Semi-supervised shield tunneling surrounding rock grade identification method

Abstract

The present invention provides a semi-supervised shield tunneling surrounding rock grade identification method, which relates to the field of surrounding rock grade identification in underground engineering shield construction. The method includes the following steps: 1) collecting shield tunneling parameters and surrounding rock grades at the target site; 2) using a feature extraction method and an abnormal cycle segment screening method to divide the data set into labeled data and unlabeled data; 3) dividing the labeled data into a training set and a test set, and using the shield tunneling parameters and surrounding rock grades in the training set to construct an initial model; 4) using a semi-supervised algorithm, combining the initial model and unlabeled data to construct a surrounding rock grade identification model, and using the test set to evaluate the model effect; 5) inputting the shield parameters to be identified into the model, and outputting the identification results. Compared with the existing technology, the present invention can effectively utilize a small amount of surrounding rock grades at the survey location and a large amount of tunneling data during the shield tunneling process. The established model has a high accuracy rate and is suitable for surrounding rock grade identification during shield tunneling.

Description

Technical Field

[0001] The present invention relates to the field of surrounding rock grade identification in underground engineering shield construction, and in particular to a semi-supervised shield surrounding rock grade identification method. Background Art

[0002] Shield tunneling, as an underground excavation technique, is widely used in tunneling projects. Surrounding rock stability is crucial during shield construction, directly impacting both construction safety and project quality. Understanding the correct surrounding rock grade during construction can provide effective guidance for shield parameter adjustments and construction decisions.

[0003] Currently, the surrounding rock grade is obtained by taking samples on site and conducting indoor rock mechanics tests on the samples. However, sampling for indoor testing is costly and inefficient. Due to time and economic cost constraints, testing can only be performed at very few locations, and surrounding rock grade test data is scarce. In recent years, emerging machine learning methods have been introduced to the problem of surrounding rock classification. However, existing machine learning surrounding rock classification models require sufficient training data, that is, sufficient data with surrounding rock grades (labeled data). For this reason, the surrounding rock at all locations between two measuring points is generally assumed to be the grade of the previous measuring point, in order to supplement the labeled data to support model training. As a result, the model is trained and tested based on rough and subjective labeled data. If only accurate labeled data from survey locations is used, a large amount of data from non-survey locations will be wasted. Effectively utilizing a small amount of surrounding rock grades at survey locations and a large amount of tunneling data during shield tunneling is of great significance for improving the accuracy of surrounding rock identification models. Summary of the Invention

[0004] To address these shortcomings, the present invention proposes a semi-supervised shield rock grade identification method for identifying the surrounding rock grade during different cycles of shield construction. This method includes collecting shield tunneling parameters and surrounding rock grade, extracting features and screening for abnormal cycles, dividing training and test sets, performing machine learning modeling, and constructing a model using a semi-supervised algorithm. This method aims to effectively utilize accurately labeled data from surveyed locations and a large number of shield tunneling parameters to identify the surrounding rock grade of shield tunneling. This method can more accurately identify the surrounding rock grade during different cycles of shield construction, thereby guiding the adjustment of construction parameters and subsequent surrounding rock treatment.

[0005] To this end, the above-mentioned purpose of the present invention is achieved through the following technical solutions:

[0006] A semi-supervised shield tunneling surrounding rock grade identification method, the method comprising the following steps:

[0007] S1. Collect shield tunneling parameters and surrounding rock grades of the target site, and match limited surrounding rock grades with shield tunneling parameters based on the surveyed locations;

[0008] S2. Using a feature extraction method and an abnormal cycle segment screening method, shield tunneling parameters are selected based on their correlation with surrounding rock grades. The shield tunneling parameters of the abnormal cycle segments are then removed from the selected shield tunneling parameters, while the shield tunneling parameters of the normal cycle segments are retained. The retained shield tunneling parameters of the normal cycle segments are divided into labeled data and unlabeled data; the labeled data are shield parameters with corresponding surrounding rock grades, while the unlabeled data are shield parameters without corresponding surrounding rock grades.

[0009] S3. Divide the labeled data into a training set and a test set, and use a machine learning algorithm to build an initial model based on the shield tunneling parameters and surrounding rock grade in the training set;

[0010] S4. Use a semi-supervised algorithm to train the initial model in combination with unlabeled data to obtain a trained model. Input the shield parameters in the test set into the trained model for model evaluation. If the evaluation result does not meet the standard, return to S2 to rebuild and train the model. If the evaluation result meets the standard, a surrounding rock grade identification model is obtained.

