Rock mass condition prediction method and device, equipment and storage medium

Abstract

The invention provides a rock mass condition prediction method and device, equipment and a storage medium. Comprising the following steps: acquiring original tunneling data of a tunnel boring machine and a rock mass state grade corresponding to the original tunneling data as original data; performing feature screening and feature construction according to the original data to form a model training data set; according to the model training data set, a plurality of machine learning model training algorithms are adopted to carry out model training, and in the model training process, a northern eagle optimization algorithm improved through a crisscross strategy is adopted to optimize hyper-parameters in the model training process; taking the machine learning model with the best model performance as a rock mass condition prediction model according to the model performance of the plurality of machine learning models obtained by training; acquiring original tunneling data of the tunnel boring machine; performing feature screening and feature construction on the original tunneling data to form a tunneling data set; and inputting the tunneling data set into the rock mass condition prediction model to predict rock mass conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of TBM intelligent sensing construction, and in particular relates to a rock mass condition prediction method and device, equipment and storage medium. Background Art

[0002] Compared with the drilling and blasting method, the tunnel boring machine (TBM) construction method has the characteristics of high hole-forming efficiency, high efficiency, environmental protection, and little impact on the surrounding environment. However, when TBM is excavating in strata with complex rock properties and variable structures, it is very easy to get stuck or equipment damaged due to large deformation of the rock mass. Therefore, real-time understanding of the tunnel rock mass conditions is of great significance for construction safety, strengthening support, and reducing construction risks.

[0003] Most of the existing rock mass condition information comes from preliminary drilling surveys, advance drilling during construction, TRT or TSP geophysical exploration, which requires special detection equipment and professionals. In addition, most of the methods cannot advance the TBM normally during the detection process, which will affect the progress of the project construction to a certain extent. How to provide a more convenient rock mass condition prediction method has become a top priority. Summary of the invention

[0004] The present invention provides a rock mass condition prediction method and device, equipment and storage medium, which can form a rock mass condition prediction model with high prediction accuracy.

[0005] In one aspect, a rock mass condition prediction method is provided, comprising: Acquire original excavation data of the tunnel boring machine and the rock mass state grade corresponding to the original excavation data as original data; Perform feature screening and feature construction based on the original data to form a model training data set; According to the model training data set, multiple machine learning model training algorithms are used for model training. During the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyperparameters in the model training process. According to the model performances of the multiple machine learning models obtained through training, the machine learning model with the best model performance is used as the rock mass condition prediction model; Obtaining raw tunneling data from the tunnel boring machine; Perform feature screening and feature construction on the original excavation data of the tunnel boring machine to form an excavation data set; The excavation data set is input into the rock mass condition prediction model to predict the rock mass conditions during the excavation process.

[0006] Optionally, feature screening and feature construction are performed based on the original data to form a model training data set, including: The isolation forest algorithm is used to remove outliers in the original data and fill in the empty values ​​in the original data to form the first tunneling parameter data set; Perform feature screening on the first excavation parameter data set, calculate the variance of a single feature and the correlation between features, remove features whose variance is lower than a variance threshold or whose correlation is higher than a correlation threshold, and obtain a second excavation parameter data set; According to the data in the first excavation parameter data set, feature construction is performed to obtain an excavation performance feature data set, wherein the excavation performance feature data set includes a thrust penetration index, a torque penetration index and a specific penetration rate; The second excavation parameter data set and the excavation performance characteristic data set are combined to obtain a model training data set.

[0007] Optionally, feature construction is performed according to the data in the first excavation parameter data set to obtain an excavation performance feature data set, wherein the excavation performance feature data set includes a thrust penetration index, a torque penetration index and a specific penetration, including: According to the formula Perform feature construction to obtain the field thrust penetration index , Indicates thrust, Indicates the number of hobs. represents the penetration; According to the formula Perform feature construction to obtain the on-site torque penetration index , Indicates the cutter head torque, Indicates the cutter head diameter; According to the formula Perform feature construction to obtain specific penetration .

