Joint roughness prediction method and joint roughness prediction model training method

Abstract

The application embodiment provides a joint roughness prediction method and a joint roughness prediction model training method. The joint roughness prediction model training method comprises the following steps: obtaining m initial training samples, each of which comprises n roughness parameter indexes and a roughness label; performing principal component analysis on the m initial training samples to obtain k principal component feature vectors; processing the m initial training samples based on the k principal component feature vectors to obtain m target training samples; and training a pre-established initial machine learning model by using the m target training samples to obtain a joint roughness prediction model. The application embodiment has high precision and wide applicability for predicting roughness coefficients of different joints.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rock engineering, and in particular to a joint roughness prediction method and a joint roughness prediction model training method. BACKGROUND

[0002] Rock joints are natural defects formed in natural rocks under the long-term action of external forces and geological forces, which seriously affect the mechanical behavior and structural stability of the rock. A large number of engineering practices show that the joint shear strength is very important for the prediction of natural geological disasters (such as fault movement, landslide and earthquake) and the stability evaluation of large geological engineering structures (such as underground caverns, mines, foundations and waste storage).

[0003] The joint roughness coefficient (JRC) is an important factor that determines the joint shear strength. In related technologies, there is a way to determine JRC by using a statistical parameter method, that is, to calculate JRC by establishing a relationship between roughness statistical parameters and JRC. However, the evaluation index system composed of a large number of roughness statistical parameters is a nonlinear system with a large amount of overlapping information. Affected by the nonlinear relationship and the diversity of joints, the existing JRC determination method has the problems of complex relationship fitting and poor applicability. SUMMARY

[0004] The embodiments of the present application provide a joint roughness prediction method and a joint roughness prediction model training method to solve the problems of complex relationship fitting and poor applicability of the JRC determination method in related technologies.

[0005] To solve the above technical problems, the present application is implemented as follows:

[0006] In a first aspect, the embodiments of the present application provide a joint roughness prediction model training method, comprising:

[0007] Obtaining m initial training samples, each of which includes n roughness parameter indicators and a roughness label, m and n are integers greater than 1;

[0008] Performing principal component analysis on the m initial training samples to obtain k principal component feature vectors, k is a positive integer;

[0009] Processing the m initial training samples based on the k principal component feature vectors to obtain m target training samples, each of which includes k target indicators and a roughness label, the target indicators being obtained by processing the roughness parameter indicators through the principal component feature vectors;

[0010] The pre-established initial machine learning model is trained using the m target training samples to obtain a joint roughness prediction model, wherein the initial machine learning model includes an initial XGBoost model and an intelligent optimization algorithm model, the initial XGBoost model is used to receive the target training samples, output a predicted roughness, and adjust network parameters according to the predicted roughness and the roughness label, and the intelligent optimization algorithm model is used to adjust hyperparameters of the initial XGBoost model.

[0011] In a second aspect, the embodiments of the present application also provide a joint roughness prediction method, characterized in that the method comprises:

[0012] Obtaining a to-be-predicted sample, the to-be-predicted sample including n roughness parameter indicators, n being an integer greater than 1;

[0013] Processing the to-be-predicted sample using k principal component feature vectors to obtain a preprocessed sample, the k principal component feature vectors being obtained based on principal component analysis, k being a positive integer;

[0014] Outputting the preprocessed sample to a joint roughness prediction model to obtain a joint roughness coefficient, the joint roughness prediction model being trained by the method of the first aspect.

[0015] In a third aspect, the embodiments of the present application also provide a joint roughness prediction model training device, comprising:

[0016] A first obtaining module is configured to obtain m initial training samples, each initial training sample including n roughness parameter indicators and a roughness label, m and n being integers greater than 1;

[0017] A principal component analysis module is configured to perform principal component analysis on the m initial training samples to obtain k principal component feature vectors, k being a positive integer;

[0018] A first processing module is configured to process the m initial training samples based on the k principal component feature vectors to obtain m target training samples, each target training sample including k target indicators and a roughness label, the target indicators being obtained by processing the roughness parameter indicators through the principal component feature vectors;

[0019] A training module is configured to train a pre-established initial machine learning model using the m target training samples to obtain a joint roughness prediction model, wherein the initial machine learning model includes an initial XGBoost model and an intelligent optimization algorithm model, the initial XGBoost model is used to receive the target training samples, output a predicted roughness, and adjust network parameters according to the predicted roughness and the roughness label, and the intelligent optimization algorithm model is used to adjust hyperparameters of the initial XGBoost model.

