence, thus affecting the accuracy and reliability of the results. Therefore, it is necessary to address the issue of varying feature dimensions.

[0108] This example uses Z-score normalization to normalize the training set. This process subtracts the mean of each eigenvalue and divides it by its standard deviation, resulting in a mean of 0 and a standard deviation of 1 for the processed data. This maintains the distribution of the original data, thereby eliminating the impact of different dimensions on model training. To ensure consistency in data processing and fairness in model evaluation, the test set is normalized using the same Z-score normalization parameters (i.e., the same mean and standard deviation) as the training set.

[0109] 3.4) K-fold cross-validation training model

[0110] K-fold cross-validation further divides the training set into a cross-validation training set and a cross-validation verification set to train the model. In this embodiment, K is taken as 5. First, the 5-fold cross-validation divides the training set into 5 subsets, selects 1 subset as the cross-validation verification set, and the other 4 subsets as the cross-validation training set. Then, the model is trained and the model evaluation index of the model trained this time on the cross-validation verification set is obtained. Finally, the above steps are repeated until each subset has served as a cross-validation verification set once, and the model evaluation index of the model trained 5 times on the cross-validation verification set is obtained, as shown in Table 4.

[0111] Table 4 5-fold cross validation set model evaluation index table

[0112] 50% off round Accuracy Average precision Average recall Average F1 score 1st Fold 93.6％ 94.1％ 94.1％ 93.7％ 2nd Fold 93.3％ 93.4％ 93.4％ 93.3％ 3rd fold 95.5％ 95.5％ 95.4％ 95.4％ 4th fold 94.6％ 95.1％ 94.7％ 94.6％ 50% off 94.3％ 94.4％ 94.0％ 94.0％

[0113] As can be seen from the table above, the five models obtained through 5-fold cross validation all have good model evaluation indicators. In this embodiment, the model obtained by the optimal 3rd fold training is selected as the surrounding rock level advanced prediction model.

[0114] 3.5) Model Evaluation

[0115] The trained rock mass level advance prediction model was applied to the test set, and the obtained accuracy was 93.5%, the average precision was 84.9%, the average recall was 92.1%, and the average F1 score was 87.3%. The confusion matrix of the test set results is as follows: Figure 10 The results show that the surrounding rock grade advance prediction model has a good effect in the advance prediction of surrounding rock grade, and based on this, it can be further used for refined advance prediction of surrounding rock grade.

[0116] Method for dividing surrounding rock into three-dimensional space units

[0117] According to the non-uniform characteristics of the cycle, the entire cycle is divided into multiple blocks in the horizontal direction. In the process of horizontal block division, considering the differences in the tunnel excavation design contours and the uniformity of the number of blocks, it is proposed to use the tunnel centerline as the vertical reference line, offset 2 to 4 meters to the left or right, until the left and right boundaries of the tunnel contour, and use the track bottom surface (pit bottom surface) as the horizontal reference line, offset 2 to 4 meters up or down, until the upper and lower boundaries of the tunnel contour. Finally, the entire cycle is divided into 18 blocks in the horizontal direction, numbered A-1 to A-18, among which the maximum size of a single block is 2 to 4 meters, and the longitudinal area of ​​a single block is 3 to 8 meters. 2 The number of blastholes in a single block is 10 to 20. Figure 11 shown.

[0118] The drilling parameters collected by the drilling rig for each cycle are then simultaneously divided according to the aforementioned horizontal segmentation rules. Ultimately, each cycle is divided into 18 segments, along with the drilling parameters within each segment. After segmentation, each segment can be considered a "micro-cycle," and the statistical characteristics of the drilling parameters for each "micro-cycle" can be directly calculated using the same method used to extract the statistical characteristics of drilling parameters for the entire cycle.

[0119] Refined advanced prediction of surrounding rock level

[0120] According to the above-mentioned method of dividing the surrounding rock into three-dimensional spatial units, each excavation cycle can be divided into 18 blocks, and 24 drilling parameter statistical characteristics of each block can be obtained.

[0121] During the actual tunnel excavation process, the four most recently excavated cycles and one drilled but not yet blasted cycle are selected and arranged in order of excavation. The statistical characteristics of the drilling parameters of the blocks at the same position in these five cycles are selected to form the sequence data of the cycle block level. The data is input into the above-trained rock mass level advance prediction model to obtain the rock mass level of the 18 blocks of the next unexcavated cycle, thus achieving refined advance prediction of the rock mass level and providing suggestions for the adjustment of the on-site construction plan. Figure 12 shown.

[0122] Case analysis of refined advanced prediction of surrounding rock level

[0123] In order to verify the effectiveness of the refined advance prediction method for tunnel surrounding rock grade based on drilling parameters, an analysis was conducted using the excavation cycle of a double-track railway tunnel as an example.

