A Method and System for Predicting Rock Shear Strength Based on Improved Machine Learning Algorithm

Abstract

This invention relates to the field of data analysis technology, specifically to a method and system for predicting rock shear strength based on an improved machine learning algorithm. A rock shear strength prediction system based on an improved machine learning algorithm includes a data acquisition module and a rock shear strength analysis module. This invention predicts rock shear strength using readily available longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength, eliminating the need for complex triaxial testing. Furthermore, a physical embedding input layer is incorporated into the rock shear strength analysis model, expanding the input features using prior knowledge of physical rules, thus solving the problem of low prediction accuracy caused by data sparsity and making the prediction more accurate. A knowledge enhancement adjustment layer is also included, using rock shear strength prediction methods under different scenarios as prior knowledge to enhance features, improving the accuracy of rock shear strength prediction while maintaining interpretability.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and specifically to a method and system for predicting rock shear strength based on an improved machine learning algorithm. Background Technology

[0002] Rock shear strength parameters (internal friction angle and cohesion) are crucial fundamental parameters in geotechnical engineering design, traditionally obtained primarily through triaxial compression tests or direct shear tests. However, these conventional testing methods have several limitations: First, during the test, personnel must closely observe the rock failure process under high pressure, bearing significant safety risks, such as equipment explosion or high-speed ejection of rock fragments due to pressure system failure; second, test results under field conditions often exhibit uncertainty due to equipment limitations, environmental interference (such as temperature fluctuations and vibration), and human operational deviations, reducing the accuracy and reliability of the parameters; third, standard triaxial tests are typically time-consuming (a single test may take several days to complete), costly (requiring specialized equipment and technicians), and have stringent requirements for sample preparation.

[0003] In engineering practice, especially during the early investigation phase of a project, geotechnical engineers often face constraints such as tight deadlines and limited budgets, making it impossible to conduct sufficient triaxial tests. For example, in the feasibility study phase of tunnels, slopes, and underground engineering, design teams often need to quickly assess rock mass stability and develop preliminary support schemes without complete triaxial test data. In such cases, engineers are forced to rely on empirical formulas or simplified assumptions to estimate the internal friction angle and cohesion, and the uncertainty of this method significantly increases engineering risk. Summary of the Invention

[0004] This invention predicts rock shear strength using readily available longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength, eliminating the need for complex triaxial testing. Furthermore, the rock shear strength analysis model incorporates a physical embedding input layer, expanding the input features using prior knowledge of physical rules to address the low prediction accuracy caused by data sparsity. Physical rules also constrain the predictions, making them more accurate. A knowledge enhancement adjustment layer further strengthens the features by using rock shear strength prediction methods under different scenarios as prior knowledge, improving the accuracy of rock shear strength prediction while maintaining interpretability. A scale mapping layer adaptively transforms the scale differences between the rock sample and the rock mass, addressing the significant discrepancies in the predicted shear strength between the rock sample and the actual rock mass. In summary, the rock shear strength analysis model proposed in this application not only possesses high interpretability but also improves the accuracy of rock shear strength prediction under the guidance of professional knowledge in the field of rock shear analysis.

[0005] This invention provides a method for predicting the shear strength of rocks based on an improved machine learning algorithm, comprising:

[0006] The rock analysis vector and scale analysis vector are fed into the rock shear strength analysis model for processing, and the rock shear strength data is output.

[0007] The rock shear strength analysis model comprises a physical embedding input layer, a knowledge enhancement adjustment layer, a scale mapping layer, and an output layer. The physical embedding input layer obtains theoretical and reference rock shear strength data based on physical rules and rock analysis vectors, and constructs physically embedded rock analysis vectors based on these data. The knowledge enhancement adjustment layer processes the physically embedded rock analysis vectors through several expert analysis units, outputting corresponding expert-enhanced rock analysis vectors. Feature fusion is then performed on all expert-enhanced rock analysis vectors to construct knowledge-enhanced rock analysis vectors. The scale mapping layer performs scale transformation on the knowledge-enhanced rock analysis vectors based on the scale analysis vectors, constructing scale-mapped rock analysis vectors for more accurate rock shear strength prediction. The output layer performs a fully connected operation on the scale-mapped rock analysis vectors to output the rock shear strength data.

