Rockburst grade prediction method and system based on HPO-SVM model

Abstract

The invention discloses a rockburst grade prediction method and system based on an HPO-SVM model, and belongs to the field of underground engineering rockburst disaster prediction.The method comprises the steps that a sample data set of rockburst cases is constructed and analyzed, the sample data set is divided into a training set and a test set, and rockburst main control factors and rockburst grades are determined; establishing a rockburst prediction model based on a hunter prey optimization algorithm and a support vector machine algorithm, and training the rockburst prediction model based on the training set to obtain a trained rockburst prediction model; selecting a plurality of evaluation indexes of the rockburst prediction model, inputting the test set into the trained rockburst prediction model to obtain a predicted value, and comparing and analyzing the actual value and the predicted value based on the evaluation indexes; and selecting the rockburst project instance as a verification set, and predicting the rockburst intensity grade of the rockburst project instance by adopting the rockburst prediction model. According to the method, the rockburst level of the underground engineering can be accurately predicted, and the safety and the construction efficiency of the underground engineering can be improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of rock burst disaster prediction of underground engineering, and particularly relates to a rock burst grade prediction method and system based on an HPO-SVM model. BACKGROUND

[0002] At present, the rock burst tendency prediction and evaluation methods can be summarized into seven categories. The first category is single-parameter prediction and evaluation method, mainly including Hoek criterion, Russenes criterion, Turchaninov criterion, Kidybinski energy criterion, Motycaka energy ratio method, Qinling tunnel criterion, Erlangshan highway tunnel criterion, surrounding rock strength ratio, Gu-Tao criterion and rock burst explosion potential criterion, etc. The second category is multi-parameter prediction and evaluation method, mainly including rock tensile and compressive strength, maximum principal stress of surrounding rock, maximum tangential and radial stress of surrounding rock, rock mass integrity coefficient, lateral pressure coefficient and rock elastic energy index, etc. The third category is prediction and evaluation method based on numerical evaluation index, mainly including rock burst potential ratio (BPI), energy release rate (ERR), excess shear stress (ESS), local energy release density (LERD), failure approach index (FAI), rock burst energy release rate (RBERR), local energy release rate (LERR), brittle shear ratio (BSR), cumulative elastic strain energy after mining, energy density criterion, ratio of actual energy storage of rock mass to limit energy, relative energy release index (RERI) and unit time relative energy release rate index (URLERI), etc. The fourth category is prediction and evaluation method based on uncertainty theory, mainly including set pair analysis method, fuzzy matter element model, entropy weight cloud model, principal component analysis, fuzzy comprehensive evaluation method, interval fuzzy comprehensive evaluation method, extension evaluation, rough set, normal cloud model, unascertained measure theory, multi-dimensional connection cloud model, fuzzy comprehensive evaluation method, dynamic weight gray classification model, attribute recognition model, rough set of reduced concept lattice, efficacy coefficient method, matter element extension model, gray clustering method, improved CRITIC method and multi-dimensional cloud model, etc. The fifth category is prediction and evaluation method based on ranking idea, mainly including principal component analysis based on mutual information, information entropy and TOPSIS method, fuzzy MABAC method, TOPSIS method, TODIM method and similarity measurement based on Vague set, etc. The sixth category is prediction and evaluation method based on machine learning algorithm, mainly including Gaussian process, abstract ant colony clustering algorithm, genetic algorithm, support vector machine, random forest, synthetic minority over-sampling technique, linear discriminant analysis, quadratic discriminant analysis, partial least squares discriminant analysis, Naive Bayes, K-nearest neighbors, Bayesian network, multilayer perceptron neural network, regression model, etc. The seventh category is rock burst prediction method based on field measurement, mainly including electromagnetic radiation method, acoustic emission method or microseismic method, microgravity method, rheological method, rebound method and drilling cuttings method, etc.

