A rock mechanics parameter while-drilling prediction method based on a concrete test piece

Abstract

The application discloses a rock mechanics parameter prediction method based on a concrete test piece while drilling, and belongs to the technical field of coal mine rock burst prevention and control. The technical measures are as follows: the concrete test piece is used to replace the original rock test piece to collect vibration signals and mechanics parameters in a room, the characteristic values of the vibration signals are extracted and de-noised, the characteristic values after dimension reduction are fully combined, the combined samples are used as input parameters, the mechanics parameters are used as output values, and the neural network, the neural network optimized by the genetic algorithm and the neural network optimized by the particle swarm are used for training, and finally, the best neural network prediction model is obtained. The combined samples corresponding to the best neural network prediction model are used as input parameters, and a rock mechanics parameter prediction model is re-established. The model is simple, a large number of original test pieces and excessive characteristic values of the parameters while drilling are not needed, the consumption of calculation resources is reduced, and the prediction accuracy is improved.

Description

Technical Field

[0001] This invention relates to the field of coal mine rockburst prevention technology, and in particular to an indoor test method for intelligent prediction of rock mechanical parameters based on drilling parameters. Background Technology

[0002] During tunnel excavation, rock fractures, rock pillar collapses, and rockburst disasters occur frequently. Reasonable and effective tunnel support methods are a powerful means to ensure safe coal mining. In recent years, bolt support technology has developed rapidly and has become the most frequently used tunnel support method. However, the setting of bolt support parameters is generally based on past support experience in similar coal mines, which results in poor accuracy of the support parameters and often leads to insufficient or excessive support strength.

[0003] During the drilling process, bolt drilling rigs generate a large number of drilling-while-drilling parameters. Using these parameters to perceive the roadway's structural conditions is a simple and easy-to-operate method in the coal mining industry. However, scientific and quantitative research on drilling-while-drilling measurements in coal mines has been lacking. Utilizing bolt drilling rig parameters to accurately perceive changes in rock mechanics parameters in real time is of great significance for setting roadway support parameters.

[0004] Currently, common methods for predicting rock mechanical parameters using drilling parameters involve: acquiring a large amount of rock in the field, obtaining drilling parameters through triaxial tests and uniaxial loading tests in the laboratory, and then establishing a relationship between the drilling parameters and the mechanical parameters. When applied in the field, the drilling parameters of the original rock are obtained, and the mechanical parameters of the original rock are inverted using the relationship. For example, Chinese Patent Publication No. CN116291271A discloses a method and system for inverting rock and soil parameters based on drilling tests. Alternatively, the lithology of the original rock can be obtained beforehand, and then a predictive model of the lithology and mechanical parameters of the original rock can be established through laboratory tests. For example, Chinese Patent Publication No. CN115659783A discloses a method for predicting rock mechanical parameters in complex formations based on an intelligent fusion strategy. Both of these methods require the collection of a large amount of original rock, which is then cut into cubes of the required size. The strength range of the specimens is limited by the amount of rock collected, which is time-consuming and labor-intensive, and the strength range of the collected original rock is also limited. In addition, in order to improve the accuracy of predictions, a wide variety of drilling parameters are used when establishing the relationship between drilling parameters and mechanical parameters or predictive models, which wastes computing resources.

[0005] Therefore, it is necessary to provide a method to analyze the influence of rock drilling parameters on mechanical parameters in the laboratory using concrete specimens, and to screen out the drilling parameters that have a greater impact on mechanical parameters based on the degree of influence, thereby simplifying the input parameters of the prediction model and establishing a rapid prediction model. Summary of the Invention

[0006] To simplify the prediction model, reduce prediction costs, and improve prediction accuracy, this invention provides a method for predicting rock mechanical parameters while drilling based on concrete specimens.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] A method for predicting rock mechanical parameters while drilling based on concrete specimens, characterized by the following steps:

[0009] Step 1: Specimen Preparation

[0010] According to the "Specification for Mix Proportion Design of Ordinary Concrete" JGJ55-2011, four materials are used: water, cement, sand and gravel. Cube concrete specimens of different strengths are prepared according to different proportions, and the strength range is required to cover the strength range of the surrounding rock in the underground roadway.

