Health state assessment method and system for rock anchor rod foundation of wind turbine generator

Abstract

The invention provides a health state evaluation method and system for a wind turbine generator rock anchor rod foundation, and relates to the technical field of health monitoring and intelligentization of wind turbine generator foundation structures, and the method comprises the steps: obtaining multi-source monitoring time sequence data of a target wind turbine generator rock anchor rod foundation, including anchor cable axial force, tower cylinder strain, concrete strain, environment temperature and wind load; carrying out data preprocessing on the multi-source monitoring time sequence data; based on the preprocessed multi-source monitoring time sequence data, key influence factors of the health state are analyzed and recognized, and feature time sequence data are constructed; inputting the characteristic time sequence data into a time sequence prediction model, and performing spatial-temporal characteristic extraction and anchor cable axial force trend prediction; based on the prediction result of the anchor cable axial force trend, the uncertainty is quantified through Monte Carlo simulation, and the basic health state is evaluated; according to the method, high-precision and real-time basic health state prediction and evaluation are realized through a multi-modal fusion architecture and random uncertainty quantification.

Description

Technical Field

[0001] This invention relates to the field of health monitoring and intelligent technology for wind turbine foundation structures, specifically to a method and system for assessing the health status of wind turbine rock anchor foundations. Background Technology

[0002] Wind turbine rock anchor foundations are subjected to multiple factors such as dynamic wind loads, temperature changes, and corrosion over long periods in complex marine or mountainous environments. Their health status directly affects the safe operation of the entire turbine. Existing health status assessment technologies for wind turbine rock anchor foundations have the following shortcomings: 1. Limitations of existing time series prediction models: Existing models such as Informer and Autoformer cannot effectively integrate time dependence and spatial correlation when processing multi-source heterogeneous monitoring data of wind power foundations, resulting in insufficient prediction accuracy; 2. Insufficient feature extraction capability: Traditional models lack the ability to deeply mine the coupled features of multiple physical quantities such as anchor cable axial force and strain, making it difficult to capture complex nonlinear spatiotemporal evolution laws; 3. Lack of uncertainty quantification: Existing methods are mostly deterministic predictions, lacking quantitative analysis of the uncertainty of prediction results, and thus unable to provide a basis for risk assessment for engineering decisions; 4. Conflict between real-time performance and accuracy: High-precision models have high computational complexity and are difficult to meet the needs of real-time monitoring; simplified models are fast but lack accuracy and cannot accurately identify early signs of degradation.

[0003] Therefore, existing technologies are insufficient in terms of reliability, real-time performance, and accuracy, and cannot guarantee the long-term safe operation of wind power foundations. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a health status assessment method and system for wind turbine rock anchor foundations. Through a multimodal fusion architecture and random uncertainty quantification, it achieves high-precision, real-time prediction and assessment of foundation health status.

[0005] According to some embodiments, the present invention adopts the following technical solution: A method for assessing the health status of rock anchor foundations for wind turbine units, comprising: Acquire multi-source monitoring time-series data of the rock anchor foundation of the target wind turbine, including anchor cable axial force, tower strain, concrete strain, ambient temperature and wind load, and perform data preprocessing on the multi-source monitoring time-series data; Based on preprocessed multi-source monitoring time-series data, key influencing factors of health status are analyzed and identified, and characteristic time-series data are constructed. Input the characteristic time series data into the time series prediction model to perform spatiotemporal feature extraction and anchor cable axial force trend prediction; Based on the predicted results of anchor cable axial force trends, the uncertainty is quantified through Monte Carlo simulation to assess the basic health status.

[0006] According to some embodiments, the present invention adopts the following technical solution: A health status assessment system for wind turbine rock anchor foundations includes: The data acquisition module is configured to acquire multi-source monitoring time-series data of the target wind turbine rock anchor foundation, including anchor cable axial force, tower strain, concrete strain, ambient temperature and wind load, and perform data preprocessing on the multi-source monitoring time-series data. The feature construction module is configured to: analyze and identify key influencing factors of health status based on preprocessed multi-source monitoring time-series data, and construct feature time-series data. The trend prediction module is configured to input characteristic time series data into the time series prediction model to perform spatiotemporal feature extraction and anchor cable axial force trend prediction. The health assessment module is configured to assess the basic health status by quantifying uncertainty through Monte Carlo simulation based on the predicted results of anchor cable axial force trends.