[0011] S5. Input the shield parameters of the tunneling section to be identified into the model and output the identification results.

[0012] As a preferred embodiment of the present invention, the shield tunneling parameters include actual cutterhead speed, set cutterhead speed, cutterhead torque, penetration and cutterhead thrust; each cycle segment in the shield tunneling parameters is output in seconds, and the surrounding rock grade is the surrounding rock grade obtained in units of cycle segments; the cycle segment is a complete working cycle of the shield machine from starting, tunneling to stopping during shield construction.

[0013] As a preferred embodiment of the present invention, the feature extraction method includes a stable segment extraction method, a parameter mean variance calculation and a box plot analysis method. The stable segment extraction method regards the part of the cycle segment where the cutterhead thrust is greater than or equal to 3000kN or the cutterhead thrust is greater than half of the thrust of the cycle segment as a stable segment; the parameter mean variance calculation refers to extracting the stable segment shield tunneling parameters in each cycle segment, and calculating the mean and variance of each shield tunneling parameter after extraction; the box plot analysis method regards the mean and variance of each of the above parameters as features, performs correlation analysis with the corresponding surrounding rock grade, and retains the shield tunneling parameters whose correlation is greater than or equal to the preset value.

[0014] As a preferred embodiment of the present invention, the training set and the test set are divided randomly in proportion in units of cycle segments, and the machine learning algorithm is a random forest algorithm, which requires hyperparameter tuning: adjusting the hyperparameters includes adjusting the maximum depth and the number of base learners, and using K-fold cross validation combined with the Bayesian parameter tuning method to optimize the hyperparameters; wherein the cross validation further divides the training set into K subsets of similar size; each time one of the subsets is selected as the validation set, and the remaining K-1 subsets are used as the training set; in each cross validation, the model is trained on each parameter combination, and the performance is evaluated on the validation set; through multiple cross validations, the average performance under different parameter combinations can be obtained; further, the Bayesian method is used to continuously update the posterior distribution of the parameters using prior information and training set data, so as to find the optimal parameter combination.

[0015] The present invention also provides a surrounding rock grade identification system for implementing the above-mentioned semi-supervised shield surrounding rock grade identification method, comprising:

[0016] The data collection module is used to collect shield tunneling parameters and surrounding rock grades of the target site, and to match limited surrounding rock grades with shield tunneling parameters based on the survey location;

[0017] The data screening module is used to select shield tunneling parameters based on the correlation between shield tunneling parameters and surrounding rock grade using feature extraction and abnormal cycle segment screening methods. The abnormal cycle segments in the selected shield tunneling parameters are then removed, and the normal cycle segments are retained. The retained normal cycle segments are divided into labeled data and unlabeled data.

[0018] The model building module is used to divide the labeled data into training and test sets, and use machine learning algorithms to build an initial model based on the shield tunneling parameters and surrounding rock grade in the training set;

[0019] The model training module is used to train the initial model using a semi-supervised algorithm in combination with unlabeled data to obtain a trained model. The shield parameters in the test set are then input into the trained model for model evaluation. If the evaluation result does not meet the standard, the model is rebuilt and trained. If the evaluation result meets the standard, a surrounding rock grade identification model is obtained.

[0020] The surrounding rock grade identification module is used to input the shield parameters of the excavation section to be identified into the surrounding rock grade identification model and output the identification results.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The shield rock grade output by the method of the present invention is used to guide shield construction, adjusting shield construction parameters based on the quality of the surrounding rock grade. The confidence level is used to guide subsequent decision-making. High-confidence identification results allow for more confident rock treatment; low-confidence identification results may require more caution or additional investigation.

[0023] The semi-supervised shield tunneling surrounding rock grade identification method provided by this invention effectively combines the limited surrounding rock grade required for the project, obtained through surrounding rock integrity and parameter analysis, with the extensive shield parameters acquired by a real-time parameter acquisition system during shield tunneling. This method avoids training and testing the model based on crude and subjective labeled data. Furthermore, a semi-supervised algorithm is employed to iteratively train and expand the dataset using pseudo-labels, avoiding the waste of data from non-surveyed locations by exclusively using accurately labeled data from surveyed locations. This method achieves surrounding rock grade identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a step diagram of the semi-supervised shield surrounding rock grade identification method provided by the present invention;

[0025] Figure 2 This is a flow chart of the semi-supervised algorithm of the semi-supervised shield surrounding rock grade identification method provided by the present invention;

[0026] Figure 3 This is the TPI value of each grade of different models of the semi-supervised shield surrounding rock grade identification method provided by the present invention. DETAILED DESCRIPTION

[0027] The present invention will be further described and illustrated below in conjunction with specific embodiments. The embodiments are merely illustrative of the present disclosure and do not limit its scope. The technical features of the various embodiments of the present invention may be combined accordingly, provided that there is no conflict between them.