[0011] Step 2: Data Acquisition

[0012] The prepared specimens were cored using a core extractor to obtain standard rock cores. During the core extraction process, a dynamic signal analyzer was connected to collect vibration signals and assign a number to each vibration signal. The purpose of the numbering is to facilitate differentiation and ensure that the vibration signals of all specimens are used later.

[0013] Then, the standard rock cores were subjected to uniaxial loading tests indoors to monitor the axial and lateral deformation of the specimens and obtain the mechanical parameters corresponding to concrete specimens of different strengths, such as uniaxial compressive strength, Young's modulus and Poisson's ratio.

[0014] Step 3: Data Preprocessing

[0015] 3.1: Feature Value Extraction from Raw Data

[0016] The vibration signals monitored in step two are analyzed and processed by the DHDAS dynamic signal acquisition and analysis system, and the vibration signal waveforms of each specimen are exported in sequence to generate Excel data. The Excel data of the same type of rock are preprocessed using MATLAB to extract the time domain feature value a1, and the time domain signal is Fourier transformed to extract the frequency domain feature value b1.

[0017] 3.2: Wavelet Denoising

[0018] The original vibration waveform signal was denoising sequentially using the MATLAB wavelet toolbox. Denoising was achieved by changing parameters such as threshold processing rules, soft threshold, and wavelet packet decomposition maturity within the wavelet denoising toolbox. The denoised waveform signal was saved and the denoised data was exported.

[0019] 3.3: Feature Value Extraction of Noise-Reduced Data

[0020] The exported denoised data is preprocessed using MATLAB to extract the denoised time-domain feature value a2; the denoised time-domain signal is then subjected to a series of calculations and transformations, including Fourier transform, to extract the denoised frequency-domain feature value b2.

[0021] 3.4: Data Processing

[0022] The vibration signal characteristic values ​​a1 and b1 of all original rock specimens, the noise-reduced vibration signal characteristic values ​​a2 and b2, as well as the uniaxial compressive strength and elastic modulus of the original rock specimens are compiled into the same file;

[0023] Step 4: Building an Intelligent Prediction Model

[0024] 4.1: Conduct comparative experiments using different combinations of vibration signal feature value parameters as input parameters for the neural network, analyze the influence of different input parameter combinations on the accuracy of neural network prediction results, obtain vibration signal feature value parameter combinations corresponding to different output parameters, and establish the optimal neural network model;

[0025] 4.2: Neural Network Optimization

[0026] The optimal neural network model established in step 4.1 was optimized using both the genetic algorithm and the particle swarm optimization algorithm, resulting in the optimal neural network model optimized by the genetic algorithm and the optimal neural network model optimized by the particle swarm optimization algorithm.

[0027] 4.3: Based on the evaluation metrics RMSE and R, the three optimal neural network models obtained in steps 4.1 and 4.2 are evaluated. 2 The values ​​are then optimized to finally select the best neural network prediction model for the concrete specimen. Based on the best neural network prediction model, the corresponding input parameter combination sample is found. This combination sample is the best combination of vibration signal feature values. It is considered that the feature values ​​contained in this combination sample have a greater impact on the mechanical parameters, while the other feature values ​​have a smaller impact.

[0028] Step 5: Prediction of Rock Mechanical Parameters

[0029] Based on the optimal combination of vibration signal characteristic values ​​determined in step four, some characteristic values ​​that have little impact on mechanical parameters are eliminated, and characteristic values ​​with a large impact are optimized. These characteristic values ​​with a large impact on the original rock are then used to re-establish the rock mechanical parameter prediction model, thereby predicting the mechanical parameters of the rock.