[0007] According to some embodiments, the present invention adopts the following technical solution: A computer program product includes a computer program that, when executed by a processor, implements the aforementioned method for assessing the health status of a wind turbine rock anchor foundation.

[0008] According to some embodiments, the present invention adopts the following technical solution: A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the aforementioned method for assessing the health status of a wind turbine rock anchor foundation.

[0009] According to some embodiments, the present invention adopts the following technical solution: An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned method for assessing the health status of a wind turbine rock anchor foundation.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention innovatively integrates LSTM, TCN, adaptive sparse attention mechanism (ASSA), and channel attention mechanism to construct a multi-scale spatiotemporal feature extraction architecture; it employs XGBoost-SHAP feature selection to optimize computational efficiency; and it uses Monte Carlo random simulation to achieve uncertainty quantification and probabilistic risk assessment. Experimental results show that compared with traditional methods, the prediction accuracy is improved by 73.6% and the computational efficiency is improved by 1.86 times, providing a reliable intelligent solution for the long-term safe operation of wind power foundations. Attached Figure Description

[0011] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0012] Figure 1 This is a flowchart of the method in Example 1. Figure 2 This is a structural diagram of the time series prediction model in Example 1.

[0013] Figure 3 This is a schematic diagram of the adaptive attention mechanism in Example 1. Detailed Implementation

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0016] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0017] Example 1 One embodiment of the present invention provides a health status assessment method for wind turbine rock anchor foundations. Through an innovative multimodal fusion architecture and random uncertainty quantification, it achieves high-precision, real-time prediction and assessment of the foundation's health status. Figure 1 As shown, the specific steps are as follows: Step S1: Obtain multi-source monitoring time-series data of the target wind turbine rock anchor foundation, including anchor cable axial force, tower strain, concrete strain, ambient temperature and wind load, and perform data preprocessing on the multi-source monitoring time-series data.

[0018] 1. Collect real-time monitoring data of wind turbine foundations in a wind farm reinforcement project. The data collection method is as follows: Four anchor cable monitoring points (MS-1 to MS-4) were set up to collect axial force data using vibrating wire anchor cable gauges; five strain monitoring points (D1-D5) were set up on the outer surface of the tower to collect tower strain using strain gauges; two strain gauges (C1-C2) were set up on the interface between new and old concrete to collect concrete strain using concrete strain gauges at a frequency of 15 minutes / time; ambient temperature and wind load were obtained from the wind power system at a frequency of 15 minutes / time.

[0019] 2. The data preprocessing is based on the identified outliers and missing values, performing outlier filtering and missing value repair; wherein, the outlier identification is performed by using iterative rolling difference-Z-Score to identify abrupt changes in the time series data, and outliers are marked by a set dynamic threshold; the missing value repair is performed by using a residual network to repair the missing data.

[0020] To ensure the quality and integrity of the input data, this embodiment designs an anomaly detection, filtering and data repair method based on iterative rolling difference-Z-Score and residual network. It efficiently processes outliers (such as mutation points) and missing values ​​in time series data, providing high-quality input data for the STA-Transformer model.

[0021] In outlier detection, iterative rolling difference-Z-Score is used to identify abrupt changes in time-series data (such as abnormally high or low values ​​caused by sensor malfunctions). Outliers are marked by setting a dynamic threshold (based on 3-4 times the standard deviation, with a rolling window of 12 data points). The detection accuracy reaches 95%, and the false alarm rate is controlled within 2%. The algorithm of iterative rolling difference-Z-Score is shown in Table 1. Table 1 Iterative Rolling Difference-Z-Score Algorithm

[0022] In missing value data repair, this method uses iterative rolling difference-Z-Score to filter and remove outliers in the data, and then constructs a data repair neural network through multi-layer residual blocks. Each layer of this network contains a 1D convolutional layer (with a kernel size of 3), a ReLU activation function, and a batch normalization layer. It also uses a Dropout layer (with a Dropout rate of 0.1) to enhance generalization ability and prevent overfitting. The network depth is dynamically adjusted to 4-6 layers according to the data size and complexity.

