Methods and systems for early warning and diagnosis of faults in high-tower wind turbines

Abstract

This invention discloses a method and system for fault early warning and diagnosis of high-tower wind turbines. The method includes: preprocessing SCADA data of the high-tower wind turbine; decomposing the preprocessed SCADA data using an improved EEMD algorithm to extract feature information at least two time scales, and concatenating the IMF components of at least two time points into a feature matrix; inputting the feature matrix into an enhanced visual Transformer (ViT) model to obtain a prediction result of the real-time state of the wind turbine; calculating a residual matrix based on the prediction result and the SCADA data; and triggering a fault early warning when the root mean square error of the residual matrix exceeds a preset early warning threshold. The technical solution of this application improves the accuracy and real-time performance of fault early warning and diagnosis for high-tower wind turbines by improving the EEMD algorithm and enhancing the visual Transformer model, thereby increasing the system's intelligence level and reducing operation and maintenance costs.

Description

Technical Field

[0001] This invention relates to the technical field of wind power generation, and in particular to a method and system for fault early warning and diagnosis of high-tower wind turbine units. Background Technology

[0002] High-tower wind turbines, as an important form of wind power generation, typically have tower heights exceeding 100 meters, aiming to utilize higher, more stable wind speeds to improve power generation efficiency. However, with increased tower height, wind turbines face more complex failure modes and higher maintenance costs during operation. Traditional fault detection methods often rely on manual inspections and periodic maintenance, which suffer from long response times, limited detection range, and high labor costs. Therefore, there is an urgent need for an efficient and intelligent fault early warning and diagnosis method to ensure the safe and stable operation of high-tower wind turbines.

[0003] Existing methods often face challenges such as high data noise, difficulty in feature extraction, and high model complexity when processing high-dimensional, multi-scale SCADA data. Therefore, how to effectively extract useful features from SCADA data and use advanced machine learning models for fault diagnosis has become an urgent issue to be addressed in the industry. Summary of the Invention

[0004] This invention provides a method and system for fault early warning and diagnosis of high-tower wind turbines, which improves the accuracy and real-time performance of fault early warning and diagnosis of high-tower wind turbines by improving the EEMD algorithm and enhancing the visual Transformer model, thereby enhancing the intelligence level of the system and reducing operation and maintenance costs.

[0005] According to a first aspect of the present invention, a method for early warning and diagnosis of faults in high-tower wind turbines is provided, the method comprising:

[0006] Preprocessing of SCADA data for high-tower wind turbines;

[0007] The preprocessed SCADA data is decomposed using the improved EEMD algorithm to extract feature information at least two time scales, and the IMF components at at least two time points are concatenated into a feature matrix.

[0008] The feature matrix is ​​input into the enhanced visual Transformer (Vi T) model to obtain the prediction results of the real-time state of the wind turbine.

[0009] The residual matrix is ​​calculated based on the prediction results and the SCADA data. When the root mean square error of the residual matrix exceeds the preset warning threshold, a fault warning is triggered.

[0010] By analyzing the abnormal variables in the residual matrix, the fault types of high-tower wind turbines are identified and fault diagnosis is performed.

[0011] In one embodiment, the decomposition of the preprocessed SCADA data using the improved EEMD algorithm includes:

[0012] Set up a sliding window and extend the extreme values ​​based on the extreme points and historical data within the window;

[0013] The termination conditions for decomposition are dynamically adjusted, and information leakage is eliminated by backtracking through historical data.

[0014] In one embodiment, it also includes:

[0015] A multi-layer asymmetric convolution module is used to extract local features from the feature matrix, and a deformable attention mechanism is combined to improve the computational efficiency of the model.

[0016] The parameters are optimized through multiple rounds of training using the Adam optimizer and the mean squared error (MSE) loss function.

[0017] In one embodiment, the fault warning triggering mechanism includes:

[0018] By analyzing the deviation between predicted and actual values ​​in the residual matrix, the tolerance for prediction error is dynamically adjusted based on historical data and normal operating conditions.

[0019] The likelihood of computational failure is determined based on the changing trend and magnitude of the residual matrix.

[0020] In one embodiment, identifying the fault type of the high-tower wind turbine and performing fault diagnosis includes:

[0021] By analyzing the abnormal patterns in the residual matrix, the variables that cause failures are identified, and the failure type of the wind turbine is inferred.

[0022] Based on the fault type, a detailed diagnostic report of the high-tower wind turbine fault is output.

[0023] In one embodiment, it also includes:

[0024] The formula for calculating the root mean square error (RMSE) is as follows:

[0025]

[0026] Among them, y i The value at time i in the actual SCADA data. α is the predicted value, N is the total number of data points, and α is the adjustment factor.

[0027] According to a second aspect of the present invention, a fault early warning and diagnosis system for high-tower wind turbines is provided, comprising:

[0028] The preprocessing module is used to preprocess the SCADA data of high-tower wind turbines;

[0029] The extraction module is used to decompose the preprocessed SCADA data using the improved EEMD algorithm, extract feature information at least two time scales, and concatenate the IMF components at at least two time points into a feature matrix.

[0030] The prediction module is used to input the feature matrix into the enhanced vision Transformer (Vi T) model to obtain the prediction result of the real-time state of the wind turbine.

[0031] The early warning module is used to calculate the residual matrix based on the prediction results and the SCADA data. When the root mean square error of the residual matrix exceeds a preset early warning threshold, a fault early warning is triggered.

[0032] The diagnostic module is used to identify the fault types of high-tower wind turbines and perform fault diagnosis by analyzing the abnormal variables in the residual matrix.

[0033] According to a third aspect of the present invention, an electronic device is provided, comprising: a communication interface, a processor, and a memory;

[0034] The memory is used to store program instructions, which, when executed by the processor that is connected to the memory via the communication interface, implement any of the above-described methods for early warning and diagnosis of faults in high-tower wind turbines.

[0035] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a computer (e.g., a processor in a computer), implement any of the above-described methods for early warning and diagnosis of faults in high-tower wind turbines.

