A method and system for quickly predicting the fatigue life of a rock anchor beam of a pumped storage power station

Abstract

This invention belongs to the field of hydraulic structure health monitoring technology, and discloses a method and system for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations. The method involves constructing a dataset for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations; performing feature engineering processing on the dataset; building a deep learning model for rapid prediction of the fatigue life of rock anchor beams based on the feature-engineered dataset; using the built deep learning model to predict the fatigue life of the rock anchor beams; and generating a fatigue life early warning report based on the predicted fatigue life of the rock anchor beams. This invention can significantly improve the efficiency and accuracy of rock anchor beam fatigue life prediction, breaking through the technical bottleneck of the time-consuming traditional numerical simulation method. Simultaneously, it identifies the main controlling factors through sensitivity analysis of influencing factors and combines an integrated early warning report to form a management model of prediction, analysis, and optimization, providing important technical support for the safe and stable operation of rock anchor beams in pumped storage power stations.

Description

Technical Field

[0001] This invention belongs to the field of hydraulic structure health monitoring technology, specifically relating to a method and system for rapid prediction of fatigue life of rock anchor beams in pumped storage power stations. Background Technology

[0002] Pumped storage power stations, as indispensable key facilities for peak shaving, valley filling, and frequency regulation energy storage in new power systems, play a vital role in the reliability of the power grid through their safe and stable operation. Rock anchor beams, as the core load-bearing structure of the underground powerhouse, bear the heavy load supporting the main equipment and hydraulic structures. They are subjected to alternating loads coupled from multiple fields, including unit start-up and shutdown loads, surrounding rock stress release, and seepage pressure fluctuations, making them highly susceptible to fatigue damage that can gradually propagate. In severe cases, this can lead to structural cracking, reduced load-bearing capacity, and other safety hazards, directly threatening the long-term service life of the power station.

[0003] Traditional methods for predicting the fatigue life of rock anchor beams often rely on a combination of numerical simulation and on-site monitoring, such as stress-strain analysis based on the finite element method and crack propagation rate calculation based on fracture mechanics. While these methods can reflect the fatigue characteristics of structures to some extent, they suffer from drawbacks such as long modeling cycles, high computational complexity, and cumbersome parameter calibration, making it difficult to meet the needs of rapid fatigue life assessment and real-time early warning in engineering projects. Furthermore, existing prediction models often focus on single influencing factors, failing to adequately consider the coupled effects of multiple factors such as surrounding rock mechanical parameters, anchoring system performance, and load condition combinations, leading to discrepancies between the predicted results and the actual service conditions.

[0004] With the rapid penetration of deep learning technology into the field of structural health monitoring, data-driven life prediction methods have gradually become a research hotspot due to their high efficiency and high accuracy. However, the service environment of rock anchor beams in pumped storage power stations is complex and influenced by multiple factors. Key technologies such as the construction of dedicated datasets for fatigue life prediction, feature engineering optimization, and deep learning model adaptation still need to be improved.

[0005] Therefore, developing a method for predicting the fatigue life of rock anchor beams in pumped storage power stations that combines speed and accuracy, constructing a prediction model driven by multi-source heterogeneous data, and forming an integrated technical system of data processing, model prediction, early warning and optimization are of great engineering value and practical significance for improving the safe operation and maintenance level of rock anchor beam structures and ensuring the stable operation of pumped storage power stations. Summary of the Invention

[0006] To address the problems existing in the prior art, this invention provides a rapid prediction method and system for the fatigue life of rock anchor beams in pumped storage power stations. This method can significantly improve the efficiency and accuracy of fatigue life prediction for rock anchor beams, breaking through the technical bottleneck of the time-consuming traditional numerical simulation method. At the same time, it identifies the main controlling factors through sensitivity analysis of influencing factors and forms a prediction, analysis, and optimization management model by combining an integrated early warning report, providing important technical support for the safe and stable operation of rock anchor beams in pumped storage power stations.