[0008] Optionally, during the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyper parameters during the model training process, including: Step 1: Initialize the northern goshawk population , , represents the population matrix of the northern goshawk, Indicates The location of the northern goshawk, Indicates The Northern Goshawk The position of each dimension of the northern goshawk represents the value of a hyperparameter; Step 2, determining a value of a model performance indicator of the machine learning model corresponding to the northern goshawk according to the position of the northern goshawk as a fitness value; the model performance indicator includes any one of a weighted average of accuracy and recall, accuracy, and an area under the curve of a receiver operating characteristic curve; Step 3, use the formula , , Conduct prey recognition and attack operations; Indicates A northern goshawk performs prey recognition and attack operations The new location of the dimension; Indicates The fitness value of a northern goshawk; Indicates the location of a northern goshawk's prey; Indicates The first of the northern goshawk's prey The location of the dimension; Indicates The fitness value of each northern goshawk's prey; Indicates target update position of a northern goshawk during prey recognition and attack operations; Indicates the first The fitness value of the target updated position of the northern goshawk; for Random numbers within is a random number 1 or 2; Step 4, use the formula , , , to perform pursuit and escape operations; among them, Indicates After the Northern Goshawk performed the chase and escape operation The new location of the dimension; Indicates The target update position of a northern goshawk during pursuit and escape operations; Indicates The fitness value of the target updated position when a northern goshawk performs pursuit and escape operations; Indicates the current iteration number; Indicates the maximum number of iterations; Step 5, performing a horizontal cross operation on the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4; Step 6, performing a vertical cross operation on the position of the northern goshawk in step 2; Step 7, comparing the fitness values ​​of the northern goshawks before and after the longitudinal crossing in step 6, and retaining the northern goshawks with high fitness values; Step 8: Repeat steps 2 to 7 until the maximum number of iterations is met. , the coordinates of the northern goshawk with the highest fitness value are output as the hyperparameter combination of the machine learning model.

[0009] Optionally, the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4 are subjected to a horizontal cross operation, including: According to the formula , Change the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4 No. Dimensions are cross-sectionalized; among them, Indicates the first The Northern Goshawk Dimensional location; Indicates the first The Northern Goshawk Dimensional location; Indicates the first The first generation of offspring obtained by horizontal cross of northern goshawks dimension; Indicates the first The first generation of offspring obtained by horizontal cross of northern goshawks dimension; and for A random number between .

[0010] Optionally, a longitudinal cross operation is performed on the position of the northern goshawk in step 2, including: According to the formula The northern goshawk in step 2 No. Peace Dimensions are cross-linked vertically; among them, To transfer the northern goshawk from step 2 No. Peace The first generation of individuals obtained after vertical crossover dimension.

[0011] Optionally, according to the model performance of the multiple machine learning models obtained through training, the machine learning model with the best model performance is used as the rock mass condition prediction model, including: The weighted average of the accuracy and recall of the trained machine learning model, the accuracy, and the area under the curve of the receiver operating characteristic curve are accumulated to obtain the cumulative sum of the model performance indicators of the machine learning model; The machine learning model with the largest cumulative sum of the model performance indicators is used as the rock condition prediction model.

[0012] In another aspect, a rock mass condition prediction device is provided, comprising: An acquisition module is used to acquire the original excavation data of the TBM and the rock mass state level corresponding to the original excavation data as original data; The data processing module is used to perform feature screening and feature construction based on the original data to form a model training data set; The model training module is used to perform model training using multiple machine learning model training algorithms based on the model training data set; during the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyperparameters in the model training process; based on the model performance of multiple machine learning models obtained through training, the machine learning model with the best model performance is used as the rock condition prediction model.

[0013] On the other hand, an electronic device is provided, comprising the rock condition prediction device as described above.

[0014] On the other hand, a computer-readable storage medium is provided, wherein at least one program code is stored in the computer-readable storage medium, and the program code is executed by a processor to implement the rock mass condition prediction method as described in any one of the above items.

