Fault diagnosis system and method for transmission chain of wind generating set

Abstract

The invention provides a transmission chain fault diagnosis system and method of a wind generating set. The system comprises a data acquisition module, a global rotating speed characteristic matrix determination module, a transmission logic characteristic matrix determination module, a graph data coding module, a characteristic distribution correction module and a fault diagnosis result generation module. According to the technical scheme provided by the invention, real-time intelligent monitoring can be accurately carried out on the transmission chain of the wind generating set, so that the fault early warning is automatically generated when the fault of a certain part of the transmission chain is detected, the operation safety of a fan is ensured, and the accuracy of fault diagnosis is improved.

Description

Technical Field

[0001] The present application relates to the field of mechanical fault diagnosis, and in particular to a transmission chain fault diagnosis system and method for a wind turbine generator set. Background Art

[0002] With the increasing emphasis on wind energy utilization and the improvement of wind power generation technology, the proportion of wind power in the electricity market has continued to rise. Among the wind turbines that have been put into production, horizontal axis, three-blade, doubly fed wind turbines occupy a dominant position.

[0003] A wind turbine (or wind turbine for short) transmits mechanical energy from a rotor to a generator via a drive train. The drive train consists of the rotor, main shaft, gearbox, high-speed shaft, and generator. The rotor drives the generator via the main shaft through the gearbox, which in turn drives the generator via the high-speed shaft. The gearbox is a speed-increasing gearbox that increases the main shaft speed from 20-50 rpm to 1000-1500 rpm, the speed required to drive most generators.

[0004] Because large wind turbines are typically located in remote areas, the harsh natural environment and complex equipment make them vulnerable to damage, impacting production. Wind turbine downtime can have a significant impact on power grid security and the national economy, making it imperative to improve wind power reliability and ensure high linear redundancy. However, current wind farm maintenance typically relies on scheduled maintenance and troubleshooting. This approach struggles to fully and promptly detect potential faults, often resulting in prolonged downtime and significant losses. Furthermore, the complex transmission chain structure of doubly-fed wind turbines has been a frequent source of failure. Fluctuations in the rotor blades during transmission expose the transmission chain and its speed sensor to vibrations, which can easily loosen the installed speed sensor, leading to inaccurate speed. Reliable speed measurement in the wind turbine transmission chain ensures safe wind turbine operation. However, if wind turbine speed measurement is abnormal or a component in the transmission chain fails, the control system will be unable to implement the shutdown strategy, potentially leading to serious runaway accidents. Therefore, there is an urgent need to provide a system that can accurately diagnose faults in the transmission chain of a wind turbine generator set and perform real-time intelligent monitoring, so as to automatically generate a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine. Summary of the Invention

[0005] The present application provides a transmission chain fault diagnosis system and method for a wind turbine generator set, so as to at least solve the technical problem that a transmission chain fault of a wind turbine generator set cannot be accurately diagnosed.

[0006] The first embodiment of the present application provides a transmission chain fault diagnosis system for a wind turbine generator set, comprising:

[0007] A data acquisition module, used to collect the rotational speed values of various parts of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period;

[0008] A global speed characteristic matrix determination module is used to determine the global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of the various parts at multiple predetermined time points within a predetermined time period;

[0009] a transmission logic characteristic matrix determination module, configured to construct a transmission logic matrix of each part of the transmission chain, and determine a transmission logic characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix;

[0010] A graph data encoding module, configured to pass the global speed characteristic matrix and the transmission logic characteristic matrix through a graph neural network to obtain a transmission topology global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed;

[0011] a characteristic distribution correction module, configured to perform characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix;

[0012] The fault diagnosis result generating module is used to pass the corrected transmission topology global speed characteristic matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

[0013] Preferably, the global speed characteristic matrix determination module includes:

[0014] A component speed feature extraction and encoding submodule is used to arrange the speed values of each component at multiple predetermined time points within a predetermined time period according to the time dimension as a speed input vector and then pass it through a multi-scale neighborhood feature extraction module to obtain a speed feature vector corresponding to each component;

[0015] A matrixing submodule, configured to perform a two-dimensional arrangement on the plurality of speed characteristic vectors to obtain a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed;

[0016] The component speed feature extraction and encoding submodule includes:

[0017] a first-scale speed feature extraction unit, configured to pass the speed input vector through a first one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale speed feature vector, wherein the first one-dimensional convolution layer performs one-dimensional convolution processing on the speed input vector using a one-dimensional convolution kernel having a first length;

[0018] a second-scale speed feature extraction unit, configured to pass the speed input vector through a second one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale speed feature vector, wherein the second one-dimensional convolution layer performs one-dimensional convolution processing on the speed input vector using a one-dimensional convolution kernel having a second length;

[0019] The multi-scale rotational speed feature aggregation unit is configured to concatenate the first-scale rotational speed feature vector and the second-scale rotational speed feature vector to obtain the rotational speed feature vector corresponding to each part.

