A method and system for monitoring wind turbine faults

Abstract

The present invention discloses a monitoring method and system for wind turbine faults, the method comprising: obtaining mechanical component sensor data, environmental data and operation data; performing data preprocessing operations according to the mechanical component sensor data, the environmental data and the operation data to obtain preprocessed multi-dimensional unit data; performing hypergraph construction operations according to the multi-dimensional unit data to obtain hypergraph data; inputting the hypergraph data into a multi-layer hypergraph convolution model, performing multi-layer hypergraph convolution processing operations, and obtaining hypergraph convolution feature data; inputting the hypergraph convolution feature data into an anomaly score calculation model, performing anomaly score calculation, and obtaining a wind turbine fault anomaly score; when it is determined that the fault anomaly score is greater than a preset anomaly threshold, generating an alarm signal, and sending the alarm signal to a user end. The method can realize real-time and accurate detection of wind turbine faults.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbine generator sets, and in particular to a method and system for monitoring faults of wind turbine generator sets. Background Art

[0002] With the rapid development of the wind power industry, wind turbines, as important energy equipment, have a direct impact on power generation efficiency and maintenance costs due to their operational stability and failure rate. Therefore, wind turbine fault monitoring and diagnosis have become key issues. In order to ensure the long-term stable operation of wind turbines, real-time monitoring of the operating status of wind turbines and accurate identification of abnormalities before or in the early stages of a fault are necessary means to reduce maintenance costs, extend equipment life, and improve power generation efficiency.

[0003] In the existing technology, the fault detection technology of wind turbines mainly relies on sensor-based data collection and traditional statistical analysis methods, such as analysis technology based on vibration signals, temperature and speed monitoring technology, etc. Some fault detection methods based on physical models are also being used, which predict equipment faults by building mathematical models of each subsystem of the wind turbine and combining system parameters and operating data.

[0004] In summary, the existing technology cannot effectively process the high-dimensional data and nonlinear characteristics involved in wind turbines under multi-source data fusion and complex scenarios, and it is difficult to mine the relationship between different data sources, resulting in insufficient monitoring accuracy and low real-time performance. Summary of the invention

[0005] The present invention provides a method and system for monitoring wind turbine faults, so as to realize real-time and accurate detection of wind turbine faults.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a method for monitoring a fault of a wind turbine generator set, comprising:

[0007] Obtain sensor data, environmental data and operating data of mechanical components;

[0008] Performing data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data;

[0009] Performing a hypergraph construction operation according to the multi-dimensional unit data to obtain hypergraph data;

[0010] Inputting the hypergraph data into a multi-layer hypergraph convolution model, performing a multi-layer hypergraph convolution processing operation, and obtaining hypergraph convolution feature data;

[0011] Inputting the hypergraph convolution feature data into an anomaly score calculation model to perform an anomaly score calculation to obtain a fault anomaly score of the wind turbine generator set;

[0012] When it is determined that the fault anomaly score is greater than a preset anomaly threshold, an alarm signal is generated and sent to a user terminal.

[0013] As an optional implementation manner, performing data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data includes:

[0014] Performing data cleaning processing according to the mechanical component sensor data, the environmental data and the operating data to obtain clean unit data;

[0015] Performing data standardization processing on the clean unit data to obtain standard unit data;

[0016] According to the standard unit data, multi-dimensional key features are extracted to obtain multi-dimensional unit data.

[0017] As an optional implementation manner, performing a hypergraph construction operation according to the multi-dimensional unit data to obtain the hypergraph data includes:

[0018] According to the multi-dimensional unit data, the data features are used as hypergraph nodes to construct nodes, thereby obtaining hypergraph node data;

[0019] According to the multi-dimensional unit data and the hypergraph node data, a hyperedge is constructed to obtain hyperedge data;

[0020] Generating a weight matrix according to the multi-dimensional unit data and the hyperedge data to obtain hypergraph weight matrix data;

[0021] A hypergraph construction operation is performed according to the hypergraph node data, the hyperedge data and the hypergraph weight matrix data to obtain hypergraph data.

[0022] As an optional implementation, the hypergraph data is input into a multi-layer hypergraph convolution model, and a multi-layer hypergraph convolution processing operation is performed to obtain hypergraph convolution feature data, including:

[0023] Based on the hypergraph data, performing a hypergraph convolution operation on nodes in the hypergraph data to obtain node feature data;

[0024] According to the node feature data, the node features are aggregated layer by layer through a multi-layer hypergraph convolutional network to obtain node high-order feature data;

[0025] Node embedding is performed based on the node high-order feature data to obtain hypergraph convolution feature data.

[0026] As an optional implementation, the hypergraph convolution operation is performed on the nodes in the hypergraph data, and the hypergraph convolution calculation formula is as follows:

[0027] ;

[0028] in, Indicates The node feature matrix of the layer, Indicates The node feature matrix of the layer; represents the activation function; represents the node degree matrix; represents the degree matrix of the hyperedge; represents the connection matrix of the hyperedge, represents the transposed form of E; Indicates The network parameters that the layer needs to learn; represents the association matrix between nodes and hyperedges, express The transposed form of .

