Fault monitoring method and system for wind power equipment based on operation data

Abstract

The invention discloses a wind power equipment fault monitoring method and system based on operation data, and the method comprises the steps: firstly obtaining historical operation data containing a fault type, employing a preset relevance algorithm to quantify the relevance between each operation parameter and a fault, mapping the relevance to a two-dimensional plane, constructing a structured point set, and carrying out the calculation of the structured point set; and the visibility and the processability of the multi-dimensional data are enhanced. In the key influence factor screening process, the operation parameter with the maximum correlation degree is selected as a reference point, the method has definite physical fault characterization significance, can focus on core parameters highly related to a target fault mechanism, and improves the accuracy of a candidate factor set from the source, so that the key influence factors are further screened. And the exclusive fault monitoring sub-model is constructed, so that high pertinence of model input can be ensured, the recognition accuracy and classification definition of various faults are greatly improved, and the fault detection accuracy of the wind power equipment is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power equipment monitoring, and in particular to a method and system for monitoring faults of wind power equipment based on operating data. Background Art

[0002] Wind turbines are integrated systems that convert wind energy into electrical energy. Their core components include rotors (blades), mechanical energy conversion devices (such as gearboxes and bearings), generators, support structures (towers), and intelligent control systems. Based on the main axis orientation, wind turbines can be categorized as horizontal-axis wind turbines (HAWTs) or vertical-axis wind turbines (VAWTs). HAWTs account for over 95% of the global market share and are widely used in large onshore and offshore wind farms. Failures in wind turbines can have serious impacts on operational safety and economic profitability, including power generation losses, shortened equipment lifespans, increased safety hazards, increased O&M costs, and environmental pollution risks. For example, a gearbox failure can cause a wind turbine to shut down for 7-15 days, resulting in an average daily power generation loss of 30,000 to 50,000 yuan. The O&M costs of offshore wind farms are more than three times those of onshore wind farms. Furthermore, equipment failures such as bearing micropitting or electrical system overheating can cause fires or structural damage, posing risks to personnel and the environment.

[0003] However, existing fault monitoring methods rely on manual experience or full-feature input models, as wind turbine operation involves hundreds of parameters, and the correlations between each parameter and specific faults are complex and variable. This leads to unclear fault characterization and high feature redundancy. Furthermore, wind turbines often experience the coexistence of multiple faults (such as the combined effects of gear wear and bearing damage). Traditional single models struggle to distinguish the coupled features, resulting in missed detections or misjudgments. Summary of the Invention

[0004] The present invention provides a method and system for monitoring faults of wind power equipment based on operating data, so as to improve the accuracy of wind power equipment fault monitoring.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides a fault monitoring method for wind power equipment based on operation data, comprising: Acquire historical operation data of the wind power equipment, wherein the historical operation data includes a fault type and fault operation data corresponding to the fault type; For any fault type, the correlation between each operating parameter in the fault operation data and the fault type is calculated based on a preset correlation algorithm, and the correlation of each fault type is mapped to a two-dimensional plane to generate a point set of the two-dimensional plane; In the point set of the two-dimensional plane, the eigenvector with the largest correlation is determined as a reference point, a candidate circle is generated with the reference point as the center and a preset length threshold as the radius, and all eigenvectors within the candidate circle are screened as a candidate key factor set; Performing a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and using the operating parameters that pass the check as the key influencing factors of the fault type; Obtaining key influencing factors corresponding to each fault type, and constructing fault monitoring sub-models based on the key influencing factors; Each fault condition is monitored separately based on the fault monitoring sub-model to complete the fault monitoring of the wind power equipment.

[0006] The present invention first obtains historical operating data containing fault types, uses a preset correlation algorithm to quantify the correlation between each operating parameter and the fault, and maps it to a two-dimensional plane to construct a structured point set, thereby enhancing the visibility and processability of multidimensional data. In the process of screening key influencing factors, the operating parameter with the greatest correlation is selected as the reference point, which has a clear physical fault characterization meaning and can focus on the core parameters that are highly correlated with the target fault mechanism, thereby improving the accuracy of the candidate factor set from the source. At the same time, parameters with low correlation with the fault and susceptible to noise interference are excluded to improve the robustness of fault feature extraction. The construction of the candidate circle adopts a dynamic radius adjustment mechanism, which adaptively adjusts the radius size based on the data distribution characteristics, achieves a balance between noise resistance and generalization ability, and avoids the risk of misjudgment or missed judgment brought by traditional fixed thresholds. In addition, the operating parameters show clustering characteristics in two-dimensional space driven by correlation, and the candidate circle can naturally capture the local parameter clustering area, thereby further enhancing the screening efficiency and accuracy. For multiple fault concurrent scenarios, various fault-related factors are spatially separated in the two-dimensional plane. By generating candidate circles separately, the fault decoupling identification of key parameters can be achieved, and the response capability to fault linkage under complex working conditions can be improved. Finally, compared with the traditional global clustering algorithm, the candidate circle strategy only needs to calculate the distance between the point and the circle center, and the time complexity is reduced by This reduces the time required to O(n), significantly improving online processing efficiency and better adapting to the needs of high-frequency monitoring and real-time early warning for wind turbines. By extracting the most representative key influencing factors for different fault types and constructing dedicated fault monitoring sub-models based on them, the model inputs are highly targeted, significantly improving the accuracy of fault identification and classification, and enhancing the accuracy of wind turbine fault detection.

[0007] Furthermore, for any fault type, the correlation between each operating parameter in the fault operation data and the target fault label is calculated based on a preset correlation algorithm, and the correlation of each operating parameter is mapped to a two-dimensional plane to generate a point set of the two-dimensional plane, including: constructing a joint probability model based on the various operating parameters in the fault operation data; Evaluate the correlation between each operating parameter and the target fault label based on the maximum correlation entropy criterion and the joint probability model; Based on the preset dimensionality reduction algorithm, the correlation between each operating parameter and the target fault label is mapped to a two-dimensional plane to generate a point set representing the correlation between each operating parameter and the target fault label.