[0036] In summary, this invention provides a method and system for fault early warning and diagnosis of high-tower wind turbines. The method includes: preprocessing SCADA data of the high-tower wind turbine; decomposing the preprocessed SCADA data using an improved EEMD algorithm to extract feature information at least two time scales, and concatenating the IMF components of at least two time points into a feature matrix; inputting the feature matrix into an enhanced vision Transformer (ViT) model to obtain a prediction result of the wind turbine's real-time state; calculating a residual matrix based on the prediction result and the SCADA data; triggering a fault early warning when the root mean square error of the residual matrix exceeds a preset early warning threshold; and identifying the fault type of the high-tower wind turbine and performing fault diagnosis by analyzing abnormal variables in the residual matrix. The technical solution of this application uses an improved EEMD algorithm to decompose SCADA data and extract multi-scale features, which helps to capture subtle changes in the operating state of the wind turbine and improves the sensitivity of fault detection. Combined with the enhanced vision Transformer model, utilizing multi-layer asymmetric convolution modules and deformable attention mechanisms, the model's learning ability for complex data patterns is enhanced, improving the accuracy of fault diagnosis. By analyzing the residual matrix and dynamically adjusting the prediction error tolerance, combined with trend analysis, the fault early warning strategy is optimized to reduce false alarms and missed alarms, thereby improving the reliability of the early warning system. The system automatically identifies fault types and generates diagnostic reports, assisting maintenance personnel in quickly locating problems, developing targeted maintenance measures, reducing maintenance costs, and extending equipment lifespan.

[0037] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and drawings.

[0038] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0039] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 A flowchart of a fault early warning and diagnosis method for high-tower wind turbines provided as an embodiment of the present invention;

[0041] Figure 2A flowchart of another method for fault early warning and diagnosis of high-tower wind turbines provided as an embodiment of the present invention;

[0042] Figure 3 A flowchart of another method for fault early warning and diagnosis of high-tower wind turbines provided as an embodiment of the present invention;

[0043] Figure 4 A flowchart of another method for fault early warning and diagnosis of high-tower wind turbines provided as an embodiment of the present invention;

[0044] Figure 5 A flowchart of another method for fault early warning and diagnosis of high-tower wind turbines provided as an embodiment of the present invention;

[0045] Figure 6 A structural diagram of a fault early warning and diagnosis system for high-tower wind turbines provided as an embodiment of the present invention;

[0046] Figure 7 This is a structural diagram of an electronic device provided as an embodiment of the present invention. Detailed Implementation

[0047] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0048] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0049] like Figure 1As shown, the present invention provides a method for fault early warning and diagnosis of high-tower wind turbines, which includes:

[0050] In step S11, the SCADA data of the high-tower wind turbine is preprocessed;

[0051] In step S12, the preprocessed SCADA data is decomposed using the improved EEMD algorithm to extract feature information at least two time scales, and the IMF components at at least two time points are concatenated into a feature matrix.

[0052] In step S13, the feature matrix is ​​input into the enhanced vision Transformer (Vi T) model to obtain the prediction result of the real-time state of the wind turbine.

[0053] In step S14, a residual matrix is ​​calculated based on the prediction results and the SCADA data. When the root mean square error of the residual matrix exceeds a preset warning threshold, a fault warning is triggered.

[0054] In step S15, the fault types of high-tower wind turbines are identified and fault diagnosis is performed by analyzing the abnormal variables in the residual matrix.

[0055] In one embodiment, with the development of data acquisition technology and artificial intelligence algorithms, fault diagnosis methods based on SCADA (Supervisory Control and Data Acquisition) system data have gradually gained attention. SCADA systems can monitor the operating status of wind turbine units in real time and collect a large amount of operating data. By conducting in-depth analysis of this data, early warning and accurate diagnosis of unit faults can be achieved.

[0056] SCADA systems are crucial for monitoring and data acquisition during wind turbine operation, recording various operating parameters such as speed, power, temperature, and vibration in real time. However, raw SCADA data often suffers from noise interference, missing values, and outliers. If these data quality issues are not properly addressed, they will severely impact the accuracy and reliability of subsequent data analysis and model training. For example, noisy data may mask subtle changes in the wind turbine's operating status, making it difficult to effectively extract fault characteristics; missing and outliers can lead to incomplete and inconsistent data sequences, affecting the overall quality and usability of the data. Therefore, preprocessing SCADA data is the foundation and prerequisite for the entire fault diagnosis process. Its purpose is to clean and organize the raw data into a high-quality dataset suitable for subsequent analysis.

[0057] By setting reasonable thresholds and rules, obviously erroneous or unreasonable data points can be identified and removed. For example, for wind speed data, negative values ​​or values ​​exceeding the physical range can be identified as erroneous data and removed. Simultaneously, persistently abnormal data segments clearly caused by sensor malfunctions should also be marked and cleaned up. Appropriate methods should be used to fill in missing data. Common methods include mean imputation, median imputation, and interpolation. For example, for power data missing for a short period, linear interpolation can be used to fill in the missing data based on data from adjacent time points to ensure data continuity; while for data missing for a long period, it may be necessary to combine the correlation of other relevant parameters for comprehensive estimation and imputation. Statistical analysis methods or machine learning algorithms should be used to identify outliers in the data. For example, based on the box-plot method, data points exceeding 1.5 times the interquartile range can be considered outliers; or machine learning algorithms such as Isolation Forest can be used to detect outliers in high-dimensional data. For detected outliers, deletion, correction, or further analysis can be performed to determine their cause, depending on the specific circumstances.

[0058] EEMD (Ensemble Empirical Mode Decomposition) is an adaptive signal decomposition method that can decompose complex nonlinear and non-stationary signals into multiple intrinsic mode functions (IMFs). Each IMF component represents the signal's characteristics at different time scales, arranged sequentially from high frequency to low frequency, thus effectively capturing the multi-scale variation information of the signal. However, traditional EEMD algorithms have some shortcomings, such as endpoint effects and mode aliasing, which may affect the accuracy and stability of the decomposition results, thereby reducing the effectiveness of fault feature extraction. Therefore, improving the EEMD algorithm is a key step in enhancing fault diagnosis performance.