[0029] As an optional implementation manner, the inputting the hypergraph convolution feature data into an anomaly score calculation model to perform an anomaly score calculation to obtain a fault anomaly score of the wind turbine generator set includes:

[0030] The hypergraph convolution feature data is used as input data and input into the graph variational autoencoder model for feature reconstruction and calculation to obtain a reconstruction error;

[0031] Calculating a node anomaly score according to the reconstruction error to obtain a node anomaly score;

[0032] According to the node anomaly scores, score aggregation is performed to obtain the fault anomaly score of the wind turbine generator set.

[0033] As an optional implementation, the hypergraph convolution feature data is used as input data and input into a graph variational autoencoder model for feature reconstruction and calculation to obtain a reconstruction error, including:

[0034] Performing data encoding processing on the hypergraph convolution feature data to obtain potential latent variable representation;

[0035] According to the latent variable representation, decoding mapping is performed to reconstruct the original high-order feature data to obtain reconstructed feature data;

[0036] A reconstruction error is calculated based on the hypergraph convolution feature data and the reconstruction feature data to obtain a reconstruction error.

[0037] As an optional implementation, the node anomaly score is calculated according to the reconstruction error, and the calculation formula of the anomaly score is as follows:

[0038] ;

[0039] in, Represents the first The abnormal score of a node, that is, the score of the data source at a certain position in the original data, such as the score of the fan blade vibration data; Representation Node The set of neighbor nodes of node The node collection related to the data source represented; Representation Node The number of neighbor nodes; Representation Node The reconstruction error.

[0040] In a second aspect, the present invention provides a wind turbine fault monitoring system, comprising:

[0041] A data acquisition module, used to acquire sensor data, environmental data and operation data of mechanical components;

[0042] A data preprocessing module, used to perform data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data;

[0043] A hypergraph construction module, used to perform a hypergraph construction operation according to the multi-dimensional unit data to obtain hypergraph data;

[0044] A hypergraph convolution module, used to input the hypergraph data into a multi-layer hypergraph convolution model, perform multi-layer hypergraph convolution processing operations, and obtain hypergraph convolution feature data;

[0045] A score calculation module, used for inputting the hypergraph convolution feature data into an anomaly score calculation model, performing anomaly score calculation, and obtaining a fault anomaly score of the wind turbine generator set;

[0046] The abnormality determination and alarm module is used to generate an alarm signal when it is determined that the fault abnormality score is greater than a preset abnormality threshold, and send the alarm signal to the user end.

[0047] In a third aspect, the present invention further provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute any one of the wind turbine fault monitoring methods described above.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention provides a monitoring method for wind turbine faults, comprising: obtaining mechanical component sensor data, environmental data and operating data; performing data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multidimensional unit data; performing a hypergraph construction operation according to the multidimensional unit data to obtain hypergraph data; inputting the hypergraph data into a multi-layer hypergraph convolution model, performing multi-layer hypergraph convolution processing operations, and obtaining hypergraph convolution feature data; inputting the hypergraph convolution feature data into an anomaly score calculation model, performing anomaly score calculation, and obtaining a fault anomaly score of the wind turbine; when it is determined that the fault anomaly score is greater than a preset anomaly threshold, generating an alarm signal, and sending the alarm signal to a user end. The present invention models the multidimensional information in the mechanical component sensor data, environmental data and operating data as hypergraph nodes and hyperedges by constructing a heterogeneous hypergraph. By adopting the anomaly score calculation model and calculating the anomaly score of the hypergraph convolution feature data, the operating status of each node of the wind turbine can be accurately evaluated, the abnormal nodes can be determined, and the fault anomaly score of the wind turbine can be finally generated. The detection standard can be adaptively adjusted, and the sensitivity to wind turbine faults is improved, thereby achieving more accurate fault detection and real-time alarm. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic flow chart of a method for monitoring a wind turbine fault provided by an embodiment of the present invention;

[0051] Figure 2 The present invention is a flowchart of a wind turbine fault monitoring system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Reference Figure 1 The first embodiment of the present invention provides a method for monitoring a fault of a wind turbine generator system, comprising the following steps:

[0054] S11, obtaining mechanical component sensor data, environmental data and operation data;

[0055] S12, performing data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data;

[0056] S13, performing a hypergraph construction operation according to the multi-dimensional unit data to obtain hypergraph data;

[0057] S14, inputting the hypergraph data into a multi-layer hypergraph convolution model, performing a multi-layer hypergraph convolution processing operation, and obtaining hypergraph convolution feature data;

[0058] S15, inputting the hypergraph convolution feature data into an anomaly score calculation model, performing an anomaly score calculation, and obtaining a fault anomaly score of the wind turbine generator set;

[0059] S16: When it is determined that the fault abnormality score is greater than a preset abnormality threshold, an alarm signal is generated and sent to the user end.