[0020] In a fourth aspect, the embodiments of the present application further provide a joint roughness prediction device, comprising:

[0021] The second acquisition module is configured to acquire a to-be-predicted sample, the to-be-predicted sample comprising n roughness parameter indicators, n being an integer greater than 1.

[0022] The second processing module is configured to process the to-be-predicted sample using k principal component feature vectors to obtain a preprocessed sample, the k principal component feature vectors being obtained based on principal component analysis, and k being a positive integer.

[0023] The prediction module is configured to output the preprocessed sample to a joint roughness prediction model to obtain a joint roughness coefficient, the joint roughness prediction model being obtained by training the method of the first aspect.

[0024] In a fifth aspect, the embodiments of the present application further provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method described above when executing the computer program.

[0025] In a sixth aspect, the embodiments of the present application further provide a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method described above.

[0026] The joint roughness prediction model training method provided by the embodiments of the present application performs principal component analysis on the roughness parameter indicators in the initial training sample, can find out the key factors affecting the joint roughness, so that the target indicators in the target training sample obtained do not lose the fuzzy information of the joint roughness, and also maximize the preservation of the original information of the joint roughness; the target training sample is used to train the pre-established initial machine learning model to obtain a joint roughness prediction model, wherein the initial machine learning model comprises an initial XGBoost model and an intelligent optimization algorithm model, and the intelligent optimization algorithm model with strong global optimization ability, high convergence accuracy and fast speed is used to optimize the hyperparameters in the initial XGBoost model, so as to improve the prediction performance and running efficiency of the joint roughness prediction model. As a trained machine learning model, the joint roughness prediction model can break through the limitations of traditional experience regression models, has strong robustness for the nonlinear relationship between multi-dimensional variables, and has wide applicability for roughness prediction of different joints. In addition, through principal component analysis, the dimension of the target indicators in the target training sample is usually reduced, which helps to solve the data sparsification problem, improve the generalization ability of the trained joint roughness prediction model, and better adapt to the prediction of joint roughness coefficients of different types of joints. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1A flowchart of a joint roughness prediction model training method provided by an embodiment of the present application is shown.

[0028] Figure 2 A flowchart of a joint roughness prediction model training process for an application example is shown.

[0029] Figure 3 A flowchart of a joint roughness prediction method provided by an embodiment of the present application is shown.

[0030] Figure 4 A structural diagram of a joint roughness prediction model training device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0031] To make the technical problems, technical solutions and advantages of the present application clearer, specific embodiments will be described in detail below with reference to the accompanying drawings. In the following description, specific details such as specific configurations and components are provided only to help a comprehensive understanding of the embodiments of the present application. Therefore, it should be clear to those skilled in the art that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present application. In addition, descriptions of known functions and structures are omitted for clarity and brevity.

[0032] Unless otherwise defined, technical terms or scientific terms used in the present application should be understood as their ordinary meanings to those skilled in the art. The terms "first", "second" and similar terms used in the present application do not represent any order, number or importance, but are only used to distinguish different components. Similarly, the terms "one" or "a" and similar terms do not represent a quantity limitation, but represent the existence of at least one.

[0033] As shown in Figure 1 The joint roughness prediction model training method provided by an embodiment of the present application includes:

[0034] Step 101, m initial training samples are obtained, each initial training sample including n roughness parameter indicators and a roughness label, m and n being integers greater than 1;

[0035] Step 102, principal component analysis is performed on the m initial training samples to obtain k principal component feature vectors, k being a positive integer;

[0036] Step 103, the m initial training samples are processed based on the k principal component feature vectors to obtain m target training samples, each target training sample including k target indicators and a roughness label, the target indicators being obtained by processing the roughness parameter indicators through the principal component feature vectors;

[0037] In step 104, the pre-established initial machine learning model is trained using the m target training samples to obtain a joint roughness prediction model, wherein the initial machine learning model includes an initial XGBoost model and an intelligent optimization algorithm model, the initial XGBoost model is used to receive the target training samples, output the predicted roughness, and adjust the network parameters according to the predicted roughness and the roughness label, and the intelligent optimization algorithm model is used to adjust the hyperparameters of the initial XGBoost model.