[0023] The residual connection structure is adopted, and the output format is as follows: ,in, x For input features, As the features transformed by the convolutional layer, the residual connection maintains gradient flow by directly passing input information, effectively avoiding the gradient vanishing problem in deep networks, ensuring training stability, and significantly improving convergence speed, especially when processing long sequence data.

[0024] In terms of results, preprocessing reduced data noise by approximately 70%, significantly improving the signal-to-noise ratio. Combined with an electromagnetic shielding device (model EM-001), outlier data was reduced by approximately 80%, and data integrity was improved to 98%. High-quality input data enabled the STA-Transformer model to maintain excellent performance in multi-step predictions, achieving 96 prediction steps. R 2 With a value of over 0.951, it demonstrates excellent robustness and stability, providing a reliable guarantee for subsequent anchor cable health status assessment.

[0025] Step S2: Based on the preprocessed multi-source monitoring time series data, analyze and identify the key influencing factors of health status, and construct characteristic time series data.

[0026] To accurately identify the key features affecting anchor cable axial force, this embodiment employs a gradient boosting decision tree (XGBoost) combined with the SHAP (SHapley Additive ExPlanations) interpretability analysis method to construct optimized feature time-series data, significantly reducing computational overhead and improving model performance. The specific implementation process is as follows: 1. Train a high-precision prediction model based on XGBoost using data; The XGBoost model performs feature selection by predicting the axial force value of each anchor cable. Its configuration employs the following optimized parameters to achieve efficient and stable feature selection and modeling performance: the number of weak learners ranges from 50 to 200, adaptively adjusted based on validation set performance through cross-validation to ensure a balance between model capacity and generalization ability; the maximum tree depth ranges from 3 to 10, controlling model complexity to prevent overfitting while ensuring the ability to model complex feature relationships; the learning rate ranges from 0.01 to 0.3, optimizing convergence speed and ensuring efficient and stable training; and L1 and L2 regularization are introduced (default values ​​of 0.1 and 0.5, respectively) to further enhance model robustness and reduce the risk of overfitting. These configurations collectively improve the accuracy and efficiency of XGBoost in identifying key features of anchor cable axial force, providing a high-quality feature subset for subsequent STA-Transformer models.

[0027] 2. Calculate the SHAP value by parsing the model output through SHAP analysis, and quantify the contribution of each feature to the prediction; SHAP interpretability analysis quantifies the contribution of each input feature (such as anchor cable axial force, tower strain, concrete strain, ambient temperature, wind load, etc.) to the prediction of anchor cable health status by calculating SHAP values, thus providing a scientific basis for feature selection.

[0028] The analysis results show that some features have a significant impact on the prediction target. For example, the SHAP value of monitoring point MS-2 compared to MS-1 is as high as 1.6, indicating that it plays a key role in the prediction of anchor cable axial force.

[0029] 3. Based on the SHAP value sorting, determine the core features that affect the target variable.

[0030] Based on SHAP value sorting, this embodiment selects the feature subset with the highest contribution (such as the top 10 key features), eliminates redundant or low-contribution features, reduces the input feature dimension by about 50%, significantly reduces computational complexity, and improves the model's prediction accuracy and interpretability.

[0031] In the specific implementation process, the preprocessed data was divided into training set and test set in a 7:3 ratio. The XGBoost model was used in combination with SHAP analysis to identify key factors. SHAP analysis showed that the SHAP value of MS-2 to MS-1 reached 1.6, followed by D1-D5 (SHAP value 0.6). The generated key feature dataset included axial force from MS-1 to MS-4 and strain from D1 to D5. The calculation speed was improved by 1.86 times and the prediction accuracy was improved by 35.5%.

[0032] Step S3: Input the feature time series data into the time series prediction model to perform spatiotemporal feature extraction and anchor cable axial force trend prediction.

[0033] The time-series prediction model employs a Transformer based on spatiotemporal attention, such as... Figure 2 As shown, it includes a feature extraction module, an encoder, and a decoder. The feature extraction module first uses a long short-term memory network based on an adaptive attention mechanism to capture the long-term evolution trend of the anchor cable axial force and obtain a long-term feature sequence. Then, it uses a temporal convolutional network based on a channel attention mechanism to extract local temporal patterns and obtain a local feature sequence. Finally, it performs feature fusion of the long-term feature sequence and the local feature sequence through residual connections.