[0003] However, the above seven types of rock burst prediction and evaluation methods have some shortcomings. The first type of prediction and evaluation method is simple and practical, easy to understand, and the data is easy to obtain, but different thresholds may be obtained according to different field experiences, and even contradictory threshold division methods. The second type of prediction and evaluation method comprehensively considers the influence of multiple factors and is simple and easy to operate, but the threshold division of rock burst grade has a certain subjectivity, and the meaning is not clear enough after being integrated into a single index. The third type of prediction and evaluation method can simulate the brittle failure behavior of rock mass and the process of energy accumulation, transfer and release, and can determine the rock burst grade, occurrence position and blast crater depth, but it is difficult to determine the specific threshold of each risk grade. The fourth type of prediction and evaluation method considers the comprehensive influence of multiple factors and the uncertainty in the evaluation process, and can handle qualitative and quantitative information at the same time, but it is difficult to determine the threshold of each index corresponding to different rock burst grades and the membership function corresponding to each index. The fifth type of prediction and evaluation method comprehensively considers the influence of multiple factors, and can be combined with fuzzy theory to handle uncertainty problems, but it is difficult to determine the standard sample of known rock burst grade. The sixth type of prediction and evaluation method comprehensively considers the influence of multiple factors and has strong ability to handle nonlinear problems, but it needs a large amount of field rock burst case data, and it is difficult to handle uncertainty problems, and to a large extent depends on the optimization of hyperparameters, and the results are greatly affected by model parameters. The seventh type of prediction and evaluation method can obtain the most timely field measured information, and then establish the corresponding relationship between the specific phenomena occurring in the rock mass excavation process and the measured information, but the accuracy of the monitoring data is greatly affected by the field construction, and the purchase, installation and maintenance of the field monitoring equipment are high, which limits the application of rock burst prediction methods based on field measurement. SUMMARY

[0004] To solve the above technical problems, the present application provides a rock burst grade prediction method based on HPO-SVM model to solve the problems existing in the prior art.

[0005] To achieve the above purpose, in a first aspect, the present application provides a rock burst grade prediction method based on HPO-SVM model, comprising:

[0006] Constructing and analyzing sample data set of rock burst cases, dividing the sample data set into training set and test set, and determining rock burst master control factor and rock burst grade;

[0007] Based on the hunter-prey optimization algorithm and the support vector machine algorithm, a rock burst prediction model is established, and the rock burst prediction model is trained based on the training set to obtain a trained rock burst prediction model;

[0008] Selecting multiple evaluation indexes of the rock burst prediction model, inputting the test set into the trained rock burst prediction model to obtain predicted values, and comparing and analyzing actual values and predicted values based on the evaluation indexes;

[0009] The rock burst prediction model is used to predict the rock burst intensity grade of the rock burst engineering example.

[0010] Preferably, the rock burst master control factors include stress coefficient, brittleness coefficient, elastic energy index and integrity coefficient.

[0011] The rock burst grade includes no rock burst, weak rock burst, moderate rock burst and strong rock burst.

[0012] Preferably, the method further comprises statistics of the rock burst master control factors and analysis of the correlation between the rock burst master control factors.

[0013] Preferably, the process of establishing the rock burst prediction model comprises:

[0014] The penalty factor and the kernel parameter corresponding to the optimal position after multiple iterations are assigned to the support vector machine algorithm for training and testing of the sample data set.

[0015] The penalty factor and the kernel parameter corresponding to the optimal position after multiple iterations are assigned to the support vector machine algorithm for training and testing of the sample data set.

[0016] Preferably, the process of optimizing the penalty factor and the kernel parameter of the kernel function in the support vector machine algorithm by the hunter-prey optimization algorithm comprises:

[0017] The hunter-prey optimization algorithm initializes the population position.

[0018] Randomly generate hunters or prey in space

[0019] Introduce hunter search mechanism and prey escape mechanism

[0020] Optimization is performed through the process of hunters searching for the position of prey and prey escaping.

[0021] Preferably, the evaluation index includes root mean square error, mean absolute percentage error and goodness of fit.

[0022] In a second aspect, the present application further provides a rock burst grade prediction system based on the HPO-SVM model, comprising:

[0023] The data set construction module is used to construct and analyze the sample data set of the rock burst case, divide the sample data set into a training set and a test set, and determine the rock burst master control factors and the rock burst grade.

[0024] The model construction module is used to establish a rock burst prediction model based on the hunter-prey optimization algorithm and the support vector machine algorithm, train the rock burst prediction model based on the training set, and obtain a trained rock burst prediction model.

[0025] The prediction evaluation module is used for selecting a plurality of evaluation indexes of the rock burst prediction model, inputting the test set into the trained rock burst prediction model to obtain a predicted value, and comparing and analyzing an actual value and the predicted value based on the evaluation indexes.

[0026] The prediction verification module selects a rock burst engineering example as a verification set and uses the rock burst prediction model to predict a rock burst intensity grade of the rock burst engineering example.