[0007] To achieve the above objectives, the present invention provides the following solution: A method for rapid prediction of fatigue life of rock anchor beams in pumped storage power stations, the method comprising: S1. Construct a dataset for rapid prediction of fatigue life of rock anchor beams in pumped storage power stations; S2. Perform feature engineering processing on the dataset; S3. Based on the dataset after feature engineering, a deep learning model for rapid prediction of fatigue life of rock anchor beams is built. S4. Use the completed deep learning model to predict the fatigue life of rock anchor beams. S5. Based on the predicted fatigue life of the rock anchor beam, generate a fatigue life early warning report.

[0008] Preferably, in step S1, the process of constructing a dataset for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations includes four stages: determination of influencing factor characteristics, sample data collection, dataset optimization, and data expansion.

[0009] Preferably, the influencing factors include the structural parameters, rock mechanics properties, anchor parameters, contact interface characteristics, and external load characteristics. The structural parameters include at least one of the following: the geometric dimensions of the rock anchor beam structure, concrete partitioning, and concrete elastic modulus. The rock mass mechanical properties include at least one of the following: rock mass integrity coefficient and rock mass elastic modulus. The anchor bolt parameters include anchor bolt material, anchoring depth, anchor bolt diameter, and anchor bolt spacing; The contact interface characteristics include the bond strength between the anchor bolt and the rock anchor beam, the bond strength between the anchor bolt and the rock mass, and the bond strength between the rock anchor beam and the rock mass. The external load characteristics include at least one of load amplitude and load frequency.

[0010] Preferably, the sample data is collected through numerical simulation, including: A geometric model of the interaction between the rock mass, rock anchor beam, and anchor rod was constructed based on the actual dimensions on site. The geometric model is discretized using the finite element method; The structural parameters, rock mechanics properties, anchor parameters, contact interface features, and external load features are input into the discretized geometric model. Considering the bonding characteristics between rock mass and anchor, rock anchor beam and anchor, and rock anchor beam and rock mass, the simulated value of the fatigue life of the corresponding rock anchor beam is calculated. Among them, the methods for calculating the simulated values ​​of the fatigue life of the rock anchor beam, considering the bond characteristics between the rock mass and the anchor bolt, the rock anchor beam and the anchor bolt, and the rock anchor beam and the rock mass, include: ; ; ; ; In the formula, This refers to the bond stress between the anchor bolt and the rock mass, and between the rock anchor beam; This refers to the relative slippage between the anchor bolt and the rock mass, or between the rock anchor beam and the rock anchor. ; ; ; These are parameters related to the rock mass, rock anchor beam strength, and steel bar diameter. The bonding characteristics between the rock anchor beam and the rock mass are determined through the contact friction relationship. Based on the SN curve of the anchor rod, the stress amplitude of the anchor rod element is mapped to the element fatigue life, which is used as the simulated value of the fatigue life of the corresponding rock anchor beam.

[0011] Preferably, the initial sample set undergoes preliminary preprocessing and screening to remove samples with outliers in fatigue life, thus optimizing the dataset. This includes: The Grubbs criterion was used to remove outliers in the fatigue life of rock anchor beams. The Grubbs statistic... The calculation formula is: ; in, The maximum standard deviation of a single sample. The standard deviation of the sample set For the first The logarithm of the fatigue life of a sample. This is the arithmetic mean of the logarithmic values ​​of the fatigue life of all samples. Grubbs statistic If the threshold is exceeded, the sample is removed.

[0012] Preferred methods for data augmentation include: For the effective sample set after screening by Grubbs criterion, a generative adversarial network is used to generate synthetic samples related to the fatigue life of rock anchor beams to supplement the dataset size and match the need for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations. The generative adversarial network (GAN) consists of a generator and a discriminator. Through minimax game optimization, it generates synthetic samples that match the characteristics of real rock anchor beam fatigue life samples. The objective function is: ; in, The standard objective function for generative adversarial networks, For generator, For discriminator, To ensure that the reconstruction loss function minimizes the feature deviation between the synthetic sample and the real sample, To fit the structural similarity loss function of the fatigue life sample characteristics of rock anchor beams, , For weight parameters, The output of the generator, For the input random noise, This is the effective sample set after being screened by the Grubbs criterion.