[0020] Furthermore, the first scale speed feature extraction unit uses the first convolution layer of the multi-scale neighborhood feature extraction module to calculate the speed of rotation using the formula Perform one-dimensional convolution encoding on the speed input vector to obtain the first-scale neighborhood speed feature vector; wherein a is the width of the first convolution kernel in the x direction, F is the first convolution kernel parameter vector, G is the local vector matrix operated with the first convolution kernel function, w is the size of the first convolution kernel, and X is the speed input vector.

[0021] Furthermore, the second-scale speed feature extraction unit uses the second convolution layer of the multi-scale neighborhood feature extraction module to calculate the speed of rotation using the formula One-dimensional convolution encoding is performed on the speed input vector to obtain the second-scale neighborhood speed feature vector; wherein b is the width of the second convolution kernel in the x direction, f is the second convolution kernel parameter vector, g is the local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents the speed input vector.

[0022] Preferably, the transmission logic characteristic matrix determination module includes:

[0023] a transmission logic topology construction submodule, configured to construct a transmission logic matrix for each part of the transmission chain, wherein the eigenvalue of each non-diagonal position in the transmission logic matrix is used to indicate whether a transmission connection exists between the corresponding two parts;

[0024] a transmission topology feature extraction submodule, configured to pass the transmission logic matrix through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix;

[0025] The step of passing the transmission logic matrix through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix includes:

[0026] Using the convolution units of each layer of the convolutional neural network model to perform convolution processing based on a two-dimensional convolution kernel on the input data to obtain a convolution feature map;

[0027] Using the pooling units of each layer of the convolutional neural network model to perform pooling processing on the convolution feature map along the channel dimension to obtain a pooled feature map; and

[0028] Using the activation units of each layer of the convolutional neural network model to perform nonlinear activation on the feature values at each position in the pooling feature map to obtain an activation feature map;

[0029] Among them, the output of the last layer of the convolutional neural network model is the transmission logic feature matrix.

[0030] Furthermore, in the transmission logic matrix, if there is a transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is one; if there is no transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is zero.

[0031] Preferably, the feature distribution correction module includes:

[0032] Eigenvector correction factor calculation unit, used to calculate the Calculate the energy aggregation factor of the wavelet-like function family of each transmission topology global speed characteristic vector in the transmission topology global speed characteristic matrix respectively; where v i represents the eigenvalue of each position of the global speed eigenvector of each transmission topology, r represents the energy aggregation factor of the wavelet-like function family of the global speed eigenvector of each transmission topology, L is the length of the global speed eigenvector of each transmission topology, and log represents the logarithmic function value with base 2;

[0033] A weighted correction unit is used to use the energy aggregation factor of the wavelet-like function family of each transmission topology global speed characteristic vector as a weighted weight, and perform weighted optimization on each transmission topology global speed characteristic vector to obtain the corrected transmission topology global speed characteristic matrix.

[0034] Preferably, the fault diagnosis result generating module uses the classifier to calculate the formula O=softmax{(W c , B c )|Project(M)}, the corrected transmission topology global speed feature matrix is processed to obtain a classification result; wherein, Project(M) represents the projection of the corrected transmission topology global speed feature matrix into a vector, W c is the weight matrix of the fully connected layer, B c Represents the bias vector of the fully connected layer.

[0035] Preferably, the system further comprises an early warning device;

[0036] The early warning device is used to automatically generate a fault early warning reminder when the classification result shows that a certain part of the transmission chain has failed.