[0015] The technical solution provided by the embodiments of the present disclosure brings the following beneficial effects: In an embodiment of the present disclosure, a rock mass condition prediction method is provided, which obtains the original excavation data of a tunnel boring machine and the rock mass state level corresponding to the original excavation data as the original data, performs feature screening and feature construction on the original data to form a model training data set, and the data in the model training data set is more conducive to the model performance of the trained machine learning model than the original data. In the model training process, the hyperparameters in the model training process are optimized by adopting the Northern Goshawk optimization algorithm improved by the vertical and horizontal cross strategy, ensuring that the hyperparameter combination used in the machine learning model training process is optimal, thereby further improving the model performance of the machine learning model. Finally, by evaluating the model performance indicators of the machine learning model, the model with the best performance is used as the rock mass condition prediction model to ensure the prediction accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are 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.

[0017] Figure 1 A method flow chart of a rock mass condition prediction method provided by an embodiment of the present disclosure; Figure 2 A complete cycle diagram of TBM excavation operation parameters provided by an embodiment of the present disclosure; Figure 3 A flowchart of a raw data processing process provided by an embodiment of the present disclosure; Figure 4 A flowchart of a model training provided in an embodiment of the present disclosure; Figure 5 A structural block diagram of a rock mass condition prediction device provided in an embodiment of the present disclosure; Figure 6 A structural block diagram of an electronic device provided in an embodiment of the present disclosure.

[0018] The reference numerals are as follows: 21: Acquisition module; 22: Data processing module; 23: Model building module; 31: processor; 32: memory. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are 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.

[0020] In essence, the excavation operation parameters generated by the TBM in real time are the result of the interaction between the TBM and the rock mass. Therefore, the TBM excavation operation parameters can represent the current excavation status of the TBM in real time and reflect the geological conditions of the surrounding rock. Therefore, by deeply exploring the TBM excavation operation parameters and establishing a machine learning model based on the operation parameters, the rock mass conditions can be predicted and output in real time, which is conducive to the TBM operators and construction personnel to make real-time decisions and judgments, select appropriate support and prevention measures, and reduce the increase of construction risks. It is of great significance to the safe construction of TBM tunnel projects.

[0021] Figure 1A method flow chart of a rock mass condition prediction method provided by an embodiment of the present disclosure. Figure 1 ,include: S11. Obtaining original excavation data of the TBM and the rock mass state grade corresponding to the original excavation data as original data.

[0022] The original excavation status data of the Tunnel Boring Machine (TBM) includes the main excavation operation parameters (cutter head speed, penetration, propulsion force, torque, etc.), mechanical motor current and output power parameters, mechanical pressure and pressure parameters, mechanical oil level and temperature, etc. The original excavation data output by different models of TBM will be different.

[0023] In the embodiment of the present disclosure, regarding the rock mass condition grade, if it is a highway tunnel excavated by TBM, the rock mass condition grade is divided into six levels, namely I, II, III, IV, V, and VI, based on the hardness and integrity of the rock mass and with reference to the "Highway Tunnel Design Code" (JTG 3370.1-2018); if it is a water transfer tunnel excavated by TBM, the rock mass condition grade is divided into five categories, namely I, II, III, IV, and V, based on the rock strength, integrity, structural surface state, groundwater and structural surface occurrence and with reference to the "Geophysical Investigation Code for Water Conservancy and Hydropower Engineering" (GB 50487-2008).

[0024] S12. Perform feature screening and feature construction based on the original data to form a model training data set.

[0025] In one example, step S12 includes: Step 1: Use the isolation forest algorithm to remove outliers in the original data and fill in the empty values ​​in the original data to form the first tunneling parameter data set.

[0026] In the disclosed embodiment, the isolation forest method steps are as follows: The first step is to select a certain number of small-volume samples.

[0027] In the second step, an isolation tree is constructed for each group of samples. A feature is randomly selected at each node, and a split value is randomly selected between the maximum and minimum values ​​of the feature. The samples are divided into left and right subtrees based on the split value. The termination condition is: the tree reaches a limited height or the number of node samples reaches a certain number. In the third step, repeat the first and second steps to construct a specific number of isolated trees to form an isolation forest; The fourth step is to calculate the average path length required for the sample from the root node to the node where the sample is isolated in the isolation forest. The samples with shorter average path length are identified as abnormal samples and removed. In addition, in the case of null values ​​in the recorded features, the average value of the feature variables of adjacent time is selected to fill the null values, and finally the first tunneling parameter data set is formed.