[0013] Preferably, in step S2, the method for performing feature engineering on the dataset includes: S21. Feature standardization: If there are non-linear feature parameters in the dataset, first use a kernel function to map the non-linear features to a high-dimensional linear space; then use the Z-score standardization method to convert all feature parameters in the dataset into standard normal distribution data with a mean of 0 and a variance of 1. S22. Feature dimensionality reduction: Principal component analysis is used to reduce the dimensionality of the standardized feature parameters, retaining the principal component features with a cumulative variance contribution rate of more than 90%, and obtaining the dimensionality-reduced feature matrix. S23. Feature partitioning: The dimensionality-reduced feature matrix is ​​randomly divided into training set, validation set and test set according to the proportion.

[0014] Preferably, in step S3, the deep learning model is a bidirectional long short-term memory network model based on an attention mechanism, and the construction process includes: S31. Construct the input layer, using the processed feature matrix as the input data of the input layer. The number of neurons in the input layer is consistent with the dimension of the reduced feature. S32. Construct a bidirectional long short-term memory network layer. Set up two bidirectional long short-term memory network layers. The number of neurons in the first bidirectional long short-term memory network layer is 64, and the number of neurons in the second bidirectional long short-term memory network layer is 32. Use the Dropout technique to suppress overfitting. The Dropout coefficient is set to 0.2. S33. Construct an attention layer and use a scaled dot product attention mechanism to assign weights to the output features of the bidirectional long short-term memory network layer. S34. Construct a fully connected layer. Set up two fully connected layers. The first fully connected layer has 16 neurons, and the second fully connected layer has 1 neuron and serves as the output layer to output the predicted fatigue life value of the rock anchor beam joint surface.

[0015] The present invention also provides a rapid prediction system for the fatigue life of rock anchor beams in pumped storage power stations. The system is used to implement the aforementioned method and includes: a dataset construction module, a dataset processing module, a model construction module, a prediction module, and an early warning module. The dataset construction module is used to construct a dataset for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations. The dataset processing module is used to perform feature engineering processing on the dataset; The model building module is used to build a deep learning model for rapid prediction of fatigue life of rock anchor beams based on the dataset after feature engineering. The prediction module is used to predict the fatigue life of rock anchor beams using the constructed deep learning model. The early warning module is used to generate a fatigue life early warning report based on the predicted fatigue life of the rock anchor beam.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention develops a method for predicting the fatigue life of rock anchor beams in pumped storage power stations that combines speed and accuracy. It constructs a prediction model driven by multi-source heterogeneous data and forms a technical system for data processing, model prediction, early warning, and optimization. This method can significantly improve the intelligence and precision of the safe operation and maintenance of rock anchor beam structures, effectively ensure the long-term stable operation of pumped storage power stations, and has important engineering application value and practical guiding significance for enhancing the full life cycle safety management and control capabilities of the core load-bearing structure of the power station and reducing operation and maintenance costs. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the rapid prediction method for the fatigue life of rock anchor beams in pumped storage power stations according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the cross-section of the rock anchor beam according to an embodiment of the present invention; Figure 3 This is a finite element mesh diagram of the rock anchor beam-anchor rod-foundation model according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the rock anchor beam grid according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the anchor mesh according to an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Example 1 This invention provides a method for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations, the method comprising: S1. Construct a dataset for rapid prediction of fatigue life of rock anchor beams in pumped storage power stations; S2. Perform feature engineering processing on the dataset; S3. Based on the dataset after feature engineering, a deep learning model for rapid prediction of fatigue life of rock anchor beams is built. S4. Use the completed deep learning model to predict the fatigue life of rock anchor beams. S5. Based on the predicted fatigue life of the rock anchor beam, generate a fatigue life early warning report. The early warning report shall include at least the predicted fatigue life value, the sensitivity ranking of each influencing factor, and targeted operation and maintenance optimization suggestions.

[0022] In this embodiment, in S1, the process of constructing the dataset includes four stages in sequence: determining the characteristics of influencing factors, collecting sample data, optimizing the dataset, and expanding the data, as detailed below: S11. Determine the characteristics of factors affecting the fatigue life of rock anchor beams in pumped storage power stations. The characteristics of these factors include structural parameters, rock mechanics properties, anchor parameters, contact interface characteristics, and external load characteristics.