[0037] A second embodiment of the present application provides a method for diagnosing transmission chain faults of a wind turbine generator set, the method comprising:

[0038] Collecting the rotational speed values of various parts of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period;

[0039] Determining a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of the various parts at a plurality of predetermined time points within a predetermined time period;

[0040] Constructing a transmission logic matrix of each part of the transmission chain, and determining a transmission logic characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix;

[0041] The global speed characteristic matrix and the transmission logic characteristic matrix are passed through a graph neural network to obtain a transmission topology global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed;

[0042] Performing characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix;

[0043] The corrected transmission topology global speed characteristic matrix is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

[0044] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects:

[0045] The present application proposes a transmission chain fault diagnosis system and method for a wind turbine generator set, wherein the system includes: a data acquisition module for collecting the speed values of each part of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period; a global speed feature matrix determination module for determining the global speed feature matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of each part at multiple predetermined time points within a predetermined time period; a transmission logic feature matrix determination module for constructing the transmission logic matrix of each part of the transmission chain and determining the transmission logic feature matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix; a graph data encoding module for passing the global speed feature matrix and the transmission logic feature matrix through a graph neural network to obtain a transmission topology global speed feature matrix corresponding to the wind turbine generator set to be diagnosed; a feature distribution correction module for performing feature distribution correction on the transmission topology global speed feature matrix to obtain a corrected transmission topology global speed feature matrix; and a fault diagnosis result generation module for passing the corrected transmission topology global speed feature matrix through a classifier to obtain a classification result, wherein the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set. The technical solution proposed in this application can accurately perform real-time intelligent monitoring of the transmission chain of the wind turbine generator set, so as to automatically generate a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine and improving the accuracy of fault diagnosis.

[0046] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0048] Figure 1 This is a diagram of an application scenario of a transmission chain fault diagnosis system and method for a wind turbine generator set according to an embodiment of the present application;

[0049] Figure 2 This is a first structural diagram of a transmission chain fault diagnosis system for a wind turbine generator set provided according to one embodiment of the present application;

[0050] Figure 3 This is a structural diagram of a global speed characteristic matrix determination module provided according to one embodiment of the present application;

[0051] Figure 4 This is a structural diagram of a component speed feature extraction and encoding submodule according to one embodiment of the present application;

[0052] Figure 5 This is a structural diagram of a transmission logic characteristic matrix determination module provided according to one embodiment of the present application;

[0053] Figure 6 A structural diagram of a feature distribution correction module provided according to one embodiment of the present application;

[0054] Figure 7 This is a second structural diagram of a transmission chain fault diagnosis system for a wind turbine generator set provided according to one embodiment of the present application;

[0055] Figure 8 This is a flow chart of a method for diagnosing transmission chain faults of a wind turbine generator set according to one embodiment of the present application;

[0056] Figure 9 A detailed flow chart of a method for diagnosing transmission chain faults of a wind turbine generator set according to one embodiment of the present application;

[0057] Figure 10 A schematic diagram of the architecture of a transmission chain fault diagnosis method for a wind turbine generator set according to an embodiment of the present application is provided;

[0058] Reference numerals:

[0059] Data acquisition module 1, global speed feature matrix determination module 2, transmission logic feature matrix determination module 3, graph data encoding module 4, feature distribution correction module 5, fault diagnosis result generation module 6, component speed feature extraction encoding submodule 2-1, matrixization submodule 2-2, first scale speed feature extraction unit 2-1-1, second scale speed feature extraction unit 2-1-2, multi-scale speed feature aggregation unit 2-1-3, transmission logic topology construction submodule 3-1, transmission topology feature extraction submodule 3-2, feature vector correction factor calculation unit 5-1, weighted correction unit 5-2, early warning device 7. DETAILED DESCRIPTION

[0060] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0061] Application Overview

[0062] A wind turbine (or wind turbine for short) transmits mechanical energy from a rotor to a generator via a drive train. The drive train consists of the rotor, main shaft, gearbox, high-speed shaft, and generator. The rotor drives the generator via the main shaft through the gearbox, which in turn drives the generator via the high-speed shaft. The gearbox is a speed-increasing gearbox that increases the main shaft speed from 20-50 rpm to 1000-1500 rpm, the speed required to drive most generators.

[0063] Because large wind turbines are typically located in wide, marginal areas, the harsh natural environment and complex equipment make them vulnerable to damage, impacting production. Wind turbine downtime can have a significant impact on grid security and the national economy, making it imperative to improve the reliability and high-linearity redundancy of wind power generation. However, current wind farm maintenance typically relies on scheduled maintenance and troubleshooting. This approach struggles to fully and promptly detect potential faults, often resulting in long-term and significant losses for operators.