[0060] It should be noted that in the existing technology, the fault detection technology of wind turbines mainly relies on sensor-based data collection and traditional statistical analysis methods, such as analysis technology based on vibration signals, temperature and speed monitoring technology, etc. The existing technology cannot effectively process the high-dimensional data and nonlinear characteristics involved in wind turbines under multi-source data fusion and complex scenarios, and it is difficult to mine the relationship between different data sources, resulting in insufficient monitoring accuracy and low real-time performance.

[0061] In the present invention, a data representation method based on a hypergraph is established by multi-dimensional data fusion of mechanical component sensor data, environmental data and operation data. Through a multi-layer hypergraph convolutional network, node features of different dimensions are aggregated layer by layer to generate high-order feature data, and combined with an anomaly score calculation model, the fault anomaly score of the wind turbine can be accurately calculated. Different from the existing monitoring technology based on a single signal, the present invention can be more refined and dynamic, and can identify potential faults in advance and issue early warning signals in time, thereby significantly improving the safety and stability of wind turbine operation.

[0062] In step S11, obtain mechanical component sensor data, environmental data and operating data

[0063] Among them, the mechanical component sensor data includes: vibration data, which is used to monitor the operating status of mechanical components, such as the vibration characteristics of blades, gears, and bearings, to help identify faults caused by friction, wear or looseness; speed data, which is used to monitor the rotation speed of wind turbines, to help analyze speed fluctuations under different loads and wind speeds, and to determine whether there are operating abnormalities; torque data, which is used to monitor torque changes in the transmission system, to help identify torque anomalies caused by wind speed fluctuations or mechanical fatigue; gearbox oil data, which is used to monitor the temperature and pressure of the gearbox oil, to determine whether the lubrication system is normal, and whether there is oil leakage or oil deterioration; temperature data, which is used to monitor temperature changes of key components, such as the temperature of bearings, gearboxes and generators, to detect overheating or cooling system abnormalities. Environmental data include: wind speed data, which is used to monitor changes in wind speed around wind turbines, helping to evaluate the operating status of wind turbines under different wind speed conditions, and the impact of load changes caused by wind speed fluctuations on the equipment; wind direction data, which is used to monitor changes in wind direction to ensure that the wind turbines are always facing the wind source. Wind direction deviations may lead to reduced power generation efficiency or additional stress on mechanical components; air temperature data, the ambient temperature has a great impact on the operation of wind turbines, especially in extreme climate conditions (such as high or low temperatures), which may cause equipment overheating, reduced cooling system efficiency or material aging; humidity data, which is used to monitor humidity changes, helps to determine whether the equipment may be affected by moisture, and has an important impact on the stability and failure risk of electrical equipment and lubrication systems; air pressure data, changes in air pressure may affect wind speed and air density, and thus affect the power generation efficiency of wind turbines. Sharp changes in air pressure may indicate extreme weather. Operational data include: power generation data, which is used to monitor the actual power generation of wind turbines in different time periods, helping to evaluate the operating efficiency of the units and whether there are abnormal power fluctuations, indicating possible failures or unstable operations; load data, including the actual load borne by mechanical components and electrical systems, especially the load changes under different wind speed and wind direction conditions, which helps to determine whether the unit is operating under overload; voltage data, which is used to monitor changes in output voltage during the power generation process of wind turbines. Abnormal voltage fluctuations may indicate potential failures in the electrical system, such as abnormal working conditions of generators, cables or transformers.

[0064] In step S12, data preprocessing is performed according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data, including:

[0065] Performing data cleaning processing according to the mechanical component sensor data, the environmental data and the operating data to obtain clean unit data;

[0066] Performing data standardization processing on the clean unit data to obtain standard unit data;

[0067] According to the standard unit data, multi-dimensional key features are extracted to obtain multi-dimensional unit data.

[0068] It should be noted that in the prior art, data preprocessing operations often ignore the comprehensive processing of data quality and multi-dimensional features, which easily leads to the mixing of noise data and affects the accuracy of fault detection. Especially in the field of wind turbine fault detection, mechanical component sensor data, environmental data and operation data usually have high noise levels and complex time correlations. The prior art cannot effectively remove these noises or perform reasonable feature extraction, resulting in poor robustness of the detection model to the data and difficulty in coping with complex working condition changes.

[0069] Among them, data cleaning processing includes three steps: anomaly detection, missing value processing, and outlier correction. The noise, missing data and outliers in the mechanical component sensor data, environmental data and operating data are processed to ensure that the clean unit data obtained accurately reflects the actual operating status of the wind turbine. Data cleaning can not only remove invalid information that may affect the model judgment, but also improve the overall consistency of the data and enhance the reliability of subsequent processing steps. Data standardization processing mainly targets different scales of various types of data (such as unit and magnitude differences such as temperature, power, and speed). Through standardization technology, the data of each dimension is mapped to the same numerical range to obtain standard unit data. This processing can eliminate the magnitude differences between different data sources, ensure that the model has consistent perception capabilities for the features of each data dimension, and is conducive to the training and optimization of subsequent algorithms. Multidimensional key feature extraction mainly combines the spatiotemporal characteristics of mechanical component sensor data, environmental data and operating data, and uses multidimensional feature extraction algorithms (such as principal component analysis, time series feature extraction, etc.) to extract multidimensional key features related to wind turbine failures from the data to form multidimensional unit data. These features include the health status of mechanical components, the changing trend of environmental conditions, load fluctuations during unit operation, etc., which can provide more discriminative input data for the fault detection model. Exemplarily, in the embodiment of the present invention, the multidimensional feature extraction algorithm used is principal component analysis. Of course, according to different actual application scenarios and user needs, the multidimensional feature extraction algorithm can also be a method such as time series feature extraction, which is not limited by the present invention.