[0124] The tunnel excavation method is the full-face method, using a fully computerized three-arm drilling rig for drilling and blasting the face, and manual and handheld pneumatic drilling and blasting for the invert. The current cycle excavation footage is 3m, ranging from DK3+428.4 to DK3+431.4. Drilled holes have not yet been blasted. The drilling parameters of the current excavation cycle and the four most recent excavation cycles are used to perform a refined advance prediction of the surrounding rock level according to the surrounding rock three-dimensional spatial unit division method and the surrounding rock level refined advance prediction strategy. The refined advance prediction results of the surrounding rock level for the next unexcavated cycle are obtained, as shown in the following figure: Figure 13 As shown, the mileage range of the next trenchless cycle is DK3+431.4~DK3+434.4.

[0125] According to the on-site geological data after the excavation of cycle DK3+431.4~DK3+434.4, the tunnel face area contains two types of rock: conglomerate and slate. The conglomerate part is located on the right side of the tunnel face area, occupies a large area, and the rock is hard, belonging to hard rock, and is relatively complete, which can be judged as Class III. The slate part is located on the left side of the tunnel face area, occupies a small area, and the rock is weak, belonging to relatively soft rock, and is relatively complete, which can be judged as Class IV. Figure 14 shown.

[0126] After the excavation of cycle DK3+431.4~DK3+434.4, the obvious non-uniformity characteristics are shown. Figure 13 and Figure 14 By comparison, the conglomerate part is basically consistent with the Grade III area, and the slate part is basically consistent with the Grade IV area, which proves that the refined advance prediction method of tunnel surrounding rock grade based on drilling parameters is reliable.

[0127] This embodiment studies a method for advanced prediction of the tunnel surrounding rock grade based on drilling parameters, and draws the following conclusions.

[0128] (1) Based on a large number of collected drilling parameters and combined with the long short-term memory network (LSTM), a tunnel surrounding rock grade advance prediction model based on drilling parameters was constructed. The model accuracy was 93.5%, the average precision was 84.9%, the average recall was 92.1%, and the average F1 score was 87.3%, indicating that the model has a good effect in the advance prediction of surrounding rock grade.

[0129] (2) Based on the three-dimensional spatial characteristics of drilling parameters, a three-dimensional spatial unit division method for surrounding rock and a refined advance prediction strategy for surrounding rock level were proposed. The constructed tunnel surrounding rock level advance prediction model based on drilling parameters was used to achieve refined advance prediction of surrounding rock level. The surrounding rock level of the next unexcavated cycle was successfully predicted in advance, which can, to a certain extent, determine the geological conditions farther away from the tunnel face.

[0130] (3) Taking actual engineering as an example, through the analysis of classic cases, the reliability of the tunnel surrounding rock level refined advance prediction method based on drilling parameters is proved.

[0131] In this embodiment, the refined advance prediction results of the surrounding rock level are obtained by the refined advance prediction method of the tunnel surrounding rock level based on drilling parameters. This can grasp some surrounding rock level conditions in front of the face in advance, pre-evaluate potential construction risks, adjust excavation footage and blasting parameters, ensure construction safety and efficiency, and provide suggestions for timely adjustment of on-site construction plans.

[0132] The above is a schematic description of the present invention and its embodiments, which is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. Therefore, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs a structure and embodiment similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A refined advance prediction method for tunnel surrounding rock grade based on drilling parameters, characterized by: The following steps are involved: Step 1: Use a drilling rig to collect drilling parameters in a continuous cycle; Step 2: Extracting statistical characteristics of drilling parameters; In step 2, specifically: The mean, standard deviation, three quartiles, and coefficient of variation were selected as the statistical features of drilling parameters for extraction; The mean is the sum of all data in the data set divided by the number of data. The calculation formula is shown in formula (1): Where μ represents the mean value of the drilling parameters in the current cycle, x i Indicates the drilling parameters of each blasthole point in the current cycle, and n indicates the number of all blasthole points in the current cycle; The standard deviation is the square root of the average of the sum of the squares of the differences between each data point and the mean in the data set. The standard deviation reflects the degree of dispersion of the data set. The larger the standard deviation, the more dispersed the data distribution is, and the smaller the standard deviation, the more concentrated the data distribution is. The calculation formula is shown in formula (2): Where σ represents the standard deviation of the drilling parameters of the current cycle, μ represents the mean value of the drilling parameters of the current cycle, and x i Indicates the drilling parameters of each blasthole point in the current cycle, and n indicates the number of all blasthole points in the current cycle; Quartiles are the values ​​at the three dividing points after all the data in the data set are arranged from small to large, and then divided into four equal parts. The three quartiles are the first quartile Q1, the second quartile Q2, and the third quartile Q3. Among them, the first quartile Q1 is equal to the 25th percentile number after all the data in the data set are arranged from small to large; the second quartile Q2 is equal to the 50th percentile number after all the data in the data set are arranged from small to large; the third quartile Q3 is equal to the 75th percentile number after all the data in the data set are arranged from small to large. The coefficient of variation is the ratio of the standard deviation to the mean. The coefficient of variation can eliminate the influence of the mean of the data set and only reflects the degree of dispersion of the data set relative to its mean. It is a relative indicator used to measure the degree of dispersion of the data set and is suitable for comparing the dispersion of different data sets. The calculation formula is shown in formula (3): Where c v represents the coefficient of variation of the drilling parameters of the current cycle, σ represents the standard deviation of the drilling parameters of the current cycle, and μ represents the mean of the drilling parameters of the current cycle; Step 3: Use the long short-term memory network model to train an advanced prediction model for the tunnel surrounding rock level based on drilling parameters; In the tunnel surrounding rock grade advance prediction model based on drilling parameters in step 3, the drilling parameter statistical characteristics of four excavated cycles are added to the drilling parameter statistical characteristics of one drilled hole but not yet blasted cycle to form a set of cyclic sequence data with a step length of 5 to advance predict the surrounding rock grade of the next unexcavated cycle; Step 4: Divide the entire cycle into multiple blocks horizontally through the surrounding rock three-dimensional space unit division method to perform refined advanced prediction of the surrounding rock level.