[0008] This method precisely quantifies the correlation between operating parameters and target fault labels using a joint probability model and the maximum correlation entropy criterion. Using a dimensionality reduction algorithm, this correlation is mapped onto a two-dimensional plane, forming a visual point set. The joint probability model simultaneously characterizes the statistical dependencies between multiple parameters, while the maximum correlation entropy criterion ensures that the selected parameters and fault labels have the greatest information gain. Furthermore, the dimensionality reduction mapping visually demonstrates the strength of the correlation between each parameter and the fault in two dimensions, providing a more reliable mathematical foundation and visualization support for subsequent key factor screening, significantly improving the accuracy and interpretability of initial feature extraction.

[0009] Furthermore, determining the feature vector with the greatest correlation as a reference point, calculating the distribution characteristics of the point set based on the reference point, and determining the screening threshold based on the distribution characteristics includes: Determine the eigenvector with the largest correlation as a reference point, and calculate the Euclidean distance distribution of the point set based on the reference point; The standard deviation of the point set is calculated based on the Euclidean distance distribution, and the product of a preset scale factor and the standard deviation is used as a screening threshold.

[0010] This method identifies the most correlated eigenvector as a reference point. Based on this reference point, the Euclidean distance distribution of the statistical point set is calculated, and a screening threshold is adaptively generated by multiplying the standard deviation by a preset scaling factor. This method dynamically adjusts the screening radius based on the actual "convergence" of the data, balancing noise reduction and coverage. It also avoids the problems of excessively loose or excessively tight fixed thresholds, achieving precise capture of highly correlated parameters and natural exclusion of noisy parameters, significantly improving the robustness and versatility of key factor screening.

[0011] Furthermore, the performing of a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and using the operating parameters that pass the check as the key influencing factors of the fault type, includes: Screening out candidate parameter sequences of respective candidate key factors from the historical operation data based on the candidate key factor set; Generate overlapping sliding windows on each candidate parameter sequence based on a preset window length and a preset step size; For any candidate parameter sequence, calculate the local correlation value between the candidate key factor and the target fault label in the sliding window; count the mean and standard deviation of the local correlation values ​​of all sliding windows on the candidate parameter sequence; when the mean is greater than or equal to the preset minimum correlation threshold and the standard deviation is less than or equal to the preset maximum fluctuation threshold, determine that the candidate key factor has passed the verification and use it as the key influencing factor of the fault type.

[0012] After screening the candidate key factors, the present invention introduces a sliding window to perform a secondary verification on them. By calculating the local correlation in multiple overlapping windows and counting their mean and standard deviation, the minimum correlation threshold and the maximum fluctuation threshold are set for strict judgment. It can eliminate short-term occasional or highly volatile pseudo-correlation parameters and only retain influencing factors with continuous and stable correlation, thereby significantly reducing the risk of false alarms and missed alarms and enhancing the temporal stability and diagnostic credibility of fault influencing factors.

[0013] Furthermore, the acquisition of key influencing factors corresponding to each fault type and the construction of fault monitoring sub-models based on the key influencing factors include: Obtain the key influencing factors corresponding to each fault type, extract the time series characteristics of the key influencing factors from the corresponding fault operation data, and construct a multidimensional feature vector; Matching a target prediction algorithm based on the multidimensional feature vector, and constructing an initial fault monitoring sub-model based on the target prediction algorithm; Training data is generated based on the multidimensional feature vector, and the initial fault monitoring sub-model is trained based on the training data to obtain a fault monitoring sub-model.

[0014] The present invention extracts the time series characteristics of the key influencing factors that have passed the secondary verification for each type of fault, constructs a multi-dimensional feature vector, and constructs an initial monitoring sub-model based on the target prediction algorithm that matches the fault mechanism. Each sub-model uses the most representative features as input, which not only ensures the specificity and diagnostic accuracy of the model, but also lays a highly targeted foundation for subsequent model training and optimization, and improves the coverage capability and recognition performance of multiple fault types.

[0015] Furthermore, generating training data based on the multidimensional feature vector, and training the initial fault monitoring sub-model based on the training data to obtain the fault monitoring sub-model includes: Dividing the multidimensional feature vector into fixed-length sequence segments in a sliding window manner according to time sequence, and labeling each sequence segment with a corresponding fault type or normal operating condition label to generate training data; The training data is divided into a training subset and a validation subset according to a preset ratio, and the initial fault monitoring sub-model is trained based on the training subset and the validation subset to obtain a fault monitoring sub-model.

[0016] This method uses a sliding window approach to segment multidimensional feature vectors into fixed-length sequence segments and label them, then divides them into training and validation sets for model training. This time-series slicing enhances sample diversity, effectively alleviating the problems of scarce fault samples and class imbalance. Furthermore, the separation of training and validation sets and the cross-validation mechanism further ensure the stability of the resulting fault monitoring submodel.

[0017] Furthermore, the fault monitoring sub-model is used to monitor each fault condition to complete the fault monitoring of the wind power equipment, including: Collecting the operating data of key influencing factors corresponding to each fault type in real time, and performing denoising, interpolation and normalization preprocessing on the operating data to generate input vectors for each fault detection sub-model; Inputting the input vector into the corresponding fault monitoring sub-model to obtain a prediction score of the fault type; An alarm is issued for the fault type based on the prediction score and a preset alarm threshold, thereby completing fault monitoring of the wind power equipment.

[0018] The present invention collects and preprocesses key influencing factor data in real time, inputs it into respective fault monitoring sub-models for online prediction, and triggers alarms based on the prediction scores and preset thresholds. It also supports the logical fusion and dynamic incremental update of multi-model outputs, thereby constructing a full-process closed loop from data collection, preprocessing, model reasoning to alarm output. It can not only realize real-time and accurate monitoring of various faults, but also ensure that the system continuously adapts to environmental changes and fault evolution during operation through dynamic adjustment and incremental training, comprehensively improving the timeliness of early warning and operation and maintenance efficiency of wind power equipment.