[0059] Methods such as mirror expansion and polynomial fitting expansion are employed to expand the endpoints of the signal to reduce decomposition errors at these endpoints. For example, for a one-dimensional power signal, mirror-symmetric expansion is performed at both the beginning and end points, ensuring smoother and more accurate IMF components near the endpoints and avoiding distortion caused by endpoint effects. By optimizing the white noise addition strategy and decomposition parameters, mode aliasing between different IMF components is reduced. For instance, in the ensemble averaging process of EEMD, the amplitude and number of white noise additions are appropriately adjusted to more clearly separate the characteristic information of different frequency bands for each IMF component, improving the physical meaning and interpretability of the decomposition results.

[0060] After completing the improved EEMD decomposition, feature information at least two time scales is extracted from the multiple IMF components obtained from the decomposition. For example, the first few IMF components in the high-frequency band (such as IMF1 and IMF2) are selected to reflect the rapid changes in the operating status of the wind turbine, such as instantaneous fault impacts; at the same time, IMF components in the low-frequency band (such as IMF3 and IMF4) are selected to characterize the long-term trend changes in the wind turbine, such as slow wear and fatigue of the unit. By concatenating the IMF components from at least two time points into a feature matrix, the feature information from different time points can be fully utilized to construct a multi-dimensional feature space that comprehensively reflects the operating status of the wind turbine. For example, for the vibration signal of a wind turbine, assuming that the improved EEMD decomposition is performed at two time points t1 and t2 respectively, and the corresponding IMF components are obtained, the IMF1 and IMF2 components at time t1 and the IMF1 and IMF2 components at time t2 are arranged and combined in a certain order to form a feature matrix, which is used as the feature vector for subsequent model input, so that the model can capture the vibration characteristic change pattern of the wind turbine at different times, thereby more accurately predicting its state and identifying potential faults.

[0061] The core advantage of the Transformer model lies in its ability to capture global dependencies between elements at different positions in sequential data through a self-attention mechanism, unlike traditional convolutional neural networks (CNNs) which are limited by local receptive fields. When processing multidimensional feature sequence data of wind turbines, global modeling capability is crucial for accurately grasping the complex relationships between various components and the dynamic changes in operating status. For example, during wind turbine operation, gearbox temperature changes may have complex nonlinear relationships with multiple factors such as generator load and speed. The Transformer model's self-attention mechanism can automatically learn and weight the interactions between different factors, thereby providing a more comprehensive understanding of the wind turbine's operating characteristics.

[0062] To better suit wind turbine fault diagnosis tasks, the ViT model in the vision domain was enhanced and improved. On one hand, the input encoding method was optimized, effectively spatially reorganizing and encoding the multi-dimensional feature matrix of the wind turbine, making it more compatible with the Transformer model's requirements for visual feature processing. On the other hand, hyperparameters such as depth and width were adjusted to balance the model's expressive power and computational complexity, ensuring efficient operation while maintaining prediction accuracy. For example, in input encoding, the feature matrix, originally arranged in a time series, was divided into blocks, and positional embedding vectors were added, enabling the model to simultaneously perceive the temporal order and spatial structure information of the features. Regarding model structure adjustment, experiments were conducted to compare the model's performance on wind turbine fault diagnosis datasets under different depth and width settings, ultimately determining a model architecture that effectively learns feature representations without overfitting.

[0063] The constructed feature matrix is ​​input into the enhanced visual Transformer (ViT) model. Through training on a large amount of historical normal operation and fault data, the model can automatically extract key patterns and regularities from the feature matrix, thereby predicting the real-time status of the wind turbine. The prediction results are typically output as a probability distribution, representing the probability of the wind turbine being in different operating states (such as normal state and different types of fault states). For example, assuming the model outputs a probability of 0.8 for the wind turbine being in a normal state, 0.1 for a gearbox fault state, and 0.1 for a generator fault state, then based on the principle of maximizing probability, it can be preliminarily determined that the wind turbine is currently in normal operation. However, this also suggests the need for further monitoring and analysis of the relevant parameters of the gearbox and generator to confirm whether there are any potential fault hazards.

[0064] Based on the prediction results of the enhanced visual Transformer (Vi T) model and the original SCADA data, a residual matrix is ​​calculated. Each element in the residual matrix represents the difference between the predicted value and the actual measured value at the corresponding location. The specific calculation method can be a simple difference calculation, or it can involve normalization or other mathematical transformations to highlight the degree of deviation between the prediction and the actual value. For example, for wind turbine rotational speed data, at a certain time t, the model predicts the rotational speed as ω_pre(t), while the actual rotational speed acquired by the SCADA system is ω_rea l(t). The corresponding residual value can be expressed as e(t) = ω_rea l(t) - ω_pre(t). Arranging the residual values ​​of all relevant parameters at different times in matrix form constitutes the residual matrix. The residual matrix can intuitively reflect the inconsistency between the model prediction and the actual operating state. This inconsistency is often a potential signal of wind turbine failure.

[0065] A fault warning is triggered when the root mean square error (RMSE) of the residual matrix exceeds a preset warning threshold. The formula for calculating the root mean square error (RMSE) is as follows:

[0066]

[0067] Among them, y i The value at time i in the actual SCADA data. α is the predicted value, N is the total number of data points, and α is the adjustment factor.

[0068] Setting the warning threshold is crucial in the fault early warning process. It requires comprehensive consideration of factors such as the fluctuation range of wind turbines during normal operation, the statistical characteristics of model prediction errors, and empirical data from actual fault cases. For example, by analyzing a large amount of historical normal operation data, the RMSE distribution of the residual matrix can be calculated, and its mean and standard deviation can be determined. Then, the warning threshold can be set as the mean plus a certain multiple of the standard deviation (e.g., the mean plus 3 times the standard deviation). During the operation of the wind turbine, once the RMSE of the residual matrix exceeds this threshold, it indicates that the operating status of the wind turbine may have changed abnormally, and there may be a risk of failure, requiring timely further inspection and diagnosis.

[0069] After triggering a fault warning, an in-depth analysis is conducted on the outlier variables in the residual matrix. Outlier variables typically refer to those parameters whose residual values ​​significantly deviate from the normal range, and are often closely related to specific fault types in wind turbine units. By analyzing the distribution characteristics, trends, and interrelationships of these outlier variables, a preliminary judgment can be made about the component or system where the fault may occur. For example, if the residual values ​​of gearbox oil temperature, oil pressure, and gearbox vibration-related parameters in the residual matrix all show large deviations simultaneously and exhibit certain correlated change patterns, such as consistently positive oil temperature residuals, consistently negative oil pressure residuals, and increasing vibration residual amplitude, it can be inferred that there may be potential faults in the gearbox, such as poor lubrication or gear wear.