[0030] Furthermore, in step three, the parameters representing the time-domain feature value a1 and the denoised time-domain feature value a2 both include 13 parameters: mean, standard deviation, root mean square deviation, maximum value, minimum value, peak-to-peak value, skewness, kurtosis, amplitude factor, waveform factor, impulse factor, margin factor, and energy; the parameters representing the frequency-domain feature value b1 and the denoised frequency-domain feature value b2 both include 5 parameters: centroid frequency, mean square frequency, root mean square frequency, frequency variance, and frequency standard deviation.

[0031] Furthermore, the specific operation process of step 4.1 is as follows:

[0032] 4.1.1: Perform pre-dimensionality reduction on the input parameters before noise reduction.

[0033] The experimental data was divided into training and testing sets to train the neural network. The drilling parameters, which are the vibration signal features of the concrete specimens, were the input values, and the mechanical parameters were the output values. Before training, the 18 input parameters of the neural network were reduced in dimensionality. The target dimension was set to n, where n < 18, and n = 6 was preferred, to obtain n kinds of vibration signal feature values, that is, n kinds of input parameters.

[0034] 4.1.2: Establishing a full combination control experiment with n-dimensional input parameters

[0035] For the n input parameters after dimensionality reduction in 4.1.1, arrange them into all combinations without considering the order, resulting in a total of 2^n-1 combination samples. Set different numbers of hidden layers and hidden layer nodes for the corresponding neural networks, train the training set, and output the predicted values ​​of the mechanical parameters corresponding to the training set; input the drilling parameters of the test set into the trained neural network, and output the predicted values ​​of the mechanical parameters corresponding to the test set.

[0036] 4.1.3: Calculate the model evaluation metrics RMSE and R0 for the training and test sets respectively, based on the predicted and actual values ​​of the mechanical parameters. 2 Values, based on the model evaluation metrics RMSE and R², from the training and test sets. 2 The optimal neural network model is selected based on the values, and for ease of distinction, it is referred to as the optimal neural network model before noise reduction.

[0037] 4.1.4: Repeat steps 4.1.1 to 4.1.3, perform the same operations on the denoised input parameters and build a neural network model in the same way to obtain the best neural network model after denoising;

[0038] 4.1.5: Based on the model evaluation metrics RMSE and R of the optimal neural network model before and after noise reduction. 2Determine the accuracy of the prediction results before and after denoising. If the accuracy of the prediction results after denoising is lower than that before denoising, the wavelet denoising parameters in 3.2 need to be adjusted or the denoising method needs to be changed. Repeat steps 3.2-3.4 and 4.1.4 to obtain new prediction results until the accuracy of the prediction results after denoising is higher than that before denoising. At this point, the neural network model that meets the accuracy requirements of the prediction results is the best neural network model.

[0039] The positive effects achieved by the present invention through the above technical solution are:

[0040] 1. This invention utilizes readily available concrete specimens instead of original rock specimens to obtain the influence of rock drilling parameters on mechanical parameters through indoor simulation experiments. Based on the degree of influence, key drilling parameters for predicting mechanical parameters are optimized. In this way, when predicting the mechanical parameters of the original rock, it is not necessary to obtain a large number of original specimens or too many drilling parameter feature values. The model established using simplified drilling parameter feature values ​​is simple, which not only reduces the consumption of computing resources but also improves the accuracy of prediction.

[0041] 2. In predicting the parameters of the original rock, this invention first uses concrete specimens to simulate the rock, and then conducts uniaxial compression tests on the rock cores in the laboratory using a standardized testing machine, which yields more accurate rock mechanical parameters. Concrete specimens can be obtained in batches according to experimental requirements, and the types and quantities of specimens can be planned in advance. Therefore, the established training and test sets are targeted and controllable.

