Rock compressive strength prediction system and method based on grey correlation analysis and GWO-BILSTM

Abstract

The invention discloses a rock compressive strength prediction system and method based on grey correlation analysis and GWO-BILSTM, and the system is characterized in that a data collection unit is used for collecting Schmidt rebound index, Leeb hardness, Rockwell hardness, longitudinal wave velocity, density and uniaxial compressive strength measured values, and taking the above factors as original data; a gray correlation analysis module, a model optimization module and a compressive strength prediction module are arranged in the data processing unit, and the gray correlation analysis module is used for performing normalization processing on data and calculating a gray correlation degree; the model optimization module is used for optimizing the compressive strength prediction module; the compressive strength prediction module is used for predicting the compressive strength of the rock and outputting a result to the visual output module; the method comprises: collecting data; correlation degree analysis and input feature screening; establishing a rock compressive strength prediction model based on a bidirectional long-short-term memory network, and performing optimization by using a GWO algorithm; and carrying out actual prediction. The system and the method can realize efficient and accurate prediction of the compressive strength of the rock.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent prediction of rock mechanical parameters, and specifically relates to a rock compressive strength prediction system and method based on grey relational analysis and GWO-BILSTM. Background Technique

[0002] The uniaxial compressive strength of rock is an important parameter of rock and is an essential basic parameter for tunnel excavation scheme selection, tunnel layout and support, and prevention and control of tunnel water inrush disasters. In traditional technologies, the uniaxial compressive strength of rock is generally obtained by means of laboratory uniaxial compressive strength tests, which generally require obtaining the rock strength at the construction site. Usually, it needs to go through five major steps: core drilling of rock at the construction site, rock transportation, processing and grinding of standard parts, testing by a testing machine, and data processing. However, the sampling process of rock is relatively difficult, and the cost of obtaining standard specimens is relatively high. At the same time, the determination time of the uniaxial compressive strength of rock is long, and it is difficult to meet the requirements of the construction site for real-time acquisition of the uniaxial compressive strength of rock. When using existing machine learning algorithms to predict the uniaxial compressive strength of rock, although the accuracy of the uniaxial compressive strength (UCS) prediction is improved to a certain extent, due to the nature of intelligent algorithms and the limitations of search strategies, sometimes it will fall into the local optimum situation and cannot reach the global optimum. In order to reduce the sampling difficulty, reduce the cost of specimens, and effectively shorten the determination time of the compressive strength to meet the requirements of the construction site for rock strength testing, it is urgent to provide a rock compressive strength prediction system and method based on grey relational analysis and GWO-BILSTM. Summary of the Invention

[0003] Aiming at the problems existing in the above-mentioned prior art, the invention provides a rock compressive strength prediction system and method based on grey relational analysis and GWO-BILSTM. The system has a simple structure and high intelligence. It can effectively eliminate redundant data, can realize the efficient and accurate prediction of rock compressive strength, can effectively reduce the sampling difficulty and cost. At the same time, it can greatly shorten the determination time of the compressive strength and can meet the requirements of the construction site for rock strength testing. The method has a simple implementation process and high intelligence. It can realize the efficient and accurate prediction of the uniaxial compressive strength of rock in coal mine working faces, can effectively solve the problems of low efficiency of traditional laboratory testing methods and easy falling into local optimum of traditional machine learning models, can significantly improve the accuracy of rock strength prediction and the engineering application efficiency, and is applicable to coal mine mining safety assessment and support design.

[0004] In order to achieve the above object, the invention provides a rock compressive strength prediction system based on grey relational analysis and GWO-BILSTM, including a data acquisition unit, a data processing unit and a visualization output unit;

[0005] The data acquisition unit is used to collect the measured values of Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity, density and uniaxial compressive strength of rock samples, and take the above six factors as the original data;

[0006] The data processing unit is built-in with a grey relational analysis module, a model optimization module and a compressive strength prediction module. The grey relational analysis module is used to normalize the original data, calculate the grey relational degree between each factor and UCS, and then screen out the factors with a relational degree ≥ 0.6 as input data and send them to the compressive strength prediction module; the model optimization module is used to optimize the prediction accuracy of the compressive strength prediction module; the compressive strength prediction module is used to input the input data into the built-in prediction model to predict the rock compressive strength, and output the prediction result to the visualization output module;

[0007] The visualization output unit is used to display the prediction result in real time.

[0008] As a preference, the data acquisition unit includes a Schmidt rebound instrument, a Richter hardness tester, a Rockwell hardness tester, an ultrasonic detector, a density measurement component and a uniaxial compression testing machine. The Schmidt rebound instrument is used to collect the Schmidt rebound index of rock samples, the Richter hardness tester is used to collect the Richter hardness of rock samples, the Rockwell hardness tester is used to collect the Rockwell hardness of rock samples, the ultrasonic detector is used to collect the longitudinal wave velocity of rock samples, the density measurement component includes a balance and a vernier caliper and is used to collect the density of rock samples, and the uniaxial compression testing machine is used to collect the measured value of the uniaxial compressive strength of rock samples.

[0009] Furthermore, in order to facilitate the intuitive observation of the prediction result and improve the user's interactive experience at the same time, the visualization output unit is a touch screen display with an interactive graphical interface.