[0070] In step S13, the hypergraph construction operation is performed according to the multi-dimensional unit data to obtain the hypergraph data, including:

[0071] According to the multi-dimensional unit data, the data features are used as hypergraph nodes to construct nodes, thereby obtaining hypergraph node data;

[0072] According to the multi-dimensional unit data and the hypergraph node data, a hyperedge is constructed to obtain hyperedge data;

[0073] Generating a weight matrix according to the multi-dimensional unit data and the hyperedge data to obtain hypergraph weight matrix data;

[0074] A hypergraph construction operation is performed according to the hypergraph node data, the hyperedge data and the hypergraph weight matrix data to obtain hypergraph data.

[0075] It should be noted that the node construction of the hypergraph model is based on the multi-dimensional unit data, taking various data features (such as sensor data of mechanical parts, temperature and humidity in environmental data, and power and speed in operating data) as nodes in the hypergraph, performing node construction operations, and obtaining hypergraph node data. Among them, each node represents an independent data feature, and the potential relationship between nodes will be characterized by hyperedges in the future. Hyperedges are based on hypergraph theory, and multiple hypergraph nodes are connected through the analysis of multi-dimensional unit data. Hyperedges can express complex multivariate data associations, and multiple data features can be associated simultaneously in a hyperedge to form hyperedge data. This operation enables the model to capture the complex associations between multiple data dimensions in wind turbines, such as the potential connection between temperature changes and speed fluctuations and fault signals.

[0076] Among them, the weight matrix is ​​constructed based on the strength of the association between nodes and the structure of the hyperedge, and can quantify the contribution of the nodes in the hyperedge. The weight matrix not only considers the direct relationship between nodes, but also introduces the high-order interaction effect between nodes, which helps to accurately capture the abnormal patterns in fault data. Hypergraph is a generalized graph structure used to represent the complex association relationship between multiple entities. The hypergraph completes the overall construction of the hypergraph by combining the hypergraph node data, the hyperedge data and the hypergraph weight matrix to obtain the complete hypergraph data. This hypergraph structure can not only efficiently express the complex association between multi-source data, but also provide rich information input for subsequent hypergraph convolution operations, so that the model can still maintain efficient fault detection capabilities in high-dimensional complex data scenarios. Exemplarily, in an embodiment of the present invention, the hypergraph can represent data from different sources (mechanical parts, environment and operating status) as nodes, and construct hyperedges according to the association between them. By constructing a hypergraph model and applying a hypergraph convolutional network, the potential abnormal relationship in the data can be effectively mined, and the abnormal score of each node and the overall abnormal score can be calculated. Hypergraph can improve the processing capabilities of high-dimensional and complex data, ultimately achieving more accurate and efficient fault detection.

[0077] In step S14, the hypergraph data is input into a multi-layer hypergraph convolution model, and a multi-layer hypergraph convolution processing operation is performed to obtain hypergraph convolution feature data, including:

[0078] Based on the hypergraph data, performing a hypergraph convolution operation on nodes in the hypergraph data to obtain node feature data;

[0079] According to the node feature data, the node features are aggregated layer by layer through a multi-layer hypergraph convolutional network to obtain node high-order feature data;

[0080] Node embedding is performed based on the node high-order feature data to obtain hypergraph convolution feature data.

[0081] It should be noted that the hypergraph convolution operation is a feature extraction method based on the hypergraph structure. By performing convolution calculations on the nodes in the hypergraph and their associated hyperedges, high-order features between nodes are extracted. The hypergraph convolution operation introduces the structural information of the hypergraph into the convolution calculation, and performs weighted aggregation on the features of the node, its directly associated hyperedges and related nodes. This process is similar to traditional graph convolution, but can handle complex structures associated with multiple nodes. Layer-by-layer aggregation of node features is a technology that gradually fuses node features through a multi-layer network structure. It is mainly used in graph convolution networks (GCNs) and their extended models (such as hypergraph convolution networks). Its core idea is that in each layer of convolution operation, the node features not only depend on the node itself, but also aggregate the information of its neighboring nodes. As the number of convolution layers increases, the node gradually integrates more feature information of its neighbors to form a global understanding of the entire graph or hypergraph. In a multi-layer hypergraph convolution network, as the number of convolution layers increases, the features of the node are not only affected by the nodes directly related to it, but also gradually aggregate a wider range of node features. This aggregation method can capture deeper relationships between nodes and extract high-order features of nodes. After convolution layer by layer, the high-order features of the nodes are further processed to generate the embedding vectors of the nodes. These embedding vectors are compact representations of the positions and features of the nodes in the hypergraph, which are used for subsequent fault detection and anomaly score calculation. In this way, the constructed multi-layer hypergraph convolutional network can effectively extract the complex structural information in the hypergraph and generate feature data that can reflect the operating status of the wind turbine, laying the foundation for subsequent fault anomaly detection.