[0027] In a third aspect, the present application further discloses a computer device comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to realize the steps of the method in the first aspect.

[0028] In a fourth aspect, the present application further discloses a computer readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to realize the steps of the method in the first aspect.

[0029] In a fifth aspect, the present application further discloses a computer program product comprising a computer program, wherein the computer program is executed by a processor to realize the steps of the method in the first aspect.

[0030] Compared with the prior art, the present application has the following advantages and technical effects:

[0031] The present application provides a rock burst grade prediction method based on an HPO-SVM model, comprising the following steps: firstly, constructing and analyzing sample data sets of rock burst cases, dividing the sample data sets into a training set and a test set, and determining rock burst main control factors and rock burst grades; secondly, based on a hunter-prey optimization algorithm and a support vector machine algorithm, establishing a rock burst prediction model, training the rock burst prediction model based on the training set, and obtaining a trained rock burst prediction model; thirdly, selecting a plurality of evaluation indexes of the rock burst prediction model, inputting the test set into the trained rock burst prediction model to obtain a predicted value, and comparing and analyzing an actual value and the predicted value based on the evaluation indexes; and finally, selecting a rock burst engineering example as a verification set and using the rock burst prediction model to predict a rock burst intensity grade of the rock burst engineering example.

[0032] The present embodiment determines the rock burst main control factor and the rock burst grade by constructing and analyzing the rock burst case sample data set, and provides a rich data basis and a clear prediction target for the model. Combining the Hunter-Prey Optimization (HPO) algorithm with the Support Vector Machine (SVM) can establish an accurate rock burst prediction model, effectively improve the accuracy of rock burst grade prediction, and make the prediction result closer to the actual rock burst situation. The global optimization capability of the HPO algorithm can avoid the SVM from falling into local optimization in the parameter optimization process, ensure that the model can achieve better performance in the training process, and thus improve the prediction capability of the model.

[0033] The present application selects multiple evaluation indexes to evaluate the rock burst prediction model, which can comprehensively and objectively evaluate the performance of the model. By inputting the test set into the trained rock burst prediction model, the predicted value is obtained, and the actual value and the predicted value are compared and analyzed based on the evaluation index, so that the prediction error, accuracy and other performance indexes of the model can be accurately measured, which provides a basis for the improvement and optimization of the model, and ensures that the model has good prediction effect.

[0034] The present application selects rock burst engineering examples as the verification set, and uses the trained rock burst prediction model to predict the rock burst intensity grade of these examples, which can verify the applicability and effectiveness of the model in actual engineering. By comparing with the actual rock burst situation, the reliability and practicality of the model are further proved, which provides an effective technical means for rock burst prediction of underground engineering, and helps to improve the safety and construction efficiency of underground engineering. BRIEF DESCRIPTION OF DRAWINGS

[0035] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application and are incorporated in and constitute a part of this application. The embodiments of this application and its description are used to explain the application and do not limit the application. In the drawings:

[0036] Figure 1 The rock burst intensity grade distribution histogram of the embodiment of the present application;

[0037] Figure 2 The rock burst prediction evaluation index distribution histogram of the embodiment of the present application; wherein (a) is the integrity coefficient histogram, (b) is the stress coefficient histogram, (c) is the brittleness coefficient histogram, and (d) is the elastic energy index histogram;

[0038] Figure 3 The index data box chart of the embodiment of the present application; wherein (a) is the integrity coefficient box chart, (b) is the stress coefficient box chart, (c) is the brittleness coefficient box chart, and (d) is the elastic energy index box chart;

[0039] Figure 4 A rock burst prediction index correlation matrix diagram of an embodiment of the present application;

[0040] Figure 5 A diagram showing the positions of the hunter searching for the prey and the prey escaping, wherein (a) is a diagram showing the searching behavior of the hunter, and (b) is a diagram showing the escaping behavior of the prey;

[0041] Figure 6 A flowchart of a rock burst grade prediction method based on the HPO-SVM model of an embodiment of the present application;

[0042] Figure 7 A comparison diagram of evaluation results of an embodiment of the present application, wherein (a) is a comparison diagram of predicted values and actual values of a training set, (b) is a diagram showing the prediction error of the training set, (c) is a comparison diagram of predicted values and actual values of a test set, and (d) is a diagram showing the prediction error of the test set. DETAILED DESCRIPTION

[0043] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict. The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0044] It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a group of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0045] Embodiment one

[0046] In view of the technical problems in the background art, the purpose of the present application is to propose a rock burst grade prediction method based on the HPO-SVM model, summarize and analyze typical rock burst engineering case data information, determine the main control factors of rock burst occurrence; based on the support vector machine (SVM) and the hunter prey optimization algorithm (HPO), establish a multi-factor integrated rock burst risk intelligent prediction model, and realize accurate and effective prediction and identification of rock burst grade.