[0028] Step 2, perform feature screening on the first excavation parameter data set, calculate the variance of a single feature and the correlation between features, remove features whose variance is lower than a variance threshold or whose correlation is higher than a correlation threshold, and obtain a second excavation parameter data set.

[0029] In one example, step 2 includes: The first step is to use the formula The variance of a single feature is calculated, where Indicates the number of data involved in the calculation of a certain feature. A piece of data representing a certain feature, Indicates the mean of the data involved in the calculation of a certain feature.

[0030] The second step is to calculate the Pearson correlation coefficient formula Calculate the correlation between features. and The data variables representing the two features respectively (that is, one of the data involved in the calculation of the feature), and Represents the average value of the data variable (that is, the average value of the data involved in the calculation of the feature).

[0031] In the third step, low variance features (variance is lower than the variance threshold) and high correlation features (correlation is higher than the correlation threshold) are removed.

[0032] For example, the variance threshold may be 1.5, and the correlation threshold may be 0.8.

[0033] Of course, the above threshold is only an example provided by the embodiment of the present disclosure, and the variance threshold and correlation threshold may be selected, and the present disclosure does not limit this.

[0034] Step 3: construct features based on the data in the first excavation parameter data set to obtain an excavation performance characteristic data set, wherein the excavation performance characteristic data set includes a thrust penetration index, a torque penetration index and a specific penetration rate.

[0035] In one example, step 3 includes: The first step is to use the formula Perform feature construction to obtain the field thrust penetration index , Indicates thrust, Indicates the number of hobs. Indicates penetration.

[0036] The second step is to use the formula Perform feature construction to obtain the on-site torque penetration index , Indicates the cutter head torque, Indicates the cutter head diameter.

[0037] The third step is to use the formula Perform feature construction to obtain specific penetration .

[0038] In the embodiment of the present disclosure, the feature is constructed by the above formula, and the original data is calculated to obtain , and , the tunneling performance characteristics of the tunnel boring machine can be described, the richness of the training data can be improved, and the prediction accuracy of the trained model can be improved.

[0039] Step 4: merge the second excavation parameter data set and the excavation performance characteristic data set to obtain a model training data set.

[0040] In the embodiments of the present disclosure, by removing outliers, filling in null values, screening features and constructing features for the original data, on the one hand, the integrity of the original data can be ensured, and on the other hand, it can ensure that the training data uses better data, which is beneficial to the prediction accuracy of the trained model.

[0041] S13. Based on the model training data set, multiple machine learning model training algorithms are used to perform model training. During the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyperparameters in the model training process.

[0042] In one example, step S13 includes: Step 1: Initialize the northern goshawk population , , represents the population matrix of the northern goshawk, Indicates The location of the northern goshawk, Indicates The Northern Goshawk The position of each dimension of the northern goshawk represents the value of a hyperparameter.

[0043] Step 2, according to the position of the northern goshawk, determine the value of a model performance indicator of the machine learning model corresponding to the northern goshawk as the fitness value, and the model performance indicator includes any one of the weighted average of precision and recall (F1), accuracy (Acurracy), and the area under the curve (AUC) of the receiver operating characteristic curve.

[0044] Step 3, use the formula , , Conduct prey recognition and attack operations; Indicates A northern goshawk performs prey recognition and attack operations The new location of the dimension; Indicates The fitness value of a northern goshawk; Indicates the location of a northern goshawk's prey; Indicates The first of the northern goshawk's prey The location of the dimension; Indicates The fitness value of each northern goshawk's prey; Indicates target update position of a northern goshawk during prey recognition and attack operations; Indicates the first The fitness value of the target updated position of the northern goshawk; for Random numbers within A random number 1 or 2.

[0045] Step 4, use the formula , , , to perform pursuit and escape operations; among them, Indicates After the Northern Goshawk performed the chase and escape operation The new location of the dimension; Indicates The target update position of a northern goshawk during pursuit and escape operations; Indicates The fitness value of the target updated position when a northern goshawk performs pursuit and escape operations; Indicates the current iteration number; Indicates the maximum number of iterations.

[0046] Step 5, performing a horizontal cross operation on the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4.