[0010] In the present invention, through the setting of the data acquisition unit, it is convenient to obtain various factor data as the original data. By building a grey relational analysis module in the data processing unit, the grey relational degree of each factor can be calculated efficiently and accurately. In this way, redundant features can be eliminated, which is beneficial to reducing the computational amount in the subsequent prediction process and avoiding the adverse impact of redundant features on the prediction result. By building a compressive strength prediction module in the data processing unit, it is convenient to use the factors screened by the grey relational analysis module as input data for efficient and accurate prediction of the compressive strength. By building a model optimization module in the data processing unit, it is convenient to optimize the compressive strength prediction module in real time during use, so as to effectively ensure the prediction accuracy of the compressive strength prediction module. Through the setting of the visualization output unit, it is convenient to display the prediction result in real time, and thus the application convenience and engineering practicability of the system can be effectively improved.

[0011] The system has a simple structure and a high degree of intelligence. It can effectively eliminate redundant data, achieve efficient and accurate prediction of the compressive strength of rocks, effectively reduce the sampling difficulty and cost. At the same time, it can greatly shorten the determination time of the compressive strength, meet the requirements of rock strength testing at the construction site, and is suitable for large-scale popularization and application.

[0012] The present invention also provides a method for predicting the compressive strength of rocks based on grey relational analysis and GWO-BILSTM. A system for predicting the compressive strength of rocks based on grey relational analysis and GWO-BILSTM is adopted, including the following steps:

[0013] Step 1: Prepare various rock specimens according to the set standards, and respectively collect six factors including the Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity, density, and measured value of uniaxial compressive strength as a group of original data samples, and construct several groups of data samples as the original data set;

[0014] Step 2: Perform standardization processing on the original data set to obtain standardized data, and calculate the grey relational degrees between the five factors of Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity, and density and the measured value of uniaxial compressive strength. Carry out correlation analysis on each factor through the grey relational analysis method, clarify the influence degree of each factor on the measured value of uniaxial compressive strength, and arrange them according to this quantified relationship. Select the factors with a correlation degree ≥ 0.6, and find out the main factors affecting the uniaxial compressive strength of rocks as input features, and use the corresponding measured value of uniaxial compressive strength as the output feature. Take the input feature and the corresponding output feature as a group of standard data samples, construct several groups of standard data samples, and then divide the several groups of standard data samples into a training set and a test set;

[0015] Step 3: Establish a prediction model for the compressive strength of rocks based on a bidirectional long short-term memory network, and use the GWO algorithm to globally optimize the hyperparameters of the prediction model for the compressive strength of rocks to obtain a prediction model for the compressive strength of rocks optimized by the GWO algorithm; Use the training set to train the prediction model for the compressive strength of rocks optimized by the GWO algorithm to obtain a prediction model for the compressive strength of rocks, and use the test set to test the performance of the prediction model for the compressive strength of rocks, and finally obtain a prediction model for the compressive strength of rocks with excellent performance;

[0016] Step 4: Collect the main factors with a correlation degree ≥ 0.6 in Step 2 and affecting the uniaxial compressive strength of rocks as input data, input them into the prediction model for the compressive strength of rocks, use the prediction model for the compressive strength of rocks to make a prediction, and output the predicted value of the uniaxial compressive strength to the visual output unit, and perform real-time display of the predicted value of the uniaxial compressive strength through the visual output unit.

[0017] Further, to ensure the prediction accuracy of the prediction model, in Step 1, multiple rock specimens including sandstone specimens, limestone specimens, mudstone specimens, and coal specimens are used to obtain the Schmidt rebound index through the Schmidt rebound test, the Richter hardness through the Richter hardness measurement test, the Rockwell hardness through the Rockwell hardness measurement test, the longitudinal wave velocity through the rock ultrasonic test, the density through the density measurement test, and the measured value of the uniaxial compressive strength through the rock uniaxial compression test.

[0018] Further, to more effectively eliminate redundant data and, at the same time, to effectively ensure the prediction accuracy, in Step 2, the process of calculating the grey relational grade is as follows:

[0019] S21: Set the reference sequence X0, X0 = (x0(1), x0(2), … x0(m));

[0020] S22: Set the comparison sequence X1, where n is the number of factor indicators and m is the number of factor objects;

[0021] S23: Standardize the data using the range normalization formula in Equation (1) or the mean-standard deviation normalization formula in Equation (2) to obtain the standardized data X1′,

[0022]

[0023] In the formula, x is the original data, i is the factor object, k is the factor indicator, max(x) and min(x) are respectively the maximum and minimum values of the data in the original data sequence; x i (k) is the kth value of the comparison sequence; x i ′(k) is the kth value of the standardized data; is the mean;

[0024] S24: Calculate the grey relational grade between the reference sequence and each comparison sequence;

[0025] S24-1: Calculate the difference value Δ i (k) between the reference sequence and each comparison sequence using Equation (3);

[0026] Δ i (k) = |x0(k) - x i (k)| (3);

[0027] In the formula, x0(k) is the kth value of the reference sequence;

[0028] S24-2: Calculate the minimum value Δ min and the maximum value Δmax ;

[0029] Δ min = min i,k Δ i (k)(4);

[0030] Δ max = max i,k Δ i (k)(5);

[0031] S24 - 3: Calculate the grey relational coefficient ξ i (k) as shown in formula (6);

[0032]

[0033] In the formula, ρ is the discrimination coefficient, and its value ranges from 0.5 to 1;

[0034] S24 - 4: Use formula (7) to calculate the average value of the grey relational coefficients of the sub - factors of all mother factors to obtain the grey relational degree r i ;

[0035]

[0036] S24 - 5: Sort according to the magnitudes of the calculated grey relational degrees. If r1 > r2, it means that the comparison sequence r1 has a higher correlation degree with the reference sequence r0 compared to the comparison sequence r2. If the grey relational degree of r1 is the largest, it represents that the influencing factor corresponding to r1 plays a dominant role among all influencing factors.