[0047] In this embodiment, the internal relationship between the rock burst incubation mechanism and the main control factors of rock burst is first clarified, and reference is made to the relevant literature and cases of underground engineering rock burst at home and abroad. The main role is to study the occurrence location, scale and law of underground engineering rock burst and the main characteristics, as a training and test sample library of the rock burst intelligent prediction model, and then determine the key factors controlling the occurrence of rock burst. At the same time, statistical analysis is made on the main control factors of rock burst, the correlation between the main control factors of rock burst is clarified, the HPO-SVM rock burst intelligent prediction model is constructed, the accuracy and engineering applicability of the model are verified in combination with rock burst engineering cases, so as to provide a new way to solve the problem of rock burst tendency evaluation.

[0048] As Figure 6 shown in the embodiment, a rock burst grade prediction method based on HPO-SVM model is provided, comprising:

[0049] Step S1, constructing and analyzing sample data set of rock burst cases, dividing the sample data set into training set and test set, and determining rock burst main control factors and rock burst grade;

[0050] Step S1 specifically includes:

[0051] Step S11, constructing rock burst case sample data set;

[0052] Specifically, referring to the relevant literature and cases of deep underground engineering rock burst at home and abroad, after removing some index missing cases, a total of 112 rock burst cases are collected as training and test samples of intelligent prediction model. The rock burst main control factors: stress coefficient σ θ / σ c , brittleness coefficient σ c / σ t , elastic energy index W et and integrity coefficient K v are selected, and the rock burst grade is divided into four grades of no, weak (slight), moderate and strong rock burst (see Appendix Figure 1 ).

[0053] According to the analysis of rock burst intensity frequency in the data set (see Appendix Figure 1 ), there are 12 groups of grade I rock burst (no rock burst), accounting for 10.71% of the total samples, 29 groups of grade II rock burst (slight rock burst), accounting for 25.89% of the total samples, 48 groups of grade III rock burst (moderate rock burst), accounting for 42.86% of the total samples, and 23 groups of grade IV rock burst (moderate rock burst), accounting for 20.54% of the total samples. Considering that the grade III rock burst sample is the most, the data classification of various rock bursts is not balanced, and for small category data, the model training is used to avoid the problem of unbalanced data set processing.

[0054] Step S12, rock burst data set index analysis;

[0055] Specifically, in order to ensure the prediction effect of the model, statistical analysis is carried out on the four indexes in the rock burst data set. The basic information of data sample index is shown in Table 1.

[0056] Table 1

[0057]

[0058] As can be seen from Table 1, the skewness of integrity coefficient K v is 2.869, the kurtosis is 6.682, and it presents a stable right bias trend with obvious long tail on the right; the skewness of stress coefficient σ θ / σ c The skewness is 1.643 and the kurtosis is 6.215, showing a stable right-skewed trend with a long tail on the right; the brittleness coefficient σ c / σ t The skewness is 1.632 and the kurtosis is 4.161, which is similar to the stress coefficient and shows a gentle right-skewed trend with an obvious long tail on the right. The elastic energy index W et The skewness is 0.690 and the kurtosis is 1.344<3, showing a steep right-skewed trend with a long tail on the right.

[0059] Combined with the prediction samples, the frequency histograms of the four indicators are further obtained, which can more intuitively identify the distribution characteristics of each indicator data (see Appendix Figure 2 ). Figure 2 It can be seen that the integrity coefficient K v , stress coefficient σ θ / σ c The frequency distribution curves of indicators such as and are all close to normal distribution, while some indicators in a small number of cases have individual abnormal values, which is also consistent with the randomness and particularity of rockburst disasters. Therefore, the collected rockburst data set can be used as a rockburst prediction sample to construct a training set and test set for rockburst intensity prediction. In view of this, we further analyze the abnormalities of each indicator data and draw box plots of different indicator data according to different rockburst levels (I to IV) (see Appendix Figure 3 ). Figure 3 It can be seen that each sample data has relatively obvious outliers at different rockburst levels, and the brittleness coefficient σ c / σ t The degree of dispersion is the largest; with increasing rockburst intensity, the values ​​of all indicators, except the brittleness coefficient, show a gradually increasing trend. To preserve the outlier characteristics of rockburst data samples and consider the particularities of the surrounding rock mass during rockbursts in actual projects, this paper does not delete these outliers. Instead, it retains all samples with some outliers as a dataset to establish an intelligent rockburst prediction risk assessment model.