[0047] In one example, step 5 includes: According to the formula , The parent individual and the offspring individuals formed in step 4 No. Dimensions are cross-sectionalized; among them, and Represents the parent individuals and offspring individuals No. dimension, and Represents the parent individuals , offspring individuals The first generation of individuals obtained by horizontal crossover dimension, and for A random number between .

[0048] In the disclosed embodiment, the above formula can be used to implement a horizontal crossover operation between parent individuals and offspring individuals, further improving the richness of the northern goshawk population.

[0049] Step 6, perform a vertical cross operation on the position of the northern goshawk in step 2.

[0050] In one example, step 6 includes: According to the formula The parent individual No. Peace The dimensions are vertically crossed, where For the parent individual No. Peace The offspring obtained after vertical crossover dimension.

[0051] In the disclosed embodiment, the above formula is used to perform a vertical crossover operation on the parent individuals of the northern goshawk to improve the richness of the group.

[0052] Step 7, compare the fitness values ​​of the northern goshawks before and after the longitudinal cross in step 6, and retain the northern goshawks with high fitness values.

[0053] Step 8: Repeat steps 2 to 7 until the maximum number of iterations is met. , the coordinates of the northern goshawk with the highest fitness value are output as the hyperparameter combination of the machine learning model.

[0054] In the disclosed embodiment, by utilizing the vertical and horizontal cross strategy to improve and optimize the Northern Goshawk optimization algorithm, the ability of the Northern Goshawk algorithm to jump out of the local optimal value can be enhanced, thereby improving the model performance of the trained machine learning model.

[0055] In step S13, the data in the model training data set is first divided into a training set (80%) and a test set (20%), and then a ten-fold cross-training method is used to perform model training using machine learning algorithms such as Light Gradient Boosting Machine (LightGBM) algorithm, Extreme Gradient Boosting (XGBoost) algorithm, Random Forest (RF) algorithm, Support Vector Machine (SVM) algorithm, and Extra-Trees (ET). During the training process, hyperparameters are optimized, for example, hyperparameters are optimized with F1 as the optimization target, and a hyperparameter combination is obtained for model training.

[0056] S14. Based on the model performance of multiple machine learning models obtained through training, the machine learning model with the best model performance is used as the rock condition prediction model.

[0057] In one example, step S14 includes: In the first step, the model performance indicators of the trained machine learning model are accumulated to obtain the cumulative sum of the model performance indicators of the machine learning model. The model performance indicators include the weighted average of accuracy and recall (F1), accuracy (Acurracy), and the area under the curve (AUC) of the receiver operating characteristic curve.

[0058] In the second step, the machine learning model with the largest sum of the model performance indicators is used as the rock condition prediction model.

[0059] In the embodiments of the present disclosure, the model performance index is used to evaluate the model performance of the trained machine learning model. By determining the size of F1, Acurracy, and AUC in the model performance index, and determining the sum of F1, Acurracy, and AUC, the model performance of multiple machine learning models finally obtained by training is evaluated, and the one with the best performance can be used for rock condition prediction.

[0060] S15. Obtaining original tunnel boring data of the tunnel boring machine.

[0061] S16. Perform feature screening and feature construction on the original tunnel boring machine excavation data to form a tunnel boring machine data set.

[0062] See step S12.

[0063] S17. Inputting the excavation data set into the rock mass condition prediction model to predict the rock mass condition during the excavation process.

[0064] In an embodiment of the present disclosure, a rock mass condition prediction method is provided, which obtains the original excavation data of a tunnel boring machine and the rock mass state level corresponding to the original excavation data as the original data, performs feature screening and feature construction on the original data to form a model training data set, and the data in the model training data set is more conducive to the model performance of the trained machine learning model than the original data. In the model training process, the hyperparameters in the model training process are optimized by adopting the Northern Goshawk optimization algorithm improved by the vertical and horizontal cross strategy, ensuring that the hyperparameter combination used in the machine learning model training process is optimal, thereby further improving the model performance of the machine learning model. Finally, by evaluating the model performance indicators of the machine learning model, the model with the best performance is used as the rock mass condition prediction model to ensure the prediction accuracy of the model.