[0082] Wherein, the hypergraph convolution operation is performed on the nodes in the hypergraph data, and the hypergraph convolution calculation formula is as follows:

[0083] ;

[0084] in, Indicates The node feature matrix of the layer, Indicates The node feature matrix of the layer; represents the activation function; represents the node degree matrix; represents the degree matrix of the hyperedge; represents the connection matrix of the hyperedge, represents the transposed form of E; Indicates The network parameters that the layer needs to learn; represents the association matrix between nodes and hyperedges, express The transposed form of .

[0085] It should be noted that the node feature matrix represents the feature vector of each node, and the initial node feature matrix Usually the original features of the node, the node feature matrix of the subsequent layer is the feature matrix of the previous layer. Obtained through the hypergraph convolution operation. The node feature matrix is ​​used to represent the features of the node and is the input and output of the hypergraph convolution operation. The node feature matrix of each layer contains the feature representation of the node in the current layer and is used for subsequent feature extraction and classification tasks. Node degree matrix Represents the degree of each node (that is, the number of hyperedges connected to the node), through the association matrix between nodes and hyperedges Calculate and get the node degree matrix Used to normalize node features to ensure that nodes with different degrees have the same weights during feature updates in hypergraph convolution operations. Indicates the number of nodes connected by each hyperedge, through the association matrix between hyperedges and nodes E The degree matrix of the hyperedge is Used to normalize hyperedge features to ensure that hyperedges of different degrees have the same weight during feature aggregation in hypergraph convolution operations. Hyperedge connection matrix Indicates which nodes are connected by which hyperedges, which is defined by the structure of the hypergraph. The connection matrix of the hyperedge It is used to represent the connection relationship between nodes and hyperedges, and is a key part of feature aggregation in hypergraph convolution operations. and The product of can aggregate node features to hyperedges. The association matrix between nodes and hyperedges is Indicates which hyperedges each node is associated with, which is also defined through the structure of the hypergraph. It is used to represent the relationship between nodes and hyperedges and is a key part of feature aggregation in hypergraph convolution operations. and The product of can map the hyperedge features back to the node features. The learnable parameter matrix Indicates The learnable parameter matrix of the layer is learned during the network training process. It is used to perform linear transformation on node features and extract higher-level feature representations. All are optimized through the back-propagation algorithm to minimize the loss function. Represents the activation function, which is obtained by selecting a suitable activation function type, usually a nonlinear function such as ReLU and Sigmoid, which is used to introduce nonlinear transformations and enhance the expressiveness of the model. Through the activation function, the features after linear transformation can be mapped to a nonlinear space, so as to better capture the complex patterns of the data. Exemplarily, in an embodiment of the present invention, the activation function used is the ReLU function. Of course, according to different actual application scenarios and user needs, the activation function can also be set to a nonlinear function such as Sigmoid, LeakReLU, etc., which is not limited by the present invention.

[0086] In step S15, the hypergraph convolution feature data is input into an anomaly score calculation model to perform an anomaly score calculation to obtain a fault anomaly score of the wind turbine generator set, including:

[0087] The hypergraph convolution feature data is used as input data and input into the graph variational autoencoder model for feature reconstruction and calculation to obtain a reconstruction error;

[0088] Calculating a node anomaly score according to the reconstruction error to obtain a node anomaly score;

[0089] According to the node anomaly scores, score aggregation is performed to obtain the fault anomaly score of the wind turbine generator set.

[0090] It should be noted that the graph variational autoencoder (GVAE) is a generative model that combines a variational autoencoder and a graph neural network. It is used to learn the potential representation of graph data and generate new graph structures. By mapping the graph data to the latent space through the encoder and reconstructing the graph data through the decoder, GVAE can perform feature reconstruction and anomaly detection, and is widely used in tasks such as graph generation, graph embedding, and anomaly detection. The graph variational autoencoder mainly includes an encoder and a decoder. Among them, the encoder maps the input graph data (such as the node feature matrix and the adjacency matrix) to the mean and variance vectors in the latent space. The decoder reconstructs the original graph data from random sampling points in the latent space. The graph variational autoencoder model (VAE) learns the potential representation of the data through the encoder and decoder structure, and reconstructs the features for feature reconstruction and calculation to obtain the reconstruction error. The reconstruction error refers to the difference between the original data and the reconstructed data during the feature reconstruction process. In the graph variational autoencoder (GVAE), the reconstruction error is used to measure the difference between the graph data reconstructed by the decoder and the original graph data. The reconstruction error reflects the difference between the original feature and the reconstructed feature. The larger the difference, the more likely the node is abnormal, so the anomaly score can be calculated. Wind turbines are usually composed of multiple mechanical components and complex sensor systems, so the anomaly score of a single node cannot fully describe the state of the entire system. In order to give a system-level fault anomaly score, the anomaly scores of each node need to be aggregated. The aggregation method can adopt weighted average, mean, or other suitable statistical methods to comprehensively analyze the abnormal conditions of each node to obtain the fault anomaly score of the entire wind turbine. Exemplarily, in an embodiment of the present invention, the aggregation method adopted is weighted average. Of course, according to different actual application scenarios and user needs, the aggregation method can also adopt other suitable statistical methods such as mean, and the present invention is not limited to this.