[0037] Furthermore, in order to effectively combine the GWO algorithm and the bidirectional long - short - term memory network so that the prediction model can have more accurate prediction accuracy, in step three, the process of globally optimizing the hyperparameters of the rock compressive strength prediction model using the GWO algorithm is as follows:

[0038] S31: Initialize the GWO algorithm, set the wolf population size, maximum number of iterations, and optimization dimension;

[0039] S32: Set the search range of the optimization parameters required by the rock compressive strength prediction model and generate the initial positions of the grey wolf population;

[0040] S33: Calculate the fitness value of each individual in the population one by one, and use the root - mean - square error as the evaluation index to measure the effect of the current parameter combination;

[0041] S34: Sort the individuals according to the fitness values, define the optimal solution as Alpha wolf, the sub - optimal solution and the third - optimal solution as Beta wolf and Delta wolf respectively;

[0042] S35: Guide the position update of other individuals according to formula (8) by means of the position information of Alpha, Beta, and Delta wolves;

[0043] X αi,k = X α,k - A1|C1X α,k - X i,k |(8);

[0044] In the formula, X αi,k represents the new position of gray wolf individual i; X α,k represents the position of Alpha wolf; A1 represents the direction and step coefficient; C1 represents the random perturbation coefficient; X i,k represents the current position of gray wolf individual i;

[0045] S36: Repeat S33 to S35 until the set number of iterations is reached, and finally obtain the optimal parameter combination.

[0046] Furthermore, in order to obtain the optimal parameter combination, in S31 of step three, the population size of the wolf pack is set to 20, the maximum number of iterations is set to 50, and the optimization dimension is set to 3.

[0047] Furthermore, in order to obtain the optimal parameter combination, in S33 of step three, calculate the fitness value F(x) according to formula (9);

[0048]

[0049] In the formula, k is the measured value of the uniaxial compressive strength; ni is the number of samples in the i-th fold validation set; is the predicted value of the j-th sample in the i-th fold of the model; y i,j is the true value of the j-th sample in the i-th fold.

[0050] Furthermore, in order to ensure the prediction accuracy, in step three, the bidirectional LSTM layer of the bidirectional long short-term memory network adopts a sequence-to-sequence structure, and the forward and backward LSTM hidden states are fused by concatenation, as shown in formula (10);

[0051]

[0052] In the formula, h t is the fusion output, is the forward LSTM output, is the backward LSTM output.

[0053] The present invention proposes a method for predicting the uniaxial compressive strength of rocks based on grey relational analysis and GWO-BILSTM. By screening out the key influencing factors of the uniaxial compressive strength (UCS) of rocks through grey relational analysis and removing redundant factors, the data dimension can be significantly reduced, which can effectively simplify the network structure of the rock compressive strength prediction model. At the same time, it is beneficial to significantly reduce the computing power, and thus the prediction results can be obtained more efficiently. Moreover, the removal of redundant factors can avoid the adverse effects of factors with small correlation degrees on the prediction results, further ensuring the prediction accuracy. By using the GWO algorithm to optimize the hyperparameters of the bidirectional long short-term memory network (BILSTM), a high-precision rock compressive strength prediction model can be constructed, which can significantly improve the prediction accuracy of the rock compressive strength prediction model.

[0054] Compared with the prior art, the present invention has the following advantages:

[0055] 1. High prediction accuracy: The intelligent prediction method for the uniaxial compressive strength of rocks proposed by the present invention, which combines the grey wolf optimization algorithm (GWO) and the bidirectional long short-term memory network (BILSTM), can effectively extract the non-linear relationships in multi-source rock parameters and achieve high-precision prediction. In the experiments of the present invention, the overall mean absolute percentage error (MAPE) of the model on rock samples such as coal samples, limestone, sandy mudstone, and sandstone is 14%, and the prediction errors of multiple samples are lower than 5%, showing good prediction stability and reliability, significantly superior to traditional methods, and having strong engineering application value.

[0056] 2. Strong model generalization ability and wide applicability to lithologies: The prediction system constructed by the present invention is trained and tested for various lithologies such as coal samples, sandstone, limestone, and sandy mudstone, and it is verified that it shows good prediction consistency under different lithologies, and has strong applicability and popularization.

[0057] 3. High computational efficiency and strong real-time performance: The present invention adopts an optimized neural network structure and combines the fast convergence characteristics of the grey wolf optimization algorithm to achieve rapid prediction of rock strength, meeting the requirements of on-site operations for real-time performance and efficiency.

[0058] 4. Can replace high-cost tests and reduce engineering risks: The method of the present invention can effectively reduce the preparation of a large number of samples and equipment operation required for traditional experiments, reduce the test cost and personnel risks, and is especially suitable for the evaluation of rock mechanical properties in complex and dangerous working environments such as coal mines.