[0038] The joint roughness prediction model training method provided by the embodiments of the present application can perform principal component analysis on the roughness parameter indicators in the initial training samples, can find out the key factors affecting the joint roughness, so that the target indicators in the target training samples obtained do not lose the fuzzy information of the joint roughness, and the original information of the joint roughness is retained to the greatest extent; the target training samples are used to train the pre-established initial machine learning model to obtain a joint roughness prediction model, wherein the initial machine learning model includes an initial XGBoost model and an intelligent optimization algorithm model, the intelligent optimization algorithm model with strong global optimization ability, high convergence accuracy and high speed is used to optimize the hyperparameters in the initial XGBoost model, so as to improve the prediction performance and running efficiency of the joint roughness prediction model, and the joint roughness prediction model as a trained machine learning model can break through the limitations of traditional experience regression models, has strong robustness for the nonlinear relationship between multi-dimensional variables, and has wide applicability for roughness prediction of different joints; in addition, through principal component analysis, the dimension of the target indicators in the target training samples can be reduced, which helps to solve the data sparseness problem, improve the generalization ability of the trained joint roughness prediction model, and better adapt to the prediction of the joint roughness coefficient (JRC) of different types of joints.

[0039] The joint roughness prediction model training method provided by the embodiments of the present application can be applied to electronic devices such as personal computers, industrial control computers, mobile terminals or servers, which are not limited here. For ease of description, the electronic device will be mainly described as the execution subject of the method in the following embodiments.

[0040] In step 101, the electronic device can obtain m initial training samples, each of which includes n roughness parameter indicators and a roughness label.

[0041] For the initial training samples, they can be obtained by measuring or statistically analyzing the joint correlation data. For example, a plurality of digitized joint profiles can be pre-established, each of which can include 10 standard contour lines, and based on these standard contour lines, roughness statistical parameters such as average relative height, maximum relative height or height standard deviation can be obtained, which can be used as the roughness parameter indicators mentioned above. Each digitized joint profile can correspond to the n roughness parameter indicators mentioned above, and each digitized joint profile can be measured by various existing quantitative methods to obtain the corresponding JRC, which can be used as the corresponding roughness label. Combining the n roughness parameter indicators and the roughness label corresponding to the digitized joint profile, the corresponding initial training sample can be obtained.

[0042] In some examples, the n roughness parameter indicators can include at least two of the following indicators: average relative height, maximum relative height, height standard deviation, average inclination, inclination standard deviation, root mean square of slope, roughness profile index and profile structure function.

[0043] In combination with some application scenarios, the n roughness parameter indicators can include the above-mentioned 8 indicators, wherein the average relative height is denoted as R ave , the maximum relative height is denoted as R max , the height standard deviation is denoted as SD h , the average inclination is denoted as i ave , the inclination standard deviation is denoted as SD i , the root mean square of slope is denoted as Z2, the roughness profile index is denoted as R p , and the profile structure function is denoted as SF. The roughness characteristics of the rock joint profile are characterized by these indicators. The definitions of the above roughness parameter indicators are as follows:

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] In the formulas (1)-(8), h aveis the average height of the joint profile, L is the length of the joint profile in the horizontal direction, y max is the maximum height coordinate of the joint profile in the height direction, y min is the minimum height coordinate of the joint profile in the height direction, N is the total number of nodes into which the joint profile is divided, n is the number of nodes (and is not the same concept as the roughness parameter index n above), (x n ,y n ) is the coordinate of the nth node.

[0053] The index system composed of roughness statistical parameters (corresponding to n roughness parameter indexes) is a nonlinear system with a large amount of overlapping information, which is difficult to apply due to the complexity of the calculation formula. Therefore, parameter selection and calculation method can be used to achieve the purpose of reasonably predicting JRC. In view of this, the present application introduces principal component analysis to realize parameter dimension reduction and effectively strip cross information.

[0054] Specifically, in step 102, the electronic device can perform principal component analysis on the m initial training samples to obtain k principal component feature vectors.

[0055] In some embodiments, the step of principal component analysis can use a principal component analysis method (PCA), such as a PCA algorithm based on eigenvalue decomposition of the covariance matrix, or a PCA algorithm based on singular value decomposition (SVD) of the covariance matrix, etc.

[0056] In one embodiment, the above step 102 can include:

[0057] Step S1, constructing an original data matrix of m*n order according to the m initial training samples and the n roughness parameter indexes included in each initial training sample;

[0058] Step S2, constructing a correlation coefficient matrix according to the original data matrix;

[0059] Step S3, solving the eigenvalues and eigenvectors of the correlation coefficient matrix;

[0060] Step S4, determining k principal component feature vectors according to the eigenvalues and eigenvectors of the correlation coefficient matrix.