[0019] In a second aspect, the present invention provides a fault monitoring system for wind power equipment based on operating data, comprising: a data acquisition module, a correlation calculation module, a distribution determination module, a candidate factor screening module, a key factor screening module, a model building module, and a fault detection module; The data acquisition module is used to obtain historical operation data of the wind power equipment, wherein the historical operation data includes the fault type and the fault operation data corresponding to the fault type; The correlation calculation module is used to calculate the correlation between each operating parameter in the fault operation data and the target fault label based on a preset correlation algorithm for any fault type, and map the correlation of each operating parameter to a two-dimensional plane to generate a point set of the two-dimensional plane; The distribution determination module is configured to determine the feature vector with the greatest correlation as a reference point, calculate the distribution characteristics of the point set based on the reference point, and determine a screening threshold based on the distribution characteristics; The candidate factor screening module is configured to generate a candidate circle with the reference point as the center and the screening threshold as the radius, and screen all feature vectors within the candidate circle as a candidate key factor set; The key factor screening module is used to perform a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and use the operating parameters that pass the check as the key influencing factors of the fault type; The model building module is used to obtain key influencing factors corresponding to each fault type and build fault monitoring sub-models based on the key influencing factors; The fault detection module is used to monitor each fault condition based on the fault monitoring sub-model to complete the fault monitoring of the wind power equipment.

[0020] Furthermore, the distribution determination module is configured to determine the feature vector with the greatest correlation as a reference point, calculate the distribution characteristics of the point set based on the reference point, and determine a screening threshold based on the distribution characteristics, including: Determine the eigenvector with the largest correlation as a reference point, and calculate the Euclidean distance distribution of the point set based on the reference point; The standard deviation of the point set is calculated based on the Euclidean distance distribution, and the product of a preset scale factor and the standard deviation is used as a screening threshold.

[0021] Furthermore, the candidate factor screening module is configured to perform a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and use the operating parameters that pass the check as the key influencing factors of the fault type, including: Screening out candidate parameter sequences of respective candidate key factors from the historical operation data based on the candidate key factor set; Generate overlapping sliding windows on each candidate parameter sequence based on a preset window length and a preset step size; For any candidate parameter sequence, calculate the local correlation value between the candidate key factor and the target fault label in the sliding window; count the mean and standard deviation of the local correlation values ​​of all sliding windows on the candidate parameter sequence; when the mean is greater than or equal to the preset minimum correlation threshold and the standard deviation is less than or equal to the preset maximum fluctuation threshold, determine that the candidate key factor has passed the verification and use it as the key influencing factor of the fault type. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1A schematic flow chart of a method for fault monitoring of wind power equipment based on operating data provided by an embodiment of the present invention; Figure 2 A structural schematic diagram of a fault monitoring system for wind power equipment based on operation data is provided. DETAILED DESCRIPTION

[0023] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0024] The terms "first," "second," and the like in the specification, claims, and drawings of this application are used to distinguish between different objects, not to describe a particular order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements, but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0025] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0026] Example 1 See also Figure 1 , Figure 1 A schematic flow chart of a method for monitoring a wind turbine fault based on operating data provided by an embodiment of the present invention. The method includes steps 101 to 107, which are as follows: Step 101: Acquire historical operation data of a wind power device, wherein the historical operation data includes a fault type and corresponding fault operation data under the fault type; In this embodiment, during the daily operation of a wind farm, wind turbine operating parameters are collected in real time using equipment such as a SCADA system, a vibration monitoring system, an oil analyzer, and temperature sensors, and stored in a historical database. This historical operating data includes at least a fault type tag and the corresponding fault operating data, which includes various operating parameters.

[0027] In this embodiment, whenever a maintenance shutdown or automatic alarm occurs, the operation and maintenance system generates a corresponding fault type label based on the fault symptoms and maintenance report, such as "gearbox gear wear," "early bearing cracks," or "generator stator winding overheating." These fault type labels can be manually reviewed or confirmed by an automatic classification module based on a fault knowledge base to ensure accuracy and consistency.

[0028] In this embodiment, for each fault type, multiple operating parameter sequences of the unit are collected and recorded within a preset time window before and after the fault occurs (for example, 24 hours before and after the fault). The parameters include but are not limited to: mechanical vibration signals (peak acceleration, spectral energy distribution, 0.5 times the meshing frequency sideband amplitude, etc.); oil monitoring indicators (oil viscosity, water content, metal wear debris concentration, etc.); temperature data (gearbox oil temperature, bearing temperature, generator stator winding temperature, etc.); electrical parameters (stator current harmonic content, power factor, generator power output, etc.); environmental and operating conditions (wind speed, wind direction, unit speed, yaw angle, etc.).

[0029] In this embodiment, each operating parameter is continuously recorded at a fixed sampling frequency (eg, 1 Hz or 10 Hz) and marked with a timestamp to form complete time series data.

[0030] In this embodiment, the synchronous storage of fault type labels and fault operation data creates a structured historical operation data table. Each record contains fields such as "fault type," "collection time period," "operation parameter name," and "parameter value sequence," laying the data foundation for subsequent correlation calculation and feature extraction based on correlation algorithms.

[0031] Step 102: For any fault type, calculate the correlation between each operating parameter in the fault operation data and the target fault label based on a preset correlation algorithm, and map the correlation of each operating parameter to a two-dimensional plane to generate a point set on the two-dimensional plane; In this embodiment, for any fault type, the correlation between each operating parameter in the fault operation data and the target fault label is calculated based on a preset correlation algorithm, and the correlation of each operating parameter is mapped to a two-dimensional plane to generate a point set on the two-dimensional plane, including: constructing a joint probability model based on the various operating parameters in the fault operation data; Evaluate the correlation between each operating parameter and the target fault label based on the maximum correlation entropy criterion and the joint probability model; Based on the preset dimensionality reduction algorithm, the correlation between each operating parameter and the target fault label is mapped to a two-dimensional plane to generate a point set representing the correlation between each operating parameter and the target fault label.

[0032] In this example, taking the "gearbox gear wear" fault type of a wind turbine as an example, the system first extracts a multi-source sensor data sequence containing fault labels from the historical operation database. The data includes several (M) operating parameters, such as cabin temperature, gearbox oil temperature, 0.5 times the meshing frequency sideband amplitude in the main shaft vibration envelope spectrum, and lubricant metal particle concentration. For any fault type, a joint probability model is constructed using kernel density estimation (KDE) or the multivariate Gaussian distribution assumption, using these operating parameters and the fault label as input. , and calculate the marginal distribution ; Then, based on the maximum correlation entropy criterion, the mutual information formula is used to quantify the correlation between each operating parameter and the fault label.