[0070] By combining the mechanistic model of wind turbine units, historical fault data, and expert experience, a mapping relationship between fault types and abnormal features of the residual matrix is ​​established. Machine learning algorithms (such as decision trees, support vector machines, and deep neural networks) or data mining techniques are used to classify and identify abnormal patterns in the residual matrix, thereby accurately diagnosing specific fault types. For example, a fault sample database is constructed based on historical fault data, containing residual matrix feature samples corresponding to different types of faults. Then, a deep neural network is used to learn and train on these samples. When a new fault warning appears, the current residual matrix is ​​input into the trained neural network model, which can automatically output the most likely fault type, such as gearbox fault or generator stator winding short-circuit fault. Furthermore, the severity and development trend of the fault can be further evaluated and analyzed by combining the mechanistic model of the wind turbine unit, providing maintenance personnel with detailed fault diagnosis information and guiding them to perform targeted maintenance operations.

[0071] The fault diagnosis method for high-tower wind turbines comprehensively utilizes a series of advanced technologies, including SCADA data preprocessing, improved EEMD algorithm decomposition and feature extraction, enhanced visual Transformer (ViT) model state prediction, residual matrix calculation and fault early warning, and fault type identification and diagnosis, to construct a systematic, efficient, and accurate fault diagnosis framework for wind turbines. It can fully leverage the massive amounts of SCADA data generated during wind turbine operation, deeply mining the multi-scale feature information contained within the data to achieve accurate prediction of the real-time status of wind turbines and early warning and precise diagnosis of faults. This effectively improves the operational reliability and economy of high-tower wind turbines, possessing significant practical application value and broad market prospects. In practical applications, this method can be further optimized and customized according to different wind turbine models, operating environments, and specific needs to better meet the fault diagnosis requirements of the wind power industry and promote the healthy and sustainable development of the wind power industry.

[0072] The technical solution in this embodiment employs an improved EEMD algorithm to decompose SCADA data and extract multi-scale features, which helps capture subtle changes in the operating status of wind turbines and improves the sensitivity of fault detection. Combined with an enhanced visual Transformer model, utilizing multi-layer asymmetric convolution modules and deformable attention mechanisms, the model's ability to learn complex data patterns is enhanced, improving the accuracy of fault diagnosis. By analyzing the residual matrix, the prediction error tolerance is dynamically adjusted, and combined with trend analysis, the fault early warning strategy is optimized to reduce false alarms and missed alarms, improving the reliability of the early warning system. The system automatically identifies fault types and generates diagnostic reports, assisting maintenance personnel in quickly locating problems, developing targeted maintenance measures, reducing maintenance costs, and extending equipment lifespan.

[0073] In one embodiment, such as Figure 2 As shown, step S12 includes the following steps S21-S22:

[0074] In step S21, a sliding window is set, and extreme value extension is performed based on the extreme points and historical data within the window;

[0075] In step S22, the termination conditions of the decomposition are dynamically adjusted, and information leakage is eliminated by backtracking historical data.

[0076] In one embodiment, during the fault early warning and diagnosis process of high-tower wind turbines, an improved Empirical Mode Decomposition (EEMD) algorithm is used to decompose the preprocessed data when processing SCADA data. This process includes setting a sliding window for extreme value extension and dynamically adjusting the decomposition termination conditions to eliminate information leakage. When processing time-series data, the sliding window technique helps capture local features of the data by applying a fixed-size window to the data and moving it step-by-step for analysis. In EEMD decomposition, the window size should be set according to the characteristics of the data and the analysis objectives. A window that is too small may not capture enough contextual information, while a window that is too large may increase the computational load. The step size determines the speed at which the window moves. A smaller step size can obtain more detailed local features but requires more computation; a larger step size is more computationally efficient but may ignore some detailed information.

[0077] Within each sliding window, local extrema are identified, and these extrema are used for extremum extension to construct upper and lower envelopes. Local maxima and minima are identified by comparing the magnitudes of adjacent data points. Using methods such as cubic spline interpolation, all local maxima are connected (constructing the upper envelope) and all local minima are connected (constructing the lower envelope). The mean of the upper and lower envelopes is calculated as the local trend, and subtracting this mean yields the local oscillation component. By performing the above processing within each window, local features of the data can be effectively captured, improving the accuracy and reliability of EEMD decomposition.

[0078] During EEMD decomposition, termination conditions need to be set to determine when to stop decomposing the signal. Common termination conditions include Standard Deviation (SD), the S-Number criterion, and the threshold method. Standard Deviation means stopping decomposition when the standard deviation of two consecutive decomposition results is less than a preset threshold. The S-Number criterion means stopping decomposition when the number of zero-crossing points and extreme points is equal or the difference does not exceed 1 in multiple consecutive decomposition results. The threshold method involves setting a threshold for the change in the mean of the upper and lower envelopes; decomposition stops when the change is less than a preset threshold.

[0079] In practical applications, using historical data backtracking to set decomposition termination conditions can lead to information leakage, meaning the model acquires future information during training, impacting prediction performance. To eliminate information leakage, the entire dataset is decomposed using EEMD, ensuring all data uses the same number of IMFs, before being divided into training, validation, and test sets. This avoids inconsistent IMFs obtained across different datasets, preventing performance degradation. Furthermore, ensuring that only current and historical data are used during model training and validation prevents future information from affecting the model's generalization ability.

[0080] Assuming that the SCADA data of a high-tower wind turbine exhibits significant seasonal fluctuations, selecting a window size that encompasses the complete seasonal variation when using the sliding window technique for EEMD decomposition can effectively capture seasonal characteristics. By performing extreme value extension within each window to construct upper and lower envelopes, endpoint effects are reduced, improving decomposition accuracy. During the decomposition process, termination conditions are dynamically adjusted, such as based on changes in the standard deviation of historical data, ensuring the adaptability and accuracy of the decomposition process. These methods effectively eliminate information leakage and improve the reliability of fault early warning models.