[0008] As a preferred aspect, the theoretical rock shear strength data and reference rock shear strength data are obtained through a physical embedding input layer based on physical rules and rock analysis vectors, and a physical embedding rock analysis vector is constructed based on the theoretical rock shear strength data and reference rock shear strength data, specifically including the following steps:

[0009] The theoretical internal friction angle and theoretical cohesion are calculated based on the Mohr-Coulomb criterion, and the theoretical internal friction angle and theoretical cohesion are combined to form theoretical rock shear strength data.

[0010] Traverse the historical rock analysis database and calculate the similarity between historical rock analysis data and rock analysis vectors in the database. If the similarity between historical rock analysis data and rock analysis vectors in the database is higher than the similarity threshold, use the historical rock shear strength data corresponding to the historical rock analysis data as the reference rock shear strength data. The reference rock shear strength data includes the reference internal friction angle and reference cohesion. Continue until all historical rock analysis data in the historical rock analysis database has been traversed, and then output all reference rock shear strength data.

[0011] The rock analysis vector, theoretical rock shear strength data, and all reference rock shear strength data are concatenated end-to-end to construct a reconstructed rock analysis vector. A physical correlation matrix is ​​then built based on this reconstructed rock analysis vector. The i-th row and j-th column of the physical correlation matrix stores the physical correlation strength value between the i-th data point and the j-th data point in the reconstructed rock analysis vector, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., N, and N is the total number of data points in the reconstructed rock analysis vector. A self-attention mechanism is then applied to the reconstructed rock analysis vector based on the physical correlation matrix to strengthen it, thus constructing a physically embedded rock analysis vector.

[0012] As a preferred aspect, a self-attention mechanism is applied to the reconstructed rock analysis vector based on the physical correlation matrix to construct a physically embedded rock analysis vector. Specifically, this includes the following steps: performing matrix multiplication operations on the reconstructed rock analysis vector with the value weight matrix and the key weight matrix respectively to construct the corresponding rock analysis value vector V and rock analysis key vector K; performing matrix multiplication operations on the physical correlation matrix with the query weight matrix to construct the corresponding rock analysis query vector Q; and implementing the self-attention mechanism using the following formula: G = softmax(QK) T / D 0.5 ), where G is the physical embedding rock analysis vector, T is the matrix transpose operation, and D is the dimension of the rock analysis bond vector K.

[0013] As a preferred aspect, the knowledge-enhanced adjustment layer processes the physically embedded rock analysis vectors through several expert analysis units, outputting corresponding expert-enhanced rock analysis vectors, and performs feature fusion based on all expert-enhanced rock analysis vectors to construct knowledge-enhanced rock analysis vectors, specifically including the following steps:

[0014] For each expert analysis unit, perform the following steps: Calculate the forget gate for the physically embedded rock analysis vector: f t =σ(W f U+b f ), where f t Let W be the forgetting gate vector, σ be the sigmoid function, and W be the forgetting gate vector. f Let b be the forget gate weight matrix. f Here, U is the forget gate bias value, and U is the physically embedded rock analysis vector. The input gate is calculated for the physically embedded rock analysis vector: r t =σ(W r U+b r ), where r t = represents the input gate vector, W r Let b be the input gate weight matrix. r The input gate bias value is used to complete feature transfer using the following formula: H = f t*U+r t *U, where H is the corresponding expert-enhanced rock analysis vector, and * is the dot product operation;

[0015] The physically embedded rock analysis vectors are fed into the expert weight analysis network for processing to obtain the expert weights corresponding to each expert analysis unit. All expert-enhanced rock analysis vectors and their corresponding expert weights are then weighted and fused through the feature fusion unit in the knowledge enhancement adjustment layer to obtain the knowledge-enhanced rock analysis vectors.