[0061] For convenience of description, the jth roughness parameter index in the ith initial training sample can be denoted as x ij , wherein i = 1, 2, 3, …, m; j = 1, 2, 3, …, n. In step S1, the m*n order original data matrix X m×n can be denoted as:

[0062]

[0063] In some embodiments, before performing principal component analysis, a correlation coefficient between any two of the n roughness parameter indicators can be calculated, and if the correlation coefficient between any two of the n roughness parameter indicators is greater than a coefficient threshold, then an original data matrix of mxn order can be constructed according to the m initial training samples and the n roughness parameter indicators included in each initial training sample.

[0064] In combination with some examples, the correlation coefficient between any two of the roughness parameter indicators can be a KMO (Kaiser-Meyer-Olkin) index, and the KMO can be calculated as follows:

[0065]

[0066] wherein i and j in formula (10) represent two different roughness parameter indicators (different from the definition in formula (9)), r ij is the correlation between the two roughness parameter indicators, p ij is the partial correlation between the two roughness parameter indicators.

[0067] When the value of the KMO is greater than a coefficient threshold, such as 0.6 or other preset values, it can be considered that the above two roughness parameter indicators can be subjected to principal component analysis. Similarly, for other combinations of two roughness parameter indicators, the KMO can also be calculated in a similar manner, and the calculation result of the KMO can be used to determine whether to perform principal component analysis.

[0068] In an embodiment, when constructing the correlation coefficient matrix according to the original data matrix in step S2, the original data matrix X m×n may be subjected to normalization processing to overcome the problem of large numbers drowning small numbers caused by the magnitude and dimension differences of the original data, and to ensure the calculation accuracy and speed up the subsequent model training.

[0069] In an embodiment, the normalization processing of the original data matrix X m×n includes two layers of processing processes of normalization and standardization, and the roughness parameter indicators subjected to the standardization processing have a mean value of 0 and a variance of 1, and the normalization formula is shown in formula (11):

[0070]

[0071] wherein x ij * is the normalized value of x ij , μ jis the mean, and σ i is the variance.

[0072] The original data matrix X m×n is normalized and standardized to obtain a standardized matrix X * m×n The specific expression can be:

[0073]

[0074] In an embodiment, step S2 of constructing the correlation coefficient matrix according to the original data matrix can be calculated according to the following formula:

[0075]

[0076] where R m×n is the correlation coefficient matrix, X * is the standardized matrix, and X *T is the transpose matrix of the standardized matrix.

[0077] Of course, in other feasible embodiments, the coefficients used in calculating the correlation coefficient matrix in formula (13) can also be adjusted as needed.

[0078] In step S3, the electronic device can solve the eigenvalues and eigenvectors of the correlation coefficient matrix.

[0079] In an embodiment, the electronic device can use the Lagrange multiplier method to solve the eigenvalues and eigenvectors of R m×n :

[0080] |R m×n - λE | = 0 (14)

[0081] where E is the identity matrix.

[0082] The eigenvalues obtained for R m×n can be denoted as λ1, λ2, λ3, …, λ n , and the eigenvectors obtained can be denoted as u1, u2, u3, …, u n .

[0083] Step S4, the electronic device can determine k principal component eigenvectors according to the eigenvalues and eigenvectors of the correlation coefficient matrix.

[0084] In an embodiment, step S4 can include:

[0085] arranging all the eigenvalues of the correlation coefficient matrix in descending order, and arranging the variance contribution rate corresponding to each eigenvalue in descending order;

[0086] According to the arrangement order of all eigenvalues, the variance contribution rates are sequentially accumulated;

[0087] In a case where the accumulated variance contribution rate value is greater than the preset threshold value for the first time, k eigenvalues corresponding to the accumulated k variance contribution rates are obtained;

[0088] The k eigenvectors corresponding to the k eigenvalues are determined as k principal component eigenvectors.

[0089] In a case where all eigenvalues are arranged in descending order, for the jth eigenvalue λ j , the corresponding variance contribution rate η j can be calculated as follows:

[0090]

[0091] According to the arrangement order of all eigenvalues, the variance contribution rates are sequentially accumulated, and k variance contribution rates can obtain η j (k), and the formula is as follows:

[0092]

[0093] When the value of η j (k) is greater than the preset threshold value for the first time, the k eigenvalues used for accumulation can be extracted, and k eigenvectors corresponding to the k eigenvalues are obtained, respectively denoted as u1, u2, u3, …, u k The k eigenvectors can be determined as k principal component eigenvectors.