[0034] Specifically, in order to effectively capture the evolution of anchor cable axial force on both short and long time scales, this embodiment designs a multi-scale temporal feature extraction module, combining the complementary advantages of LSTM and TCN, and achieving feature fusion through residual connections.

[0035] 1. A long short-term memory network based on an adaptive attention mechanism is used to capture the long-term evolution trend of anchor cable axial force.

[0036] LSTM achieves selective memorization through input gates, forget gates, and output gates, with the hidden layer dimension set to d. model It can effectively preserve key historical information and alleviate the gradient vanishing problem in long sequence modeling; the input of LSTM is feature time series data, and the output is a time feature sequence. To enhance the feature extraction capability of complex spatiotemporal modeling in the health status assessment of wind turbine rock anchor foundations, an adaptive attention (ASSA) mechanism is adopted. This mechanism focuses on key time points or patterns in the time series, reducing noise interference (such as outliers or irrelevant fluctuations). Its dynamic weighting mechanism adaptively adjusts the attention allocation based on the characteristics of the time series (such as periodicity or trend), thereby improving the model's ability to capture long-term dependencies or complex patterns. Figure 3 As shown, specifically: First, the soft attention branch captures global feature associations through the Softmax function, ensuring the semantic continuity of complex time-series data, which is particularly suitable for modeling long-term dependencies in anchor cable axial force data.

[0037] Secondly, the hard attention branch uses the squared operation of the ReLU activation function to highlight the high correlation feature and enhance the model's sensitivity to key signals (such as sudden stress changes).

[0038] Its core formula is as follows:

[0039] in: These are query, key, and value matrices, with dimensions of [missing information]. , The sequence length is given. d For the attention head dimension, This is a scaling factor to stabilize the gradient.

[0040] Third, adaptive weight balancing is achieved through... , Learnable weights dynamically adjust the contribution ratio of soft and hard attention, calculated through exponential normalization: in: For learnable parameters, Optimization is achieved through gradient descent, enabling the model to flexibly adapt to different task requirements.

[0041] Finally, spatial perception capability is significantly enhanced by introducing relative position coding, as shown in the formula: Where Table is a learnable relative position bias table, and Index... i,j This serves as a location index matrix, thereby enhancing the model's ability to model the spatial relationships between monitoring points.

[0042] After the LSTM output data is enhanced by the adaptive attention mechanism, the combination of these mechanisms enables the adaptive attention mechanism to exhibit excellent accuracy and computational efficiency in complex spatiotemporal dynamic modeling, providing reliable technical support for anchor cable health status prediction.

[0043] 2. A Temporal Convolutional Network (TCN) based on channel attention mechanism is used to extract local temporal patterns.

[0044] TCN utilizes causal convolution and dilated convolution* combined with residual connections to address the gradient vanishing and parallel computation difficulties of traditional recurrent neural networks (RNNs) in long sequence tasks. The causal convolution kernel size is 3, and the number of layers is configurable (usually 3), capturing long sequence dependencies through dilated convolution. Each convolutional layer in TCN is followed by a ReLU activation function and a dropout layer (with a dropout rate of 0.1) to enhance the model's generalization ability. The input to TCN is the same as LSTM, and the output is a sequence of local features with the same shape [B, L, d]. model [It excels at capturing short-term fluctuation patterns.]

[0045] The Channel Attention Mechanism (SENet) is embedded after each causal dilated convolution layer of the TCN to dynamically adjust the importance of channel features, enhancing the ability to express key features in time series prediction. SENet uses a "compression-activation" process to channel-weight the local features extracted by the TCN, compensating for the TCN's insufficient modeling of global features, and complementing the global attention mechanism of the Transformer and the long-term dependency modeling of the LSTM.

[0046] 3. Feature Fusion Strategy: To comprehensively utilize the long-term feature sequences of LSTM (including ASSA adaptive attention enhancement) and the local feature sequences of TCN (including SENet channel attention enhancement), feature fusion is achieved through residual connections. The fusion formula is as follows:

[0047] in, The output of the Transformer embedding layer (containing time stamp information). Local features extracted by TCN The temporal features extracted by LSTM.