[0060] Step S13: Correlation analysis of rockburst data sets.

[0061] Specifically, the correlation between the various indicators has a greater impact on the rockburst risk prediction model. A high correlation may affect the precision and accuracy of the intelligent prediction. Therefore, combined with the 112 groups of rockburst samples in the data set, the correlation between the four indicators selected by this invention is analyzed to further explore the rationality and scientificity of the rockburst indicators. Based on this, a rockburst prediction indicator correlation matrix (Pearson correlation coefficient) is established, see Appendix Figure 4 .

[0062] Pearson correlation coefficient (Pearson) can well measure the correlation degree of two random variables, which is a value between-1 and 1, when the linear relationship between two variables is enhanced, the correlation coefficient tends to 1 or-1; when one variable increases, the other variable also increases, indicating that they are positively correlated, the correlation coefficient is greater than 0; on the contrary, it is negative correlation, and the correlation coefficient is less than 0; if the correlation coefficient is equal to 0, it indicates that there is no linear correlation between them.

[0063] By the attached Figure 4 It can be seen that the absolute value of the correlation coefficient between each characteristic parameter is generally small, all below 0.5; the maximum value is only-0.41 of the correlation coefficient between the integrity coefficient and the elastic energy index. According to the description of the correlation coefficient in the existing literature, only when the Pearson correlation coefficient is greater than 0.6, the correlation between two indexes is strong. Therefore, the correlation between each index considered in the application is small, which can be used as an evaluation index, and provides a basis for the intelligent prediction model of rock burst. The application selects the four indexes to classify and predict the rock burst intensity grade. After preparing the required rock burst data according to the above content, the input data is normalized to improve the accuracy and efficiency of modeling and eliminate the interference of data dimension and size.

[0064] Step S2, based on the hunter prey optimization algorithm and the support vector machine algorithm, a rock burst prediction model is established, and the rock burst prediction model is trained based on the training set, to obtain a trained rock burst prediction model;

[0065] In step S2, based on support vector machine (SVM) and hunter prey optimization algorithm (HPO), a multi-factor comprehensive integrated rock burst risk intelligent prediction model is established, the data samples are trained and tested, the optimization ability of support vector machine is enhanced, and the prediction accuracy is improved.

[0066] Support vector machine (SVM) is a machine learning method, which is usually used as a classifier in industry, and is also helpful for the development and application of deep learning algorithm. The earliest academician Feng Xiaoting proposed a prediction method based on support vector machine for rock burst problems of tunnel, VCR stope and carbonized stope. The method is scientific and feasible, and has a wide application prospect. At present, SVM has developed rapidly in many rock burst research fields such as classification prediction, regression prediction and time series prediction; as a typical binary classification model, SVM separates positive and negative planes through hyperplane, and introduces linear classification criterion with maximum interval, and gives nonlinear ability to linearizer by kernel trick (nonlinear mapping).

[0067] Traditional neural network prediction requires a large number of data samples, and the existing rock burst case database is difficult to achieve. SVM support vector machine is used for prediction of limited sample data, with strong adaptability, strong generalization, short training time, and small probability of being trapped in local search. However, due to the limitations of the algorithm itself, parameter optimization is prone to local optimization, and global optimization cannot be achieved. Therefore, it is necessary to find other optimization algorithms for search to jump out of local optimization in order to improve the efficiency and accuracy of evaluation.

[0068] Hunter-Prey Optimization (HPO) is a new parameter optimization algorithm proposed by Naruei et al. in 2022. The idea of this algorithm comes from the behavior of predators hunting prey in a group in nature. The basic assumption is that hunters may choose prey far from the group when searching for prey; when hunters chase prey, they will escape the attack of predators and reach a safe place. This hunting process is accompanied by the update of the position of the prey by the hunter. Finally, the optimal position of the prey is found, and the search process is completed. HPO has the advantages of fast convergence speed and strong optimization ability. The implementation steps of the algorithm are as follows:

[0069] (1) Initialize the population position, randomly generate the hunter or prey in the space, as shown in equation (1):

[0070] x n =rand(1,d).*(u-l)+l (1)

[0071] where x n represents the position of the population member, d represents the dimension of the problem, and u and l represent the upper and lower bounds of the problem variable, respectively.