[0094] In step 103, the electronic device can process m initial training samples based on the k principal component eigenvectors to obtain m target training samples. In an embodiment, step 103 can specifically include:

[0095] According to the n roughness parameters included in each initial training sample, a row vector is determined, each principal component eigenvector is taken as a column vector, and a target training sample is obtained by dot multiplication.

[0096] In some embodiments, the electronic device can directly use the original data matrix X m×n and the principal component eigenvectors to calculate the target training samples, or the electronic device can use the standardized matrix X m×n obtained by normalization and the principal component eigenvectors to calculate the target training samples. * m×n

[0097] As an exemplary illustration, the subsequent description will mainly be based on the standardized matrix X * ​m×n and the principal component feature vectors to calculate the target training sample.

[0098] For the k principal component feature vectors u1, u2, u3, …, uk, the principal component matrix Z can be calculated as follows. k The principal component score coefficient matrix U can be obtained by combining the column vectors respectively. n×k * m×n and U n×k m×k , the formula is as follows:

[0099] Z m×k = X * m×n U n×k (17)

[0100] It can be seen that the principal component matrix Z m×k has m rows of data, each row corresponds to a target training sample, and each row has k data, which corresponds to the k target indicators in a target training sample. Based on formula (17), it can be seen that the n roughness parameters in each row of X * m×n Each column in U n×k corresponds to a principal component feature vector and can be determined as a column vector. By multiplying the row vector and the column vector, the corresponding target training sample can be obtained. Correspondingly, the target indicator is obtained by processing the normalized roughness parameter indicator through the principal component feature vector, and the target training sample can also include the roughness label described above.

[0101] In step 104, the electronic device can use the m target training samples to train the pre-established initial machine learning model to obtain the joint roughness prediction model.

[0102] The initial machine learning model includes an initial XGBoost model and an intelligent optimization algorithm model. In some examples, the intelligent optimization algorithm can be a White Shark Optimizer (WSO), a whale algorithm, or a differential evolution algorithm, etc. For simplicity of description, the intelligent optimization algorithm will be mainly described as WSO in the following. The initial machine learning model including the initial XGBoost model and the WSO algorithm can be referred to as an untrained WSO-XGB model, and the joint roughness prediction model is referred to as a fully trained WSO-XGB model.

[0103] ​​During the training of the WSO-XGB model, the initial XGBoost model itself receives target training samples, outputs predicted roughness, and adjusts its network parameters based on the predicted roughness and roughness labels. The WSO model can adjust the hyperparameters of the initial XGBoost model, such as adjusting at least one of the maximum number of iterations, depth, and learning rate. The parameters of the WSO model, such as the population size or the maximum number of iterations, can be preset before training.

[0104] In some implementations, step 104, training a pre-established initial machine learning model using m target training samples to obtain a joint roughness prediction model, includes:

[0105] Divide the m target training samples into a training set and a test set;

[0106] The initial machine learning model is trained using the training set to obtain the joint roughness prediction model;

[0107] The performance of the joint roughness prediction model was evaluated using a test set.

[0108] Using the initial machine learning model as an untrained WSO-XGB model as an example, the training process will be explained below with a specific application example. Figure 2 As shown, Figure 2 This is a flowchart illustrating the training process of a joint roughness prediction model.

[0109] A database related to joints can be established in electronic devices. To create a comprehensive database, in addition to 10 standard profiles, multiple digitized joint profiles can be collected from existing publicly available databases. The projected lengths of these joints range from 72 to 119.6 mm, and the joint profiles cover a wide range of rock types, including sandstone, limestone, marble, granite, gneiss, slate, dolomite, and siltstone. These eight rock types correspond to igneous, metamorphic, and sedimentary rocks, respectively. The JRC values ​​for these profiles are obtained by substituting the direct shear test results into existing JRC calculation formulas. Based on the above digitized joint profiles, the average relative height (R) is calculated. ave ), maximum relative height (R) max Standard deviation of height (SD) h ), mean tilt angle (i ave ), Standard deviation of tilt angle (SD) i ), root mean square slope (Z2), roughness profile index (R) p), profile structure function (SF), etc. The above eight roughness statistical parameters can be used as roughness parameter indicators of the initial training samples, and the calculated JRC value can be used as the roughness label of the initial training samples.

[0110] The principal component analysis is performed on the set of initial training samples to obtain target training samples, and the specific implementation process is described in detail in the above embodiment, which will not be repeated here. In this application example, the number of target indicators in the target training samples can be two, denoted as F1 and F2.