[0081] In fault early warning and diagnosis of high-tower wind turbines, an improved EEMD algorithm is used to decompose SCADA data. Combined with sliding window and dynamic termination condition adjustment techniques, this effectively extracts multi-scale features of the data and improves the model's predictive performance. By reasonably setting the sliding window, extreme value extension method, and decomposition termination conditions, and combining historical data analysis, information leakage can be effectively eliminated, ensuring the model's generalization ability and reliability.

[0082] In one embodiment, such as Figure 3 As shown, it also includes the following steps S31-S32:

[0083] In step S31, a multi-layer asymmetric convolution module is used to extract local features from the feature matrix, and a deformable attention mechanism is combined to improve the computational efficiency of the model.

[0084] In step S32, the parameters are optimized through multiple rounds of training using the Adam optimizer and the mean squared error (MSE) loss function.

[0085] In one embodiment, a multi-layer asymmetric convolution module is used in the fault diagnosis method for high-tower wind turbines to extract local features from the feature matrix. An asymmetric convolution module typically refers to a convolution kernel with different sizes in different directions, such as a combination of 1×3 and 3×1 kernels. This expands the receptive field of the convolution operation and captures richer local feature information without significantly increasing computational load. In wind turbine fault diagnosis scenarios, the feature matrix contains multi-dimensional operating parameter data from different sensors, such as temperature, pressure, and vibration. Local feature patterns may exist at different time scales and spatial dimensions. Multi-layer asymmetric convolution modules can extract these local features layer by layer. For example, for the vibration signal feature matrix of a wind turbine gearbox, the asymmetric convolution module can capture minute impact features during gear meshing and periodic fluctuations in shaft rotation. These local features are often sensitive indicators of early faults.

[0086] Traditional attention mechanisms require comprehensive pairwise computation across the entire feature sequence. While they can capture global dependencies, they are computationally intensive and inefficient when handling large-scale feature data. Deformable attention mechanisms, however, introduce learnable offsets to adjust sampling positions, allowing attention calculations to focus more on key feature regions and reducing unnecessary computational overhead. In wind turbine fault diagnosis applications, this can more quickly filter out highly fault-related features from large SCADA data feature matrices, improving diagnostic speed. For example, when processing feature matrices composed of multi-point vibration monitoring data from wind turbine blades, deformable attention mechanisms can rapidly locate potentially problematic blade positions and their corresponding vibration frequency components without requiring equally complex calculations for all data points, thus significantly improving computational efficiency.

[0087] Using the Adam (Adaptive Moment Estimation) optimizer to optimize model parameters is a crucial step in this fault diagnosis method. The Adam optimizer is an optimization algorithm based on an adaptive learning rate. It combines the advantages of traditional stochastic gradient descent (SGD) with momentum acceleration characteristics, and can automatically adjust the learning rate according to different parameters. During the training of the wind turbine fault diagnosis model, facing complex nonlinear feature mapping relationships and a large amount of training data, the Adam optimizer can quickly converge to a better parameter solution. For example, in the early stages of model training, the Adam optimizer can quickly adjust parameters with a large learning rate, causing the model loss function to decrease rapidly. As training progresses, it automatically reduces the learning rate for fine-tuning the parameters, avoiding oscillations near the minimum of the loss function caused by an excessively large learning rate, thus ensuring the stability and efficiency of model training.

[0088] The Mean Squared Error (MSE) loss function, used as the objective function for model training, measures the difference between the model's predicted output and the actual operating state labels of wind turbines. In fault diagnosis scenarios, the MSE loss function directly reflects the model's accuracy in predicting the normal operating state and various fault states of wind turbines. By continuously optimizing the model parameters through multiple rounds of training to minimize the MSE loss function value, the model gradually learns the precise mapping relationship between the feature matrix and the operating state of the wind turbine. For example, during training, if the model's predicted value for a certain type of fault state has a large error compared to the actual value, the MSE loss function will give a large loss value, prompting the Adam optimizer to adjust the model parameters, making subsequent predictions increasingly closer to the reality, ultimately improving the accuracy and reliability of the entire fault diagnosis model.

[0089] In the high-tower wind turbine fault diagnosis method of this embodiment, a multi-layer asymmetric convolution module combined with a deformable attention mechanism is used to achieve efficient local feature extraction and computational acceleration of the wind turbine operating state feature matrix. Simultaneously, a multi-round training parameter optimization strategy based on the Adam optimizer and MSE loss function ensures that the model can accurately learn complex fault modes, thereby improving the overall performance of wind turbine fault diagnosis. This innovative technical solution provides strong support for the intelligent maintenance and reliable operation of wind turbines, and in practical applications, it can effectively reduce downtime due to faults and improve power generation efficiency and economic benefits.

[0090] In one embodiment, such as Figure 4 As shown, step S14 includes the following steps S41-S42:

[0091] In step S41, by analyzing the deviation between the predicted and actual values ​​in the residual matrix, the tolerance for prediction error is dynamically adjusted based on historical data and normal operating conditions.

[0092] In step S42, the probability of failure is calculated based on the changing trend and magnitude of the residual matrix.

[0093] In one embodiment, the fault early warning triggering mechanism for high-tower wind turbines requires in-depth analysis of historical data. This historical data contains rich information about the wind turbine's operation under different conditions, including the normal operating parameter ranges under various environmental conditions (such as different wind speeds, wind directions, and temperatures) and the abnormal data characteristics corresponding to past fault cases. Through the mining and statistical analysis of a large amount of historical data, a relatively reasonable prediction error baseline range can be preliminarily determined.

[0094] The operating status of wind turbines is not static. Even under similar external environments, the parameters of wind turbines during normal operation will exhibit certain dynamic changes due to factors such as wear and tear, aging, and short-term load fluctuations. Therefore, it is essential to continuously collect and evaluate data under normal operating conditions to promptly capture these subtle changes and provide real-time data for dynamically adjusting the tolerance of prediction errors.