[0072] (2) Introduce the hunter search mechanism, and the hunter selects the prey far from the group as the hunting target, as shown in equation (2).

[0073] x i,j (t+1)=x i,j (t)+0.5[(2WZP pos(j) -x i,j (t))+(2(1-W)Zτ (j) -x i,j (t))] (2)

[0074] where x i,j (t) represents the current position of the hunter, x i,j (t+1) represents the next position of the hunter, P pos(j) represents the position of the target prey, τ (j) represents the average position between the hunter and the prey, and Z is an adaptive parameter, and W is a balance parameter in the algorithm exploration, and its calculation process is shown in equation (3).

[0075]

[0076] Wherein, t represents the number of iterations, t max represents the maximum number of iterations.

[0077] (3) Introducing prey escape mechanism. The prey moves to the global optimal position, avoiding the search of the hunter, and escaping from hunting, as shown in equation (4).

[0078] x i,j (t+1)=WZcos(2πR)(T pos(j) -x i,j (t)) (4)

[0079] Wherein, T pos(j) is the target position of the prey (global optimal position), and R is a random number in the range of [-1, 1].

[0080] Combining equation (2) and equation (4) obtains the selection mechanism of the algorithm to the predatory behavior, sets the adjustment parameter δ, compares the random number R1 between [0, 1] and δ; if R1 is less than δ, the hunter search mechanism is executed. Otherwise, the prey escape mechanism is triggered until the condition is met and the optimization is completed. The present application sets the adjustment parameter δ to 0.1. The schematic diagram of the hunter searching for the position of the prey and the escape process of the prey is shown in FIG. 1. Figure 5 .

[0081] The basic idea of the HPO-SVM prediction model is to use the HPO algorithm to optimize the penalty factor c and the kernel parameter g of the kernel function in the SVM, and assign the optimal position c and g values after multiple iterations to the SVM, train and test the constructed data samples, enhance the optimization ability of the support vector machine, and improve the prediction accuracy. The technical route of the rock burst prediction model based on HPO-SVM is shown in FIG. 2, and the specific modeling process is as follows: Figure 6

[0082] (1) Data set division. Machine learning generally divides data into two parts of training set and test set. First, 1 to 112 groups of disorder are generated, which are used to randomly extract the training set and test set, and the training set and test set are divided according to the ratio of 8:2.

[0083] (2) Data input. The mapminmax function in the MTALAB 2016b software is used to normalize the data set, and the normalized data is determined to be in the range (0, 1) interval.

[0084] (3) Selection of kernel function. The mapping function, i.e. the kernel function, is introduced. The Gaussian radial basis kernel function is selected.

[0085] (4) Cross-validation. The 5-fold cross-validation model is adopted.​

[0086] (5) HPO optimization. The penalty factor c in SVM is set to minimize the objective function, and the size of the function parameter g determines the degree of roughness of SVM classification. The optimal c and g are searched by HPO, and the range of c and g is set to (0.1, 5000); the maximum iteration number is 50; the initial population number is 30.

[0087] (6) Assign the optimized parameters to the model.

[0088] (7) Train and test the model, identify the prediction data and de-normalize.

[0089] Step S3, select a plurality of evaluation indexes of the rockburst prediction model, input the test set into the trained rockburst prediction model to obtain a predicted value, and compare and analyze the actual value and the predicted value based on the evaluation indexes.

[0090] In step S3, root mean squared error (RMSE), mean absolute percentage error (MAPE), and goodness of fit (R 2 ) are used as evaluation indexes of the HPO-SVM model to compare and analyze the actual value and the evaluation value of the training set and the test set.

[0091] The model uses root mean squared error (RMSE), mean absolute percentage error (MAPE), and goodness of fit (R 2 ) as evaluation indexes of the HPO-SVM model. The calculation formulas of RMSE, MAPE, and R 2 are as follows:

[0092]

[0093]

[0094] Wherein, N is the number of data samples; y i is the rockburst grade prediction value; is the rockburst grade measured value; is the rockburst grade measured average value. The smaller the values of RMSE and MAPE are, the better.