[0111] To avoid overfitting in the model training process, the target training samples can be randomly divided into two groups, i.e., a training set and a test set, wherein the training set can account for 90%, and the test set can account for 10%. Of course, the ratio of the training set and the test set can be adjusted according to actual needs.

[0112] The training set can be input into the initial XGBoost model (corresponding to the XGB model), and the network parameters can be adjusted according to the roughness label carried in the training set and the predicted roughness output by the initial XGBoost model itself. The WSO model can adjust the hyperparameters of the initial XGBoost model. The WSO model and the initial XGBoost model can constitute an untrained WSO-XGB model. When the loss function in the WSO-XGB model converges during the training process of the WSO-XGB model, it is considered that the training requirement is met, and the final WSO-XGB model, i.e., the joint roughness prediction model described above, is obtained, which can be used to predict JRC in actual application. The test set can be used to evaluate the performance of the joint roughness prediction model.

[0113] Based on the above application example, the joint roughness prediction model training method provided by the embodiments of the present application acquires the roughness statistical parameters of the joint, and establishes a database of initial training samples. The initial training samples contain eight input variables and one output variable (JRC). Principal component analysis is introduced to reduce the dimensionality of the indicator system, and two independent principal components (corresponding to target indicators) are extracted to solve the overlap problem of roughness representation. Then, the WSO algorithm is introduced to optimize the XGBoost model, and a hybrid machine learning prediction model (i.e., WSO-XGB model) is proposed to predict JRC. Based on the extraction of the two principal components, the established database is used to train and test the WSO-XGB model. Compared with the empirical model, the WSO-XGB model is data-driven, not only can extract features, but also has strong learning ability, so that the prediction result is more reliable, the prediction accuracy of JRC is effectively improved, and the model has good generalization ability.

[0114] As Figure 3As shown in the embodiments of this application, a method for predicting joint roughness is also provided, including:

[0115] Step 301: Obtain the sample to be predicted. The sample to be predicted includes n roughness parameter indices, where n is an integer greater than 1.

[0116] Step 302: Process the sample to be predicted using k principal component eigenvectors to obtain preprocessed samples. The k principal component eigenvectors are obtained based on principal component analysis, and k is a positive integer.

[0117] Step 303: Output the preprocessed sample to the joint roughness prediction model to obtain the joint roughness coefficient. The joint roughness prediction model is trained by the joint roughness prediction model training method provided in the above embodiment.

[0118] In this embodiment, the joint roughness prediction model is trained based on the joint roughness prediction model training method provided in the above embodiment. For the same reasons as the above training method, the obtained joint roughness prediction model can effectively improve the prediction accuracy of JRC.

[0119] like Figure 4 As shown in the embodiments of this application, a joint roughness prediction model training device is also provided, including:

[0120] The first acquisition module 401 is used to acquire m initial training samples, each initial training sample including n roughness parameter indices and roughness labels, where m and n are both integers greater than 1;

[0121] Principal component analysis module 402 is used to perform principal component analysis on m initial training samples to obtain k principal component feature vectors, where k is a positive integer.

[0122] The first processing module 403 is used to process m initial training samples based on k principal component feature vectors to obtain m target training samples. Each target training sample includes k target indicators and roughness labels. The target indicators are obtained by processing the roughness parameter indicators through the principal component feature vectors.

[0123] Training module 404 is used to train a pre-established initial machine learning model using m target training samples to obtain a joint roughness prediction model. The initial machine learning model includes an initial XGBoost model and an intelligent optimization algorithm model. The initial XGBoost model is used to receive target training samples, output predicted roughness, and adjust network parameters according to the predicted roughness and roughness label. The intelligent optimization algorithm model is used to adjust the hyperparameters of the initial XGBoost model.

[0124] Optionally, the principal component analysis module 402 may be specifically used for:

[0125] constructing an original data matrix of m*n order according to the m initial training samples and the n roughness parameter indexes included in each initial training sample;

[0126] constructing a correlation coefficient matrix according to the original data matrix;

[0127] solving eigenvalues and eigenvectors of the correlation coefficient matrix;

[0128] determining k principal component eigenvectors according to the eigenvalues and eigenvectors of the correlation coefficient matrix.

[0129] Optionally, the principal component analysis module 402 can be further configured to:

[0130] normalizing and standardizing the original data matrix to obtain a standardized matrix;

[0131] determining the correlation coefficient matrix R according to the following formula m×n :

[0132]

[0133] wherein X * is the standardized matrix, and X *T is the transpose matrix of the standardized matrix.