[0064] Due to the complex transmission chain structure of doubly-fed wind turbines, they have always been a frequent source of wind turbine failure. Fluctuations in the transmission of the rotor blades expose the transmission chain and its speed sensor to vibrations, which can easily loosen the installed speed sensor, leading to inaccurate speed. Reliable speed measurement of the wind turbine transmission chain ensures safe operation. If an anomaly in wind turbine speed measurement occurs, or if a component in the transmission chain fails, the control system will be unable to enter a shutdown state according to the control strategy, potentially resulting in a serious runaway accident.

[0065] Therefore, a transmission chain fault diagnosis system and method for a wind turbine generator set is expected, which can perform real-time intelligent monitoring of the transmission chain of the wind turbine generator set, so as to automatically generate a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine.

[0066] Currently, deep learning and neural networks are widely used in fields such as computer vision, natural language processing, and speech signal processing. Furthermore, deep learning and neural networks have demonstrated capabilities approaching or even surpassing those of humans in areas such as image classification, object detection, semantic segmentation, and text translation.

[0067] In recent years, the development of deep learning and neural networks has provided new solutions and solutions for transmission chain fault diagnosis of wind turbines.

[0068] Accordingly, since the transmission chain of the wind turbine generator set includes a wind wheel, a main shaft, a gearbox, a high-speed shaft and a generator, and the wind wheel drives the gearbox through the main shaft, and the gearbox drives the generator through the high-speed rotating shaft. Therefore, there are transmission correlation characteristics between the various components in the transmission chain of the wind turbine generator set. Based on this, in the technical solution of the present application, an artificial intelligence monitoring technology based on deep learning is used to extract the multi-scale neighborhood correlation characteristics of each part of the transmission chain of the wind turbine generator set in time sequence, and further construct a transmission logic topology matrix between the various components, and use the transmission logic correlation characteristics between the various components of the transmission chain to assist in the transmission chain fault diagnosis of the wind turbine generator set, thereby improving the accuracy of fault diagnosis. In this way, the transmission chain of the wind turbine generator set can be accurately monitored in real time and intelligently, so as to automatically generate a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine.

[0069] Specifically, in the technical solution of the present application, first, a speed sensor is used to obtain the speed values of each component of the wind turbine's transmission chain at multiple predetermined time points within a predetermined time period. Here, the wind turbine's transmission chain includes a rotor, a main shaft, a gearbox, a high-speed shaft, and a generator. During actual transmission chain operation, the rotor drives the gearbox via the main shaft, and the gearbox drives the generator via the high-speed shaft. Then, considering that the speed values of each component of the wind turbine's transmission chain have different periodic distribution patterns over different time spans, in order to more accurately extract the temporal dynamic feature information of the speed values of each component, the technical solution of the present application further arranges the speed values of each component at multiple predetermined time points within the predetermined time period according to the time dimension into speed input vectors, which are then processed through a multi-scale neighborhood feature extraction module to obtain multiple speed feature vectors. Specifically, the multi-scale neighborhood feature extraction module encodes the speed input vectors of each component of the wind turbine to extract dynamic multi-scale neighborhood correlation features of the speed values of each component over different time spans, thereby obtaining multiple speed feature vectors.

[0070] Next, considering that there are transmission-related characteristics between the various parts of the transmission chain, for example, the wind wheel drives the gearbox through the main shaft, and the gearbox drives the generator through the high-speed rotating shaft. Therefore, if the transmission logic topology characteristics of the various parts of the transmission chain can be used to optimize the speed characteristic expression of the various parts, it is obvious that the accuracy of fault diagnosis can be improved. Therefore, in the technical solution of the present application, a transmission logic matrix of the various parts of the transmission chain is further constructed, wherein the eigenvalues of each position on the non-diagonal position in the transmission logic matrix are used to indicate whether there is a transmission connection between the corresponding two components, that is, specifically, in the transmission logic matrix, if there is a transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is one, and if there is no transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is zero. Then, the transmission logic matrix is passed through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix. That is, the transmission logic matrix is feature mined using a convolutional neural network model as a feature extractor that has excellent performance in extracting implicit correlation features, so as to extract the correlation features of each position in the transmission logic matrix, that is, the transmission logic topology features between the various parts of the transmission chain, thereby obtaining a transmission logic feature matrix.