[0023] The specific composition of each influencing factor is as follows: The structural parameters include at least one of the following: the geometric dimensions of the rock anchor beam structure, concrete zoning, and concrete elastic modulus. Rock mass mechanical properties include at least one of the following: rock mass integrity coefficient and rock mass elastic modulus; Anchor bolt parameters include anchor bolt material, anchorage depth, anchor bolt diameter, and anchor bolt spacing; The contact interface characteristics include the bond strength between the anchor bolt and the rock anchor beam, the bond strength between the anchor bolt and the rock mass, and the bond strength between the rock anchor beam and the rock mass. External load characteristics include at least one of load amplitude and load frequency; S12. Collect sample data. Through numerical simulation, construct a geometric model of the interaction between the rock mass, rock anchor beam, and anchor rod based on the actual on-site dimensions (the constructed geometric model includes the rock mass, rock anchor beam, and anchor rod). Discretize the geometric model using the finite element method. Input the structural parameters, rock mass mechanical properties, anchor rod parameters, contact interface characteristics, and external load characteristics (these parameters have certain ranges and are randomly generated within a certain range, so many sets can be generated, and the sample is expected to approximate a normal distribution). Considering the bonding characteristics between the rock mass and anchor rod, rock anchor beam and anchor rod, and rock anchor beam and rock mass, calculate the stress and deformation of the overall rock mass-rock anchor beam-anchor rod model. The bonding relationship between the anchor rod and the rock mass, and between the rock anchor beam and the rock mass, is as follows: In the formula, This refers to the bond stress between the anchor bolt and the rock mass, and between the rock anchor beam; This refers to the relative slippage between the anchor bolt and the rock mass, or between the rock anchor beam and the rock anchor. ; ; . These are parameters related to the rock mass, the strength of the rock anchor beam, and the diameter of the reinforcing steel; values ​​can be obtained from relevant specifications. The bond characteristics between the rock anchor beam and the rock mass are determined through contact friction. Based on the SN curve of the anchor bolt, the stress amplitude of the anchor bolt element is mapped to the element fatigue life, which is then used as the simulated value of the corresponding rock anchor beam fatigue life.

[0024] S13. Dataset optimization: Preliminary preprocessing and screening of the initial sample set to remove samples with outliers in fatigue life; specifically, the Grubbs criterion is used to remove outliers in the fatigue life of rock anchor beams, and the Grubbs statistic is used. Calculated using the following formula: ; in, The maximum standard deviation of a single sample. The standard deviation of the sample set For the first The logarithm of the fatigue life of a sample. This is the arithmetic mean of the logarithms of fatigue life for all samples. Grubbs' statistic. If the threshold is exceeded, the sample is removed.

[0025] S14. Data expansion: The number of samples is supplemented by a sample generation method based on generative adversarial networks to match the fatigue life of rock anchor beams in pumped storage power stations, thereby obtaining a dataset for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations.

[0026] For the valid sample set filtered by the Grubbs criterion, a generative adversarial network (GAN) is used to generate synthetic samples related to the fatigue life of rock anchor beams, supplementing the dataset size and meeting the need for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations. The GAN consists of a generator and a discriminator, and is optimized through a minimax game to generate synthetic samples consistent with the characteristics of real rock anchor beam fatigue life samples. The objective function is: ; in, The standard objective function for generative adversarial networks, For generator, For discriminator, To ensure that the reconstruction loss function minimizes the feature deviation between the synthetic sample and the real sample, To fit the structural similarity loss function of the fatigue life sample characteristics of rock anchor beams, , For weight parameters, The output of the generator, For the input random noise, This is the effective sample set after being screened by the Grubbs criterion.

[0027] In this embodiment, S2, the feature engineering process sequentially includes three steps: feature standardization, feature dimensionality reduction, and feature partitioning, as detailed below: S21. Feature Standardization Process. If the dataset contains non-linear feature parameters, first use a kernel function to map the non-linear features to a high-dimensional linear space; then use the Z-score standardization method to convert all feature parameters in the dataset into standard normal distribution data with a mean of 0 and a variance of 1. S22. Feature Dimensionality Reduction. Principal component analysis is used to reduce the dimensionality of the standardized feature parameters, retaining the principal component features with a cumulative variance contribution rate exceeding 90%, resulting in the dimensionality-reduced feature matrix. S23. Feature partitioning. The dimensionality-reduced feature matrix is ​​randomly divided into training, validation, and test sets according to a set ratio.