[0016] As a preferred aspect, the knowledge-enhanced rock analysis vector is scaled based on the scale analysis vector by the scale mapping layer to construct the scale-mapped rock analysis vector. Specifically, the steps are as follows: the knowledge-enhanced rock analysis vector and the scale analysis vector are concatenated and fed into the scale mapping network for processing, and the scale transformation matrix is ​​output. Then, the knowledge-enhanced rock analysis vector and the scale transformation matrix are subjected to a dot product operation to construct the scale-mapped rock analysis vector.

[0017] As a preferred approach, training the rock shear strength analysis model involves the following steps:

[0018] Several rock shear strength analysis training samples were obtained. Each rock shear strength analysis training sample included a rock analysis vector, a scale analysis vector, and the corresponding rock shear strength data. All rock shear strength analysis training samples were divided into several scenario rock shear strength analysis training sets according to the scenario corresponding to the expert analysis unit. For each scenario rock shear strength analysis training set, the following training was performed: a pre-trained model was constructed, which included a physical embedding input layer, a knowledge enhancement adjustment layer, a scale mapping layer, and an output layer. Only one expert analysis unit was set in the knowledge enhancement adjustment layer. The pre-trained model was trained using the scenario rock shear strength analysis training sets. The training target was the rock shear strength data in the rock shear strength analysis training samples.

[0019] All pre-trained models are acquired, and the expert analysis units in all pre-trained models are concatenated in parallel. These concatenated units are then combined with the expert weight analysis network and feature fusion unit to form a new knowledge enhancement adjustment layer. Based on this new knowledge enhancement adjustment layer, a complete rock shear strength analysis model is constructed. All rock shear strength analysis training samples are combined to form a rock shear strength analysis training set. The complete rock shear strength analysis model is then trained using this training set, with the training objective being the rock shear strength data in the rock shear strength analysis training samples.

[0020] This invention also provides a rock shear strength prediction system based on an improved machine learning algorithm, comprising:

[0021] The data acquisition module combines the longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength of the rock sample into a rock analysis vector, and combines the scale of the rock sample with the scale of the rock mass into a scale analysis vector.

[0022] The rock shear strength analysis module is used to input rock analysis vectors and scale analysis vectors into the rock shear strength analysis model for processing and output rock shear strength data.

[0023] The present invention has the following advantages:

[0024] This invention predicts rock shear strength using readily available longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength, eliminating the need for complex triaxial testing. Furthermore, the rock shear strength analysis model incorporates a physical embedding input layer, expanding the input features using prior knowledge of physical rules to address the low prediction accuracy caused by data sparsity. Physical rules also constrain the predictions, making them more accurate. A knowledge enhancement adjustment layer further strengthens the features by using rock shear strength prediction methods under different scenarios as prior knowledge, improving the accuracy of rock shear strength prediction while maintaining interpretability. A scale mapping layer adaptively transforms the scale differences between the rock sample and the rock mass, addressing the significant discrepancies in the predicted shear strength between the rock sample and the actual rock mass. In summary, the rock shear strength analysis model proposed in this application not only possesses high interpretability but also improves the accuracy of rock shear strength prediction under the guidance of professional knowledge in the field of rock shear analysis. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the rock shear strength analysis model used in an embodiment of the present invention.

[0026] Figure 2 This is a schematic diagram of the rock shear strength prediction system based on an improved machine learning algorithm used in an embodiment of the present invention. Detailed Implementation

[0027] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this invention.

[0028] Example 1: A method for predicting rock shear strength based on an improved machine learning algorithm, comprising:

[0029] The rock analysis vector is composed of the longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength corresponding to the rock sample. The scale analysis vector is composed of the scale of the rock sample and the scale of the rock mass. It should be noted that, in order to indirectly analyze the rock shear strength, this application selects the easily obtainable longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength as inputs, avoiding the complex work of triaxial testing. The scale is characterized by the corresponding diameter. The rock analysis vector and the scale analysis vector are fed into the rock shear strength analysis model for processing, and the rock shear strength data is output. The rock shear strength data includes the internal friction angle and cohesion. The internal friction angle and cohesion are key characteristic parameters characterizing the rock shear strength.