[0091] The hypergraph convolution feature data is used as input data and input into the graph variational autoencoder model for feature reconstruction and calculation to obtain the reconstruction error, including:

[0092] Performing data encoding processing on the hypergraph convolution feature data to obtain potential latent variable representation;

[0093] According to the latent variable representation, decoding mapping is performed to reconstruct the original high-order feature data to obtain reconstructed feature data;

[0094] A reconstruction error is calculated based on the hypergraph convolution feature data and the reconstruction feature data to obtain a reconstruction error.

[0095] It should be noted that the working process of the graph variational autoencoder (GVAE) consists of two main steps: encoding and decoding, which are used to capture the complex relationship between nodes and perform anomaly detection based on this. By calculating the reconstruction error, it is determined whether the node is abnormal. The data encoding process encodes the input hypergraph convolution feature data, and the encoder maps it into potential low-dimensional latent variable representations (latent variables). These latent variable representations contain the complex relationship between hypergraph nodes and their local features, and condense the multi-dimensional information between nodes. The encoding process generates a probability distribution (mean and variance) of node features to model the complexity of node features. When decoding and mapping and reconstructing features, the decoder maps it back to the original high-dimensional space based on the latent variable representation obtained by encoding, that is, reconstructs the input feature data. This process is similar to reverse restoration, which aims to regenerate features similar to the original hypergraph convolution feature data from the compressed latent variable representation. It is worth noting that the reconstructed high-order feature data should be as consistent as possible with the original input data, but if the relationship or features between nodes are abnormal, there will be obvious differences between the reconstructed features and the original data. The reconstruction error reflects the degree of fit of the model to the node features and their relationships. A larger error indicates that the node has a potential anomaly in the overall operating state of the wind turbine. The reconstruction error is completed by calculating the difference between the original hypergraph convolution feature data and the reconstructed feature data. The size of the error can be measured by a commonly used error function (such as square error, Gaussian error, etc.). Exemplarily, in an embodiment of the present invention, the error function used is the square error. Of course, according to different actual application scenarios and user needs, the aggregation method can also use other suitable error functions such as the Gaussian error function, and the present invention is not limited to this.

[0096] In step S15, the node anomaly score is calculated based on the reconstruction error, and the calculation formula of the anomaly score is as follows:

[0097] ;

[0098] in, Represents the first The abnormal score of a node, that is, the score of the data source at a certain position in the original data, such as the score of the fan blade vibration data; Representation Node The set of neighbor nodes of node The node collection related to the data source represented; Representation Node The number of neighbor nodes; Representation Node The reconstruction error.

[0099] It should be noted that Represents the first The abnormality score of a node is used to measure the abnormality of node i. This score can be understood as a numerical indicator that reflects whether the node behaves abnormally in the wind turbine. For example, A node may represent a specific sensor data source, such as fan blade vibration data or temperature data, through the score It can be evaluated whether the working status of the sensor is normal. Representation Node The set of neighbor nodes of node i There may be certain correlations between data sources with strong correlations, such as multiple sensors. i When looking at the abnormality score, we cannot just look at i The node itself also needs to be combined with the performance of its surrounding neighbor nodes, because the failure may be propagated among multiple related sensors or data sources. Representation Node The number of neighbor nodes of node i The number of other nodes that are directly connected. By weighting the reconstruction error of neighboring nodes and averaging them, the node can be reflected more stably. i The anomaly score is calculated to avoid excessive errors caused by a single node anomaly. Representation Node The reconstruction error of each neighbor node j When reconstructing features through the graph variational autoencoder model (GVAE), a reconstruction error is generated. This error value reflects the neighboring nodes j The abnormality of the node j If the reconstruction error of the neighboring nodes is large, it means that it may also be abnormal. After averaging the reconstruction errors of the neighboring nodes through the formula, we can get the node i This method can capture the correlation between nodes, so that the anomaly score depends not only on a single node but also on the overall network structure between nodes.

[0100] It is worth noting that the above formula is calculated by the node i The mean reconstruction error of all neighbor nodes of node i The advantage of this is that if a node has a small reconstruction error, but its neighboring nodes have a large reconstruction error, the node i It will still be considered as a potential anomaly, thus improving the accuracy of fault detection.

[0101] In step S16, when it is determined that the fault anomaly score is greater than a preset anomaly threshold, an alarm signal is generated and sent to the user end.