[0033] In this embodiment, the mutual information formula is: (1) In this embodiment, the obtained M-dimensional mutual information vector is then processed by a dimensionality reduction algorithm such as principal component analysis (PCA) or t-SNE to obtain the point set in a two-dimensional plane. The visual distribution map is composed of The spatial position directly reflects the parameter's effect on the fault type. The correlation strength and its joint feature aggregation with other parameters.

[0034] In this embodiment, the nonlinear statistical dependence is quantitatively characterized, and a clear geometric basis is provided for subsequent reference point selection and candidate circle screening through dimensionality reduction mapping.

[0035] In this embodiment, a joint probability model and the maximum correlation entropy criterion are used to precisely quantify the correlation between operating parameters and target fault labels. A dimensionality reduction algorithm is then used to map the correlation results onto a two-dimensional plane, forming a visualized set of points. The joint probability model simultaneously characterizes the statistical dependencies between multiple parameters, while the maximum correlation entropy criterion ensures that the selected parameters and fault labels have the greatest information gain. Furthermore, the dimensionality reduction mapping visually demonstrates the strength of the correlation between each parameter and the fault in two-dimensional space, providing a more reliable mathematical foundation and visualization support for subsequent key factor screening, significantly improving the accuracy and interpretability of initial feature extraction.

[0036] Step 103: determining the feature vector with the largest correlation as a reference point, calculating the distribution characteristics of the point set based on the reference point, and determining a screening threshold based on the distribution characteristics; In this embodiment, determining the feature vector with the greatest correlation as a reference point, calculating the distribution characteristics of the point set based on the reference point, and determining the screening threshold based on the distribution characteristics include: Determine the eigenvector with the largest correlation as a reference point, and calculate the Euclidean distance distribution of the point set based on the reference point; The standard deviation of the point set is calculated based on the Euclidean distance distribution, and the product of a preset scale factor and the standard deviation is used as a screening threshold.

[0037] In this embodiment, after mapping a two-dimensional point set, the point with the highest mutual information value or other correlation metric is first identified and used as the reference point. The Euclidean distances from all other points in the set to the circle center are then calculated, with the reference point as the center. The distribution characteristics of the resulting distances are statistically analyzed to determine the standard deviation σ of the Euclidean distances. Finally, a scaling factor k is pre-set based on experience or engineering requirements, and the product of the scaling factor and the standard deviation, r = k·σ, is used as the screening threshold.

[0038] In this embodiment, the threshold can be adaptively adjusted according to the overall "convergence and divergence" of the point set, and can ensure that the candidate circle can not only cover parameters that are highly relevant to the target fault mechanism, but also effectively exclude discrete noise points, providing a reliable geometric and statistical basis for the subsequent screening of key influencing factors based on the candidate circle.

[0039] In this embodiment, the most correlated eigenvector is determined as a reference point. Based on this reference point, the Euclidean distance distribution of the point set is statistically analyzed, and a screening threshold is adaptively generated by multiplying the standard deviation by a preset scaling factor. This method dynamically adjusts the screening radius based on the actual "convergence" of the data, balancing noise reduction and coverage. It also avoids the issues of excessively loose or excessively tight fixed thresholds, achieving precise capture of highly correlated parameters and natural exclusion of noisy parameters, significantly improving the robustness and versatility of key factor screening.

[0040] Step 104: generating a candidate circle with the reference point as the center and the screening threshold as the radius, and screening all feature vectors within the candidate circle as a candidate key factor set; In this embodiment, after the two-dimensional plane point set mapping and threshold calculation are completed, a corresponding candidate circle is constructed on the two-dimensional plane, with the reference point corresponding to the maximum mutual information as the circle center and the threshold r = k⋅σ as the circle radius. Specifically, the coordinates of each eigenvector in the mapped point set are traversed, and the Euclidean distance between it and the reference point is calculated. When this distance is less than or equal to the preset radius r, the corresponding operating parameter is considered to have sufficient statistical correlation and physical representation significance with the target fault label, and the eigenvector corresponding to this parameter is then included in the candidate key factor set.

[0041] In this embodiment, the candidate circle screening can effectively cover a set of operating parameters that are highly relevant to the core fault mechanism, while automatically excluding discrete points that are weakly associated with the fault or are greatly affected by noise, providing a refined and information-rich set of candidate factors for subsequent sliding window secondary verification and sub-model construction.

[0042] Step 105: performing a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and using the operating parameters that pass the check as the key influencing factors of the fault type; In this embodiment, the secondary verification of each operating parameter in the candidate key factor set based on a preset sliding window and taking the operating parameters that pass the verification as the key influencing factors of the fault type include: Screening out candidate parameter sequences of respective candidate key factors from the historical operation data based on the candidate key factor set; Generate overlapping sliding windows on each candidate parameter sequence based on a preset window length and a preset step size; For any candidate parameter sequence, calculate the local correlation value between the candidate key factor and the target fault label in the sliding window; count the mean and standard deviation of the local correlation values ​​of all sliding windows on the candidate parameter sequence; when the mean is greater than or equal to the preset minimum correlation threshold and the standard deviation is less than or equal to the preset maximum fluctuation threshold, determine that the candidate key factor has passed the verification and use it as the key influencing factor of the fault type.

[0043] In this embodiment, after completing the initial screening of candidate key factors, the continuous time series value sequence for each candidate key factor is first extracted from the historical operating data, recorded as a candidate parameter sequence. Subsequently, based on a preset sliding window length L (e.g., 100 sampling points) and step size Δ (e.g., 20 sampling points), multiple sets of overlapping sliding windows are sequentially generated on each candidate parameter sequence, with each window corresponding to a contiguous subsequence of operating data.

