Rock uniaxial compressive strength prediction method of multi-algorithm fusion machine learning model

Abstract

The invention discloses a rock uniaxial compressive strength prediction method based on a multi-algorithm fusion machine learning model, and the method comprises the following steps: carrying out the calculation and analysis of a gray correlation degree analysis method, and obtaining a characteristic parameter which is higher in correlation with the rock uniaxial compressive strength, and taking the characteristic parameter as an input characteristic parameter of a rock uniaxial compressive strength prediction model; according to the method, a Newton-Raphson optimization algorithm (NRBO), a convolutional neural network (CNN) and a long short-term memory network (LSTM) are fused, and an NRBO-CNN-LSTM model for predicting the uniaxial compressive strength of rock is established. And dividing a training set and a prediction set, and carrying out training prediction on the NRBO-CNN-LSTM prediction model. According to the rock uniaxial compressive strength prediction method provided by the invention, the defects of low accuracy, large manpower and material resource consumption and the like of the existing empirical formula method and indoor experiment method can be avoided, and the acquisition of the rock uniaxial compressive strength is more efficient and convenient.

Description

Technical Field

[0001] The present invention relates to the field of mining technology, and in particular to a method for predicting the uniaxial compressive strength of rock by integrating a multi-algorithm fusion machine learning model. Background Art

[0002] During mineral mining, rock strength characteristics are crucial indicators for evaluating rock mass stability and assessing engineering safety risks. Conventional methods for determining rock mass strength characteristics include point load strength tests, uniaxial compression tests, and Brazilian splitting tests. However, as mining depth increases, the geological environment of a mine becomes increasingly complex. In these complex geological conditions, obtaining standard rock samples for conducting indoor rock mechanics tests to determine rock strength characteristics is often difficult. Poor rock quality, such as developed joints and fissures, rock fragmentation, and core cake formation, is often encountered. In these circumstances, obtaining intact rock samples for rock mechanics testing is difficult, making it difficult to obtain rock strength information. Furthermore, the long cycle and high cost of indoor testing can lead to untimely updates of mine rock characteristics, hindering construction progress. Rapid determination of rock mass uniaxial compressive strength is crucial for the construction and support design of mining and geotechnical engineering projects. Failure to update rock uniaxial compressive strength not only impacts project progress and economic benefits but, more seriously, poses a direct threat to construction safety and the safety of workers and property. Therefore, an efficient method for determining rock strength characteristics is of great significance to mining efficiency, underground construction safety, support system optimization, and ensuring the safety of workers.

[0003] Traditional methods for evaluating the uniaxial compressive strength of rock primarily include prediction methods based on classical uniform isotropic elastic wave theory and a series of prediction methods based on the acoustic response mechanism of rock mechanical parameters. However, the significant variability in the uniaxial compressive strength of complex rocks and the complex rock physical response patterns result in low prediction accuracy from traditional rock mechanical calculation methods, posing a significant challenge for evaluating geomechanical parameters in complex formations. Most previously proposed rock uniaxial compressive strength prediction models are single machine learning models, suffering from low accuracy and poor generalization. Convolutional neural networks (CNNs) have unique convolution kernels that can extract characteristic information from data and better identify the relationship between rock uniaxial compressive strength and other input characteristic parameters. Long short-term memory networks (LSTMs) have memory units that can be used to rapidly build network frameworks to identify patterns between parameters. The Newton-Raphson method (NRBO) offers the advantages of faster convergence and the ability to quickly approach the extreme value of the objective function. As an optimization algorithm, the learning rate, regularization coefficient, and number of hidden layer nodes in the prediction model are optimized to find the optimal parameters that best fit the prediction model. To this end, a rock uniaxial compressive strength prediction method based on a multi-algorithm fusion machine learning model of NRBO-CNN-LSTM was proposed. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the background technology and propose a method for predicting the uniaxial compressive strength of rock by integrating multiple algorithms with a machine learning model.