[0028] In this embodiment, in S3, the deep learning model is a bidirectional long short-term memory network model that incorporates an attention mechanism, and its construction and training process includes the following steps: S31. Construct the input layer. Use the feature matrix processed in step S2 as the input data of the input layer. The number of neurons in the input layer is consistent with the dimension of the reduced feature. S32. Construct a bidirectional long short-term memory network layer. Set up two bidirectional long short-term memory network layers. The number of neurons in the first bidirectional long short-term memory network layer is 64, and the number of neurons in the second bidirectional long short-term memory network layer is 32. Use the Dropout technique to suppress overfitting. The Dropout coefficient is set to 0.2. S33. Construct an attention layer and use a scaled dot product attention mechanism to assign weights to the output features of the bidirectional long short-term memory network layer to strengthen the feature information that has a significant impact on fatigue life. S34. Construct a fully connected layer. Set up two fully connected layers. The first fully connected layer has 16 neurons, and the second fully connected layer has 1 neuron and serves as the output layer to output the predicted fatigue life value of the rock anchor beam joint surface. S35. Model training and optimization: The mean squared error is used as the loss function, and the adaptive moment estimator optimizes the model parameters iteratively. The learning rate is set to 0.001, the number of iterations is 100, and the batch size is 32. The model parameters are adjusted in real time using the validation set. If the model overfits during training, an early stopping strategy and L2 regularization are used for optimization (the patience parameter of the early stopping strategy is set to 10) until the model's loss value on the validation set tends to stabilize.

[0029] In this embodiment, step S4, the fatigue life prediction of the rock anchor beam further includes a model accuracy verification step, specifically: The coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE) are used as accuracy evaluation metrics to validate the prediction results of the deep learning model on the test set; when If the model prediction accuracy meets the requirements, the final fatigue life prediction result of the rock-anchor-beam interface is output; otherwise, return to step S3 to adjust the model parameters and retrain.

[0030] Example 2 like Figure 1 As shown, this invention discloses a method for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations, comprising the following steps: Step 1: Construct a dataset for rapid prediction of fatigue life of rock anchor beams in pumped storage power stations The first phase of the rock anchor beam of a pumped storage power station was poured using C25 grade concrete, and the second phase was poured using C30 grade concrete. The beam cross-sectional dimensions are as follows: Figure 2 As shown; the anchor bolts are made of threaded steel, and the rock mass is composed of dolomite and diorite.

[0031] S11. Determine the characteristics of influencing factors (values ​​based on engineering experience). Structural parameters include: elastic modulus of the first-stage concrete is 28 GPa, and elastic modulus of the second-stage concrete is 30 GPa; rock mass mechanical properties include: elastic modulus of rock mass is 10~40 GPa; anchor bolt parameters include: anchor bolt material (HRB335, HRB400, HRB500), anchoring depth range 7.0~11.0 m, anchor bolt diameter range 25~40 mm, and anchor bolt spacing range 0.6~1.2 mm; contact interface characteristics include: bond strength between anchor bolt and rock anchor beam is 2.5~3.5 MPa; bond strength between anchor bolt and rock mass is 2.0~3.0 MPa; a thin layer is set between the rock anchor beam and rock mass contact interface, and the elastic modulus of the thin layer is 0.0001~0.1 times the elastic modulus of the rock anchor beam; external load characteristics include: wheel load amplitude range 300~900 kN.

[0032] S12. Collect sample data. Construct a three-dimensional finite element model of the rock mass-rock anchor beam-anchor bolt system based on the actual dimensions on site, such as... Figures 3-5 As shown, the horizontal direction of the surrounding rock extends from the sidewall into the interior of the surrounding rock to three times the maximum anchoring depth of the system anchor bolts; the vertical direction extends from the top surface of the rock anchor beam and upwards and downwards by four times the cross-sectional height of the rock anchor beam; the axial width is consistent with the spacing between the rock anchor beams. Based on the parameter combination of the above influencing factors, considering the bond-slip relationship between the anchor bolt and the rock mass, the anchor bolt and the rock anchor beam, and the contact relationship between the rock anchor beam and the rock mass, 100 sets of numerical simulations were carried out. The stress distribution and fatigue life simulation values ​​of the anchor bolts at the rock anchor beam interface were obtained through simulation calculations, forming 100 sets of initial simulation sample data.