[0065] Figure 2 A complete cycle diagram of TBM excavation operation parameters provided in an embodiment of the present disclosure.

[0066] Figure 3 A flowchart of a raw data processing process provided by an embodiment of the present disclosure. Figure 3 , Figure 3 The processing process of the original data is shown, and the data sequence number represents the processing step.

[0067] Figure 4 A flowchart of a model training provided by an embodiment of the present disclosure. Figure 4 , Figure 4 The process of model training is shown in Figure 2.

[0068] Figure 5 This is a structural block diagram of a rock mass condition prediction device provided by an embodiment of the present disclosure. Figure 5 ,include: The acquisition module 21 is used to acquire the original excavation data of the TBM and the rock mass state level corresponding to the original excavation data as the original data; and to acquire the original excavation data of the tunnel boring machine.

[0069] The data processing module 22 is used to perform feature screening and feature construction based on the original data to form a model training data set; and perform feature screening and feature construction on the original excavation data of the tunnel boring machine to form an excavation data set.

[0070] The model training module 23 is used to perform model training using multiple machine learning model training algorithms according to the training set in the model training data set. During the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyperparameters in the model training process; the model performance of the multiple machine learning models obtained by the training is tested using the test set, and the machine learning model with the best model performance is used as the rock mass condition prediction model; The prediction module 24 is used to input the excavation data set into the rock mass condition prediction model to predict the rock mass conditions during the excavation process.

[0071] Optionally, the data processing module 22 is used to: The isolation forest algorithm is used to remove outliers in the original data and fill in the empty values ​​in the original data to form the first tunneling parameter data set; Perform feature screening on the first excavation parameter data set, calculate the variance of a single feature and the correlation between features, remove features whose variance is lower than a variance threshold or whose correlation is higher than a correlation threshold, and obtain a second excavation parameter data set; According to the data in the first excavation parameter data set, feature construction is performed to obtain an excavation performance feature data set, wherein the excavation performance feature data set includes a thrust penetration index, a torque penetration index and a specific penetration rate; The second excavation parameter data set and the excavation performance characteristic data set are combined to obtain a model training data set.

[0072] Optionally, the data processing module 22 is used to: According to the formula Perform feature construction to obtain the field thrust penetration index , Indicates thrust, Indicates the number of hobs. represents the penetration; According to the formula Perform feature construction to obtain the on-site torque penetration index , Indicates the cutter head torque, Indicates the cutter head diameter; According to the formula Perform feature construction to obtain specific penetration .

[0073] Optionally, the model training module 23 is used to: Step 1: Initialize the northern goshawk population , , represents the population matrix of the northern goshawk, Indicates The location of the northern goshawk, Indicates The Northern Goshawk The position of each dimension of the northern goshawk represents the value of a hyperparameter; Step 2, determining a value of a model performance indicator of the machine learning model corresponding to the northern goshawk according to the position of the northern goshawk as a fitness value; the model performance indicator includes any one of a weighted average of accuracy and recall, accuracy, and an area under the curve of a receiver operating characteristic curve; Step 3, use the formula , , Conduct prey recognition and attack operations; Indicates A northern goshawk performs prey recognition and attack operations The new location of the dimension; Indicates The fitness value of a northern goshawk; Indicates the location of a northern goshawk's prey; Indicates The first of the northern goshawk's prey The location of the dimension; Indicates The fitness value of each northern goshawk's prey; Indicates target update position of a northern goshawk during prey recognition and attack operations; Indicates the first The fitness value of the target updated position of the northern goshawk; for Random numbers within is a random number 1 or 2; Step 4, use the formula , , , to perform pursuit and escape operations; among them, Indicates After the Northern Goshawk performed the chase and escape operation The new location of the dimension; Indicates The target update position of a northern goshawk during pursuit and escape operations; Indicates The fitness value of the target updated position when a northern goshawk performs pursuit and escape operations; Indicates the current iteration number; Indicates the maximum number of iterations; Step 5, performing a horizontal cross operation on the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4; Step 6, performing a vertical cross operation on the position of the northern goshawk in step 2; Step 7, comparing the fitness values ​​of the northern goshawks before and after the longitudinal crossing in step 6, and retaining the northern goshawks with high fitness values; Step 8: Repeat steps 2 to 7 until the maximum number of iterations is met. , the coordinates of the northern goshawk with the highest fitness value are output as the hyperparameter combination of the machine learning model.