[0005] The present invention proposes a method for predicting the uniaxial compressive strength of rock based on a multi-algorithm fusion machine learning model, comprising the following steps:

[0006] Step 1: Statistically analyze the parameters used in the traditional rock uniaxial compressive strength model, preliminarily select parameters related to the rock uniaxial compressive strength, use the grey correlation analysis method to calculate the correlation between other physical parameters and the rock uniaxial compressive strength, determine the final machine learning prediction model input characteristic parameters, and establish a rock uniaxial compressive strength database for the machine learning prediction model;

[0007] Step 2: Use standardization or maximum and minimum normalization scaling methods to scale the values of the final selected feature parameters. At the same time, divide the processed data set into training and test sets according to a certain ratio. The training set will be used to train the machine learning model, and the trained model will be used to make predictions on the test set.

[0008] Step 3: Establish a machine learning model for the uniaxial compressive strength of rock and evaluate the model accuracy. The Newton-Raphson optimized convolutional neural network-long short-term memory network (NRBO-CNN-LSTM) machine learning model is trained on the training set and predicted on the test set. The performance of the model is evaluated using the following three indicators: mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R 2 ), the calculation formula is as follows:

[0009]

[0010] Where n is the total number of observations, y i is the actual observed value, is the predicted value, is the mean of the observed values.

[0011] R 2 This value reflects the quality of the regression model's fit. A value closer to 1 indicates a better fit, while a value closer to 0 indicates a poorer fit. MAE calculates the mean of the absolute errors between the actual and predicted values, reflecting the actual nature of the errors. RMSE calculates the root mean square of the deviations between the actual and predicted values and is more sensitive to large errors (outliers). Both metrics maintain the same scale as the actual values. Generally speaking, a larger R² value indicates a smaller MAE and RMSE value, indicating a higher accuracy of the trained model.

[0012] In the process of establishing a machine learning model, it is necessary to perform hyperparameter optimization based on the parameter characteristics involved in each model, select the optimal model hyperparameters, the number of hidden layers and neurons in the neural network model, the learning rate, regularization coefficient and number of hidden layer nodes in the Newton-Raphson method optimization part.

[0013] At the same time, a 5-fold cross-validation is performed, that is, the training set is further subdivided into 5 subsets without repetition, 4 of which are used as training subsets and the remaining one is used as a validation subset. This cycle is repeated 5 times to obtain 5 training models and corresponding model performance evaluations.

[0014] Compared with the prior art, the present invention has the following beneficial effects:

[0015] 1. Compared with traditional models, the rock uniaxial compressive strength prediction model established based on the machine learning algorithm does not rely on prior assumptions or the specific expression form of the model in advance. It directly constructs a complex nonlinear mapping between the uniaxial compressive strength of rock and multiple parameters. Therefore, the rock uniaxial compressive strength prediction model established by the machine learning model has higher precision and accuracy. 2. The machine learning model is established based on a larger data set, with stronger generalization ability and better applicability. 3. The rock mechanics parameter model established based on the machine learning algorithm is more scalable. With the continuous enrichment of the experimental database in the future, the rock uniaxial compressive strength prediction model can be continuously improved and optimized. Since more data can be used for training the machine learning model, the accuracy, robustness and generalization ability of the model in predicting the uniaxial compressive strength of rock will be further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the machine learning prediction model for the uniaxial compressive strength of rock;

[0017] Figure 2 It is the grey correlation analysis result diagram of rock uniaxial compressive strength and other physical parameters;

[0018] Figure 3 is the compressive strength box plot in the established rock uniaxial compressive strength database;

[0019] Figure 4 It is a box plot of the input feature parameter data of the four selected machine learning prediction models;

[0020] Figure 5 This is a comparison chart of the prediction results of NRBO-CNN-LSTM in the test set;

[0021] Figure 6 It is a linear correlation graph of the actual value and the predicted value on the application validation set. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. Figure 1 It should be understood that the specific embodiments described herein are only used to illustrate the present invention and are not intended to limit the present invention.