[0071] Furthermore, using the multiple speed feature vectors of each component as feature representations of nodes, and the transmission logic feature matrix as the feature representation of edges between nodes, a global speed feature matrix obtained by two-dimensionally permuting the multiple speed feature vectors and the transmission logic feature matrix are processed through a graph neural network to obtain a transmission topology global speed feature matrix. Specifically, the graph neural network uses learnable neural network parameters to perform graph-structured data encoding on the global speed feature matrix and the transmission logic feature matrix to obtain the transmission topology global speed feature matrix containing irregular transmission logic topology features and speed feature information of each component. The transmission topology global speed feature matrix is then passed through a classifier to obtain a classification result, which is used to indicate whether a fault exists in the wind turbine's transmission chain. Specifically, the transmission topology global speed feature matrix is used as the classification feature matrix for classification processing in the classifier to obtain a classification result indicating whether a fault exists in the wind turbine's transmission chain. This enables accurate diagnosis and detection of transmission chain faults in the wind turbine.

[0072] In particular, in the technical solution of the present application, when the transmission topology global speed feature matrix is passed through a classifier to obtain a classification result, since each transmission topology global speed feature vector of the transmission topology global speed feature matrix is a topological expression of the multi-scale correlation characteristics of the timing speed of each part of the transmission chain, and each part of the transmission chain has a certain degree of independence in the transmission logic and speed timing characteristics, this makes the information expression consistency between the various transmission topology global speed feature vectors relatively poor. Directly classifying the transmission topology global speed feature matrix through a classifier may lead to inductive deviations between the various transmission topology global speed feature vectors.

[0073] Therefore, for each transmission topology global speed eigenvector, the energy aggregation factor of its wavelet-like function family is calculated and expressed as:

[0074]

[0075] where log represents the logarithmic function with base 2, and L is the length of the global speed characteristic vector of each transmission topology.

[0076] That is, considering that for high-dimensional manifolds, information representation tends to concentrate on high-frequency components and therefore tends to be distributed at the edges of the manifold, a family of wavelet-like functions, acting as a separable transform for separating edges along the feature distribution dimension, can be used to convert the latent state of high-dimensional features into frequency components and express the amount of information in a wavelet-like energy manner. Thus, by using these functions as weighting coefficients to weight the global speed feature vectors of each transmission topology and then classifying the transmission topology global speed feature matrix through a classifier, the degree of information aggregation within the feature expression space of each transmission topology global speed feature vector in the transmission topology global speed feature matrix can be improved, thereby enhancing the consistency of information expression between the various transmission topology global speed feature vectors, thereby avoiding inductive bias and improving classification accuracy. This allows for accurate real-time intelligent monitoring of the wind turbine's transmission chain, automatically generating a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine.

[0077] Based on this, the present application provides a transmission chain fault diagnosis system for a wind turbine generator set, which includes:

[0078] The data acquisition module 1 is used to collect the rotational speed values of each part of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period;

[0079] A global speed characteristic matrix determination module 2 is configured to determine a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of the various parts at a plurality of predetermined time points within a predetermined time period;

[0080] a transmission logic characteristic matrix determination module 3, configured to construct a transmission logic matrix of each part of the transmission chain, and determine a transmission logic characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix;

[0081] A graph data encoding module 4 is configured to pass the global speed characteristic matrix and the transmission logic characteristic matrix through a graph neural network to obtain a transmission topology global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed;

[0082] A characteristic distribution correction module 5 is used to perform characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix;

[0083] The fault diagnosis result generating module 6 is used to pass the corrected transmission topology global speed characteristic matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

[0084] Figure 1 The figure shows an application scenario diagram of the transmission chain fault diagnosis system and method of a wind turbine generator set according to an embodiment of the present application. Figure 1 As shown, in this application scenario, by deploying in the transmission chain of the wind turbine generator set (here, the transmission chain of the wind turbine generator set includes the wind wheel, the main shaft, the gear box, the high-speed shaft and the generator, for example, Figure 1 The speed sensors (eg, Figure 1 G1-Gn) are used to collect the rotational speed values of the transmission chain of the wind turbine generator set at multiple predetermined time points within a predetermined time period. At the same time, a transmission logic matrix of each part of the transmission chain is constructed, wherein the characteristic values of each position on the non-diagonal position in the transmission logic matrix are used to indicate whether there is a transmission connection between the corresponding two parts. Then, the collected rotational speed values of the various parts at multiple predetermined time points within a predetermined time period and the transmission logic matrix are input into a server deployed with a transmission chain fault diagnosis algorithm for the wind turbine generator set (for example, Figure 1 S) as shown in , wherein the server is capable of using the transmission chain fault diagnosis algorithm of the wind turbine generator set to process the speed values of the various parts at multiple predetermined time points within a predetermined time period and the transmission logic matrix to generate a fault diagnosis result of the transmission chain of the wind turbine generator set.