[0134] Optionally, the principal component analysis module 402 can be further configured to:

[0135] arranging all eigenvalues of the correlation coefficient matrix in descending order, and determining the variance contribution rate corresponding to each eigenvalue;

[0136] adding the variance contribution rates in sequence according to the arrangement order of all eigenvalues;

[0137] in a case where the added value of the variance contribution rate is greater than a preset threshold for the first time, obtaining k eigenvalues corresponding to k added variance contribution rates;

[0138] determining k eigenvectors corresponding to the k eigenvalues as k principal component eigenvectors.

[0139] Optionally, the principal component analysis module 402 can be further configured to:

[0140] obtaining a correlation coefficient between any two roughness parameter indexes;

[0141] in a case where the correlation coefficient between any two roughness parameter indexes is greater than a coefficient threshold, constructing an original data matrix of m*n order according to the m initial training samples and the n roughness parameter indexes included in each initial training sample.

[0142] Optionally, the first processing module 403 can be specifically used for:

[0143] The row vector is determined according to the n roughness parameters included in each initial training sample, the principal component feature vector is taken as a column vector, and the target training sample is obtained by dot multiplication.

[0144] Optionally, the training module 404 can be specifically used for:

[0145] The m target training samples are divided into a training set and a test set;

[0146] The pre-established initial machine learning model is trained using the training set to obtain a joint roughness prediction model;

[0147] The performance of the joint roughness prediction model is evaluated using the test set.

[0148] Optionally, the n roughness parameter indicators include at least two of the following indicators: average relative height, maximum relative height, height standard deviation, average dip angle, dip angle standard deviation, root mean square of slope, roughness profile index, and profile structure function;

[0149] The intelligent optimization algorithm model includes a white shark optimization algorithm;

[0150] The hyperparameters include at least one of the maximum number of iterations, depth, and learning rate.

[0151] The joint roughness prediction model training device provided by the embodiments of the present application is a device permission corresponding to the joint roughness prediction model training method of the above embodiments. The method embodiments can be applied to the device embodiments and achieve the same technical effects. Here, it is not repeated.

[0152] The embodiments of the present application also provide a joint roughness prediction device, comprising:

[0153] The second acquisition module is configured to acquire a to-be-predicted sample, the to-be-predicted sample including n roughness parameter indicators, n being an integer greater than 1;

[0154] The second processing module is configured to process the to-be-predicted sample using k principal component feature vectors to obtain a preprocessed sample, the k principal component feature vectors being obtained based on principal component analysis, and k being a positive integer;

[0155] The prediction module is configured to output the preprocessed sample to a joint roughness prediction model to obtain a joint roughness coefficient, the joint roughness prediction model being trained by the joint roughness prediction model training method described above.

[0156] The joint roughness prediction device provided by the embodiments of the present application is a device right corresponding to the joint roughness prediction method of the above embodiments. The method embodiments can be applied to the device embodiments and achieve the same technical effects. Details are not described here.

[0157] The embodiments of the present application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor implements the joint roughness prediction model training method or the joint roughness prediction method when executing the computer program.

[0158] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the joint roughness prediction model training method or the joint roughness prediction method.

[0159] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified. In actual application, the above functions can be completed by different functional units and modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiments can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of software functional unit. In addition, the specific names of each functional unit and module are only for easy distinction, and do not limit the protection scope of the present application. The specific working process of the units and modules in the above system can refer to the corresponding process in the above method embodiments, which will not be described here.

[0160] In the above embodiments, the description of each embodiment has its own emphasis. The parts not described or recorded in detail in a certain embodiment can be referred to the related description of other embodiments.

[0161] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0162] In the embodiments of the present application, it should be understood that the disclosed apparatus / terminal device and method can be implemented in other manners. For example, the embodiments of the apparatus / terminal device described above are merely schematic; for example, the division of the modules or units is merely logical function division; an actual implementation can be another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between different units, can be indirect couplings or communication connections through some interfaces, devices or units, and can be electrical, mechanical or in other forms.

[0163] The units described as separated components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0164] In addition, each functional unit in the embodiments of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit. The integrated unit can be implemented in the form of hardware, or in the form of software function units.