[0028] like Figure 1 As shown, a semi-supervised shield tunneling surrounding rock grade identification method includes the following steps:

[0029] S1. Collect shield tunneling parameters and surrounding rock grades of the target site; match the limited surrounding rock grades with shield tunneling parameters based on the survey location.

[0030] In an optional embodiment of the present invention, 5,868 sets of measured tunneling parameters and 294 sets of measured surrounding rock grades were collected for a shield tunneling project. These included 45 sets of Class II surrounding rock, 144 sets of Class III surrounding rock, 87 sets of Class IV surrounding rock, and 18 sets of Class V surrounding rock. Based on actual engineering data, the surrounding rock was further divided into qualified surrounding rock, including Classes II-III, and unqualified surrounding rock, including Classes IV-V.

[0031] Among them, the shield tunneling parameters include five parameters: actual cutterhead speed output in seconds for each cycle segment, set cutterhead speed, cutterhead torque (kN·m), penetration (mm / r), cutterhead thrust (kN) and propulsion speed (mm / min); the surrounding rock grade is the surrounding rock grade obtained based on the "GB50487-2008 Code for Geological Investigation of Water Conservancy and Hydropower Engineering" in units of cycle segments.

[0032] S2. Using feature extraction and abnormal cycle segment screening methods, shield tunneling parameters are selected based on their correlation with surrounding rock grades. Abnormal cycle segments are then removed from the selected shield tunneling parameters, while normal cycle segments are retained. The dataset is then divided into labeled data including shield parameters and corresponding surrounding rock grades and unlabeled data containing only shield parameters.

[0033] Among them, the feature extraction method includes a stable segment extraction method, a parameter mean variance calculation and a box plot analysis method. The stable segment extraction method is to regard the part of the cycle segment where the cutterhead thrust is greater than or equal to 3000kN or the cutterhead thrust is greater than half of the thrust of the cycle segment as a stable segment; the parameter mean variance calculation refers to extracting the stable segment shield tunneling parameters in each cycle segment, and calculating the mean and variance of each shield tunneling parameter after extraction; the box plot analysis method is to regard the mean and variance of each of the above parameters as features, perform correlation analysis with the corresponding surrounding rock grade, and retain the shield tunneling parameters with a correlation greater than or equal to a preset value.

[0034] In an optional embodiment of the present invention, correlation analysis revealed that the correlation between the mean actual cutterhead speed, the mean set cutterhead speed, the mean cutterhead torque, the mean penetration, the variance of penetration, the mean cutterhead thrust, and the variance of cutterhead thrust and the surrounding rock grade was greater than 0.2. Therefore, these parameters were set as input features.

[0035] Among them, the abnormal cycle segment screening method includes analyzing the cutter disc thrust and cutter disc torque curves of each cycle segment, eliminating cycle segments with abnormal conditions such as too short time, missing data, mechanical shutdown and abnormal parameter increase, and retaining normal cycle segments for subsequent analysis.

[0036] In a specific embodiment of the present invention, the above four types of abnormal cycle segments all exist. According to the above standards, a total of 395 groups of cycle segments are considered as abnormal cycle segments. After eliminating them, a total of 5473 cycle segments remain.

[0037] S3. Divide the labeled data into a training set and a test set, and use a machine learning algorithm to build an initial model based on the shield tunneling parameters and surrounding rock grade in the training set.

[0038] Among them, the training set and the test set are divided randomly in a ratio of 6:4 based on the cycle segment. The machine learning algorithm is a random forest algorithm, which requires hyperparameter tuning. The hyperparameters to be adjusted include the maximum depth and the number of base learners. K-fold cross validation combined with the Bayesian parameter tuning method is used for parameter optimization. Specifically, cross validation further divides the training set into K subsets of similar size. Each time, one of the subsets is selected as the validation set, and the remaining K-1 subsets are used as training sets. In each cross validation, the model is trained on each parameter combination, and the performance is evaluated on the validation set. Through multiple cross validations, the average performance under different parameter combinations can be obtained; further, the Bayesian method is used to continuously update the posterior distribution of the parameters using prior information and training set data to find the optimal parameter combination.