[0059] The implementation process of this method is simple and the degree of intelligence is high. It can realize the efficient and accurate prediction of the uniaxial compressive strength of rock in the working face of coal mines, effectively solving the problems of low efficiency of traditional laboratory testing methods and traditional machine learning models that are prone to fall into local optimality. It significantly improves the accuracy of rock strength prediction and engineering application efficiency, and is suitable for coal mine mining safety assessment and support design. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of the method of the present invention;

[0061] Figure 2 It is a structural principle block diagram of the system in the present invention;

[0062] Figure 3 It is a schematic diagram of the interface of the visual output unit in the present invention. DETAILED DESCRIPTION

[0063] The present invention will be further described below in conjunction with the accompanying drawings.

[0064] like Figure 1 and Figure 2 As shown, the present invention provides a rock compressive strength prediction system based on grey correlation analysis and GWO-BILSTM (GWO is a grey wolf optimization algorithm, BILSTM is a rock compressive strength prediction model based on a bidirectional long short-term memory network), including a data acquisition unit, a data processing unit and a visualization output unit;

[0065] The data acquisition unit is used to collect the Schmidt rebound index (MPa), Leeb hardness (HL), Rockwell hardness (HR), longitudinal wave velocity (m / s), density (g / mm 3 ) and the measured value of uniaxial compressive strength (UCS) (MPa), and the above five factors are used as raw data; as a preferred embodiment, the raw data can be stored in a standardized Excel format;

[0066] The data processing unit has built-in grey correlation analysis module, model optimization module and compressive strength prediction module. The grey correlation analysis module is used to normalize the original data, and to calculate the grey correlation between each factor and UCS, and then screen out the factors with correlation ≥ 0.6 as input data to send to the compressive strength prediction module; the model optimization module is used to optimize the prediction accuracy of the compressive strength prediction module; the compressive strength prediction module is used to input the input data into the built-in prediction model to predict the rock compressive strength, and output the prediction results to the visualization output module;

[0067] The visualization output unit is used to display the prediction results in real time.

[0068] As a preference, the data acquisition unit includes a Schmidt rebound hammer, a Leeb hardness tester, a Rockwell hardness tester, an ultrasonic detector, a density measurement component, and a uniaxial compression testing machine. The Schmidt rebound hammer is used to collect the Schmidt rebound index of the rock sample. The Leeb hardness tester is used to collect the Leeb hardness of the rock sample. The Rockwell hardness tester is used to collect the Rockwell hardness of the rock sample. The ultrasonic detector is used to collect the longitudinal wave velocity of the rock sample. The density measurement component includes a balance and a vernier caliper, and is used to collect the density of the rock sample. The uniaxial compression testing machine is used to collect the measured value of the uniaxial compressive strength of the rock sample.

[0069] As a preference, the model of the integrated Schmidt rebound hammer is ZC3-A, the model of the ultrasonic detector is HC-U81, and the uniaxial compression testing machine is an RMT rock mechanics testing machine. The range of the Schmidt rebound index of the collected rock is 0 - 100 MPa, the range of the longitudinal wave velocity is 500 - 6000 m / s, the density range is 1.4 - 2.8 g / mm 3 and the UCS range is 10 - 120 MPa.

[0070] As Figure 3 shown, in order to facilitate the intuitive observation of the prediction results while improving the user's interactive experience, the visualization output unit is a touch screen display, which has an interactive graphical interface. It supports data input, user permission management, user registration and login, visualization of prediction results (it can also display the prediction comparison curve and error distribution of the training set and test set in real time, and monitor the RMSE (root mean square error) curve during the training process in real time. The results can be compared by means of a scatter plot + error histogram), batch data import and data export (supporting the export of prediction results in Excel / CSV format, which includes the predicted value of USC, error analysis, and confidence interval), etc. As a preference, the interactive graphical interface of the visualization output unit can be developed based on MATLAB GUI, which has a good human-computer interaction experience. It can directly input the original rock parameters, automatically complete the prediction and generate stress-strain curves and error analysis diagrams, and is easy to operate, suitable for use by engineering and technical personnel.

[0071] As a preference, the data processing unit is an industrial computer.

[0072] In the present invention, through the setting of the data acquisition unit, it is convenient to obtain various factor data as the original data. By building a grey relational analysis module into the data processing unit, the grey relational degrees of each factor can be calculated efficiently and accurately. In this way, redundant features can be removed, which is beneficial to reducing the computational amount in the subsequent prediction process and avoiding the adverse effects of redundant features on the prediction results. By building a compressive strength prediction module into the data processing unit, it is convenient to use the factors screened by the grey relational analysis module as input data for efficient and accurate prediction of the compressive strength. By building a model optimization module into the data processing unit, it is convenient to optimize the compressive strength prediction module in real time during use, thus effectively ensuring the prediction accuracy of the compressive strength prediction module. Through the setting of the visual output unit, it is convenient to display the prediction results in real time, thereby effectively improving the application convenience and engineering practicability of the system.

[0073] The system has a simple structure and a high degree of intelligence. It can effectively remove redundant data, achieve efficient and accurate prediction of the rock compressive strength, effectively reduce the sampling difficulty and cost, and greatly shorten the determination time of the compressive strength. It can meet the requirements of rock strength testing at the construction site and is suitable for large-scale popularization and application.