[0102] It should be noted that the process of generating alarm signals is to timely identify potential wind turbine failure risks. By calculating the reconstruction error of each node and aggregating it into the fault anomaly score of the wind turbine, the system can monitor the operating status of the unit in real time. When the fault anomaly score exceeds the preset threshold, it means that the system detects that the wind turbine may have a fault, such as mechanical wear, sensor failure, or other equipment abnormalities. At this time, the system will immediately generate an alarm signal. The alarm signal will contain detailed information about the abnormal situation, such as the fault anomaly score, the time when the abnormal event occurred, the associated sensor data characteristics, etc., to ensure that the user end can quickly take corresponding maintenance measures. This operation improves the response speed of wind turbine fault detection through a real-time feedback mechanism, reduces the impact of potential faults on the operating efficiency of the unit, and thus improves the overall reliability of the wind turbine.

[0103] Among them, the preset abnormal threshold is a reference value set by the system when performing fault detection, which is used to determine whether the state of the wind turbine is abnormal. Specifically, the fault abnormality score quantifies the abnormality of the current operating state of the unit. When the score exceeds the preset threshold, the system will consider that the current operating state has abnormal characteristics, thereby triggering an alarm signal. The setting of the preset abnormal threshold is usually optimized and adjusted based on historical fault data, the operating requirements of the unit, and specific application scenarios. A threshold that is too low may lead to a high false alarm rate (i.e., the normal operating state is mistaken for an abnormality), while a threshold that is too high may lead to missed alarms (i.e., the real fault is not detected in time). Therefore, it is crucial to reasonably set the abnormal threshold to balance the sensitivity and accuracy of detection, ensuring that the system can respond quickly to potential faults and avoid unnecessary false alarm warnings. Exemplarily, in an embodiment of the present invention, the preset abnormal threshold is set to 85. When the fault abnormality score is greater than the preset abnormal threshold, the system will generate an alarm signal and send it to the user end. Of course, according to different actual application scenarios, the preset abnormal threshold can also be set to 80, 90, etc., and the present invention is not limited to this.

[0104] Reference Figure 2 The second embodiment of the present invention provides a wind turbine fault monitoring system, comprising:

[0105] A data acquisition module, used to acquire sensor data, environmental data and operation data of mechanical components;

[0106] A data preprocessing module, used to perform data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data;

[0107] A hypergraph construction module, used to perform a hypergraph construction operation according to the multi-dimensional unit data to obtain hypergraph data;

[0108] A hypergraph convolution module, used to input the hypergraph data into a multi-layer hypergraph convolution model, perform multi-layer hypergraph convolution processing operations, and obtain hypergraph convolution feature data;

[0109] A score calculation module, used for inputting the hypergraph convolution feature data into an anomaly score calculation model, performing anomaly score calculation, and obtaining a fault anomaly score of the wind turbine generator set;

[0110] The abnormality determination and alarm module is used to generate an alarm signal when it is determined that the fault abnormality score is greater than a preset abnormality threshold, and send the alarm signal to the user end.

[0111] It should be noted that a wind turbine fault monitoring system provided in an embodiment of the present invention is used to execute all process steps of a wind turbine fault monitoring method in the above embodiment, and the working principles and beneficial effects of the two correspond one to one, so they will not be repeated.

[0112] In summary, the present invention provides a method for monitoring wind turbine faults, including: obtaining mechanical component sensor data, environmental data and operating data; performing data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multidimensional unit data; performing a hypergraph construction operation according to the multidimensional unit data to obtain hypergraph data; inputting the hypergraph data into a multi-layer hypergraph convolution model, performing multi-layer hypergraph convolution processing operations, and obtaining hypergraph convolution feature data; inputting the hypergraph convolution feature data into an anomaly score calculation model, performing anomaly score calculation, and obtaining a fault anomaly score of the wind turbine; when it is determined that the fault anomaly score is greater than a preset anomaly threshold, generating an alarm signal, and sending the alarm signal to a user end. The present invention models the multidimensional information in the mechanical component sensor data, environmental data and operating data as hypergraph nodes and hyperedges by constructing a heterogeneous hypergraph. By adopting the anomaly score calculation model and calculating the anomaly score of the hypergraph convolution feature data, the operating status of each node of the wind turbine can be accurately evaluated, the abnormal nodes can be determined, and the fault anomaly score of the wind turbine can be finally generated. The detection standard can be adaptively adjusted, and the sensitivity to wind turbine faults is improved, thereby achieving more accurate fault detection and real-time alarm.

[0113] The embodiment of the present invention further provides a terminal device. The terminal device includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a wind turbine fault monitoring program. When the processor executes the computer program, the steps in the above-mentioned wind turbine fault monitoring method embodiments are implemented, such as Figure 1 Alternatively, the processor implements the functions of each module / unit in the above-mentioned system embodiments when executing the computer program.

[0114] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0115] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, and a smart tablet. The terminal device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the above components are merely examples of terminal devices and do not constitute a limitation on the terminal device. The terminal device may include more or fewer components than the above components, or may combine certain components, or different components. For example, the terminal device may also include input and output devices, network access devices, buses, etc.