[0074] Optionally, the model training module 23 is used to: According to the formula , Change the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4 No. Dimensions are cross-sectionalized; among them, Indicates the first The Northern Goshawk Dimensional location; Indicates the first The Northern Goshawk Dimensional location; Indicates the first The first generation of offspring obtained by horizontal cross of northern goshawks dimension; Indicates the first The first generation of offspring obtained by horizontal cross of northern goshawks dimension; and for A random number between .

[0075] Optionally, the model training module 23 is used to: According to the formula The northern goshawk in step 2 No. Peace Dimensions are cross-linked vertically; among them, To transfer the northern goshawk from step 2 No. Peace The first generation of individuals obtained after vertical crossover dimension.

[0076] Optionally, the model training module 23 is used to: The weighted average of the accuracy and recall of the trained machine learning model, the accuracy, and the area under the curve of the receiver operating characteristic curve are accumulated to obtain the cumulative sum of the model performance indicators of the machine learning model; The machine learning model with the largest cumulative sum of the model performance indicators is used as the rock condition prediction model.

[0077] Figure 6 This is a structural block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 6 , electronic devices may include Figure 5 The rock mass condition prediction device generally includes a processor 31 and a memory 32 .

[0078] The processor 31 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 31 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 31 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. The memory 32 may include one or more computer-readable storage media, which may be non-transitory. The memory 32 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices and flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 32 is used to store at least one instruction, which is used to be executed by the processor 31 to implement the rock mass condition prediction method performed by an electronic device provided in the method embodiment of the present application.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rock mass condition prediction method, characterized in that: include: Acquire original excavation data of the tunnel boring machine and the rock mass state grade corresponding to the original excavation data as original data; Perform feature screening and feature construction based on the original data to form a model training data set; According to the model training data set, multiple machine learning model training algorithms are used for model training. During the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyperparameters in the model training process. According to the model performances of the multiple machine learning models obtained through training, the machine learning model with the best model performance is used as the rock mass condition prediction model; Obtaining raw tunneling data from the tunnel boring machine; Perform feature screening and feature construction on the original excavation data of the tunnel boring machine to form an excavation data set; The excavation data set is input into the rock mass condition prediction model to predict the rock mass conditions during the excavation process.

2. The rock mass condition prediction method according to claim 1, characterized in that: Perform feature screening and feature construction based on the original data to form a model training data set, including: The isolation forest algorithm is used to remove outliers in the original data and fill in the empty values ​​in the original data to form the first tunneling parameter data set; Perform feature screening on the first excavation parameter data set, calculate the variance of a single feature and the correlation between features, remove features whose variance is lower than a variance threshold or whose correlation is higher than a correlation threshold, and obtain a second excavation parameter data set; According to the data in the first excavation parameter data set, feature construction is performed to obtain an excavation performance feature data set, wherein the excavation performance feature data set includes a thrust penetration index, a torque penetration index and a specific penetration rate; The second excavation parameter data set and the excavation performance characteristic data set are combined to obtain a model training data set.

3. The rock mass condition prediction method according to claim 2, characterized in that: According to the data in the first excavation parameter data set, feature construction is performed to obtain an excavation performance feature data set, wherein the excavation performance feature data set includes a thrust penetration index, a torque penetration index and a specific penetration, including: According to the formula Perform feature construction to obtain the field thrust penetration index , Indicates thrust, Indicates the number of hobs. represents the penetration; According to the formula Perform feature construction to obtain the on-site torque penetration index , Indicates the cutter head torque, Indicates the cutter head diameter; According to the formula Perform feature construction to obtain specific penetration .