[0039] In a specific embodiment of the present invention, after screening for abnormal cycle segments, the labeled data includes 266 sets of data, of which the training set includes 159 sets of data, the test set includes 107 sets of data, and the unlabeled data includes 5207 sets of data. In order to make full use of the data, this embodiment removes the labels from the test set data, treats it as unlabeled data, and adds it to the training. During the Bayesian parameter tuning process, the maximum depth is set between 2 and 4, the number of base learners is set between 100 and 200, 10 tunings are performed, and 5-fold cross-validation is used. After parameter tuning, the final selected hyperparameters are max_depth = 3, n_estimators = 173.

[0040] S4. Use a semi-supervised algorithm to train the initial model in combination with unlabeled data to obtain a trained model. Input the shield parameters in the test set into the trained model for model evaluation. If the evaluation result does not meet the standard, return to S2 to rebuild and train the model. If the evaluation result meets the standard, a surrounding rock grade identification model is obtained.

[0041] The semi-supervised algorithm inputs unlabeled data into the initial model to obtain recognition results and confidence levels. High-confidence samples are treated as pseudo-labels and added to the training set, and iterative training is performed to obtain a new model. Low-confidence samples are treated as unlabeled data and added to the new model. This method is repeated multiple times. When no new high-confidence samples are treated as pseudo-labels in a given iteration, the iteration is terminated, resulting in a trained model. The shield tunneling parameters in the test set are input into the surrounding rock grade output by the trained model. The trained model is evaluated based on the surrounding rock grade corresponding to the shield tunneling parameters in the test set and the surrounding rock grade output by the model. The model evaluation index determines whether the trained model meets the requirements. If the evaluation result does not meet the requirements, the model is retrained in S2. If the evaluation result meets the requirements, a surrounding rock grade recognition model is obtained. High-confidence samples are samples with a confidence level above the manually set confidence threshold, and low-confidence samples are samples with a confidence level below the manually set confidence threshold.

[0042] A specific embodiment of the present invention tried multiple confidence thresholds and analyzed the number of iterations corresponding to each confidence threshold and the number of new pseudo-labels added in each iteration. Table 1 shows the number of iterations and recognition effects corresponding to different probability thresholds. The results show that when the probability threshold is very large, the model has no new pseudo-labels and the performance does not improve; when the probability threshold is large, the number of new pseudo-labels is small and the effect of improving model performance is limited; when the probability threshold is 95%, the accuracy is greatly improved; as the probability threshold continues to decrease, the model accuracy decreases due to the addition of inaccurate pseudo-labels. Therefore, it is reasonable to choose 95% as the probability threshold.

[0043] Table 1

[0044]

[0045] The confidence judgment method is to calculate the probability of each sample being divided into each level in each decision tree, and regard the probability of the final output level as the confidence.

[0046] In an optional embodiment of the present invention, the shield tunneling parameters in the test set are input into the model, and the confidence level of each sample and whether the classification is correct are output. Table 2 shows the model confidence range and its actual accuracy. As can be seen from the table, the accuracy of each confidence range given by the model is within the confidence range, and the accuracy of high-confidence samples is higher, indicating that the confidence provided by the model is reliable.

[0047] Table 2

[0048]

[0049] The accuracy evaluation indicators include the accuracy rate of the overall model accuracy evaluation indicator, the recall rate of each level of accuracy evaluation indicator, and the precision rate of each level of accuracy evaluation indicator. A higher recall rate indicates that the model is less likely to identify the target cycle segment as another label; a higher precision rate indicates that the model is better at finding all cycle segments belonging to the target label. The rock breaking indicator is the torque penetration index (TPI). A higher TPI indicates better surrounding rock quality during shield tunneling.

[0050] In a specific embodiment of the present invention, 107 test sets of shield tunneling parameters were input into the prediction set, resulting in 107 recognition results. These results were then compared with measured results to evaluate model accuracy. The recognition performance of traditional supervised and unsupervised models was compared, and the performance of models constructed using only the rising segment parameters and the entire cycle segment parameters was compared. The accuracy of different models is shown in Table 3, and the recognition performance of different models constructed using the stable segment parameters at various levels is shown in Table 4. As shown in Table 3, when the input parameters are stable segment parameters, the proposed model performs well, outperforming the traditional supervised model, which in turn outperforms the unsupervised model. However, when the input parameters are the entire cycle segment parameters, the overall accuracy of each model decreases, and the accuracy ranking of each model is consistent with that of the stable segment. As shown in Table 4, each model performs better in the classification of Class II-III surrounding rock than Class IV-V surrounding rock. In terms of the recognition performance of Class II-III and Class IV-V surrounding rock grades, the proposed model performs best overall across all three evaluation metrics, while the unsupervised model performs worst. Especially in identifying the IV-V surrounding rock grades, the semi-supervised model is significantly better than the unsupervised and supervised models.