[0095] In classification prediction, the primary criterion for model prediction is accuracy (precision), which is the ratio of the number of samples with accurate predictions to the total sample size, and the ratio of the number of samples with incorrect predictions to the total sample size as the error rate (prediction error). Output the actual values ​​of the training set and the test set and compare them with the evaluation values. The evaluation results of HPO-SVM are compared (see Appendix). Figure 7 After the samples to be evaluated are input into the program, the rockburst evaluation results for Level I, Level II, Level III, and Level IV are output as labels 1, 2, 3, and 4, respectively. The evaluation metrics for the training and test sets are shown in Table 2.

[0096] Table 2

[0097]

[0098] By the attached Figure 7 As shown in Table 2, the prediction accuracy of the HPO-SVM rockburst prediction model was tested using 23 groups of rockburst prediction samples. When the training set data was 90 groups and the test set data was 22 groups, the training set and test set fitting goodness of fit R of the HPO rockburst prediction model was 0. 2 The results are 0.98715 and 0.94683 respectively, the RMSE are 0.106 and 0.20851 respectively, the MAPE are 0.5618% and 1.087% respectively, and the accuracy rates are 98.88% and 95.65% respectively.

[0099] Step S4: Select a rockburst engineering example as a validation set, and use the rockburst prediction model to predict the rockburst intensity level of the rockburst engineering example.

[0100] In step S4, to further verify the applicability, effectiveness, and accuracy of the HPO-SVM model in predicting rockburst, 15 typical rockburst engineering cases were selected as a validation set, and the HPO-SVM model was used to predict the rockburst intensity levels of the above engineering cases.

[0101] Step S4 specifically includes:

[0102] Step S41: Typical rockburst engineering case index data and intensity levels;

[0103] Specifically, to further verify the applicability, effectiveness, and accuracy of the HPO-SVM model in predicting rockbursts, 15 typical rockburst engineering cases were selected as the validation set. The HPO-SVM model was used to predict the rockburst intensity levels of these engineering cases. The detailed data and the intensity levels of the on-site rockbursts are shown in Table 3.

[0104] Step S42: Intelligent prediction of rockburst risk in deep underground engineering based on HPO-SVM.

[0105] Specifically, to better verify the effectiveness and engineering applicability of the proposed prediction model in distinguishing the intensity of rockburst, the HPO-SVM rockburst risk intelligent evaluation model is used to predict 15 rockburst engineering examples, and a typical empirical criterion is selected for comparative analysis, and the accuracy of the rockburst verification cases is shown in Table 4.

[0106] As shown in Table 4, the HPO-SVM model proposed in the present application predicts the intensity of rockburst of 15 test data, and the correct accuracy is 86.67%, only 2 are misjudged (i.e. the difference between moderate rockburst and intense rockburst), and the accuracy of the prediction model is not as good as that of Shang Yankun's rockburst criterion and Zhou's energy criterion.

[0107] In summary, the prediction results of the rockburst risk evaluation model proposed in the present application are more consistent with the actual cases, and have good accuracy and engineering applicability.

[0108] Table 3

[0109]

[0110]

[0111] Table 4

[0112]

[0113] Compared with the prior art, the present application has the following advantages:

[0114] The present embodiment provides a rockburst grade prediction method based on HPO-SVM model, selects 112 typical rockburst engineering case data, establishes an intelligent rockburst prediction model based on machine learning method (SVM) and combined with a new parameter optimization algorithm (HPO). The machine learning method used in the present embodiment has the advantages of strong adaptability, strong generalization, short training time, small probability of being trapped in local search, etc. In addition, the new parameter optimization algorithm can achieve global optimization, and has the advantages of fast convergence speed and strong optimization ability.

[0115] At the same time, with the help of Pearson correlation coefficient principle, it is found that there is no strong correlation between the rockburst indexes. The existing classical rockburst prediction method and the rockburst grade prediction method proposed in the present embodiment are used to verify the accuracy and applicability of some typical rockburst engineering examples at home and abroad, and compared with the actual rockburst intensity grade, it is found that the prediction results of the rockburst risk evaluation model are more consistent with the actual cases, and the accuracy is higher, which shows that the prediction method in the present embodiment has certain applicability in evaluating the rockburst damage degree and rockburst grade, and provides certain reference basis for solving the rockburst prediction and evaluation problem of underground engineering.