[0048] Residual connections not only preserve the original input information, but also optimize feature representation through adaptive learning, significantly improving the model's ability to represent complex spatiotemporal patterns.

[0049] Step S4: Based on the predicted results of the anchor cable axial force trend, the uncertainty is quantified through Monte Carlo simulation to assess the basic health status.

[0050] To comprehensively assess the uncertainty of anchor cable health status and provide a scientific probabilistic risk assessment, this embodiment uses the Monte Carlo simulation method. Based on the anchor cable axial force value predicted by the STA-Transformer model, it calculates the probability of future failure, providing a reliable basis for engineering decisions on wind turbine rock anchor foundations.

[0051] Anchor cable health status is determined by a state function. Quantification is performed, among which The design prestress value (set to 200kN) represents the expected axial force under normal working conditions. This refers to the axial force value predicted by STA-Transformer based on multi-source monitoring data (such as axial force, strain, and environmental factors). The failure criterion is set as follows. (i.e., 140kN), indicating that the axial force is significantly lower than the safety threshold, which may lead to structural risks.

[0052] To quantify and predict uncertainty, Modeled as a normal distribution The model predicts the values, and the variance is... Calibrate using confidence intervals or historical prediction errors to ensure consistency with measured data.

[0053] Monte Carlo simulations generate a large number of data through 100,000 random samples. Sample, calculate failure probability ( N =100,000), in The system returns 1 if the condition is met, and 0 otherwise. 0 indicates that the current sample meets the failure condition; 1 indicates that the current sample does not meet the failure condition (i.e., the system is operating normally or the prediction result is within an acceptable range). The sampling process uses a highly efficient pseudo-random number generator to ensure uniform sample distribution and statistical representativeness. Simulations evaluated the failure probabilities at 24, 48, 72, and 96 steps (corresponding to 6 hours, 12 hours, 18 hours, and 24 hours), which were 0%, 0%, 0.01%, and 0.02%, respectively, demonstrating that the system maintains extremely high reliability during long-term operation.

[0054] The Kolmogorov-Smirnov test was used to compare the simulated cumulative distribution function (CDF) with the median of the measured data, with the deviation controlled within 5%, ensuring the accuracy and reliability of the assessment results. This method, combined with the high-precision predictions of the STA-Transformer, provides scientific support for the probabilistic assessment of anchor cable health status, significantly improving the accuracy and economy of engineering decision-making.

[0055] The Kolmogorov-Smirnov test (KS test) and Monte Carlo simulation complement each other in the probabilistic assessment of anchor cable health, jointly providing reliable statistical support for the predictions of the STA-Transformer model. Monte Carlo simulation generates a large number of prediction samples by adding perturbations to the input or model parameters through 100,000 random samples (N=100,000), quantifying the uncertainty of the model output (such as anchor cable tension or stress), and statistically satisfying failure conditions (e.g., returning 0 if the predicted value exceeds a threshold, otherwise returning 1).

[0056] Example 2 One embodiment of the present invention provides a health status assessment system for wind turbine rock anchor foundations, comprising: The data acquisition module is configured to acquire multi-source monitoring time-series data of the target wind turbine rock anchor foundation, including anchor cable axial force, tower strain, concrete strain, ambient temperature and wind load, and perform data preprocessing on the multi-source monitoring time-series data. The feature construction module is configured to: analyze and identify key influencing factors of health status based on preprocessed multi-source monitoring time-series data, and construct feature time-series data. The trend prediction module is configured to input characteristic time series data into the time series prediction model to perform spatiotemporal feature extraction and anchor cable axial force trend prediction. The health assessment module is configured to assess the basic health status by quantifying uncertainty through Monte Carlo simulation based on the predicted results of anchor cable axial force trends.

[0057] Example 3 One embodiment of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method for assessing the health status of a wind turbine rock anchor foundation.

[0058] Example 4 In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided for storing computer instructions. When the computer instructions are executed by a processor, they implement the aforementioned method for assessing the health status of a wind turbine rock anchor foundation.

[0059] Example 5 One embodiment of the present invention provides an electronic device, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the aforementioned method for assessing the health status of a wind turbine rock anchor foundation.