[0051] Table 3

[0052]

[0053] Table 4

[0054]

[0055] The average values ​​of torque penetration index of all cycle sections under each model and surrounding rock classification are as follows: Figure 3 As shown, Class II-III surrounding rock has higher mean TPI values, indicating relatively good rock quality and lower crushing difficulty, while Class IV-V surrounding rock has lower mean TPI values, indicating poorer rock quality and higher crushing difficulty. This result is consistent with the physical property that higher rock grade indicates higher mean TPI values. The TPI difference between the two surrounding rock types is 2.05 for the semi-supervised model, significantly higher than both the supervised and unsupervised models. This indicates that the semi-supervised model is more effective than other models in capturing the difference in rock crushing difficulty between Class II-III and Class IV-V surrounding rock.

[0056] S5. Input the shield parameters that need to be identified into the model and output the identification results and confidence levels.

[0057] The outputted shield rock grade is used to guide shield construction, adjusting shield construction parameters based on the quality of the surrounding rock grade. The confidence level is used to guide subsequent decision-making. High-confidence identification results allow for more confident rock treatment, while low-confidence identification results may require more caution or additional investigation.

[0058] The above-described embodiments merely illustrate several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. Persons skilled in the art will readily appreciate that variations and modifications may be made without departing from the scope of the present invention, all of which fall within the scope of protection of the present invention.

Claims

1. A semi-supervised shield surrounding rock grade identification method, characterized in that: The following steps are involved: S1. Collect shield tunneling parameters and surrounding rock grades of the target site, and map some shield tunneling parameters to surrounding rock grades based on the surveyed locations; S2. Using a feature extraction method and an abnormal cycle segment screening method, shield tunneling parameters are selected based on their correlation with surrounding rock grades. The shield tunneling parameters of the abnormal cycle segments are then removed from the selected shield tunneling parameters, while the shield tunneling parameters of the normal cycle segments are retained. The retained shield tunneling parameters of the normal cycle segments are divided into labeled data and unlabeled data; the labeled data are shield parameters with corresponding surrounding rock grades, while the unlabeled data are shield parameters without corresponding surrounding rock grades. S3. Divide the labeled data into a training set and a test set, and use a machine learning algorithm to build an initial model based on the shield tunneling parameters and surrounding rock grade in the training set; S4. Use a semi-supervised algorithm to train the initial model in combination with unlabeled data to obtain a trained model. Input the shield parameters in the test set into the trained model for model evaluation. If the evaluation result does not meet the standard, return to S2 to rebuild and train the model. If the evaluation result meets the standard, a surrounding rock grade identification model is obtained. S5. Input the shield parameters of the tunneling section to be identified into the model and output the identification results.

2. The semi-supervised shield surrounding rock grade identification method according to claim 1 is characterized in that: The shield tunneling parameters include actual cutterhead speed, set cutterhead speed, cutterhead torque, penetration and cutterhead thrust; each cycle segment in the shield tunneling parameters is output in seconds, and the surrounding rock grade is the surrounding rock grade obtained in units of cycle segments; the cycle segment is a complete working cycle of the shield machine from starting, tunneling to stopping during shield construction.

3. The semi-supervised shield surrounding rock grade identification method according to claim 1 is characterized in that: The feature extraction method includes a stable segment extraction method, a parameter mean variance calculation and a box plot analysis method. The stable segment extraction method is to regard the part of the cycle segment where the cutterhead thrust is greater than or equal to 3000kN or the cutterhead thrust is greater than half of the thrust of the cycle segment as a stable segment; the parameter mean variance calculation refers to extracting the stable segment shield tunneling parameters in each cycle segment, and calculating the mean and variance of each shield tunneling parameter after extraction; the box plot analysis method is to regard the mean and variance of each of the above parameters as features, perform correlation analysis with the corresponding surrounding rock grade, and retain the shield tunneling parameters whose correlation is greater than or equal to the preset value.