[0042] 3. The input parameter of the prediction model of this invention is the vibration signal, which is easy to obtain. When applied in underground coal mines, the anchor bolt support in underground roadways requires drilling a large number of anchor bolt holes. Therefore, the vibration signal can be monitored and acquired in real time while the drilling rig is drilling into the rock, and then the established prediction model can be used to predict the rock to be predicted. This enables real-time and efficient prediction of rock conditions while drilling is underway. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0045] The following is in conjunction with the appendix Figure 1The construction process of the embodiments of the present invention will be described in detail so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more explicit definition of the scope of protection of the present invention.

[0046] like Figure 1 As shown, the present invention provides a method for predicting rock mechanical parameters while drilling based on concrete specimens, characterized by the following steps:

[0047] Step 1: Specimen Preparation

[0048] According to the "Specification for Mix Proportioning Design of Ordinary Concrete" JGJ55-2011, four raw materials—water, cement, sand, and gravel—were used. Based on different proportions, 19 different concrete specimens of varying strengths (C10, C15, C20, C25, C30, C35, C40, C45, C50, C55, C60, C65, C70, C75, C80, C85, C90, C95, and C100) were prepared. The reason for preparing 19 different strength concrete specimens is to ensure that the prepared concrete specimens can simulate the various strengths of rocks mentioned in the rock hardness classification. Considering that the strength range of surrounding rock in typical coal mine roadways is 10-100 MPa, this invention produces specimens within this strength range, which basically conforms to the field conditions. Furthermore, when the strength of the surrounding rock in the applied roadway exceeds this range, appropriate adjustments can be made according to the specific circumstances.

[0049] The concrete specimens are 100mm*100mm*100mm cubes. Ten specimens are made for each strength grade, for a total of 190 specimens. The prepared concrete specimens are then cured with water for 28 days.

[0050] Step 2: Data Acquisition

[0051] The prepared specimens were cored using a core extractor to obtain standard rock cores (Φ50mm*100mm). During the core extraction process, a DH5909N handheld dynamic signal analyzer was connected to collect vibration signals. The standard rock cores were then subjected to uniaxial loading using a WDW-3100 micro-controlled electronic universal testing machine. The axial and lateral deformation of the specimens were monitored simultaneously to obtain the uniaxial compressive strength, Young's modulus, and Poisson's ratio of concrete specimens with different strengths.

[0052] 2.1: Device Connection

[0053] Connect the IEPE piezoelectric accelerometer to the DH5909N handheld dynamic signal analyzer and adjust the instrument to normal working condition; clamp the prepared specimen, prepare the core extractor, and connect the accelerometer to the core extractor.

[0054] 2.2: Vibration Signal Acquisition

[0055] Once the instrument is connected, create a new acquisition file in the dynamic signal analyzer. While starting the coring machine to drill for rock cores, click "Start Acquisition" to acquire complete vibration signals. Acquire the vibration signals of each specimen in sequence and assign a number to each vibration. The number is written using the concrete specimen strength and specimen order. For example, C30-1 is the first specimen with a strength of 30 MPa.

[0056] 2.3: Obtaining Rock Mechanical Parameters of Specimens

[0057] The standard rock cores were subjected to uniaxial loading using a WDW-3100 micro-controlled electronic universal testing machine. The axial and lateral deformation of the specimens were monitored at the same time, and the uniaxial compressive strength, Young's modulus and Poisson's ratio of the specimens were obtained respectively.

[0058] Step 3: Data Preprocessing

[0059] 3.1: Feature Value Extraction from Raw Data

[0060] The vibration signals monitored in step 2.2 were analyzed and processed by the DHDAS dynamic signal acquisition and analysis system, and the vibration signal waveforms of each original rock specimen were exported in sequence as Excel data. The Excel data of the same rock were preprocessed using MATLAB to extract time-domain feature values ​​a1: 13 parameters including mean, standard deviation, root mean square deviation, maximum value, minimum value, peak-to-peak value, skewness, kurtosis, amplitude factor, waveform factor, impact factor, margin factor, and energy. The time-domain signals were Fourier transformed to extract frequency-domain feature values ​​b1: 5 parameters including centroid frequency, root mean square frequency, frequency variance, and frequency standard deviation.