[0044] In this embodiment, for each sliding window, the local correlation value between the candidate key factor and the target fault label within that window is calculated using the same correlation algorithm as the initial screening. The local correlation value sequence for all windows is then summarized. The mean μ and standard deviation σ of this sequence are then calculated. If and only if μ ≥ τ_min (τ_min is a preset minimum correlation threshold, such as 0.6) and σ ≤ τ_max (τ_max is a preset maximum fluctuation threshold, such as 0.1), the candidate factor is considered to have a strong and stable temporal correlation with the fault label. This verification is passed, and the candidate is ultimately confirmed as a key influencing factor for the fault type.

[0045] In this embodiment, the sliding window secondary verification process can eliminate occasional noise interference or short-term abnormal correlation, ensuring that the selected key factors have continuous and reliable fault indication capabilities in different operating stages.

[0046] In this embodiment, a sliding window is introduced to perform secondary verification on the candidate key factors after screening. By calculating the local correlation in multiple overlapping windows and counting their mean and standard deviation, the minimum correlation threshold and the maximum fluctuation threshold are set for strict judgment. This can eliminate short-term occasional or highly volatile pseudo-correlation parameters and retain only influencing factors with continuous and stable correlation, thereby significantly reducing the risk of false alarms and missed alarms and enhancing the temporal stability and diagnostic credibility of the fault influencing factors.

[0047] Step 106: Obtain key influencing factors corresponding to each fault type, and construct fault monitoring sub-models based on the key influencing factors; In this embodiment, the acquisition of key influencing factors corresponding to each fault type and the construction of fault monitoring sub-models based on the key influencing factors include: Obtain the key influencing factors corresponding to each fault type, extract the time series characteristics of the key influencing factors from the corresponding fault operation data, and construct a multidimensional feature vector; Matching a target prediction algorithm based on the multidimensional feature vector, and constructing an initial fault monitoring sub-model based on the target prediction algorithm; Training data is generated based on the multidimensional feature vector, and the initial fault monitoring sub-model is trained based on the training data to obtain a fault monitoring sub-model.

[0048] In this embodiment, taking the "gearbox gear wear" fault type as an example, first, based on the list of key influencing factors obtained by the aforementioned sliding window verification, such as the 0.5 times meshing frequency sideband amplitude, the lubricating oil metal particle concentration, and the gearbox oil temperature M' parameter in the main shaft vibration envelope spectrum, the continuous time series of each parameter is extracted from the corresponding fault operation data at a uniform sampling frequency; then, the time series of each key influencing factor is truncated into sliding window segments of the same length, and the statistical features (such as mean, variance, maximum value, minimum value) and frequency domain features (such as power spectrum density peak, harmonic content) of each segment are calculated, and the obtained features are spliced ​​in a predetermined order to construct a multidimensional feature vector representing the fault condition. . According to the dimension of the feature vector and the nature of the fault mode, the system automatically matches the applicable target prediction algorithm, such as the support vector machine (SVM) suitable for small sample high-dimensional features, or the long short-term memory network (LSTM) suitable for features with strong time series dependence, and constructs the initial fault monitoring sub-model accordingly. Then, the historical multi-dimensional feature vector sequence and its corresponding fault label are packaged in the same sliding window manner to form a training sample set, and divided into a training set and a validation set according to a certain ratio. The selected prediction algorithm is used to perform iterative training on the training set, and finally the fault monitoring sub-model corresponding to each fault type is obtained for real-time fault identification and early warning.

[0049] In this embodiment, based on the multi-dimensional feature vectors of each fault type, a modeling algorithm suitable for the fault mechanism is selected or customized, including but not limited to: a classification algorithm based on a support vector machine (SVM) for fault identification with small sample sizes and high-dimensional features; a time series prediction algorithm based on a long short-term memory network (LSTM) for capturing fault development trends; and a Gaussian mixture model (GMM) or isolation forest based anomaly detection algorithm for anomaly alarm in an unsupervised environment.

[0050] In this embodiment, the time series characteristics of the key influencing factors that have passed the secondary verification are extracted for each type of fault, a multi-dimensional feature vector is constructed, and an initial monitoring sub-model is constructed based on the target prediction algorithm that matches the fault mechanism. Each sub-model uses the most representative features as input, which not only ensures the specificity and diagnostic accuracy of the model, but also lays a highly targeted foundation for subsequent model training and optimization, and improves the coverage capability and recognition performance of multiple fault types.

[0051] In this embodiment, generating training data based on the multidimensional feature vector, and training the initial fault monitoring sub-model based on the training data to obtain the fault monitoring sub-model includes: Dividing the multidimensional feature vector into fixed-length sequence segments in a sliding window manner according to time sequence, and labeling each sequence segment with a corresponding fault type or normal operating condition label to generate training data; The training data is divided into a training subset and a validation subset according to a preset ratio, and the initial fault monitoring sub-model is trained based on the training subset and the validation subset to obtain a fault monitoring sub-model.

[0052] In this embodiment, for the extracted multi-dimensional feature vector sequence First, the multidimensional feature vector sequence is segmented into sliding windows on the time axis with a fixed length L and step size to generate several overlapping sequence segments. , each fragment sequence fragment A multidimensional feature vector containing L adjacent moments. Subsequently, the fault type or "normal" label is labeled for each sequence segment based on the corresponding fault occurrence time or normal operating time interval, thereby constructing an input-output paired training data set. Next, the training data is randomly divided into a training subset and a validation subset according to a preset ratio (for example, 70% training, 30% validation), and iterative learning (such as gradient descent, Adam optimization) is performed on the training subset with the initial fault monitoring sub-model as the skeleton. At the same time, the model performance indicators (such as accuracy, recall rate, loss function value) are periodically evaluated on the validation subset. By adjusting the learning rate or early stopping strategy until the model performance on the validation set converges stably, a fault monitoring sub-model with high recognition accuracy and generalization ability is finally obtained.

[0053] In this embodiment, a sliding window approach is used to segment multidimensional feature vectors into fixed-length sequence segments and label them. This is then used to divide the model into training and validation sets for training. This time-series slicing enhances sample diversity, effectively alleviating the issues of scarcity and class imbalance in fault samples. Furthermore, the separation of training and validation sets and the cross-validation mechanism ensure the stability of the resulting fault monitoring sub-model.

[0054] Step 107: monitor each fault condition separately based on the fault monitoring sub-model to complete the fault monitoring of the wind power equipment.