[0074] The present invention also provides a method for predicting the rock compressive strength based on grey relational analysis and GWO-BILSTM. By using a system for predicting the rock compressive strength based on grey relational analysis and GWO-BILSTM, the method includes the following steps:

[0075] Step 1: Prepare various rock specimens according to the set standards (preferably using the ISO 14577 standard), and respectively collect six factors including the Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity, density, and the measured value of uniaxial compressive strength as a group of original data samples, and construct several groups of data samples as the original data set;

[0076] Step 2: Perform standardization processing on the original data set to obtain the standardized data, and calculate the grey relational degrees of the five factors of the Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity, and density with the measured value of uniaxial compressive strength. Carry out the correlation analysis work on each factor through the grey relational analysis method, clarify the influence degree of each factor on the measured value of uniaxial compressive strength, and arrange them according to this quantified relationship. Select the factors with a correlation degree ≥ 0.6, and find out the main factors affecting the uniaxial compressive strength of the rock as the input features, and use the corresponding measured value of uniaxial compressive strength as the output feature. Take the input feature and the corresponding output feature as a group of standard data samples, construct several groups of standard data samples, and then divide the several groups of standard data samples into a training set and a test set; In this way, the good generalization ability of the model can be ensured;

[0077] Step 3: Establish a rock uniaxial compressive strength prediction model based on a bidirectional long short-term memory network (BILSTM), and use the GWO algorithm to globally optimize the hyperparameters of the rock uniaxial compressive strength prediction model to obtain a rock uniaxial compressive strength prediction model optimized by the GWO algorithm;

[0078] The network structure of the constructed rock uniaxial compressive strength prediction model includes an input layer (the number of nodes = the number of input features after screening. When the selected factors are Schmidt rebound index, longitudinal wave velocity, and density, the number of nodes = 3;), a bidirectional LSTM layer (the number of hidden nodes is determined by the GWO algorithm), a fully connected layer, and a regression output layer. The activation function is ReLU, and the loss function is mean square error (MSE);

[0079] During the global optimization process, the optimized parameters include the learning rate (search range: 0.001 - 0.1), the number of hidden layer nodes (10 - 100), and the L2 regularization coefficient (1e-5 - 1e-2). Among them, the fitness function is the root mean square error (RMSE) between the predicted value and the measured value;

[0080] Use the training set to train the rock uniaxial compressive strength prediction model optimized by the GWO algorithm to obtain a rock uniaxial compressive strength prediction model, and use the test set to test the performance of the rock uniaxial compressive strength prediction model. Finally, obtain a rock uniaxial compressive strength prediction model with excellent performance; During the training process, use the Adam optimizer to train the rock uniaxial compressive strength prediction model. The learning rate is determined by the GWO algorithm, the batch size is set to 16, and the maximum number of training epochs is 500. At the same time, during the training process, make MAE ≤ 6.68 MPa, RMSE ≤ 7.64 MPa, and R 2 ≥ 0.94 to obtain a rock uniaxial compressive strength prediction model with high accuracy, good stability, and excellent fitting effect; As a further optimization, use the early stopping method (when the validation loss does not decrease for 10 consecutive times) to terminate the training;

[0081] Step 4: Collect the factors with a correlation degree ≥ 0.6 and being the main factors affecting the rock uniaxial compressive strength in Step 2 as input data, input them into the rock uniaxial compressive strength prediction model, use the rock uniaxial compressive strength prediction model for prediction, and output the predicted value of the uniaxial compressive strength to the visualization output unit, and perform real-time display of the predicted value of the uniaxial compressive strength through the visualization output unit.

[0082] To ensure the prediction accuracy of the prediction model, in Step 1, a variety of rock specimens, including sandstone specimens, limestone specimens, mudstone specimens, and coal specimens, are used to obtain the Schmidt rebound index through the Schmidt rebound test, the Richter hardness through the Richter hardness measurement test, the Rockwell hardness through the Rockwell hardness measurement test, the longitudinal wave velocity through the rock ultrasonic test, the density through the density measurement test, and the measured value of the uniaxial compressive strength through the rock uniaxial compression test.

[0083] Preferably, each group of sample data contains at least 5 parallel specimen results.

[0084] In order to more effectively eliminate redundant data and, at the same time, to effectively ensure the prediction accuracy, in Step 2, the process of calculating the grey relational grade is as follows:

[0085] S21: Set the reference sequence X0, X0 = (x0(1), x0(2), … x0(m));

[0086] S22: Set the comparison sequence X1, where n is the number of factor indicators and m is the number of factor objects;

[0087] S23: Standardize the data using the range normalization formula in Equation (1) or the mean-standard deviation normalization formula in Equation (2) to obtain the standardized data X1′,

[0088]

[0089] In the formula, x is the original data, i is the factor object, k is the factor indicator, max(x) and min(x) are respectively the maximum and minimum values of the data in the original data sequence; x i (k) is the kth value of the comparison sequence; x i ′(k) is the kth value of the standardized data; is the mean;

[0090] Since the factor data of the research object is non-time series, range normalization is used to standardize the factor data.