4. The rock mass condition prediction method according to claim 1, characterized in that: During the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyperparameters in the model training process, including: Step 1: Initialize the northern goshawk population , , represents the population matrix of the northern goshawk, Indicates The location of the northern goshawk, Indicates The Northern Goshawk The position of each dimension of the northern goshawk represents the value of a hyperparameter; Step 2, determining a value of a model performance indicator of the machine learning model corresponding to the northern goshawk according to the position of the northern goshawk as a fitness value; the model performance indicator includes any one of a weighted average of accuracy and recall, accuracy, and an area under the curve of a receiver operating characteristic curve; Step 3, use the formula , , Conduct prey recognition and attack operations; Indicates A northern goshawk performs prey recognition and attack operations The new location of the dimension; Indicates The fitness value of a northern goshawk; Indicates the location of a northern goshawk's prey; Indicates The first of the northern goshawk's prey The location of the dimension; Indicates The fitness value of each northern goshawk's prey; Indicates target update position of a northern goshawk during prey recognition and attack operations; Indicates the first The fitness value of the target updated position of the northern goshawk; for Random numbers within is a random number 1 or 2; Step 4, use the formula , , , to perform pursuit and escape operations; among them, Indicates After the Northern Goshawk performed the chase and escape operation The new location of the dimension; Indicates The target update position of a northern goshawk during pursuit and escape operations; Indicates The fitness value of the target updated position when a northern goshawk performs pursuit and escape operations; Indicates the current iteration number; Indicates the maximum number of iterations; Step 5, performing a horizontal cross operation on the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4; Step 6, performing a vertical cross operation on the position of the northern goshawk in step 2; Step 7, comparing the fitness values ​​of the northern goshawks before and after the longitudinal crossing in step 6, and retaining the northern goshawks with high fitness values; Step 8: Repeat steps 2 to 7 until the maximum number of iterations is met. , the coordinates of the northern goshawk with the highest fitness value are output as the hyperparameter combination of the machine learning model.

5. The rock mass condition prediction method according to claim 4, characterized in that: The position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4 are horizontally crossed, including: According to the formula , Change the position of the northern goshawk in step 2 and the position of the northern goshawk obtained in step 4 No. Dimensions are cross-sectionalized; among them, Indicates the first The Northern Goshawk Dimensional location; Indicates the first The Northern Goshawk Dimensional location; Indicates the first The first generation of offspring obtained by horizontal cross of northern goshawks dimension; Indicates the first The first generation of offspring obtained by horizontal cross of northern goshawks dimension; and for A random number between .

6. The rock mass condition prediction method according to claim 4, characterized in that: Perform a vertical cross operation on the position of the northern goshawk in step 2, including: According to the formula The northern goshawk in step 2 No. Peace Dimensions are cross-linked vertically; among them, To transfer the northern goshawk from step 2 No. Peace The first generation of individuals obtained after vertical crossover dimension.

7. The rock mass condition prediction method according to any one of claims 1 to 6, characterized in that: According to the model performance of multiple machine learning models obtained through training, the machine learning model with the best model performance is used as the rock mass condition prediction model, including: The weighted average of the accuracy and recall of the trained machine learning model, the accuracy, and the area under the curve of the receiver operating characteristic curve are accumulated to obtain the cumulative sum of the model performance indicators of the machine learning model; The machine learning model with the largest cumulative sum of the model performance indicators is used as the rock condition prediction model.

8. A rock mass condition prediction device, characterized in that: include: An acquisition module, which acquires original excavation data of the tunnel boring machine and the rock mass state level corresponding to the original excavation data as original data; Obtaining raw tunneling data from the tunnel boring machine; The data processing module is used to perform feature screening and feature construction based on the original data to form a model training data set; perform feature screening and feature construction on the original tunnel boring machine excavation data to form a tunnel boring machine excavation data set; The model training module is used to perform model training using multiple machine learning model training algorithms according to the model training data set; during the model training process, the northern goshawk optimization algorithm improved by the vertical and horizontal cross strategy is used to optimize the hyperparameters in the model training process; based on the model performance of the multiple machine learning models obtained through training, the machine learning model with the best model performance is used as the rock mass condition prediction model; The prediction module is used to input the excavation data set into the rock condition prediction model to predict the rock conditions during the excavation process.

9. An electronic device, characterized in that: It comprises the rock condition prediction device as described in claim 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one program code, and the program code is executed by a processor to implement the rock condition prediction method according to any one of claims 1 to 7.

Images