[0061] 3.2: Wavelet Denoising

[0062] Since the vibration signal acquisition process is mixed with environmental noise, in order to more accurately analyze the relationship between the vibration signal and rock mechanical parameters, the original vibration waveform signal is denoised sequentially using the MATLAB wavelet toolbox. Denoising is performed by changing parameters such as threshold processing rules, soft threshold, and wavelet packet decomposition maturity in the wavelet denoising toolbox. The denoised waveform signal is saved and the denoised data is exported.

[0063] 3.3: Feature Value Extraction of Noise-Reduced Data

[0064] The exported denoised data was preprocessed using MATLAB to extract the denoised time-domain feature values ​​a2: 13 parameters including mean, standard deviation, root mean square deviation, maximum value, minimum value, peak-to-peak value, skewness, kurtosis, amplitude factor, waveform factor, impulse factor, margin factor, and energy. The denoised time-domain signal was then transformed using Fourier transform and other calculations to extract the denoised frequency-domain feature values ​​b2: 5 parameters including centroid frequency, root mean square frequency, frequency variance, and frequency standard deviation.

[0065] 3.4: Data Processing

[0066] The vibration signal characteristic values ​​a1 and b1 of all original rock specimens, the noise-reduced vibration signal characteristic values ​​a2 and b2, as well as the uniaxial compressive strength and elastic modulus of the original rock specimens are compiled into the same file;

[0067] Step 4: Building an Intelligent Prediction Model

[0068] 4.1: Conduct comparative experiments using different combinations of vibration signal eigenvalue parameters as input parameters to analyze the influence of different input parameter combinations on prediction accuracy, obtain vibration signal eigenvalue parameter combinations corresponding to different output parameters, and establish the optimal neural network model. The specific operation is as follows:

[0069] 4.1.1 Perform pre-dimensionality reduction on the input parameters before noise reduction.

[0070] The experimental data was divided into training and testing sets to train the neural network. The drilling parameters, which are the vibration signal features of the concrete specimens, were used as input values, and the mechanical parameters were used as output values. The pre-dimensionality reduction steps for the neural network input dimension are as follows: the target dimension was set to 6, and principal component analysis was used to reduce the dimensionality of the 18 parameters, which consist of 13 parameters representing time-domain feature value a1 and 5 parameters representing frequency-domain feature value b1, to obtain a combined sample composed of 6 types of vibration signal feature values.

[0071] The total number of combinations for n parameters is 2^n-1. Therefore, the total number of combinations for 18 parameters is 2^18-1, or 262,143 combinations. This is too large a dataset to model and compare all combinations. For practical feasibility, we pre-reduce the dimensionality to 6 dimensions, resulting in a total of 2^6-1 combinations, or 63 combinations. Here, n is a suggested dimension, but it can be set according to the actual situation.

[0072] 4.1.2: Establishing a full combination control experiment with 6-dimensional input parameters

[0073] The six input parameters after dimensionality reduction in 4.1.1 were arranged into all possible combinations without regard to order, resulting in a total of 63 combination samples. Different numbers of hidden layers and hidden layer nodes were set for the corresponding neural networks. The training set was then trained, and the predicted values ​​of the mechanical parameters corresponding to the training set were output. The drilling parameters of the test set were input into the trained neural network, and the predicted values ​​of the mechanical parameters corresponding to the test set were output.

[0074] 4.1.3: Calculate the model evaluation metrics RMSE and R0 for the training and test sets respectively, based on the predicted and actual values ​​of the mechanical parameters. 2 Values, based on the model evaluation metrics RMSE and R², from the training and test sets. 2 The optimal neural network model is selected based on the values, and for ease of distinction, it is referred to as the optimal neural network model before noise reduction.