[0030] Here is a brief explanation of how to obtain the longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength. The longitudinal wave velocity is calculated by placing an ultrasonic source in one hole and a receiver in the other in adjacent boreholes of the rock sample. The density of the rock sample is measured by measuring the density of the rock sample. The uniaxial compressive strength and uniaxial tensile strength of the rock sample are measured by a point load tester.

[0031] See Figure 1The rock shear strength analysis model includes a physical embedding input layer, a knowledge enhancement adjustment layer, a scale mapping layer, and an output layer. The physical embedding input layer calculates the rock analysis vector based on physical rules, constructs corresponding theoretical rock shear strength data, and obtains reference rock shear strength data from a historical rock analysis database based on the rock analysis vector. The rock analysis vector is then expanded using the theoretical and reference rock shear strength data to construct a reconstructed rock analysis vector. Finally, a corresponding physical correlation matrix is ​​constructed to enhance the reconstructed rock analysis vector, resulting in a physically embedded rock analysis vector. It should be noted that, due to the rock... The rock analysis vector only has four data points, leading to data sparsity issues during subsequent machine learning analysis and resulting in poor actual analysis performance. Therefore, in the physical embedding input layer, the rock analysis vector is expanded using theoretical rock shear strength data and reference rock shear strength data to increase the effective feature dimensions, reduce analysis errors, and construct theoretical rock shear strength data based on physical rules to establish physical relationship constraints between effective features. In addition to data-driven approaches, physical constraint-driven approaches are also implemented based on physical correlation matrices, improving the accuracy of rock shear strength prediction. The knowledge enhancement adjustment layer uses several expert analysis units to further refine the physical... The process embeds rock analysis vectors for processing, outputting corresponding expert-enhanced rock analysis vectors. Feature fusion is then performed based on all expert-enhanced rock analysis vectors to construct knowledge-enhanced rock analysis vectors. Here, the expert analysis units correspond to the prediction of rock shear strength under different scenarios, such as scenarios where the rock is sandstone and its variants, granite, mudstone, or dolomite, or scenarios where the rock is under the influence of groundwater or high ground stress. Since the prediction method for rock shear strength varies depending on the scenario, this application sets different expert analysis units to fit the shear strength prediction under different scenarios, and uses the expert analysis units as... To enhance the features embedded in the rock analysis vector with prior knowledge, the system can improve the accuracy of rock shear strength prediction while maintaining interpretability. The scale mapping layer is used to scale the knowledge-enhanced rock analysis vector based on the scale analysis vector to construct a scale-mapped rock analysis vector. Since there are significant differences in scale between rock samples and actual rock masses, the corresponding shear strength predictions also differ significantly. Therefore, it is necessary to scale the rock analysis vector to obtain a more accurate prediction of rock shear strength. The output layer performs a fully connected operation on the scale-mapped rock analysis vector to output the rock shear strength data.

[0032] This application predicts rock shear strength using readily available P-wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength, eliminating the need for complex triaxial testing. Furthermore, the rock shear strength analysis model incorporates a physical embedding input layer, extending input features with prior knowledge of physical rules to address the low prediction accuracy issue caused by data sparsity. Physical rules also provide constraints, enhancing prediction accuracy. A knowledge enhancement adjustment layer uses rock shear strength prediction methods under different scenarios as prior knowledge to strengthen features, improving accuracy while maintaining interpretability. A scale mapping layer adaptively transforms the scale differences between rock samples and rock masses, addressing the significant discrepancies in predicted shear strength between rock samples and actual rock masses. In summary, the rock shear strength analysis model proposed in this application not only possesses high interpretability but also improves the accuracy of rock shear strength prediction under the guidance of professional knowledge in the field of rock shear analysis.