[0033] S13. Dataset Optimization. Outliers in the initial sample set are removed using the Grubbs criterion, with a significance level of [missing value]. Five outlier samples were removed; the default threshold for missing values ​​was set at 30%, and after inspection, the missing value ratio of all samples was less than 5%, so no removal was necessary, and 95 valid samples were retained.

[0034] S14. Data Augmentation. A generative adversarial network (GAN)-based sample generation method was used to augment the 95 valid samples. The generator of the GAN was a 3-layer fully connected network, and the discriminator was a 2-layer fully connected network. After 5000 iterations of training, 405 virtual samples were generated, which were then merged with the original 95 valid samples to form a final dataset of 500 samples.

[0035] Step 2: Perform feature engineering on the dataset.

[0036] S21. Feature Standardization Processing. First, the Z-score standardization method is used to standardize all feature parameters to obtain standard normal distribution data with a mean of 0 and a variance of 1.

[0037] S22. Feature Dimensionality Reduction. Principal component analysis (PCA) is used to reduce the dimensionality of the standardized feature parameters. By calculating eigenvalues ​​and eigenvectors, the top 8 principal component features with a cumulative variance contribution rate exceeding 90% are retained, forming an 8-dimensional feature matrix, thus simplifying the feature dimension.

[0038] S23. Feature partitioning. The 8-dimensional feature matrix is ​​randomly divided into a training set (350 sets), a validation set (100 sets), and a test set (50 sets) in a ratio of 7:2:1.

[0039] Step 3: Build a deep learning model based on attention mechanism bidirectional long short-term memory network for fatigue life prediction.

[0040] S31. Construct the input layer. Set the number of neurons in the input layer to 8, consistent with the dimension of the reduced features, and input the 8-dimensional feature matrix of the training set into the model.

[0041] S32. Construct a bidirectional long short-term memory (LSTM) network layer. Set up two bidirectional LSM network layers: the first bidirectional LSM network layer has 64 neurons and uses the hyperbolic tangent activation function; the second bidirectional LSM network layer has 32 neurons and uses the sigmoid activation function; add a dropout layer to both layers, with the dropout coefficient set to 0.2 to suppress model overfitting.

[0042] S33. Constructing the Attention Layer. A scaled dot product attention mechanism is used to assign weights to the output features of the second-layer bidirectional long short-term memory network. By calculating the correlation between each feature and fatigue life, higher weights are assigned to three significantly influential features: anchor diameter, bond strength at the anchor-rock anchor interface, and anchor spacing, thus strengthening key feature information.

[0043] S34. Construct fully connected layers. Set up two fully connected layers. The first fully connected layer has 16 neurons and uses the ReLU activation function; the second fully connected layer has 1 neuron and serves as the output layer, outputting the predicted fatigue life value of the rock anchor beam.

[0044] S35. Model Training and Optimization. Mean squared error was used as the loss function, and adaptive moment estimation was used as the optimizer. The learning rate was set to 0.001, the number of iterations to 100, and the batch size to 32. The model loss value was monitored in real-time using the validation set. During training, overfitting was observed after the 65th iteration. An early stopping strategy was initiated (patience parameter set to 10), and an L2 regularization term (regularization coefficient of 0.001) was added. Training continued until the 75th iteration, when the model's loss value on the validation set stabilized, and training was stopped.

[0045] Step 4: Prediction of fatigue life of rock anchor beams and verification of model accuracy.

[0046] Fifty samples from the test set were input into the trained model to obtain predicted fatigue life values ​​for rock anchor beams. The accuracy of the prediction results was verified using the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). All satisfy To meet the accuracy requirements, the final fatigue life prediction result is output.

[0047] Step 5: Generate a fatigue life early warning report.