[0055] In this embodiment, the fault monitoring sub-model is used to monitor each fault condition to complete the fault monitoring of the wind power equipment, including: Collecting the operating data of key influencing factors corresponding to each fault type in real time, and performing denoising, interpolation and normalization preprocessing on the operating data to generate input vectors for each fault detection sub-model; Inputting the input vector into the corresponding fault monitoring sub-model to obtain a prediction score of the fault type; An alarm is issued for the fault type based on the prediction score and a preset alarm threshold, thereby completing fault monitoring of the wind power equipment.

[0056] In this embodiment, during the online operation phase of a wind turbine, the system collects key influencing factors corresponding to each fault type in real time via a high-speed data bus, such as time series signals of main shaft vibration amplitude, gearbox oil temperature, and oil metal particle concentration. Subsequently, the collected raw data is firstly subjected to wavelet denoising or low-pass filtering algorithms to remove high-frequency noise. Missing data points due to sensor packet loss or network jitter are then repaired using linear interpolation or spline interpolation. The key factor data are then normalized based on the mean and standard deviation from the training phase to generate a dimensionless input vector with a consistent numerical distribution. This input vector is fed into the corresponding fault type monitoring sub-model in real time, which outputs a predicted score reflecting the fault severity. When this score is greater than or equal to a pre-calibrated alarm threshold, an alarm signal for the corresponding fault type is triggered and pushed to the monitoring center via the operation and maintenance platform, enabling online identification and early warning of wind turbine equipment faults.

[0057] In this embodiment, in a scenario where multiple faults occur simultaneously, the alarm signals of each sub-model are subjected to logical OR or weighted fusion processing to comprehensively generate the overall fault alarm status of the wind power equipment.

[0058] In this embodiment, based on subsequent maintenance results and newly added fault samples, the alarm-triggered sub-model is incrementally trained and parameter fine-tuned regularly or online to maintain monitoring accuracy and adaptability; In this embodiment, key influencing factor data are collected and preprocessed in real time, and input into respective fault monitoring sub-models for online prediction. Alarms are triggered based on the prediction scores and preset thresholds. At the same time, logical fusion and dynamic incremental updates of multi-model outputs are supported, thereby building a full-process closed loop from data collection, preprocessing, model reasoning to alarm output. This not only enables real-time and accurate monitoring of various faults, but also ensures that the system continuously adapts to environmental changes and fault evolution during operation through dynamic adjustment and incremental training, comprehensively improving the timeliness of early warning and operation and maintenance efficiency of wind power equipment.

[0059] In this embodiment, historical operating data containing fault types is first acquired. A preset correlation algorithm is then used to quantify the correlation between each operating parameter and the fault. This correlation is then mapped onto a two-dimensional plane to construct a structured point set, enhancing the visibility and processability of the multidimensional data. During the key influencing factor screening process, the operating parameter with the highest correlation is selected as the benchmark. This has clear physical fault characterization significance, allowing focus on core parameters highly correlated with the target fault mechanism, thereby improving the accuracy of the candidate factor set from the source. Parameters with low fault correlation and susceptible to noise interference are also excluded, enhancing the robustness of fault feature extraction. The candidate circle construction utilizes a dynamic radius adjustment mechanism, adaptively adjusting the radius based on data distribution characteristics. This achieves a balance between noise immunity and generalization capabilities, avoiding the risk of misjudgment or missed detection associated with traditional fixed thresholds. Furthermore, driven by correlation, the operating parameters exhibit clustering characteristics in two-dimensional space. The candidate circle naturally captures local parameter clustering regions, thereby enhancing screening efficiency and accuracy in this embodiment. For multiple fault concurrent scenarios, various fault-related factors are spatially separated in the two-dimensional plane. By generating candidate circles separately, the fault decoupling identification of key parameters can be achieved, and the response capability to fault linkage under complex working conditions can be improved. Finally, compared with the traditional global clustering algorithm, the candidate circle strategy only needs to calculate the distance between the point and the circle center, and the time complexity is reduced by This reduces the time required to O(n), significantly improving online processing efficiency and better adapting to the needs of high-frequency monitoring and real-time early warning for wind turbines. By extracting the most representative key influencing factors for different fault types and constructing dedicated fault monitoring sub-models based on them, the model inputs are highly targeted, significantly improving the accuracy of fault identification and classification, and enhancing the accuracy of wind turbine fault detection.

[0060] Please refer to Figure 2 , Figure 2 A schematic structural diagram of a fault monitoring system for wind power equipment based on operating data is provided, comprising: a data acquisition module 201, a correlation calculation module 202, a distribution determination module 203, a candidate factor screening module 204, a key factor screening module 205, a model construction module 206, and a fault detection module 207; The data acquisition module 201 is used to obtain historical operation data of the wind power equipment, wherein the historical operation data includes a fault type and corresponding fault operation data under the fault type; The correlation calculation module 202 is used to calculate the correlation between each operating parameter in the fault operation data and the target fault label based on a preset correlation algorithm for any fault type, and map the correlation of each operating parameter to a two-dimensional plane to generate a point set of the two-dimensional plane; The distribution determination module 203 is configured to determine the feature vector with the greatest correlation as a reference point, calculate the distribution characteristics of the point set based on the reference point, and determine a screening threshold based on the distribution characteristics; The candidate factor screening module 204 is configured to generate a candidate circle with the reference point as the center and the screening threshold as the radius, and screen all feature vectors within the candidate circle as a candidate key factor set; The key factor screening module 205 is configured to perform a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and use the operating parameters that pass the check as the key influencing factors of the fault type; The model building module 206 is used to obtain key influencing factors corresponding to each fault type and build fault monitoring sub-models based on the key influencing factors; The fault detection module 207 is configured to monitor each fault condition based on the fault monitoring sub-model to complete fault monitoring of the wind power equipment.