[0095] For example, when analyzing wind turbine power output data, historical data might show that the normal fluctuation range of power output within a specific wind speed range is ±5%. However, in actual operation, due to seasonal changes or minor wear and tear on internal components, the power fluctuation during normal operation may gradually increase to ±7%. In this case, relying on the fixed tolerance of ±5% from historical data to diagnose faults could lead to frequent false alarms. Therefore, dynamically adjusting the tolerance based on real-time normal operation data, appropriately widening it to ±7%, can more accurately reflect the actual operating characteristics of the wind turbine and avoid unnecessary misjudgments.

[0096] Wind turbines operate in complex and ever-changing natural environments, constantly affected by factors such as wind and sand erosion, rainwater runoff, and temperature fluctuations, causing their performance and condition to gradually change. Furthermore, the operating load of wind turbines is not constant and adjusts in real time according to factors such as wind speed and grid demand. A mechanism for dynamically adjusting the tolerance of prediction errors can respond promptly to these changes. When the wind turbine is in a stable operating period, with parameters fluctuating little and remaining within a certain range, the tolerance can be appropriately tightened to improve the sensitivity of fault warnings and facilitate the timely detection of minor anomalies. Conversely, when the wind turbine has just completed maintenance or when there are significant changes in the operating environment (such as entering a season with frequent wind speed fluctuations), parameters may experience significant fluctuations in the short term. In such cases, the tolerance should be appropriately relaxed to avoid misinterpreting normal adjustments as faults. For example, regarding the gearbox oil temperature parameter of a wind turbine, the normal range of oil temperature may rise in the high-temperature environment of summer, and the tolerance for prediction error should also be increased accordingly. However, in the period after maintenance, due to the different degrees of break-in of various components, the oil temperature may fluctuate in the short term. At this time, the dynamic adjustment of the tolerance can effectively avoid false alarms while not overlooking real signs of faults.

[0097] The residual matrix contains information about the deviation between predicted and actual values. By comprehensively analyzing its changing trends and magnitudes, the probability of a fault can be accurately calculated. The changing trend reflects the direction and rate of deviation evolution over time. For example, if the residual value continues to increase, it indicates that the wind turbine's operating state may be gradually deviating from its normal trajectory, posing a potential fault risk. The magnitude represents the absolute size of the deviation; a larger residual magnitude means that the difference between the current state and the normal mode is more significant, and the probability of a fault is higher. Specifically, when the residual value of a key parameter of the wind turbine (such as generator speed) shows an upward trend at multiple consecutive time points, and the increase exceeds a certain threshold, it can be considered that the system corresponding to that parameter may have an anomaly. For example, if the residual value gradually rises from near zero to more than 1.5 times the tolerance limit, and the upward trend continues, this strongly suggests that there may be a problem with the generator speed control mechanism, such as control algorithm parameter drift or actuator failure.

[0098] In calculating the probability of failure, it is necessary to comprehensively consider various factors, including the residual characteristics of different parameters in the residual matrix, the correlation between parameters, and the operating conditions of the wind turbine. Different parameters have different sensitivities to the operating state of the wind turbine. Anomalies in some parameters may directly indicate serious failures, while slight fluctuations in others may simply be random disturbances during normal operation. For example, for gearbox vibration parameters and oil temperature parameters, changes in the residuals of vibration parameters may more directly reflect the mechanical failure status of the gearbox, and therefore have a higher weight in the corresponding failure probability calculation. However, changes in the residuals of oil temperature parameters may be affected by various factors (such as ambient temperature and load), and have a relatively lower weight. Simultaneously, the correlation between the gearbox and other components (such as the generator and blades) must be considered. When the gearbox malfunctions, it may have a chain reaction on the operation of the generator; therefore, these interconnected effects need to be comprehensively evaluated when calculating the probability of failure.

[0099] Furthermore, the operating conditions of wind turbines also affect the assessment of the likelihood of failure. Under high-load operating conditions, the residual tolerance of certain parameters can be appropriately relaxed, and the threshold for calculating the likelihood of failure can be adjusted accordingly to adapt to the complex operating environment.

[0100] The fault early warning triggering mechanism for high-tower wind turbines in this embodiment effectively adapts to the dynamic changes in the operating status of wind turbines by dynamically adjusting the prediction error tolerance and employing a precise strategy based on the changing trend and magnitude of the residual matrix to calculate the probability of faults. This improves the accuracy and reliability of fault early warning. It not only fully leverages the value of historical and normal operating data but also comprehensively considers various actual operating factors, providing strong support for early fault warning and preventative maintenance of wind turbines. This helps reduce downtime due to faults, ensures the safe and stable operation of wind turbines, and enhances the overall economic benefits and operational management level of wind farms.

[0101] In one embodiment, such as Figure 5 As shown, step S15 includes the following steps S51-S52:

[0102] In step S51, by analyzing the abnormal patterns in the residual matrix, the variables that have failed are identified, and then the failure type of the wind turbine is inferred.

[0103] In step S52, a detailed diagnostic report of the high-tower wind turbine fault is output based on the fault type.

[0104] In one embodiment, the residual matrix is ​​the core data foundation in the fault diagnosis process of high-tower wind turbines. Through systematic analysis of the residual matrix, the variables causing the fault can be accurately located, and the specific fault type can be further inferred. Anomalies in the residual matrix typically manifest as significant deviations from the normal range of residual values, specific trends, and abnormal correlations with the residuals of other variables. For example, when a gearbox in a wind turbine malfunctions, the residual values ​​of gearbox oil temperature, oil pressure, and vibration-related parameters may simultaneously increase significantly, and the trends of these residual values ​​may exhibit certain correlations, such as a continuous increase in oil temperature residuals, a decrease in oil pressure residual fluctuations, and an increase in vibration residual amplitude with changes in frequency components. These anomalies contrast sharply with the residual characteristics during normal gearbox operation, thus enabling effective identification.

[0105] If an abnormal pattern is detected in the residual matrix, the specific variable causing the fault can be identified. This is achieved by setting reasonable thresholds and judgment rules. For example, for the residual sequence of each monitored variable, its mean, standard deviation, and other statistical characteristics are calculated. When the residual value exceeds the mean plus a certain multiple of the standard deviation, the variable can be judged to be faulty. After identifying the faulty variable, the possible fault types are inferred based on the wind turbine's mechanistic model, fault mode database, and expert experience. Different fault types often correspond to specific combinations of variables and abnormal patterns. For example, if the generator stator temperature residual is consistently high, the stator current residual shows periodic fluctuations, and the rotor vibration residual increases, it can be inferred that the generator may have a stator winding short-circuit fault.