[0048] Based on the predicted fatigue life value of the rock anchor beam, a fatigue life early warning report is generated. The report mainly includes three parts: (1) Fatigue life prediction results: The predicted fatigue life of the rock anchor beam of the pumped storage power station is 1,422,329 load cycles, and the corresponding safety level is Level 1. (2) Sensitivity ranking of influencing factors: The weight results output by the attention mechanism are ranked as follows: anchor diameter, bond strength between anchor and rock anchor beam, anchor spacing, external load amplitude, bond strength between rock mass and rock anchor beam, anchoring depth, bond strength between anchor and rock anchor beam, and anchor material. (3) Operation and maintenance optimization suggestions: ① Regularly monitor the load amplitude to avoid overload operation; ② Regularly monitor the stress and deformation of the anchor bolts; ③ It is recommended to shorten the operation and maintenance monitoring cycle to once every 3 months.

[0049] Example 3 The present invention also provides a rapid prediction system for the fatigue life of rock anchor beams in pumped storage power stations. The system is used to implement the method described in Embodiment 1. The system includes: a dataset construction module, a dataset processing module, a model construction module, a prediction module, and an early warning module. The dataset building module is used to construct a dataset for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations. A dataset processing module is used to perform feature engineering processing on the dataset; The model building module is used to build a deep learning model for rapid prediction of fatigue life of rock anchor beams based on the dataset after feature engineering. The prediction module is used to predict the fatigue life of rock anchor beams using the completed deep learning model. The early warning module is used to generate a fatigue life early warning report based on the predicted fatigue life of the rock anchor beam.

[0050] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for quickly predicting the fatigue life of a rock-anchored beam of a pumped storage power station, characterized by, The method includes: S1. Construct a dataset for rapid prediction of fatigue life of rock anchor beams in pumped storage power stations; S2. Perform feature engineering processing on the dataset; S3. Based on the dataset after feature engineering, a deep learning model for rapid prediction of fatigue life of rock anchor beams is built. S4. Use the completed deep learning model to predict the fatigue life of rock anchor beams. S5. Based on the predicted fatigue life of the rock anchor beam, generate a fatigue life early warning report.

2. The method according to claim 1, characterized in that, In S1, the process of constructing a dataset for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations includes four stages: determination of influencing factors, collection of sample data, optimization of dataset, and data expansion.

3. The method according to claim 2, characterized in that, The influencing factors include the characteristics of the structure's own parameters, the characteristics of the rock mass's mechanical properties, the characteristics of the anchor bolt parameters, the characteristics of the contact interface, and the characteristics of the external load. The structural parameters include at least one of the following: the geometric dimensions of the rock anchor beam structure, concrete partitioning, and concrete elastic modulus. The rock mass mechanical properties include at least one of the following: rock mass integrity coefficient and rock mass elastic modulus. The anchor bolt parameters include anchor bolt material, anchoring depth, anchor bolt diameter, and anchor bolt spacing; The contact interface characteristics include the bond strength between the anchor bolt and the rock anchor beam, the bond strength between the anchor bolt and the rock mass, and the bond strength between the rock anchor beam and the rock mass. The external load characteristics include at least one of load amplitude and load frequency.

4. The method according to claim 3, characterized in that, Sample data was collected through numerical simulation, including: A geometric model of the interaction between the rock mass, rock anchor beam, and anchor rod was constructed based on the actual dimensions on site. The geometric model is discretized using the finite element method; The structural parameters, rock mechanics properties, anchor parameters, contact interface features, and external load features are input into the discretized geometric model. Considering the bonding characteristics between rock mass and anchor, rock anchor beam and anchor, and rock anchor beam and rock mass, the simulated value of the fatigue life of the corresponding rock anchor beam is calculated. Among them, the methods for calculating the simulated values ​​of the fatigue life of the rock anchor beam, considering the bond characteristics between the rock mass and the anchor bolt, the rock anchor beam and the anchor bolt, and the rock anchor beam and the rock mass, include: ； ； ； ； In the formula, This refers to the bond stress between the anchor bolt and the rock mass, and between the rock anchor beam; This refers to the relative slippage between the anchor bolt and the rock mass, or between the rock anchor beam and the rock anchor. ; ; ; These are parameters related to the rock mass, rock anchor beam strength, and steel bar diameter. The bonding characteristics between the rock anchor beam and the rock mass are determined through the contact friction relationship. Based on the SN curve of the anchor rod, the stress amplitude of the anchor rod element is mapped to the element fatigue life, which is used as the simulated value of the fatigue life of the corresponding rock anchor beam.