4. The semi-supervised shield surrounding rock grade identification method according to claim 1 is characterized in that: The abnormal cycle segment screening method is to analyze the cutter head thrust and cutter head torque curves of each cycle segment, eliminate abnormal cycle segments with a time less than a preset value, missing data, mechanical shutdown or abnormal parameter increase, and retain normal cycle segments for subsequent analysis.

5. The semi-supervised shield surrounding rock grade identification method according to claim 1 is characterized in that: The training set and the test set are divided randomly in proportion with the cycle segment as the unit. The machine learning algorithm is a random forest algorithm, which requires hyperparameter tuning: the hyperparameters including the maximum depth and the number of base learners are adjusted, and the hyperparameters are optimized by K-fold cross validation combined with the Bayesian parameter tuning method; wherein the cross validation further divides the training set into K subsets of similar size; each time one of the subsets is selected as the validation set, and the remaining K-1 subsets are used as the training set; in each cross validation, the model is trained on each parameter combination, and the performance is evaluated on the validation set; through multiple cross validations, the average performance under different parameter combinations is obtained; further, the Bayesian method is used to continuously update the posterior distribution of the parameters using the prior information and the data of the training set, so as to find the optimal parameter combination.

6. The semi-supervised shield surrounding rock grade identification method according to claim 5 is characterized in that: The semi-supervised algorithm inputs unlabeled data into the initial model to obtain recognition results and confidence levels; treats high-confidence samples as pseudo-labels and adds them to the training, iteratively training the new model; Low-confidence samples are treated as unlabeled data and added to the new model. This method is repeated multiple times. When no new high-confidence samples are considered as pseudo-labels during the iteration, the iteration is stopped. After obtaining a trained model, the shield tunneling parameters in the test set are input into the trained model to obtain the surrounding rock grade output by the model. The trained model is evaluated by testing the surrounding rock grade corresponding to the centralized shield tunneling parameters and the surrounding rock grade output by the model, and judging whether the trained model meets the standards according to the model evaluation index. If the evaluation result does not meet the standards, the model is returned to S2 to be retrained. If the evaluation result meets the standards, a surrounding rock grade identification model is obtained; the model evaluation index includes an accuracy evaluation index and a rock breaking index; the high confidence sample is a sample whose confidence is higher than the manually set confidence threshold, and the low confidence sample is a sample whose confidence is lower than the manually set confidence threshold.

7. The semi-supervised shield surrounding rock grade identification method according to claim 6 is characterized in that: The confidence judgment method is to calculate the probability of each sample being divided into each level in each decision tree, and regard the probability of the final output level as the confidence.

8. The semi-supervised shield surrounding rock grade identification method according to claim 6 is characterized in that: The accuracy evaluation indicators include the accuracy rate of the model's overall accuracy evaluation indicator, the recall rate of each level of accuracy evaluation indicator, and the precision rate of each level of accuracy evaluation indicator; the higher the recall rate, the less likely the model is to identify the target cycle segment as other labels; the higher the precision rate, the better the model can find all cycle segments belonging to the target label; the rock breaking indicator is the torque penetration index; the higher the torque penetration index, the better the quality of the surrounding rock during shield excavation.

9. A surrounding rock grade identification system for implementing the semi-supervised shield surrounding rock grade identification method according to claim 1, characterized in that: include: The data collection module is used to collect shield tunneling parameters and surrounding rock grades of the target site, and to match limited surrounding rock grades with shield tunneling parameters based on the survey location; The data screening module is used to select shield tunneling parameters based on the correlation between shield tunneling parameters and surrounding rock grade using feature extraction and abnormal cycle segment screening methods. The abnormal cycle segments in the selected shield tunneling parameters are then removed, and the normal cycle segments are retained. The retained normal cycle segments are divided into labeled data and unlabeled data. The model building module is used to divide the labeled data into training and test sets, and use machine learning algorithms to build an initial model based on the shield tunneling parameters and surrounding rock grade in the training set; The model training module is used to train the initial model using a semi-supervised algorithm in combination with unlabeled data to obtain a trained model. The shield parameters in the test set are then input into the trained model for model evaluation. If the evaluation result does not meet the standard, the model is rebuilt and trained. If the evaluation result meets the standard, a surrounding rock grade identification model is obtained. The surrounding rock grade identification module is used to input the shield parameters of the excavation section to be identified into the surrounding rock grade identification model and output the identification results.

10. A device for identifying surrounding rock grades, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the semi-supervised shield surrounding rock grade identification method according to any one of claims 1 to 8 are implemented.

Images