[0091] S24: Calculate the grey relational grade between the reference sequence and each comparison sequence;

[0092] S24-1: Calculate the difference value Δ i (k) between the reference sequence and each comparison sequence using Equation (3);

[0093] Δ i (k) = |x0(k) - x i (k)| (3);

[0094] where \(x_0(k)\) is the \(k\)th value of the reference sequence;

[0095] S24 - 2: Calculate the minimum value \(\Delta\) and the maximum value \(\Delta\) among all the difference values respectively by using formula (4) and formula (5); min and the maximum value \(\Delta\) max ;

[0096] \(\Delta\) min = min i,k \(\Delta\) i (k) (4);

[0097] \(\Delta\) max = max i,k \(\Delta\) i (k) (5);

[0098] S24 - 3: Calculate the grey correlation coefficient \(\xi\) i (k) as shown in formula (6);

[0099]

[0100] where \(\rho\) is the resolution coefficient, with a value between 0.5 and 1, preferably 0.5;

[0101] S24 - 4: Calculate the average value of the grey correlation coefficients of the sub - factors of all the mother factors by using formula (7) to obtain the grey correlation degree \(r\) i ;

[0102]

[0103] S24 - 5: Sort according to the magnitudes of the calculated grey correlation degrees. If \(r_1>r_2\), it means that the comparison sequence \(r_1\) has a higher correlation degree with the reference sequence \(r_0\) compared to the comparison sequence \(r_2\). The influencing factor corresponding to \(r_1\) has a higher proportion of influence on the uniaxial compressive strength of the rock, which is greater than \(r_2\). If the magnitude of the grey correlation degree of \(r_1\) is the largest, it represents that the influencing factor corresponding to \(r_1\) plays a dominant role among all the influencing factors. This influencing factor plays an important role in the process of uniaxial compression of the rock.

[0104] In order to effectively combine the GWO algorithm and the bidirectional long - short - term memory network so that the prediction model can have more accurate prediction accuracy, in step three, the process of globally optimizing the hyper - parameters of the rock compressive strength prediction model by using the GWO algorithm is as follows:

[0105] S31: Initialize the GWO algorithm, set the wolf pack population size, the maximum number of iterations, and the optimization dimension. Among them, the wolf pack population size, the maximum number of iterations, and the optimization dimension respectively correspond to the learning rate, the number of hidden layer nodes, and the L2 regularization coefficient of the rock compressive strength prediction model;

[0106] S32: Set the search range of the optimization parameters required for the rock compressive strength prediction model and generate the initial positions of the grey wolf population;

[0107] S33: Calculate the fitness value of each individual in the population one by one, using the root mean square error as the evaluation index to measure the effect of the current parameter combination;

[0108] S34: Sort the individuals according to the fitness values, define the optimal solution as the Alpha wolf, and set the sub-optimal solution and the third-best solution as the Beta wolf and the Delta wolf respectively;

[0109] S35: With the help of the position information of the Alpha, Beta, and Delta wolves, update the positions of other individuals according to formula (8);

[0110] X αi,k = X α,k - A1|C1X α,k - X i,k | (8);

[0111] In the formula, X αi,k represents the new position of grey wolf individual i; X α,k represents the position of the Alpha wolf; A1 represents the direction and step coefficient; C1 represents the random perturbation coefficient; X i,k represents the current position of grey wolf individual i;

[0112] S36: Repeat S33 to S35 until the set number of iterations is reached, and finally obtain the optimal parameter combination.

[0113] In order to obtain the optimal parameter combination, in S31 of step three, the population size of the wolf pack is set to 20, the maximum number of iterations is set to 50, and the optimization dimension is set to 3.

[0114] In order to obtain the optimal parameter combination, in S33 of step three, calculate the fitness value F(x) according to formula (9);

[0115]

[0116] In the formula, k is the measured value of the uniaxial compressive strength; n i is the number of samples in the i-th fold validation set; is the predicted value of the j-th sample in the i-th fold of the model; y i,j is the true value of the j-th sample in the i-th fold.

[0117] To ensure the prediction accuracy, in step three, the bidirectional LSTM layer of the bidirectional long short-term memory network adopts a sequence-to-sequence structure, and the forward and backward LSTM hidden states are fused by concatenation, as shown in formula (10);

[0118]

[0119] In the formula, h t is the fused output, is the forward LSTM output, is the backward LSTM output.

[0120] The present invention proposes a method for predicting the uniaxial compressive strength of rocks based on grey relational analysis and GWO-BILSTM. It screens out the key influencing factors of the uniaxial compressive strength (UCS) of rocks through grey relational analysis and eliminates redundant factors, which can significantly reduce the data dimension, and then can effectively simplify the network structure of the rock compressive strength prediction model. At the same time, it is beneficial to significantly reduce the computing power. Therefore, the prediction result can be obtained more efficiently. Moreover, the elimination of redundant factors can avoid the adverse effects of factors with small correlation degrees on the prediction result, further ensuring the prediction accuracy; using the GWO algorithm to optimize the hyperparameters of the bidirectional long short-term memory network (BILSTM) can construct a high-precision rock compressive strength prediction model, and then can significantly improve the prediction accuracy of the rock compressive strength prediction model.

[0121] The implementation process of this method is simple and highly intelligent. It can realize the efficient and accurate prediction of the uniaxial compressive strength of rocks in coal mine working faces, effectively solve the problems of low efficiency of traditional laboratory testing methods and the tendency of traditional machine learning models to fall into local optima, significantly improve the accuracy of rock strength prediction and the engineering application efficiency, and is applicable to coal mine mining safety assessment and support design.

[0122] Example:

[0123] Predict the uniaxial compressive strength (UCS) of the rocks in Caojiatan Coal Mine:

[0124] 1. Data collection:

[0125] Types of rock specimens: 15 groups of sandstone, 15 groups of mudstone, 15 groups of coal samples, and 5 groups of limestone, a total of 50 groups. The training set is 40 groups and the test set is 10 groups.

[0126] Testing equipment: Schmidt rebound hammer (accuracy ±2MPa), Leeb hardness tester, Rockwell hardness tester, ultrasonic detector (frequency 50kHz), RMT testing machine (loading rate 0.02mm / s), density measurement component (balance and vernier caliper).