[0033] Theoretical rock shear strength data and reference rock shear strength data are obtained through a physical embedding input layer based on physical rules and rock analysis vectors. Then, a physical embedding rock analysis vector is constructed based on the theoretical rock shear strength data and the reference rock shear strength data. Specifically, the steps include:

[0034] The theoretical internal friction angle and theoretical cohesion are calculated based on the Mohr-Coulomb criterion. The specific calculation method is as follows: theoretical internal friction angle φ_theoretical = arcsin[(UCS-UTS) / (UCS+UTS)], where UCS is the uniaxial compressive strength and UTS is the uniaxial tensile strength, and theoretical cohesion c_theoretical = UCS(1-sinφ_theoretical) / (2cosφ_theoretical). The theoretical internal friction angle and theoretical cohesion are combined to form the theoretical rock shear strength data.

[0035] The process iterates through the historical rock analysis database, calculating the similarity between historical rock analysis data and rock analysis vectors. A cosine similarity algorithm can be used for this calculation. The historical rock analysis data refers to historically collected P-wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength. If the similarity between the historical rock analysis data and the rock analysis vectors exceeds a similarity threshold (which is manually set), the corresponding historical rock shear strength data is used as reference rock shear strength data. This reference shear strength data includes reference internal friction angle and reference cohesion. It should be noted that this historical rock shear strength data is obtained from triaxial tests or output from a rock shear strength analysis model, and the historical rock shear strength data has confidence levels. This process continues until all historical rock analysis data in the database has been traversed, outputting all reference rock shear strength data. These reference rock shear strength data can provide prior knowledge of similar rock shear strength while expanding features.

[0036] The rock analysis vector, theoretical rock shear strength data, and all reference rock shear strength data are concatenated end-to-end to construct a reconstructed rock analysis vector. Based on this reconstructed vector, a physical correlation matrix is ​​built. This matrix stores the physical correlation strength values ​​between the data points in the reconstructed rock analysis vector. These physical correlation strength values ​​refer to the correlation between various physical quantities. For example, longitudinal wave velocity and density are correlated in elastic wave propagation theory, and uniaxial compressive strength and uniaxial tensile strength are correlated in brittle material failure theory. The physical correlation strength values ​​are determined by experts using a fuzzy comprehensive evaluation method. The i-th row and j-th column of the physical correlation matrix stores the physical correlation strength value between the i-th data point and the j-th data point in the reconstructed rock analysis vector, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., N, and N is the total number of data points in the reconstructed rock analysis vector. These data points refer to P-wave velocity, density, uniaxial compressive strength, uniaxial tensile strength, theoretical internal friction angle, theoretical cohesion, internal friction angle, and reference cohesion. Based on the physical correlation matrix, a self-attention mechanism is applied to the reconstructed rock analysis vector to strengthen it and construct a physically embedded rock analysis vector.

[0037] A self-attention mechanism is applied to the reconstructed rock analysis vector based on the physical correlation matrix to construct a physically embedded rock analysis vector, specifically including the following steps:

[0038] The reconstructed rock analysis vectors are multiplied by the value weight matrix and the key weight matrix respectively to construct the corresponding rock analysis value vector V and rock analysis key vector K. The physical association matrix is ​​multiplied by the query weight matrix to construct the corresponding rock analysis query vector Q. The self-attention mechanism is implemented through the following formula: G = softmax(QK) T / D 0.5 ), where G is the physically embedded rock analysis vector, T is the matrix transpose operation, and D is the dimension of the rock analysis key vector K; by performing a self-attention mechanism on the reconstructed rock analysis vector through the physical correlation matrix, the data in the reconstructed rock analysis vector that conforms to the physical correlation relationship can be strengthened, while the data that does not conform to the physical correlation relationship can be suppressed, making the rock shear strength prediction process more in line with physical constraints;

[0039] The knowledge-enhanced adjustment layer processes the physically embedded rock analysis vectors through several expert analysis units, outputting corresponding expert-enhanced rock analysis vectors. Based on all expert-enhanced rock analysis vectors, feature fusion is performed to construct the knowledge-enhanced rock analysis vectors. Specifically, the layer includes the following steps:

[0040] For each expert analysis unit, perform the following steps: Calculate the forget gate for the physically embedded rock analysis vector: f t =σ(W f U+b f ), where f t Let W be the forgetting gate vector, σ be the sigmoid function, and W be the forgetting gate vector. f Let b be the forget gate weight matrix. f Here, U is the forget gate bias value, and U is the physically embedded rock analysis vector. The input gate is calculated for the physically embedded rock analysis vector: r t =σ(W r U+b r ), where r t = represents the input gate vector, W r Let b be the input gate weight matrix. r The input gate bias value is used to complete feature transfer using the following formula: H = f t *U+r t *U, where H is the corresponding expert-enhanced rock analysis vector, and * is the dot product operation; through the forget gate and the input gate, the features that are not of interest and the features that need to be retained in the scene corresponding to the expert analysis unit can be determined respectively, so as to realize the feature analysis process of the corresponding scene;

[0041] The physically embedded rock analysis vectors are fed into an expert weight analysis network for processing to obtain the expert weights corresponding to each expert analysis unit. The expert weight analysis network is based on a multilayer perceptron, and the obtained expert weights can represent the proportion of the scene in which the corresponding rock sample is located. All expert-enhanced rock analysis vectors and their corresponding expert weights are weighted and fused through the feature fusion unit in the knowledge enhancement adjustment layer to obtain knowledge-enhanced rock analysis vectors.

[0042] The scale-mapped rock analysis vector is constructed by performing a scale transformation on the knowledge-enhanced rock analysis vector based on the scale analysis vector through a scale mapping layer. The specific steps include the following:

[0043] The knowledge-enhanced rock analysis vector and the scale analysis vector are concatenated and then fed into the scale mapping network for processing, outputting a scale transformation matrix. This scale mapping network is also based on a multilayer perceptron. The knowledge-enhanced rock analysis vector and the scale transformation matrix are then subjected to a dot product operation to construct a scale-mapped rock analysis vector. The knowledge-enhanced rock analysis vector is then processed by the scale transformation matrix to map it to the scale of the rock mass, which makes the prediction of rock shear strength more consistent with the actual scale of the rock mass.

[0044] Training the rock shear strength analysis model involves the following steps:

[0045] Several rock shear strength analysis training samples were obtained. These samples included rock analysis vectors, scale analysis vectors, and corresponding rock shear strength data. These vectors were obtained by operators through actual triaxial tests. All rock shear strength analysis training samples were divided into several scenario-based rock shear strength analysis training sets according to the scenarios corresponding to the expert analysis units. For each scenario-based training set, the following training was performed: A pre-trained model was constructed, comprising a physical embedding input layer, a knowledge enhancement adjustment layer, a scale mapping layer, and an output layer. The knowledge enhancement adjustment layer contained only one expert analysis unit. The pre-trained model was trained using the scenario-based rock shear strength analysis training sets, with the training objective being the rock shear strength data in the training samples. The training conditions were then determined: the accuracy of the pre-trained model generally met expectations. If the training conditions were met, the trained pre-trained model was output; otherwise, the pre-trained model was continued to be trained using the scenario-based rock shear strength analysis training sets.

[0046] All pre-trained models are acquired, and the expert analysis units from all pre-trained models are concatenated in parallel. These concatenated units, along with the expert weight analysis network and feature fusion unit, form a new knowledge enhancement and adjustment layer. Based on this new knowledge enhancement and adjustment layer, a complete rock shear strength analysis model is constructed. All rock shear strength analysis training samples are combined into a rock shear strength analysis training set. The complete rock shear strength analysis model is trained using this training set, with the training objective being the rock shear strength data in the training samples. The training conditions are then determined: the accuracy of the rock shear strength analysis model generally meets expectations. If the training conditions are met, the trained rock shear strength analysis model is output; otherwise, the complete rock shear strength analysis model continues to be trained using the rock shear strength analysis training set.