[0116] Example 2

[0117] Based on the same inventive concept, the embodiment also provides a rock burst grade prediction system based on an HPO-SVM model, comprising:

[0118] a data set construction module, configured to construct a sample data set of rock burst cases, divide the sample data set into a training set and a test set, and determine rock burst main control factors and rock burst grades;

[0119] a model construction module, configured to establish a rock burst prediction model based on a hunter-prey optimization algorithm and a support vector machine algorithm, and train the rock burst prediction model based on the training set to obtain a trained rock burst prediction model;

[0120] a prediction evaluation module, configured to select a plurality of evaluation indexes of the rock burst prediction model, input the test set into the trained rock burst prediction model to obtain a predicted value, and compare and analyze an actual value and the predicted value based on the evaluation indexes;

[0121] a prediction verification module, configured to select a rock burst engineering example as a verification set, and use the rock burst prediction model to predict a rock burst intensity grade of the rock burst engineering example.

[0122] The rock burst grade prediction system based on the HPO-SVM model provided in the embodiment has all the advantages of the rock burst grade prediction method based on the HPO-SVM model provided in the first embodiment.

[0123] Embodiment three

[0124] The embodiment also discloses a computer device, comprising a memory, a processor and a computer program stored in the memory, and the processor executes the computer program to implement the steps of the method in the first embodiment.

[0125] Embodiment four

[0126] The embodiment also discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the method in the first embodiment.

[0127] Embodiment five

[0128] The embodiment also discloses a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the steps of the method in the first embodiment.

[0129] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A rockburst grade prediction method based on the HPO-SVM model, characterized in that: The following steps are involved: Constructing and analyzing a sample data set of rockburst cases, dividing the sample data set into a training set and a test set, and determining the main controlling factors and rockburst levels of rockburst; Establishing a rockburst prediction model based on a hunter-prey optimization algorithm and a support vector machine algorithm, and training the rockburst prediction model based on the training set to obtain a trained rockburst prediction model; selecting a plurality of evaluation indicators of the rockburst prediction model, inputting the test set into the trained rockburst prediction model to obtain predicted values, and comparing and analyzing actual values ​​with predicted values ​​based on the evaluation indicators; A rock burst engineering example is selected as a validation set, and the rock burst prediction model is used to predict the rock burst intensity level of the rock burst engineering example.

2. The method according to claim 1, characterized in that The main controlling factors of rock burst include stress coefficient, brittleness coefficient, elastic energy index and integrity coefficient; The rockburst levels include no rockburst, weak rockburst, moderate rockburst and strong rockburst.

3. The method according to claim 1, characterized in that It also includes collecting statistics on the main controlling factors of rock burst and analyzing the correlation between the main controlling factors of rock burst.

4. The method according to claim 1, wherein The process of establishing a rockburst prediction model includes: The hunter-prey optimization algorithm is used to optimize the penalty factor and kernel parameters of the kernel function in the support vector machine algorithm; The penalty factors and kernel parameters corresponding to the optimal positions after multiple iterations are assigned to the support vector machine algorithm, and the sample data sets are trained and tested.

5. The method according to claim 4, characterized in that The process of optimizing the penalty factor and kernel parameters of the kernel function in the support vector machine algorithm using the hunter-prey optimization algorithm includes: The hunter-prey optimization algorithm initializes the population position; Hunter or prey in randomly generated space Introducing hunter search mechanism and prey escape mechanism Optimize the location where hunters search for prey and the process of prey escaping.

6. The method according to claim 1, characterized in that The evaluation indicators include root mean square error, mean absolute percentage error and goodness of fit.

7. A rockburst grade prediction system based on the HPO-SVM model, characterized in that: include: A data set construction module is used to construct and analyze a sample data set of rock burst cases, divide the sample data set into a training set and a test set, and determine the main controlling factors and rock burst levels of the rock burst; A model building module is used to establish a rockburst prediction model based on a hunter-prey optimization algorithm and a support vector machine algorithm, and train the rockburst prediction model based on the training set to obtain a trained rockburst prediction model; A prediction and evaluation module is used to select multiple evaluation indicators of the rockburst prediction model, input the test set into the trained rockburst prediction model to obtain a predicted value, and compare and analyze the actual value with the predicted value based on the evaluation indicators; The prediction and verification module selects a rock burst engineering example as a verification set and uses the rock burst prediction model to predict the rock burst intensity level of the rock burst engineering example.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory, wherein: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images