[0075] 4.1.4: Repeat steps 4.1.1 to 4.1.3, perform the same operations on the denoised input parameters and build a neural network model in the same way to obtain the best neural network model after denoising;

[0076] 4.1.5: Based on the model evaluation metrics RMSE and R of the optimal neural network model before and after noise reduction. 2 Determine the accuracy of the prediction results before and after denoising. If the accuracy of the prediction results after denoising is lower than that before denoising, the wavelet denoising parameters in 3.2 need to be adjusted or the denoising method needs to be changed. Repeat steps 3.2-3.4 and 4.1.4 to obtain new prediction results. Continue in this way until the accuracy of the prediction results after denoising is higher than that before denoising. At this point, the neural network model that meets the accuracy of the prediction results is the best neural network model.

[0077] 4.2: Neural Network Optimization

[0078] The mechanical parameters of the concrete specimens were predicted using neural network models optimized by genetic algorithm and neural network models optimized by particle swarm optimization, respectively, in the same way as in step 4.1. The best neural network model optimized by genetic algorithm and the best neural network model optimized by particle swarm optimization were selected respectively.

[0079] 4.3: Based on the evaluation metrics RMSE and R, the three optimal neural network models obtained in steps 4.1.4 and 4.2 are evaluated. 2 The values ​​are then optimized to finally select the best neural network prediction model for the concrete specimen. Based on the best neural network prediction model, the corresponding input parameter combination sample is found. This combination sample is the best combination of vibration signal feature values. It is then considered that the feature values ​​contained in this combination sample have a greater impact on the mechanical parameters, while the other feature values ​​have a smaller impact.

[0080] Step 5: Prediction of Rock Mechanical Parameters

[0081] Based on the optimal combination of vibration signal characteristic values ​​determined in step four, some characteristic values ​​that have little impact on mechanical parameters are eliminated, and characteristic values ​​with a large impact are optimized. These characteristic values ​​with a large impact on the original rock are then used to re-establish the rock mechanical parameter prediction model, thereby predicting the mechanical parameters of the rock.

[0082] It should be noted that when rebuilding the prediction model using the original rock, the input dimension of the neural network is the number of types of feature values ​​that have a significant impact on mechanical parameters, as selected in step four. The remaining modeling steps, such as setting the number of hidden layers and hidden layer nodes, are performed according to the standard modeling steps.

[0083] This invention utilizes model evaluation metrics RMSE and R 2 Selecting the optimal neural network model is an existing technique and not an innovation of this invention; therefore, it is not described in detail in the material. The above description represents preferred embodiments of this invention and is not intended to limit the technical solutions of this invention. It should be noted that those skilled in the art can make various improvements without departing from the principles described in this invention, and these improvements should also be considered within the scope of protection of this invention.