[0047] Example 2: A rock shear strength prediction system based on an improved machine learning algorithm (see [link]). Figure 2 ,include:

[0048] The data acquisition module combines the longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength of the rock sample into a rock analysis vector, and combines the scale of the rock sample with the scale of the rock mass into a scale analysis vector. It should be noted that, in order to indirectly analyze the rock shear strength, this application selects easily obtainable longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength as inputs, avoiding the complex work of triaxial testing, and the scale is characterized by the corresponding diameter.

[0049] The rock shear strength analysis module is used to input rock analysis vectors and scale analysis vectors into the rock shear strength analysis model for processing and output rock shear strength data. The rock shear strength data includes the internal friction angle and cohesion, which are key characteristic parameters characterizing the rock shear strength.

[0050] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A method for predicting the shear strength of rock based on an improved machine learning algorithm, characterized in that, include: The rock analysis vector and scale analysis vector are fed into the rock shear strength analysis model for processing, and the rock shear strength data is output. The rock shear strength analysis model comprises a physical embedding input layer, a knowledge enhancement adjustment layer, a scale mapping layer, and an output layer. The physical embedding input layer obtains theoretical and reference rock shear strength data based on physical rules and rock analysis vectors, and constructs physically embedded rock analysis vectors based on these data. The knowledge enhancement adjustment layer processes the physically embedded rock analysis vectors through several expert analysis units, outputting corresponding expert-enhanced rock analysis vectors. Feature fusion is then performed on all expert-enhanced rock analysis vectors to construct knowledge-enhanced rock analysis vectors. The scale mapping layer performs scale transformation on the knowledge-enhanced rock analysis vectors based on the scale analysis vectors, constructing scale-mapped rock analysis vectors for more accurate rock shear strength prediction. The output layer performs a fully connected operation on the scale-mapped rock analysis vectors to output the rock shear strength data. Theoretical rock shear strength data and reference rock shear strength data are obtained through a physical embedding input layer based on physical rules and rock analysis vectors. Then, a physical embedding rock analysis vector is constructed based on the theoretical rock shear strength data and the reference rock shear strength data. Specifically, the steps include: The theoretical internal friction angle and theoretical cohesion are calculated based on the Mohr-Coulomb criterion, and the theoretical internal friction angle and theoretical cohesion are combined to form theoretical rock shear strength data. Traverse the historical rock analysis database and calculate the similarity between historical rock analysis data and rock analysis vectors in the database. If the similarity between historical rock analysis data and rock analysis vectors in the database is higher than the similarity threshold, use the historical rock shear strength data corresponding to the historical rock analysis data as the reference rock shear strength data. The reference rock shear strength data includes the reference internal friction angle and reference cohesion. Continue until all historical rock analysis data in the historical rock analysis database has been traversed, and then output all reference rock shear strength data. The rock analysis vector, theoretical rock shear strength data, and all reference rock shear strength data are concatenated end-to-end to construct a reconstructed rock analysis vector. A physical correlation matrix is ​​then built based on this reconstructed rock analysis vector. The i-th row and j-th column of the physical correlation matrix stores the physical correlation strength value between the i-th data point and the j-th data point in the reconstructed rock analysis vector, where i = 1, 2, 3, ..., N, j = 1, 2, 3, ..., N, and N is the total number of data points in the reconstructed rock analysis vector. A self-attention mechanism is then applied to the reconstructed rock analysis vector based on the physical correlation matrix to strengthen it, thus constructing a physically embedded rock analysis vector.

2. The rock shear strength prediction method based on an improved machine learning algorithm according to claim 1, characterized in that, A self-attention mechanism is applied to the reconstructed rock analysis vector based on the physical association matrix to construct a physically embedded rock analysis vector. Specifically, this involves the following steps: Multiplying the reconstructed rock analysis vector with the value weight matrix and the key weight matrix respectively to construct the corresponding rock analysis value vector V and rock analysis key vector K; multiplying the physical association matrix with the query weight matrix to construct the corresponding rock analysis query vector Q; and implementing the self-attention mechanism using the following formula: G = softmax(QK) T / D 0.5 ), where G is the physical embedding rock analysis vector, T is the matrix transpose operation, and D is the dimension of the rock analysis bond vector K.