[0165] If the integrated module / unit is implemented in the form of software function units and sold or used as an independent product, it can be stored in a computer readable storage medium. Based on such an understanding, all or part of the flow of the method in the above embodiments can also be implemented by a computer program instructing related hardware to complete, and the computer program can be stored in a computer readable storage medium. The computer program, when executed by a processor, can implement the steps of the above method embodiments. The computer program includes computer program code, which can be in the form of source code, object code, executable file or some intermediate form. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0166] The above examples are only used to illustrate the technical solutions of the present application, but not limit the same; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalent ones; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A joint roughness prediction model training method, characterized in that, The method comprises the following steps: obtaining m initial training samples, each of which comprises n roughness parameter indicators and a roughness label, m and n are integers greater than 1; performing principal component analysis on the m initial training samples to obtain k principal component feature vectors, k is a positive integer; processing the m initial training samples based on the k principal component feature vectors to obtain m target training samples, each of which comprises k target indicators and the roughness label, the target indicators being obtained by processing the roughness parameter indicators through the principal component feature vectors; training a pre-established initial machine learning model using the m target training samples to obtain a joint roughness prediction model, wherein the initial machine learning model comprises an initial XGBoost model and an intelligent optimization algorithm model, the initial XGBoost model is used to receive the target training samples, output a predicted roughness, and adjust network parameters according to the predicted roughness and the roughness label, and the intelligent optimization algorithm model is used to adjust the hyperparameters of the initial XGBoost model; wherein the principal component analysis on the m initial training samples to obtain k principal component feature vectors comprises: constructing an m*n order original data matrix according to the m initial training samples and the n roughness parameter indicators included in each of the initial training samples; constructing a correlation coefficient matrix according to the original data matrix; solving the eigenvalues and eigenvectors of the correlation coefficient matrix; determining the k principal component feature vectors according to the eigenvalues and eigenvectors of the correlation coefficient matrix; determining the k principal component feature vectors according to the eigenvalues and eigenvectors of the correlation coefficient matrix; arranging all the eigenvalues of the correlation coefficient matrix in descending order, and determining the variance contribution rate corresponding to each of the eigenvalues; sequentially accumulating the variance contribution rates according to the arrangement order of the eigenvalues; in the case that the value of the accumulated variance contribution rate is greater than a preset threshold for the first time, obtaining k eigenvalues corresponding to the k accumulated variance contribution rates; determining the k eigenvectors corresponding to the k eigenvalues as the k principal component feature vectors.

2. The method of claim 1, wherein, The constructing the correlation coefficient matrix according to the original data matrix comprises: performing normalization and standardization processing on the original data matrix to obtain a standardized matrix; determining the correlation coefficient matrix R according to the following formula m×n : where X * is the normalized matrix, X *T is the transpose matrix of the normalized matrix.

3. The method of claim 1, wherein, constructing an m*n order original data matrix according to the m initial training samples and the n roughness parameter indicators included in each of the initial training samples comprises: obtaining the correlation coefficient between any two roughness parameter indicators; in the case that the correlation coefficient between any two roughness parameter indicators is greater than a coefficient threshold, constructing an m*n order original data matrix according to the m initial training samples and the n roughness parameter indicators included in each of the initial training samples.

4. The method of claim 1, wherein, processing the m initial training samples based on the k principal component feature vectors to obtain m target training samples comprises: determining a row vector according to the n roughness parameters included in each of the initial training samples, taking each of the principal component feature vectors as a column vector, and obtaining the target training samples through dot product.

5. The method of claim 1, wherein, The training of the pre-established initial machine learning model using the m target training samples to obtain a joint roughness prediction model comprises: dividing the m target training samples into a training set and a test set; training the pre-established initial machine learning model using the training set to obtain a joint roughness prediction model; and evaluating the performance of the joint roughness prediction model using the test set.

6. The method of claim 1, wherein, The n roughness parameter indicators include at least two of the following indicators: average relative height, maximum relative height, height standard deviation, average dip angle, dip angle standard deviation, root mean square of slope, roughness profile index, and profile structure function; the intelligent optimization algorithm model includes a white shark optimization algorithm; and the hyperparameters include at least one of the maximum number of iterations, depth, and learning rate.

7. A method of joint roughness prediction, characterized by, The method comprises: obtaining a to-be-predicted sample, the to-be-predicted sample including n roughness parameter indicators, n being an integer greater than 1; processing the to-be-predicted sample using k principal component feature vectors to obtain a preprocessed sample, the k principal component feature vectors being obtained based on principal component analysis, k being a positive integer; and outputting the preprocessed sample to a joint roughness prediction model to obtain a joint roughness coefficient, the joint roughness prediction model being trained by the method of any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the joint roughness prediction model training method of any one of claims 1 to 6 or the joint roughness prediction method of claim 7.

Images