[0116] 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) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0117] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module 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, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0118] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or system capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0119] It should be noted that the system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the system embodiment provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0120] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for monitoring wind turbine faults, characterized in that: Executed by the server, including: Obtain sensor data, environmental data and operating data of mechanical components; Performing data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data; Performing a hypergraph construction operation according to the multi-dimensional unit data to obtain hypergraph data; Inputting the hypergraph data into a multi-layer hypergraph convolution model, performing a multi-layer hypergraph convolution processing operation, and obtaining hypergraph convolution feature data; Inputting the hypergraph convolution feature data into an anomaly score calculation model to perform an anomaly score calculation to obtain a fault anomaly score of the wind turbine generator set; When it is determined that the fault anomaly score is greater than a preset anomaly threshold, an alarm signal is generated and sent to a user terminal; The step of inputting the hypergraph convolution feature data into an anomaly score calculation model to perform an anomaly score calculation to obtain a fault anomaly score of the wind turbine generator includes: The hypergraph convolution feature data is used as input data and input into the graph variational autoencoder model for feature reconstruction and calculation to obtain a reconstruction error; Calculating a node anomaly score according to the reconstruction error to obtain a node anomaly score; According to the node anomaly scores, score aggregation is performed to obtain a fault anomaly score of the wind turbine; The hypergraph convolution feature data is used as input data and input into the graph variational autoencoder model for feature reconstruction and calculation to obtain the reconstruction error, including: Performing data encoding processing on the hypergraph convolution feature data to obtain potential latent variable representation; According to the latent variable representation, decoding mapping is performed to reconstruct the original high-order feature data to obtain reconstructed feature data; Calculating a reconstruction error based on the hypergraph convolution feature data and the reconstruction feature data to obtain a reconstruction error; Wherein, the node anomaly score is calculated according to the reconstruction error, and the calculation formula of the anomaly score is as follows: ； in, represents the anomaly score of the 𝑖th node in the hypergraph, that is, the score of the data source at a certain position in the original data; Representation Node The set of neighbor nodes of node The node collection related to the data source represented; Representation Node The number of neighbor nodes; Representation Node The reconstruction error.

2. The method for monitoring wind turbine fault according to claim 1, characterized in that: The data preprocessing operation is performed according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data, including: Performing data cleaning processing according to the mechanical component sensor data, the environmental data and the operating data to obtain clean unit data; Performing data standardization processing on the clean unit data to obtain standard unit data; According to the standard unit data, multi-dimensional key features are extracted to obtain multi-dimensional unit data.

3. The method for monitoring wind turbine failure according to claim 1, characterized in that: The step of performing a hypergraph construction operation according to the multi-dimensional unit data to obtain hypergraph data includes: According to the multi-dimensional unit data, the data features are used as hypergraph nodes to construct nodes, thereby obtaining hypergraph node data; According to the multi-dimensional unit data and the hypergraph node data, a hyperedge is constructed to obtain hyperedge data; Generating a weight matrix according to the multi-dimensional unit data and the hyperedge data to obtain hypergraph weight matrix data; A hypergraph construction operation is performed according to the hypergraph node data, the hyperedge data and the hypergraph weight matrix data to obtain hypergraph data.

4. The method for monitoring wind turbine failure according to claim 1, characterized in that: The step of inputting the hypergraph data into a multi-layer hypergraph convolution model and performing a multi-layer hypergraph convolution processing operation to obtain hypergraph convolution feature data includes: Based on the hypergraph data, performing a hypergraph convolution operation on nodes in the hypergraph data to obtain node feature data; According to the node feature data, the node features are aggregated layer by layer through a multi-layer hypergraph convolutional network to obtain node high-order feature data; Node embedding is performed based on the node high-order feature data to obtain hypergraph convolution feature data.

5. The method for monitoring wind turbine fault according to claim 4, characterized in that: The hypergraph convolution operation is performed on the nodes in the hypergraph data, and the hypergraph convolution calculation formula is as follows: ； in, Indicates The node feature matrix of the layer, Indicates The node feature matrix of the layer; represents the activation function; represents the node degree matrix; represents the degree matrix of the hyperedge; represents the connection matrix of the hyperedge, represents the transposed form of E; Indicates The network parameters that the layer needs to learn; represents the association matrix between nodes and hyperedges, express The transposed form of .

6. A wind turbine fault monitoring system, characterized in that: A method for monitoring a wind turbine fault according to any one of claims 1 to 5, comprising: A data acquisition module, used to acquire sensor data, environmental data and operation data of mechanical components; A data preprocessing module, used to perform data preprocessing operations according to the mechanical component sensor data, the environmental data and the operating data to obtain preprocessed multi-dimensional unit data; A hypergraph construction module, used to perform a hypergraph construction operation according to the multi-dimensional unit data to obtain hypergraph data; A hypergraph convolution module, used to input the hypergraph data into a multi-layer hypergraph convolution model, perform multi-layer hypergraph convolution processing operations, and obtain hypergraph convolution feature data; A score calculation module, used for inputting the hypergraph convolution feature data into an anomaly score calculation model, performing anomaly score calculation, and obtaining a fault anomaly score of the wind turbine generator set; The abnormality determination and alarm module is used to generate an alarm signal when it is determined that the fault abnormality score is greater than a preset abnormality threshold, and send the alarm signal to the user end.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the method for monitoring wind turbine faults according to any one of claims 1 to 5.

Images