3. The rock shear strength prediction method based on an improved machine learning algorithm according to claim 2, characterized in that, The knowledge-enhanced adjustment layer processes the physically embedded rock analysis vectors through several expert analysis units, outputting corresponding expert-enhanced rock analysis vectors. Based on all expert-enhanced rock analysis vectors, feature fusion is performed to construct the knowledge-enhanced rock analysis vectors. Specifically, the layer includes the following steps: For each expert analysis unit, perform the following steps: Calculate the forget gate for the physically embedded rock analysis vector: f t =σ(W f U+b f ), where f t Let W be the forgetting gate vector, σ be the sigmoid function, and W be the forgetting gate vector. f Let b be the forget gate weight matrix. f Here, U is the forget gate bias value, and U is the physically embedded rock analysis vector. The input gate is calculated for the physically embedded rock analysis vector: r t =σ(W r U+b r ), where r t = represents the input gate vector, W r Let b be the input gate weight matrix. r The input gate bias value is used to complete feature transfer using the following formula: H = f t *U+r t *U, where H is the corresponding expert-enhanced rock analysis vector, and * is the dot product operation; The physically embedded rock analysis vectors are fed into the expert weight analysis network for processing to obtain the expert weights corresponding to each expert analysis unit. All expert-enhanced rock analysis vectors and their corresponding expert weights are then weighted and fused through the feature fusion unit in the knowledge enhancement adjustment layer to obtain the knowledge-enhanced rock analysis vectors.

4. The rock shear strength prediction method based on the improved machine learning algorithm according to claim 3, characterized in that, The scale-mapped rock analysis vector is constructed by scaling the knowledge-enhanced rock analysis vector based on the scale analysis vector through a scale mapping layer. The specific steps are as follows: the knowledge-enhanced rock analysis vector and the scale analysis vector are concatenated and fed into the scale mapping network for processing, and the scale transformation matrix is ​​output. Then, the knowledge-enhanced rock analysis vector and the scale transformation matrix are subjected to a dot product operation to construct the scale-mapped rock analysis vector.

5. The rock shear strength prediction method based on an improved machine learning algorithm according to claim 4, characterized in that, Training the rock shear strength analysis model involves the following steps: Several rock shear strength analysis training samples were obtained. Each rock shear strength analysis training sample included a rock analysis vector, a scale analysis vector, and the corresponding rock shear strength data. All rock shear strength analysis training samples were divided into several scenario rock shear strength analysis training sets according to the scenario corresponding to the expert analysis unit. For each scenario rock shear strength analysis training set, the following training was performed: a pre-trained model was constructed, which included a physical embedding input layer, a knowledge enhancement adjustment layer, a scale mapping layer, and an output layer. Only one expert analysis unit was set in the knowledge enhancement adjustment layer. The pre-trained model was trained using the scenario rock shear strength analysis training sets. The training target was the rock shear strength data in the rock shear strength analysis training samples. All pre-trained models are acquired, and the expert analysis units in all pre-trained models are concatenated in parallel. These concatenated units are then combined with the expert weight analysis network and feature fusion unit to form a new knowledge enhancement adjustment layer. Based on this new knowledge enhancement adjustment layer, a complete rock shear strength analysis model is constructed. All rock shear strength analysis training samples are combined to form a rock shear strength analysis training set. The complete rock shear strength analysis model is then trained using this training set, with the training objective being the rock shear strength data in the rock shear strength analysis training samples.

6. A rock shear strength prediction system based on an improved machine learning algorithm, characterized in that, The system applies the rock shear strength prediction method based on the improved machine learning algorithm described in any one of claims 1-5, including: The data acquisition module combines the longitudinal wave velocity, density, uniaxial compressive strength, and uniaxial tensile strength of the rock sample into a rock analysis vector, and combines the scale of the rock sample with the scale of the rock mass into a scale analysis vector. The rock shear strength analysis module is used to input rock analysis vectors and scale analysis vectors into the rock shear strength analysis model for processing and output rock shear strength data.

Images