5. The method according to claim 4, characterized in that, The initial sample set undergoes preliminary preprocessing and screening to remove samples with outliers in fatigue life, thus optimizing the dataset, including: The Grubbs criterion was used to remove outliers in the fatigue life of rock anchor beams. The Grubbs statistic... The calculation formula is: ； in, The maximum standard deviation of a single sample. The standard deviation of the sample set, For the first The logarithm of the fatigue life of a sample. This is the arithmetic mean of the logarithmic values ​​of the fatigue life of all samples. Grubbs statistic If the threshold is exceeded, the sample is removed.

6. The method according to claim 5, characterized in that, Data augmentation methods include: For the effective sample set after screening by Grubbs criterion, a generative adversarial network is used to generate synthetic samples related to the fatigue life of rock anchor beams to supplement the dataset size and match the need for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations. The generative adversarial network (GAN) consists of a generator and a discriminator. Through minimax game optimization, it generates synthetic samples that match the characteristics of real rock anchor beam fatigue life samples. The objective function is: ； in, The standard objective function for generative adversarial networks, For generator, For discriminator, To ensure that the reconstruction loss function minimizes the feature deviation between the synthetic sample and the real sample, To fit the structural similarity loss function of the fatigue life sample characteristics of rock anchor beams, , For weight parameters, The output of the generator, For the input random noise, This is the effective sample set after being screened by the Grubbs criterion.

7. The method according to claim 1, characterized in that, In step S2, the method for performing feature engineering on the dataset includes: S21. Feature standardization: If there are non-linear feature parameters in the dataset, first use a kernel function to map the non-linear features to a high-dimensional linear space; then use the Z-score standardization method to convert all feature parameters in the dataset into standard normal distribution data with a mean of 0 and a variance of 1. S22. Feature dimensionality reduction: Principal component analysis is used to reduce the dimensionality of the standardized feature parameters, retaining the principal component features with a cumulative variance contribution rate of more than 90%, and obtaining the dimensionality-reduced feature matrix. S23. Feature partitioning: The dimensionality-reduced feature matrix is ​​randomly divided into training set, validation set and test set according to the proportion.

8. The method according to claim 7, characterized in that, In step S3, the deep learning model is a bidirectional long short-term memory network model based on an attention mechanism, and the construction process includes: S31. Construct the input layer, using the processed feature matrix as the input data of the input layer. The number of neurons in the input layer is consistent with the dimension of the reduced feature. S32. Construct a bidirectional long short-term memory network layer. Set up two bidirectional long short-term memory network layers. The number of neurons in the first bidirectional long short-term memory network layer is 64, and the number of neurons in the second bidirectional long short-term memory network layer is 32. Use the Dropout technique to suppress overfitting. The Dropout coefficient is set to 0.

2. S33. Construct an attention layer and use a scaled dot product attention mechanism to assign weights to the output features of the bidirectional long short-term memory network layer. S34. Construct a fully connected layer. Set up two fully connected layers. The first fully connected layer has 16 neurons, and the second fully connected layer has 1 neuron and serves as the output layer to output the predicted fatigue life value of the rock anchor beam joint surface.

9. A rapid fatigue life prediction system for rock anchor beams in pumped storage power stations, the system being used to implement the method described in any one of claims 1-8, characterized in that, The system includes: a dataset construction module, a dataset processing module, a model construction module, a prediction module, and an early warning module; The dataset construction module is used to construct a dataset for rapid prediction of the fatigue life of rock anchor beams in pumped storage power stations. The dataset processing module is used to perform feature engineering processing on the dataset; The model building module is used to build a deep learning model for rapid prediction of fatigue life of rock anchor beams based on the dataset after feature engineering. The prediction module is used to predict the fatigue life of rock anchor beams using the constructed deep learning model. The early warning module is used to generate a fatigue life early warning report based on the predicted fatigue life of the rock anchor beam.

Images