[0085] After introducing the basic principles of the present application, various non-limiting embodiments of the present application will be described in detail with reference to the accompanying drawings.

[0086] Exemplary Systems

[0087] Figure 2FIG. 1 is a structural diagram of a transmission chain fault diagnosis system for a wind turbine generator set according to an embodiment of the present application. Figure 2 As shown, the system includes:

[0088] The data acquisition module 1 is used to collect the rotational speed values of each part of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period;

[0089] It should be noted that, since the transmission chain of the wind turbine generator set includes a wind rotor, a main shaft, a gearbox, a high-speed shaft, and a generator, and the wind rotor drives the gearbox through the main shaft, and the gearbox drives the generator through the high-speed rotating shaft. Therefore, there are transmission-related characteristics between the various components in the transmission chain of the wind turbine generator set. Based on this, in the technical solution of the present application, it is expected to accurately perform real-time intelligent monitoring of the transmission chain of the wind turbine generator set based on the high-dimensional implicit characteristics of the rotational speeds of the various components in the transmission chain of the wind turbine generator set and the transmission-related characteristics of the transmission chain of the wind turbine generator set.

[0090] Specifically, in the technical solution of the present application, the speed values of the various parts of the transmission chain of the wind turbine generator set at multiple predetermined time points within a predetermined time period are collected by speed sensors deployed on various parts of the transmission chain of the wind turbine generator set. Here, the transmission chain of the wind turbine generator set includes a wind wheel, a main shaft, a gearbox, a high-speed shaft and a generator. During the actual operation of the transmission chain, the wind wheel drives the gearbox through the main shaft, and the gearbox drives the generator through the high-speed rotating shaft.

[0091] A global speed characteristic matrix determination module 2 is configured to determine a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of the various parts at a plurality of predetermined time points within a predetermined time period;

[0092] a transmission logic characteristic matrix determination module 3, configured to construct a transmission logic matrix of each part of the transmission chain, and determine a transmission logic characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix;

[0093] A graph data encoding module 4 is configured to pass the global speed characteristic matrix and the transmission logic characteristic matrix through a graph neural network to obtain a transmission topology global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed;

[0094] It should be noted that the multiple speed feature vectors of each component are used as feature representations of the nodes, and the transmission logic feature matrix is used as the feature representation of the edges between nodes. The global speed feature matrix obtained by two-dimensionally arranging the multiple speed feature vectors and the transmission logic feature matrix are passed through a graph neural network to obtain a transmission topology global speed feature matrix. Specifically, the graph neural network uses learnable neural network parameters to perform graph structure data encoding on the global speed feature matrix and the transmission logic feature matrix to obtain the transmission topology global speed feature matrix that contains irregular transmission logic topology features and speed feature information of each component.

[0095] A characteristic distribution correction module 5 is used to perform characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix;

[0096] It should be noted that when the transmission topology global speed feature matrix is passed through a classifier to obtain a classification result, since each transmission topology global speed feature vector in the transmission topology global speed feature matrix is a topological expression of the multi-scale correlation characteristics of the time-series speed of each part of the transmission chain, and each part of the transmission chain has a certain degree of independence in terms of transmission logic and speed time-series characteristics, this results in poor consistency in the information expression between the various transmission topology global speed feature vectors. Directly classifying the transmission topology global speed feature matrix through a classifier may result in inductive deviations between the various transmission topology global speed feature vectors. Therefore, for each transmission topology global speed feature vector, its wavelet-like function family energy aggregation factor is calculated.

[0097] The fault diagnosis result generating module 6 is used to pass the corrected transmission topology global speed characteristic matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

[0098] That is, the transmission topology global speed characteristic matrix is used as a classification characteristic matrix and is classified by a classifier to obtain a classification result indicating whether a fault exists in the wind turbine's transmission chain. In this way, the wind turbine's transmission chain fault can be accurately diagnosed and detected.

[0099] It should be noted that the fault diagnosis result generating module 6 uses the classifier to calculate the formula O=softmax{(W c , B c )|Project(M)}, the corrected transmission topology global speed feature matrix is processed to obtain a classification result; wherein, Project(M) represents the projection of the corrected transmission topology global speed feature matrix into a vector, W cis the weight matrix of the fully connected layer, B c Represents the bias vector of the fully connected layer.