[0060] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0062] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method of health condition assessment of a rock anchor foundation of a wind turbine generator unit, characterized in that, The method comprises: obtaining multi-source monitoring time series data of a rock anchor foundation of a target wind turbine, including anchor cable axial force, tower drum strain, concrete strain, environmental temperature and wind load, and performing data preprocessing on the multi-source monitoring time series data; based on the preprocessed multi-source monitoring time series data, analyzing and identifying key influencing factors of the health state, and constructing feature time series data; inputting the feature time series data into a time series prediction model, performing spatio-temporal feature extraction and anchor cable axial force trend prediction; based on the prediction result of the anchor cable axial force trend, quantifying the uncertainty through Monte Carlo simulation, and evaluating the foundation health state.

2. A method of assessing the health of a rock anchor foundation of a wind turbine generator according to claim 1, wherein, The data preprocessing is based on the identified abnormal values and missing values, and the abnormal value filtering and missing value repair are performed. The abnormal value identification is performed by using a residual network to identify the mutation points in the time series data, and the abnormal values are marked by setting a dynamic threshold.

3. A method of assessing the health of a rock anchor foundation of a wind turbine generator according to claim 1, wherein, The analysis and identification of the key influencing factors of the health state are performed by using gradient boosting decision tree XGBoost combined with SHAP. A high-precision prediction model based on XGBoost is trained using data. The SHAP value is calculated by analyzing the model output, and the contribution of each feature to the prediction is quantified. Based on the SHAP value sorting, the core features affecting the target variable are determined.

4. A method of assessing the health of a rock anchor foundation of a wind turbine generator according to claim 1, wherein, The time series prediction model adopts a Transformer based on spatio-temporal attention, including a feature extraction module, an encoder and a decoder. The feature extraction module first uses a long short-term memory network based on an adaptive attention mechanism to capture the long-term evolution trend of the anchor cable axial force, obtains a long-term feature sequence, then uses a time convolution network based on a channel attention mechanism to extract local time series patterns, obtains a local feature sequence, and finally performs feature fusion of the long-term feature sequence and the local feature sequence through residual connection.

5. A method of assessing the health of a rock anchor foundation of a wind turbine generator according to claim 4, wherein, The adaptive attention mechanism is realized by soft and hard attention weighting fusion. where, are the query, key and value matrices, respectively, with dimensions , is the sequence length, d is the attention head dimension, is a scaling factor to stabilize the gradients.

6. A method of assessing the health of a rock anchor foundation of a wind turbine generator according to claim 1, wherein, The evaluation of the foundation health state through Monte Carlo simulation quantification of uncertainty is specifically as follows: State function based The failure probability is calculated by random sampling to realize multi-scale risk assessment, wherein, is a design prestress value, representing an expected axial force under normal working conditions; is a predicted axial force value.

7. A system for health condition assessment of a rock anchor foundation of a wind turbine generator unit, characterized in that, The data acquisition module is configured to obtain multi-source monitoring time series data of a rock anchor foundation of a target wind turbine, including anchor cable axial force, tower drum strain, concrete strain, environmental temperature and wind load, and perform data preprocessing on the multi-source monitoring time series data. The feature construction module is configured to analyze and identify key influencing factors of the health state based on the preprocessed multi-source monitoring time series data, and construct feature time series data. The trend prediction module is configured to input the feature time series data into a time series prediction model, perform spatio-temporal feature extraction and anchor cable axial force trend prediction. The health evaluation module is configured to quantify the uncertainty through Monte Carlo simulation based on the prediction result of the anchor cable axial force trend, and evaluate the foundation health state. The computer program is executed by the processor to realize the health state evaluation method of the rock anchor foundation of the wind turbine according to any one of claims 1-6.

8. A computer program product comprising a computer program, characterized in that, The non-transitory computer readable storage medium is used to store computer instructions, which are executed by the processor to realize the health state evaluation method of the rock anchor foundation of the wind turbine according to any one of claims 1-6.

9. A non-transitory computer-readable storage medium, comprising: The method comprises:

10. An electronic device, comprising: ​ The processor, the memory and the computer program; wherein the processor is connected with the memory, the computer program is stored in the memory, when the electronic equipment runs, the processor executes the computer program stored in the memory, so that the electronic equipment executes the health state evaluation method of the wind turbine rock anchor foundation as claimed in any one of claims 1-6.

Images