[0106] Based on the identified fault type, a fault diagnosis report for the high-tower wind turbine is generated. The report typically includes a fault overview, detailed fault analysis, fault impact assessment, and maintenance recommendations.

[0107] The fault overview provides a brief description of the time of the fault, the wind turbine number, the fault type, and the initial assessment of the fault severity, enabling maintenance personnel to quickly understand the overall situation of the fault.

[0108] Detailed fault analysis delves into the abnormal behavior of various relevant variables when a fault occurs, including the trend and magnitude of residual value changes and correlation analysis with other variables; it analyzes the specific causes that may lead to the fault in conjunction with the operating principle and structure of the wind turbine, such as component wear, poor lubrication, and the influence of external environmental factors; and it provides a description of the fault development process to help maintenance personnel fully understand the fault evolution mechanism.

[0109] Fault impact assessment evaluates the potential impact of a fault on the wind turbine's power generation efficiency, operational stability, other components, and the possible cascading effects on the entire wind farm's operation. For example, a gearbox failure may lead to a decrease in power generation, vibration transmission causing fatigue damage to adjacent components, or even force the wind turbine to be shut down for maintenance, affecting the wind farm's overall power generation and economic benefits.

[0110] The maintenance recommendations provide specific maintenance suggestions based on the type and severity of the fault, including a list of parts to be replaced, maintenance procedures, required tools and equipment, and safety precautions. They also provide estimated maintenance time and costs to help wind farm operators plan maintenance schedules and allocate resources effectively.

[0111] Fault diagnosis reports should be presented using intuitive and easy-to-understand language combined with charts and graphs. For example, line graphs can be used to show the trend of residual values, tables can list fault variables and their abnormal characteristics, and flowcharts can represent the fault inference logic and repair steps. This allows maintenance personnel to quickly and accurately obtain key information and put it into practice. Furthermore, the report should have a certain degree of traceability and verifiability. It should record the data sources, analysis methods, and judgment criteria used in the fault diagnosis process for verification and validation in subsequent maintenance processes. Simultaneously, it should provide data support for subsequent fault case analysis and model optimization, continuously improving the performance and accuracy of the fault diagnosis system.

[0112] The fault type identification and diagnosis report generation mechanism for high-tower wind turbines in this embodiment accurately locates fault variables and infers fault types through in-depth analysis of abnormal patterns in the residual matrix, thereby generating detailed and clearly structured fault diagnosis reports. This not only provides wind turbine maintenance personnel with clear fault information and maintenance guidance, helping to promptly repair faults, reduce downtime, and lower maintenance costs, but also improves the intelligence and reliability of the wind turbine fault diagnosis system by continuously accumulating fault diagnosis cases and optimizing the diagnostic model, providing a solid guarantee for the safe and stable operation of high-tower wind turbines and the sustainable development of the wind power industry.

[0113] In one embodiment, Figure 6 This is a block diagram illustrating a fault early warning and diagnosis system for high-tower wind turbines according to an exemplary embodiment. Figure 6 As shown, the high-tower wind turbine fault early warning and diagnosis system includes a preprocessing module 61, an extraction module 62, a prediction module 63, an early warning module 64, and a diagnosis module 65.

[0114] The preprocessing module 61 is used to preprocess the SCADA data of the high-tower wind turbine.

[0115] The extraction module 62 is used to decompose the preprocessed SCADA data using the improved EEMD algorithm, extract feature information at least two time scales, and concatenate the IMF components at at least two time points into a feature matrix.

[0116] The prediction module 63 is used to input the feature matrix into the enhanced visual Transformer (Vi T) model to obtain the prediction result of the real-time state of the wind turbine.

[0117] The early warning module 64 is used to calculate the residual matrix based on the prediction results and the SCADA data. When the root mean square error of the residual matrix exceeds a preset early warning threshold, a fault early warning is triggered.

[0118] The diagnostic module 65 is used to identify the fault type of the high-tower wind turbine and perform fault diagnosis by analyzing the abnormal variables in the residual matrix.

[0119] The preprocessing module 61, the extraction module 62, the prediction module 63, the early warning module 64, and the diagnosis module 65 included in the block diagram of the high-tower wind turbine fault early warning and diagnosis system are controlled to execute the high-tower wind turbine fault early warning and diagnosis method described in any of the above embodiments.

[0120] like Figure 7 As shown, the present invention provides an electronic device 700, which includes: a communication interface, a processor 701, and a memory 702;

[0121] The memory 702 stores program instructions. When executed by the processor 701, which is connected to the memory 702 via the communication interface, the program instructions preprocess the SCADA data of the high-tower wind turbine. The preprocessed SCADA data is decomposed using an improved EEMD algorithm to extract feature information at least two time scales, and the IMF components at at least two time points are concatenated into a feature matrix. The feature matrix is ​​input into an enhanced vision Transformer (ViT) model to obtain a prediction result of the wind turbine's real-time state. Based on the prediction result and the SCADA data, a residual matrix is ​​calculated. When the root mean square error of the residual matrix exceeds a preset warning threshold, a fault warning is triggered. By analyzing the abnormal variables in the residual matrix, the fault type of the high-tower wind turbine is identified and fault diagnosis is performed.

[0122] This invention provides a computer-readable storage medium storing computer program instructions. When executed by a processor, the computer program instructions preprocess SCADA data of a high-tower wind turbine. An improved EEMD algorithm is used to decompose the preprocessed SCADA data, extracting feature information at least two time scales, and concatenating the IMF components of at least two time points into a feature matrix. The feature matrix is ​​then input into an enhanced vision Transformer (ViT) model to obtain a prediction of the wind turbine's real-time state. Based on the prediction and the SCADA data, a residual matrix is ​​calculated. When the root mean square error of the residual matrix exceeds a preset warning threshold, a fault warning is triggered. By analyzing the abnormal variables in the residual matrix, the fault type of the high-tower wind turbine is identified and fault diagnosis is performed.