[0023] The first step is feature parameter selection and database establishment:

[0024] The parameters used in the traditional rock uniaxial compressive strength prediction model were statistically analyzed, and the correlation between other parameters and the rock uniaxial compressive strength was calculated using the grey correlation analysis method, such as Figure 2 As shown:

[0025] By calculating and analyzing the parameters used in the traditional model, porosity (n), Schmidt hammer rebound number (SHN), longitudinal wave velocity (Vp) and point load intensity (Is(50)) are used as input characteristic parameters in the rock uniaxial compressive strength prediction model, and the rock uniaxial compressive strength is the output characteristic in the prediction model.

[0026] According to the selected characteristic parameters, the collected data mainly includes complete rock characteristics, a total of 498 sets of data. A database was established, and preliminary analysis found that the collected sample data had some missing values and duplicate values. In order to make the prediction results as accurate as possible, the problem samples were deleted. Finally, 381 sets of data were retained, and box plots of each characteristic were drawn to visualize the data, as shown in the following figure. Figure 3-4 As shown, a box plot was used to identify outliers in the dataset. To ensure sufficient sample size and improve prediction accuracy, outliers in each parameter were processed. Outliers exceeding the upper limit were replaced with the sample mean, and outliers below the lower limit were replaced with the sample median. The input parameters and shear strength statistics are shown in Table 1.

[0027] Table 1:

[0028]

[0029] The processed dataset is divided into a training set and a test set with a ratio of 7:3. The training set contains 266 data sets and the test set contains 114 data sets. The training set will be used to train the machine learning model, and the trained model will be used to make predictions on the test set.

[0030] The second step is to establish a machine learning prediction model for the uniaxial compressive strength of rock:

[0031] The Newton-Raphson optimized convolutional neural network-long short-term memory network (NRBO-CNN-LSTM) machine learning model was trained on the training set and predicted on the test set. The performance of the model was evaluated using the following three indicators: mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R 2 ), the calculation formula is as follows:

[0032]

[0033] Where n is the total number of observations, y i is the actual observed value, is the predicted value, is the mean of the observed values.

[0034] In the process of establishing a machine learning model, it is necessary to optimize the hyperparameters according to the parameter characteristics of each model and select the most appropriate model hyperparameters. Through the hyperparameter optimization of the machine learning model, the number of hidden layers in the neural network model is selected as 1, the number of hidden layer neurons is selected as 5, the learning rate of the Newton-Raphson optimization part is selected as 0.0001, and the regularization coefficient is selected as 2. 5.2 and the number of hidden layer nodes is 1.

[0035] The NRBO-CNN-LSTM machine learning model was trained on the training set and validated on the test set. On the training set, the NRBO-CNN-LSTM machine learning model achieved a prediction accuracy of over 80% with a root mean square error of 1.5 MPa. On the test set, the NRBO-CNN-LSTM machine learning model achieved a prediction accuracy of 98% with a root mean square error of 1.3 MPa. Figure 5 shown.

[0036] During model training, a 5-fold cross-validation was performed. This involves further subdividing the training set into five subsets without duplication, with four serving as training subsets and the remaining one serving as validation subset. This cycle was repeated five times, resulting in five training models and corresponding model performance evaluations. The constructed multi-algorithm fusion machine learning model for predicting rock uniaxial compressive strength improves prediction results and achieves better accuracy.

[0037] The third step is to make predictions on the new dataset and evaluate the generalization performance of the rock uniaxial compressive strength prediction model:

[0038] Field investigations were conducted in the Longdong mining area of Daye Iron Mine, and rock samples were collected. A series of tests, including uniaxial compressive strength, point load strength, acoustic wave velocity, porosity, and Schmidt hammer rebound, were performed on the rock samples to establish an application verification data set, as shown in Table 3.

[0039] Table 3:

[0040]

[0041] The prediction software was used to verify the application validation data set, and the correlation coefficient R value between the predicted value and the actual value of the data set was 0.93. Figure 6 As shown in the figure, the prediction model is validated on the applied research dataset and shows that the multi-algorithm fusion machine learning prediction model has good generalization performance and prediction accuracy, and can provide guidance for the actual production and operation of mines.