Claims

1. A method for predicting rock mechanical parameters while drilling based on concrete specimens, characterized in that, The steps are as follows: Step 1: Specimen Preparation Cube concrete specimens with different strengths, covering the strength range of the surrounding rock in underground roadways, were prepared according to different mix proportions. Step 2: Data Acquisition The prepared specimens were sampled from standard rock cores and vibration signals were collected. Then, uniaxial loading tests were performed on the standard rock cores to obtain the mechanical parameters corresponding to concrete specimens of different strengths. The mechanical parameters included uniaxial compressive strength, Young's modulus, and Poisson's ratio. Step 3: Data Preprocessing 3.1: Feature Value Extraction from Raw Data The monitored vibration signals were analyzed and preprocessed to extract time-domain feature value a1, and the time-domain signal was Fourier transformed to extract frequency-domain feature value b1. 3.2: Wavelet Denoising The original vibration waveform signal was denoising sequentially using the MATLAB wavelet toolbox. Denoising was achieved by changing the threshold processing rules, soft threshold, and wavelet packet decomposition parameters within the wavelet denoising toolbox. The denoised waveform signal was saved and the denoised data was exported. 3.3: Feature Value Extraction of Noise-Reduced Data The exported denoised data is preprocessed using MATLAB to extract the time-domain feature value a2 after denoising, and the time-domain signal is subjected to Fourier transform to extract the frequency-domain feature value b2 after denoising. Step 4: Building an Intelligent Prediction Model 4.1: Conduct comparative experiments using different combinations of vibration signal feature values ​​as neural network input parameters to analyze the impact of different input parameter combinations on the accuracy of neural network prediction results and establish the optimal neural network model; 4.2: Neural Network Optimization The optimal neural network model established in step 4.1 was optimized using both the genetic algorithm and the particle swarm optimization algorithm, resulting in the optimal neural network model optimized by the genetic algorithm and the optimal neural network model optimized by the particle swarm optimization algorithm. 4.3: Further optimization of the three optimal neural network models was carried out to finally select the best neural network prediction model for the concrete specimen; Step 5: Prediction of Rock Mechanical Parameters Based on the optimal neural network prediction model of the concrete specimen, the best combination of vibration signal feature values ​​is found, and these vibration signal feature values ​​of the original rock are obtained to re-establish the rock mechanical parameter prediction model, thereby predicting the mechanical parameters of the rock. The parameters representing the time-domain feature value a1 and the denoised time-domain feature value a2 both include 13 parameters: mean, standard deviation, root mean square deviation, maximum value, minimum value, peak-to-peak value, skewness, kurtosis, amplitude factor, waveform factor, impulse factor, margin factor, and energy. The parameters representing the frequency-domain feature value b1 and the denoised frequency-domain feature value b2 both include 5 parameters: centroid frequency, root mean square frequency, frequency variance, and frequency standard deviation.

2. The method for predicting rock mechanical parameters while drilling based on concrete specimens as described in claim 1, characterized in that, The specific operation process of step 4.1 is as follows: 4.1.1: Perform pre-dimensionality reduction on the input parameters before noise reduction. The experimental data was divided into training and testing sets to train the neural network. The drilling parameters, which are the vibration signal features of the concrete specimens, were the input values, and the mechanical parameters were the output values. Before training, the 18 input parameters of the neural network were reduced in dimensionality, and the target dimension was set to n, where n < 18, to obtain n kinds of vibration signal feature values, that is, n kinds of input parameters. 4.1.2: Establishing a full combination control experiment with n-dimensional input parameters Arrange all n input parameters after dimensionality reduction in step 4.1.1 without considering the order, and obtain a total of 2^n-1 combination samples. Set different numbers of hidden layers and hidden layer nodes for the corresponding neural networks, train the training set, and output the predicted values ​​of the mechanical parameters corresponding to the training set; input the drilling parameters of the test set into the trained neural network, and output the predicted values ​​of the mechanical parameters corresponding to the test set. 4.1.3: Calculate the model evaluation metrics RMSE and R based on the predicted and actual values ​​of the mechanical parameters for the training and test sets, respectively. 2 Values, based on the model evaluation metrics RMSE and R², from the training and test sets. 2 The optimal neural network model is selected based on the values, and this model is called the optimal neural network model before noise reduction. 4.1.4: Repeat steps 4.1.1 to 4.1.3, perform the same operations on the denoised input parameters and build a neural network model in the same way to obtain the best neural network model after denoising; 4.1.5: Based on the model evaluation metrics RMSE and R of the optimal neural network model before and after noise reduction. 2 Determine the accuracy of the prediction results before and after denoising. If the accuracy of the prediction results after denoising is lower than that before denoising, the wavelet denoising parameters in step 3.2 need to be adjusted or the denoising method needs to be changed. Repeat steps 3.2-3.3 and 4.1.4 to obtain new prediction results until the accuracy of the prediction results after denoising is higher than that before denoising. At this point, the neural network model that meets the accuracy requirements of the prediction results is the optimal neural network model.

3. The method for predicting rock mechanical parameters while drilling based on concrete specimens as described in claim 2, characterized in that, The n=6 mentioned above.

Images