[0127] 2. Grey relational analysis:

[0128] Input factors: Schmidt rebound index, Leeb hardness, longitudinal wave velocity, density, Rockwell hardness.

[0129] Output result: The correlation degree ranking is Schmidt rebound index (0.762) > density (0.706) > longitudinal wave velocity (0.691) > Rockwell hardness (0.615). The first three items are selected as input features.

[0130] 3. GWO - BiLSTM optimization:

[0131] Optimal parameters: learning rate = 0.005, hidden nodes = 64, L2 = 1e - 4.

[0132] Training time: shortened by 30% compared with non - optimized BiLSTM (from 120s to 84s).

[0133] 4. Prediction performance:

[0134] Test set results: MAE = 6.68MPa (58% lower than BP neural network), RMSE = 7.64MPa, R 2 = 0.942.

[0135] Treatment of abnormal samples: For sandstone samples with UCS > 100MPa, the prediction error ≤ 8%.

[0136] 5. Application effect

[0137] 5.1. Precision improvement: MAE is reduced by 50% - 60% compared with traditional models (BP, SVM), and R 2 is increased to above 0.94.

[0138] The prediction stability for complex lithology (such as coal samples with developed fractures) is improved, and the error fluctuation range is reduced to ±10%.

[0139] 5.2. Efficiency optimization: The model training time is reduced by 30% after data dimensionality reduction, and the real - time prediction response time is < 1 second under GPU acceleration.

[0140] 5.3. Engineering value: In the application of the system in the 122109 working face of Caojiatan Coal Mine, it guides the optimization of support design, reduces the roof accident rate by 15%. The visual interface supports non - professional personnel operation, reducing the laboratory test cost by about 40%.

Claims

1. A rock compressive strength prediction system based on grey relational analysis and GWO-BILSTM, characterized in that, It includes a data acquisition unit, a data processing unit and a visualization output unit; The data acquisition unit is used to collect the measured values of Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity, density and uniaxial compressive strength of rock samples, and take the above six factors as the original data; The data processing unit is built-in with a grey relational analysis module, a model optimization module and a compressive strength prediction module. The grey relational analysis module is used to normalize the original data, calculate the grey relational degree between each factor and UCS, and then screen out the factors with a relational degree ≥ 0.6 as the input data and send them to the compressive strength prediction module; The model optimization module is used to optimize the prediction accuracy of the compressive strength prediction module; The compressive strength prediction module is used to input the input data into the built-in prediction model to predict the rock compressive strength, and output the prediction result to the visualization output module; The visualization output unit is used to display the prediction result in real time.

2. The rock compressive strength prediction system based on grey relational analysis and GWO-BILSTM according to claim 1, characterized in that, The data acquisition unit includes a Schmidt rebound instrument, a Richter hardness tester, a Rockwell hardness tester, an ultrasonic detector, a density measurement component and a uniaxial compression testing machine. The Schmidt rebound instrument is used to collect the Schmidt rebound index of rock samples, the Richter hardness tester is used to collect the Richter hardness of rock samples, the Rockwell hardness tester is used to collect the Rockwell hardness of rock samples, the ultrasonic detector is used to collect the longitudinal wave velocity of rock samples, the density measurement component includes a balance and a vernier caliper, which is used to collect the density of rock samples, and the uniaxial compression testing machine is used to collect the measured value of the uniaxial compressive strength of rock samples.

3. The rock compressive strength prediction system based on grey relational analysis and GWO-BILSTM according to claim 2, characterized in that, The visualization output unit is a touch screen display with an interactive graphical interface.

4. A method for predicting the compressive strength of rock based on grey relational analysis and GWO-BILSTM, which uses a system for predicting the compressive strength of rock based on grey relational analysis and GWO-BILSTM as described in any one of claims 1 to 3, is characterized in that, It includes the following steps: Step 1: Prepare a variety of rock specimens according to the set standards, and respectively collect six factors including the Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity, density and the measured value of uniaxial compressive strength as 1 group of original data samples, and construct several groups of data samples as the original data set; Step 2: Perform standardization processing on the original data set to obtain standardized data, calculate the grey relational degree between the five factors of Schmidt rebound index, Richter hardness, Rockwell hardness, longitudinal wave velocity and density and the measured value of uniaxial compressive strength, carry out the relational degree analysis work on each factor through the grey relational analysis method, clarify the influence degree of each factor on the measured value of uniaxial compressive strength, and arrange them according to this quantified relationship, screen out the factors with a relational degree ≥ 0.6, and find out the main factors affecting the uniaxial compressive strength of rocks as the input features, and take the corresponding measured value of uniaxial compressive strength as the output feature. Take the input feature and the corresponding output feature as 1 group of standard data samples, construct several groups of standard data samples, and then divide the several groups of standard data samples into a training set and a test set; Step 3: Establish a rock uniaxial compressive strength prediction model based on a bidirectional long short-term memory network, and use the GWO algorithm to globally optimize the hyperparameters of the rock uniaxial compressive strength prediction model to obtain a rock uniaxial compressive strength prediction model optimized by the GWO algorithm; use the training set to train the rock uniaxial compressive strength prediction model optimized by the GWO algorithm to obtain a rock uniaxial compressive strength prediction model, and use the test set to test the performance of the rock uniaxial compressive strength prediction model, and finally obtain a rock uniaxial compressive strength prediction model with excellent performance; Step 4: Collect the main factors with a gray relational degree ≥ 0.6 and affecting the uniaxial compressive strength of rocks in Step 2 as input data, input them into the rock uniaxial compressive strength prediction model, use the rock uniaxial compressive strength prediction model to make predictions, and output the predicted value of the uniaxial compressive strength to the visualization output unit, and display the predicted value of the uniaxial compressive strength in real time through the visualization output unit.