[0061] In this embodiment, the correlation calculation module is used to calculate the correlation between each operating parameter in the fault operation data and the target fault label based on a preset correlation algorithm for any fault type, and map the correlation of each operating parameter to a two-dimensional plane to generate a point set on the two-dimensional plane, including: constructing a joint probability model based on the various operating parameters in the fault operation data; Evaluate the correlation between each operating parameter and the target fault label based on the maximum correlation entropy criterion and the joint probability model; Based on the preset dimensionality reduction algorithm, the correlation between each operating parameter and the target fault label is mapped to a two-dimensional plane to generate a point set representing the correlation between each operating parameter and the target fault label.

[0062] In this embodiment, the distribution determination module is configured to determine the feature vector with the greatest correlation as a reference point, calculate the distribution characteristics of the point set based on the reference point, and determine the screening threshold based on the distribution characteristics, including: Determine the eigenvector with the largest correlation as a reference point, and calculate the Euclidean distance distribution of the point set based on the reference point; The standard deviation of the point set is calculated based on the Euclidean distance distribution, and the product of a preset scale factor and the standard deviation is used as a screening threshold.

[0063] In this embodiment, the key factor screening module is configured to perform a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and use the operating parameters that pass the check as the key influencing factors of the fault type, including: Screening out candidate parameter sequences of respective candidate key factors from the historical operation data based on the candidate key factor set; Generate overlapping sliding windows on each candidate parameter sequence based on a preset window length and a preset step size; For any candidate parameter sequence, calculate the local correlation value between the candidate key factor and the target fault label in the sliding window; count the mean and standard deviation of the local correlation values ​​of all sliding windows on the candidate parameter sequence; when the mean is greater than or equal to the preset minimum correlation threshold and the standard deviation is less than or equal to the preset maximum fluctuation threshold, determine that the candidate key factor has passed the verification and use it as the key influencing factor of the fault type.

[0064] In this embodiment, the model building module is used to obtain key influencing factors corresponding to each fault type and build fault monitoring sub-models based on the key influencing factors, including: Obtain the key influencing factors corresponding to each fault type, extract the time series characteristics of the key influencing factors from the corresponding fault operation data, and construct a multidimensional feature vector; Matching a target prediction algorithm based on the multidimensional feature vector, and constructing an initial fault monitoring sub-model based on the target prediction algorithm; Training data is generated based on the multidimensional feature vector, and the initial fault monitoring sub-model is trained based on the training data to obtain a fault monitoring sub-model.

[0065] In this embodiment, the model building module is used to generate training data based on the multidimensional feature vector, and train the initial fault monitoring sub-model based on the training data to obtain the fault monitoring sub-model, including: Dividing the multidimensional feature vector into fixed-length sequence segments in a sliding window manner according to time sequence, and labeling each sequence segment with a corresponding fault type or normal operating condition label to generate training data; The training data is divided into a training subset and a validation subset according to a preset ratio, and the initial fault monitoring sub-model is trained based on the training subset and the validation subset to obtain a fault monitoring sub-model.

[0066] In this embodiment, the fault detection module is used to monitor each fault condition based on the fault monitoring sub-model to complete the fault monitoring of the wind power equipment, including: Collecting the operating data of key influencing factors corresponding to each fault type in real time, and performing denoising, interpolation and normalization preprocessing on the operating data to generate input vectors for each fault detection sub-model; Inputting the input vector into the corresponding fault monitoring sub-model to obtain a prediction score of the fault type; An alarm is issued for the fault type based on the prediction score and a preset alarm threshold, thereby completing fault monitoring of the wind power equipment.

[0067] In an embodiment of the present invention, a terminal device is also provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the above-mentioned fault monitoring method for wind power equipment based on operating data is implemented.

[0068] In an embodiment of the present invention, a computer-readable storage medium is further provided, which includes a stored computer program. When the computer program runs, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned wind power equipment fault monitoring method based on operating data.

[0069] For example, a computer program may be divided into one or more modules, one or more of which are stored in a memory and executed by a processor to implement the present invention. One or more modules may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in a terminal device.

[0070] The terminal device may be a computing device such as a desktop computer, laptop, PDA, or cloud server. The terminal device may include, but is not limited to, a processor, memory, and display. Those skilled in the art will appreciate that the aforementioned components are merely examples of terminal devices and do not constitute a limitation of the terminal device. The terminal device may include more or fewer components, or a combination of certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, and the like.

[0071] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0072] The memory can be used to store computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store an operating system and at least one application required for a function (such as a sound playback function, a text conversion function, etc.); the data storage area can store data generated based on the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory can include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0073] If the module for fault monitoring of wind turbines based on operational data is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. Computer-readable media can include any entity or device capable of carrying computer program code, recording media, USB flash drives, mobile hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Persons of ordinary skill in the art can understand and implement the present invention without inventive effort.

[0074] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A fault monitoring method for wind power equipment based on operating data, characterized in that: include: Acquire historical operation data of the wind power equipment, wherein the historical operation data includes a fault type and fault operation data corresponding to the fault type; For any fault type, the correlation between each operating parameter in the fault operation data and the target fault label is calculated based on a preset correlation algorithm, and the correlation between each operating parameter is mapped to a two-dimensional plane to generate a point set of the two-dimensional plane; Determining a feature vector with the greatest correlation as a reference point, calculating a distribution feature of the point set based on the reference point, and determining a screening threshold based on the distribution feature; Generate a candidate circle with the reference point as the center and the screening threshold as the radius, and screen out all feature vectors within the candidate circle as a candidate key factor set; Performing a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and using the operating parameters that pass the check as the key influencing factors of the fault type; Obtaining key influencing factors corresponding to each fault type, and constructing fault monitoring sub-models based on the key influencing factors; Each fault condition is monitored separately based on the fault monitoring sub-model to complete the fault monitoring of the wind power equipment.

2. A fault monitoring method for wind power equipment based on operating data according to claim 1, characterized in that: For any fault type, the correlation between each operating parameter in the fault operation data and the target fault label is calculated based on a preset correlation algorithm, and the correlation of each operating parameter is mapped to a two-dimensional plane to generate a point set of the two-dimensional plane, including: constructing a joint probability model based on the various operating parameters in the fault operation data; Evaluate the correlation between each operating parameter and the target fault label based on the maximum correlation entropy criterion and the joint probability model; Based on the preset dimensionality reduction algorithm, the correlation between each operating parameter and the target fault label is mapped to a two-dimensional plane to generate a point set representing the correlation between each operating parameter and the target fault label.