[0042] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the uniaxial compressive strength of rock by integrating multiple algorithms with a machine learning model, characterized in that: The steps include: 1) Select characteristic parameters and calculate the correlation between rock physical parameters and rock uniaxial compressive strength using the grey correlation analysis method. This will screen out data types with a high correlation with rock uniaxial compressive strength. The input and output characteristic parameters of the Newton-Raphson optimized convolutional neural network-long short-term memory network (NRBO-CNN-LSTM) prediction model will be determined. 2) Collect multiple sets of rock physical parameter data and the corresponding rock uniaxial compressive strength, standardize and analyze the data sets, exclude or merge redundant parameters, and establish a rock uniaxial compressive strength database for machine learning prediction models; divide the processed data sets into training, validation, and test sets according to a certain ratio; 3) Develop an NRBO-CNN-LSTM machine learning model for predicting the uniaxial compressive strength of rock. The training set will be used to train the machine learning model, and the validation set will be used to optimize and adjust the model's hyperparameters. The trained and validated model will be used to perform predictions on the test set to evaluate its performance. Predictions will be performed on a new dataset to evaluate the generalization performance of the rock uniaxial compressive strength prediction model.

2. The NRBO-CNN-LSTM model for predicting the uniaxial compressive strength of rock according to claim 1, characterized in that: In the above 1-1), the parameters used in the traditional rock uniaxial compressive strength model are analyzed by the grey correlation analysis calculation method to select parameters related to the rock uniaxial compressive strength. At the same time, it is necessary to consider the available data to establish a database for the machine learning prediction model.

3. The NRBO-CNN-LSTM model for predicting the uniaxial compressive strength of rock according to claim 1-1) is characterized in that: The grey relational analysis calculation method is: Where minΔ ij and maxΔ ij The table shows the minimum and maximum absolute differences of all comparison sequences; ρ is the resolution coefficient, which is generally 0.1 to 0.5; r i For the degree of correlation.

4. The NRBO-CNN-LSTM model for predicting the uniaxial compressive strength of rock according to claim 1, characterized in that: In 1-2), there are two options for data scaling: standardization and normalization. Standardization rescales the features to have a standard normal distribution with a mean of 0 and a standard deviation of 1, while also preserving the shape properties of the original dataset. Another method, minimum-maximum normalization, scales the feature values to the interval [0, 1].

5. The NRBO-CNN-LSTM model for predicting the uniaxial compressive strength of rock according to claim 1-2) is characterized in that: The rock physical parameter data include point load strength, Schmidt hammer rebound number, porosity, and longitudinal wave velocity; the rock mechanical parameters include uniaxial compressive strength.

6. The NRBO-CNN-LSTM model for predicting the uniaxial compressive strength of rock according to claim 1, characterized in that: In 1-3), the NRBO-CNN-LSTM machine learning model is trained on the training set and predicted on the test set. The model performance is mainly evaluated using the following three indicators: mean absolute error (MAE), root mean square error (RMSE) and coefficient of determination (R 2 ), the calculation formula is as follows: Where n is the total number of observations, y i is the actual observed value, is the predicted value, is the mean of the observed values.

7. The NRBO-CNN-LSTM model for predicting the uniaxial compressive strength of rock according to claim 1, characterized in that: In 1-3), in order to avoid overfitting of the training model, the NRBO-CNN-LSTM machine learning model adopts 5-fold cross validation, and the performance of the model prediction accuracy and prediction error is evaluated based on the model performance indicators in the 5-fold cross validation.

8. The method for establishing and predicting the uniaxial compressive strength of rock according to claim 1, characterized in that: In 1-3), the parameter analysis of the machine learning model is performed on a new dataset, which is generated based on the features of the original dataset, so as to further verify and compare the generalization performance of each machine learning model.

9. A computer device, characterized in that: include memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, wherein the processor implements the rock uniaxial compressive strength prediction method of claims 1-8 when executing the computer-executable instructions. A computer-readable storage medium, characterized in that It stores computer instructions, which, when executed by a processor, implement the rock uniaxial compressive strength prediction method of claims 1-8.

Images