[0123] It should be understood that the specific features, operations, and details described above regarding the method of the present invention can also be similarly applied to the apparatus and system of the present invention, or vice versa. Furthermore, each step of the method of the present invention described above can be performed by a corresponding component or unit of the apparatus or system of the present invention.

[0124] It should be understood that the various modules / units of the system of the present invention can be implemented wholly or partially through software, hardware, firmware, or a combination thereof. Each module / unit can be embedded in the processor of a computer device in hardware or firmware form or independent of the processor, or it can be stored in the memory of a computer device in software form for the processor to call to execute the operation of each module / unit. Each module / unit can be implemented as an independent component or module, or two or more modules / units can be implemented as a single component or module.

[0125] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores computer instructions executable by the processor, which, when executed by the processor, instruct the processor to perform steps of the methods of embodiments of the present invention. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the methods of the present invention.

[0126] This invention can be implemented as a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0127] Those skilled in the art will understand that the method steps of this invention can be performed by a computer program instructing related hardware, such as a computer device or processor. The computer program may be stored in a non-transitory computer-readable storage medium, and its execution causes the steps of this invention to be performed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage devices, optical data storage devices, hard disks, and solid-state drives. Examples of volatile memory include random access memory (RAM) and external cache memory. (Note: The text also mentions using an improved EEMD algorithm to decompose SCADA data and extract multi-scale features, which helps capture subtle changes in wind turbine operating status and improves fault detection sensitivity. It also mentions using a reinforced visual TRANSFORMER model, utilizing multi-layer asymmetric convolution modules and deformable attention mechanisms to enhance the model's learning ability for complex data patterns and improve fault diagnosis accuracy. The system automatically identifies fault types and generates diagnostic reports, assisting maintenance personnel in quickly locating problems, developing targeted maintenance measures, reducing maintenance costs, and extending equipment lifespan.)

[0128] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for fault early warning and diagnosis of high-tower wind turbine units, characterized in that, include: Preprocessing of SCADA data for high-tower wind turbines; The preprocessed SCADA data is decomposed using an improved EEMD algorithm. The decomposition of the preprocessed SCADA data using the improved EEMD algorithm includes: setting a sliding window and extending the extreme values ​​based on the extreme points and historical data within the window; dynamically adjusting the termination conditions of the decomposition; eliminating information leakage by backtracking through historical data; extracting feature information at least two time scales; and concatenating the IMF components of at least two time points into a feature matrix. The feature matrix is ​​input into the enhanced vision Transformer (ViT) model to obtain the prediction results of the real-time state of the wind turbine. Based on the prediction results and the SCADA data, a residual matrix is ​​calculated. When the root mean square error of the residual matrix exceeds a preset warning threshold, a fault warning is triggered. The triggering mechanism for the fault warning includes: By analyzing the deviation between predicted and actual values ​​in the residual matrix, the tolerance for prediction errors is dynamically adjusted based on historical data and normal operating conditions. The probability of failure is calculated in real time based on the changing trend and magnitude of the residual matrix. By analyzing outliers in the residual matrix, fault types of high-tower wind turbines are identified and fault diagnosis is performed. This identification and diagnosis includes: By analyzing the abnormal patterns in the residual matrix, the variables that cause the fault are identified, and the fault type of the wind turbine is inferred. Based on the fault type, a detailed diagnostic report of the high-tower wind turbine fault is output. A multi-layer asymmetric convolution module is used to extract local features from the feature matrix, and a deformable attention mechanism is combined to improve the computational efficiency of the model. The parameters are optimized through multiple rounds of training using the Adam optimizer and the mean squared error (MSE) loss function.

2. The method for fault early warning and diagnosis of high-tower wind turbine units as described in claim 1, characterized in that, Also includes: The root mean square error (RMSE) is calculated as follows: Among them, y i The value at time i in the actual SCADA data. α is the predicted value, N is the total number of data points, and α is the adjustment factor.

3. A fault early warning and diagnosis system for high-tower wind turbine units, characterized in that, include: The preprocessing module is used to preprocess the SCADA data of high-tower wind turbines; The extraction module is used to decompose the preprocessed SCADA data using an improved EEMD algorithm. The decomposition of the preprocessed SCADA data using the improved EEMD algorithm includes: setting a sliding window and performing extreme value extension based on the extreme points and historical data within the window; dynamically adjusting the termination conditions of the decomposition; eliminating information leakage by backtracking through historical data; extracting feature information at least two time scales; and concatenating the IMF components of at least two time points into a feature matrix. The prediction module is used to input the feature matrix into the enhanced vision Transformer (ViT) model to obtain the prediction result of the real-time state of the wind turbine. The early warning module is used to calculate the residual matrix based on the prediction results and the SCADA data. When the root mean square error of the residual matrix exceeds a preset early warning threshold, a fault early warning is triggered. The triggering mechanism of the fault early warning includes: dynamically adjusting the tolerance of the prediction error based on historical data and normal operating status by analyzing the deviation between the predicted value and the actual value in the residual matrix; and calculating the probability of a fault in real time according to the changing trend and magnitude of the residual matrix. The diagnostic module is used to identify the fault type of the high-tower wind turbine and perform fault diagnosis by analyzing the abnormal variables in the residual matrix. The identification of the fault type of the high-tower wind turbine and the performance of fault diagnosis include: determining the variables that have caused the fault by analyzing the abnormal patterns in the residual matrix, and then inferring the fault type of the wind turbine; and outputting a detailed diagnostic report of the high-tower wind turbine fault based on the fault type. The system is also used to extract local features from the feature matrix using a multi-layer asymmetric convolution module, and to improve the computational efficiency of the model by combining a deformable attention mechanism; the Adam optimizer and mean squared error (MSE) loss function are used to optimize the parameters through multiple rounds of training.

4. An electronic device, characterized in that, include: Communication interface, processor, memory; The memory is used to store program instructions, which, when executed by the processor that is connected to the memory via the communication interface, enable the electronic device to implement the high-tower wind turbine fault early warning and diagnosis method according to any one of claims 1 to 2.

5. A computer-readable storage medium having program instructions stored thereon, characterized in that, When the program instructions are executed by the computer, the computer implements the high-tower wind turbine fault early warning and diagnosis method according to any one of claims 1 to 2.

Images