3. A fault monitoring method for wind power equipment based on operation data according to claim 2, characterized in that: The step of determining the feature vector with the greatest correlation as a reference point, calculating the distribution characteristics of the point set based on the reference point, and determining a screening threshold based on the distribution characteristics includes: Determine the eigenvector with the largest correlation as a reference point, and calculate the Euclidean distance distribution of the point set based on the reference point; The standard deviation of the point set is calculated based on the Euclidean distance distribution, and the product of a preset scale factor and the standard deviation is used as a screening threshold.

4. A fault monitoring method for wind power equipment based on operation data according to claim 3, characterized in that: The performing secondary verification on each operating parameter in the candidate key factor set based on a preset sliding window, and using the operating parameters that pass the verification as the key influencing factors of the fault type, includes: Screening out candidate parameter sequences of respective candidate key factors from the historical operation data based on the candidate key factor set; Generate overlapping sliding windows on each candidate parameter sequence based on a preset window length and a preset step size; For any candidate parameter sequence, calculate the local correlation value between the candidate key factor and the target fault label in the sliding window; count the mean and standard deviation of the local correlation values ​​of all sliding windows on the candidate parameter sequence; when the mean is greater than or equal to the preset minimum correlation threshold and the standard deviation is less than or equal to the preset maximum fluctuation threshold, determine that the candidate key factor has passed the verification and use it as the key influencing factor of the fault type.

5. The method for monitoring faults of wind power equipment based on operating data according to claim 4, characterized in that: The obtaining of key influencing factors corresponding to each fault type and constructing fault monitoring sub-models based on the key influencing factors include: Obtain the key influencing factors corresponding to each fault type, extract the time series characteristics of the key influencing factors from the corresponding fault operation data, and construct a multidimensional feature vector; Matching a target prediction algorithm based on the multidimensional feature vector, and constructing an initial fault monitoring sub-model based on the target prediction algorithm; Training data is generated based on the multidimensional feature vector, and the initial fault monitoring sub-model is trained based on the training data to obtain a fault monitoring sub-model.

6. A fault monitoring method for wind power equipment based on operation data according to claim 5, characterized in that: Generating training data based on the multidimensional feature vector, and training the initial fault monitoring sub-model based on the training data to obtain the fault monitoring sub-model, includes: The multidimensional feature vector is divided into sequence segments of fixed length in a sliding window manner according to time sequence, and each sequence segment is labeled with a corresponding fault type or normal operating condition label to generate training data; The training data is divided into a training subset and a validation subset according to a preset ratio, and the initial fault monitoring sub-model is trained based on the training subset and the validation subset to obtain a fault monitoring sub-model.

7. A fault monitoring method for wind power equipment based on operation data according to claim 6, characterized in that: The method of monitoring each fault condition based on the fault monitoring sub-model to complete the fault monitoring of the wind power equipment includes: Collecting the operating data of key influencing factors corresponding to each fault type in real time, and performing denoising, interpolation and normalization preprocessing on the operating data to generate input vectors for each fault detection sub-model; Inputting the input vector into the corresponding fault monitoring sub-model to obtain a prediction score of the fault type; An alarm is issued for the fault type based on the prediction score and a preset alarm threshold, thereby completing fault monitoring of the wind power equipment.

8. A fault monitoring system for wind power equipment based on operating data, characterized in that: include: Data acquisition module, correlation calculation module, distribution determination module, candidate factor screening module, key factor screening module, model building module and fault detection module; The data acquisition module is used to obtain historical operation data of the wind power equipment, wherein the historical operation data includes the fault type and the fault operation data corresponding to the fault type; The correlation calculation module is used to calculate the correlation between each operating parameter in the fault operation data and the target fault label based on a preset correlation algorithm for any fault type, and map the correlation of each operating parameter to a two-dimensional plane to generate a point set of the two-dimensional plane; The distribution determination module is configured to determine the feature vector with the greatest correlation as a reference point, calculate the distribution characteristics of the point set based on the reference point, and determine a screening threshold based on the distribution characteristics; The candidate factor screening module is configured to generate a candidate circle with the reference point as the center and the screening threshold as the radius, and screen all feature vectors within the candidate circle as a candidate key factor set; The key factor screening module is used to perform a secondary check on each operating parameter in the candidate key factor set based on a preset sliding window, and use the operating parameters that pass the check as the key influencing factors of the fault type; The model building module is used to obtain key influencing factors corresponding to each fault type and build fault monitoring sub-models based on the key influencing factors; The fault detection module is used to monitor each fault condition based on the fault monitoring sub-model to complete the fault monitoring of the wind power equipment.

9. A fault monitoring system for wind power equipment based on operating data according to claim 8, characterized in that: The distribution determination module is configured to determine the feature vector with the greatest correlation as a reference point, calculate the distribution characteristics of the point set based on the reference point, and determine a screening threshold based on the distribution characteristics, including: Determine the eigenvector with the largest correlation as a reference point, and calculate the Euclidean distance distribution of the point set based on the reference point; The standard deviation of the point set is calculated based on the Euclidean distance distribution, and the product of a preset scale factor and the standard deviation is used as a screening threshold.

10. A fault monitoring system for wind power equipment based on operation data according to claim 9, characterized in that: The candidate factor screening module is configured to perform secondary verification on each operating parameter in the candidate key factor set based on a preset sliding window, and use the operating parameters that pass the verification as the key influencing factors of the fault type, including: Screening out candidate parameter sequences of respective candidate key factors from the historical operation data based on the candidate key factor set; Generate overlapping sliding windows on each candidate parameter sequence based on a preset window length and a preset step size; For any candidate parameter sequence, calculate the local correlation value between the candidate key factor and the target fault label in the sliding window; count the mean and standard deviation of the local correlation values ​​of all sliding windows on the candidate parameter sequence; when the mean is greater than or equal to the preset minimum correlation threshold and the standard deviation is less than or equal to the preset maximum fluctuation threshold, determine that the candidate key factor has passed the verification and use it as the key influencing factor of the fault type.

Images