5. A rock compressive strength prediction method based on grey relational analysis and GWO-BILSTM according to claim 4, characterized in that, In Step 1, a variety of rock specimens include sandstone specimens, limestone specimens, mudstone specimens and coal specimens. The Schmidt rebound index is obtained through the Schmidt rebound test, the Richter hardness is obtained through the Richter hardness measurement test, the Rockwell hardness is obtained through the Rockwell hardness measurement test, the longitudinal wave velocity is obtained through the rock ultrasonic test, the density is obtained through the density measurement test, and the measured value of the uniaxial compressive strength is obtained through the rock uniaxial compression test.

6. A rock compressive strength prediction method based on grey relational analysis and GWO-BILSTM according to claim 5, characterized in that, In Step 2, the process of calculating the gray relational degree is as follows: S21: Set the reference sequence X0, X0 = (x0(1), x0(2), … x0(m)); S22: Set the comparison sequence X1, where n is the number of factor indicators, m is the number of factor objects; S23: Standardize the data using the range normalization formula in Equation (1) or the mean-standard deviation normalization formula in Equation (2) to obtain the standardized data X1'. Where x is the original data, i is the factor object, k is the factor index, max(x) and min(x) are respectively the maximum and minimum values of the data in the original data sequence; x i (k) is the k-th value of the comparison sequence; x i ′(k) is the k-th value of the data after normalization; is the mean value; S24: Calculate the gray relational degree between the reference sequence and each comparison sequence; S24-1: Calculate the difference value Δ(k) between the reference sequence and each comparison sequence using formula (3). i (k); Δ i (k) = |x0(k) - x i (k)| (3); In the formula, x0(k) is the k-th value of the reference sequence; S24-2: Calculate the minimum value Δ and the maximum value Δ among all the difference values by using formula (4) and formula (5) respectively. min and the maximum value Δ max ; Δ min = min i,k Δ i (k)(4); Δ max = max i,k Δ i (k)(5); S24-3: Calculate the grey correlation coefficient ξ i (k) as shown in formula (6); In the formula, ρ is the resolution coefficient, and its value ranges from 0.5 to 1; S24-4: Calculate the average value of the grey relational coefficients of the sub-factors of all parent factors using formula (7) to obtain the grey relational degree r i ; S24-5: Sort according to the calculated gray relational degree. If r1 > r2, it means that the comparison sequence r1 has a higher correlation degree with the reference sequence r0 than the comparison sequence r2. If the gray relational degree of r1 is the largest, it means that the influencing factor corresponding to r1 plays a dominant role among all influencing factors.

7. A method for predicting the compressive strength of rock based on grey relational analysis and GWO-BILSTM according to claim 6, characterized in that, In Step 3, the process of globally optimizing the hyperparameters of the rock uniaxial compressive strength prediction model using the GWO algorithm is as follows: S31: Initialize the GWO algorithm, and set the wolf population size, the maximum number of iterations and the optimization dimension; S32: Set the search range of the optimization parameters required by the rock uniaxial compressive strength prediction model, and generate the initial positions of the gray wolf population; S33: Calculate the fitness value of each individual in the population one by one, and use the root mean square error as the evaluation index to measure the effect of the current parameter combination; S34: Sort the individuals according to the fitness value, define the optimal solution as the Alpha wolf, and set the second-best solution and the third-best solution as the Beta wolf and the Delta wolf respectively; S35: With the help of the position information of the Alpha, Beta and Delta wolves, update the positions of other individuals according to formula (8); X αi,k = X α,k - A1|C1X α,k - X i,k | (8); In the formula, X αi,k represents the new position of the gray wolf individual i; X α,k represents the position of the Alpha wolf; A1 represents the direction and step coefficient; C1 represents the random perturbation coefficient; X i,k represents the current position of the gray wolf individual i; S36: Repeat S33 to S35 until the set number of iterations is reached, and finally obtain the optimal parameter combination.

8. A rock compressive strength prediction method based on grey relational analysis and GWO-BILSTM according to claim 7, characterized in that In S31 of Step 3, the population size of the wolf pack is set to 20, the maximum number of iterations is set to 50, and the optimization dimension is set to 3.

9. A rock compressive strength prediction method based on grey relational analysis and GWO-BILSTM according to claim 8, characterized in that, In S33 of Step 3, the fitness value F(x) is calculated according to formula (9); where k is the measured value of the uniaxial compressive strength; ni is the number of samples in the i-th fold validation set; is the predicted value of the j-th sample in the i-th fold of the model; y i,j is the true value of the j-th sample in the i-th fold.

10. A rock compressive strength prediction method based on grey relational analysis and GWO-BILSTM according to claim 9, characterized in that, In Step 3, the bidirectional LSTM layer of the bidirectional long short-term memory network adopts a sequence-to-sequence structure, and the forward and backward LSTM hidden states are fused by concatenation, as shown in formula (10); where h t is the fusion output, is the forward LSTM output, is the backward LSTM output.

Images