2. The method for advanced prediction of tunnel surrounding rock grade based on drilling parameters according to claim 1 is characterized by: In step 1, the drilling rig uses an impact-rotation coupling method to drill holes. The drilling process includes four actions: propulsion, impact, rotation, and flushing. During this process, a total of six drilling parameters, including feed speed, propulsion pressure, impact pressure, rotation pressure, water pressure, and water flow, as well as the corresponding drilling depth, interval time, and drilling coordinates, are collected. Among them, the feed speed and propulsion pressure correspond to the propulsion drilling action, the impact pressure corresponds to the impact drilling action, the rotary pressure corresponds to the rotary drilling action, and the water pressure and water flow correspond to the flushing drilling action.

3. The method for advanced prediction of tunnel surrounding rock grade based on drilling parameters according to claim 2 is characterized by: In step 1, during the drilling process, tunnel excavation is carried out in a cyclic manner, and 20,000 to 70,000 sets of drilling parameters are collected in each cycle. After each drilling cycle is completed, the drilling rig combines all the sets of drilling parameters for each cycle in units of cycles to form a drilling parameter file, which serves as the drilling parameters for that cycle.

4. The method for advanced prediction of tunnel surrounding rock grade based on drilling parameters according to claim 3 is characterized by: The training process of the tunnel surrounding rock grade advance prediction model based on drilling parameters is as follows: 3.1) Dataset division; The dataset is divided into a training set and a test set in a ratio of 4:

1. The training set is used to train the model, and the test set is used to evaluate the model. The dataset consists of samples, and a sample consists of two parts: sample features and sample labels. The sample features are passed to the model as model input to obtain the model output, and then the model output is compared with the sample label to achieve the purpose of training or testing the model. A sample consists of six consecutive cycles arranged in the order of excavation. The statistical characteristics of drilling parameters of the first five cycles constitute the sample characteristics, and the surrounding rock level of the last cycle constitutes the sample label. 3.2) SMOTE oversampling of training set; SMOTE oversampling creates new samples by interpolating between minority class samples; 3.3) Training set standardization; The training set was standardized using Z-score normalization. The mean of each eigenvalue was subtracted and divided by its standard deviation, so that the mean of the processed data was 0 and the standard deviation was 1, while maintaining the distribution of the original data. This eliminates the influence of different dimensions on model training. At the same time, to maintain the consistency of data processing and the fairness of model evaluation, the test set was standardized using the same Z-score normalization parameters as the training set. 3.4) K-fold cross validation training model; K is taken as 5. First, the 5-fold cross-validation is used to divide the training set into 5 subsets, and 1 subset is selected as the cross-validation verification set, and the other 4 subsets are used as cross-validation training sets. Then, the model is trained, and the model evaluation index of the model trained this time on the cross-validation verification set is obtained. Finally, the above steps are repeated until each subset has served as a cross-validation verification set once, and the model evaluation index of the model trained 5 times on the cross-validation verification set is obtained, and the optimal model is selected as the surrounding rock level advanced prediction model.

5. The method for advanced prediction of tunnel surrounding rock grade based on drilling parameters according to claim 4 is characterized by: In step 4, the method for dividing the surrounding rock three-dimensional space unit body is: Based on the non-uniform characteristics of the loop, the entire loop is divided into multiple blocks in the horizontal direction. During the horizontal block division process, the tunnel centerline is used as a vertical reference line and offset to the left or right until the left and right boundaries of the tunnel outline are reached. The horizontal reference line near the bottom of the rail is offset up or down until the upper and lower boundaries of the tunnel outline are reached. Finally, the entire loop is divided into multiple blocks in the horizontal direction. According to the above-mentioned horizontal block division rule, the drilling parameters of each cycle collected by the drilling rig are synchronously divided, and finally a plurality of blocks and drilling parameters within the range of each block are obtained for each cycle.

6. The method for advanced prediction of tunnel surrounding rock grade based on drilling parameters according to claim 5 is characterized by: In step 4, the four most recently excavated cycles and one drilled but not yet blasted cycle are selected and arranged in order of excavation. The statistical features of the drilling parameters of the blocks at the same position in these five cycles are selected to form sequence data at the cycle block level, which are input into the above-trained rock mass level advance prediction model to obtain the rock mass levels of multiple blocks in the next unexcavated cycle, thereby realizing refined advance prediction of the rock mass level.

Images