[0100] That is, the classifier first uses a fully connected layer to fully connect the global speed feature matrix of the corrected transmission topology to fully utilize the information of each position in the global speed feature matrix of the corrected transmission topology to reduce the dimension of the global speed feature matrix of the corrected transmission topology to a one-dimensional classification feature vector; then, the Softmax function value of the one-dimensional classification feature vector is calculated, that is, the probability value of the classification feature vector belonging to each classification label. In the embodiment of the present application, the classification labels include the presence of a fault in the transmission chain of the wind turbine generator set (the first label) and the absence of a fault in the transmission chain of the wind turbine generator set (the second label). Finally, the label corresponding to the larger probability value is used as the classification result.

[0101] In the embodiment of the present disclosure, Figure 3 As shown, the global speed characteristic matrix determination module 2 includes:

[0102] The component speed feature extraction and encoding submodule 2-1 is used to arrange the speed values of each component at multiple predetermined time points within a predetermined time period according to the time dimension as a speed input vector, and then pass it through the multi-scale neighborhood feature extraction module to obtain a speed feature vector corresponding to each component;

[0103] It should be noted that, considering that the speed values of each part of the transmission chain of the wind turbine generator set have different periodic regular distributions under different time spans, in order to more accurately extract the temporal dynamic feature information of the speed values of each part, in the technical solution of the present application, the speed values of each part at multiple predetermined time points within a predetermined time period are further arranged according to the time dimension as speed input vectors, and then a multi-scale neighborhood feature extraction module is used to obtain multiple speed feature vectors. In other words, the speed input vectors of each part of the wind turbine generator set are encoded by the multi-scale neighborhood feature extraction module to extract the dynamic multi-scale neighborhood correlation features of the speed values of each part under different time spans, thereby obtaining multiple speed feature vectors.

[0104] A matrixing submodule 2-2 is configured to perform a two-dimensional arrangement on the plurality of speed characteristic vectors to obtain a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed;

[0105] It should be noted that the multiple speed feature vectors represent the dynamic multi-scale neighborhood correlation characteristics of the speed values of the various parts under different time spans, but the fault diagnosis of the transmission chain of the wind turbine generator set should be analyzed on a global basis. Therefore, the multiple speed feature vectors are arranged in two dimensions according to the sample dimensions in the transmission chain to obtain a global speed feature matrix, that is, the high-dimensional implicit features of the speed of each part of the transmission chain of the wind turbine generator set are integrated into a feature matrix according to the sample dimensions in the transmission chain.

[0106] Among them, Figure 4 As shown, the component speed feature extraction encoding submodule 2-1 includes:

[0107] a first-scale speed feature extraction unit 2-1-1, configured to pass the speed input vector through a first one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale speed feature vector, wherein the first one-dimensional convolution layer performs one-dimensional convolution processing on the speed input vector using a one-dimensional convolution kernel having a first length;

[0108] It should be noted that the first scale speed feature extraction unit 2-1-1 uses the first convolution layer of the multi-scale neighborhood feature extraction module to calculate the speed feature of the first scale speed feature extraction unit 2-1-1 using the formula Perform one-dimensional convolution encoding on the speed input vector to obtain the first-scale neighborhood speed feature vector; where a is the width of the first convolution kernel in the x-direction, F is the first convolution kernel parameter vector, G is the local vector matrix operated on by the first convolution kernel function, W is the size of the first convolution kernel, and X is the speed input vector.

[0109] a second-scale speed feature extraction unit 2-1-2, configured to pass the speed input vector through the second one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale speed feature vector, wherein the second one-dimensional convolution layer performs one-dimensional convolution processing on the speed input vector using a one-dimensional convolution kernel having a second length;

[0110] It should be noted that the second-scale speed feature extraction unit 2-1-2 uses the second convolution layer of the multi-scale neighborhood feature extraction module to calculate the speed feature of the multi-scale neighborhood feature extraction module using the formula One-dimensional convolution encoding is performed on the speed input vector to obtain the second-scale neighborhood speed feature vector; wherein b is the width of the second convolution kernel in the x direction, f is the second convolution kernel parameter vector, g is the local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents the speed input vector.

[0111] The multi-scale rotational speed feature aggregation unit 2-1-3 is configured to concatenate the first-scale rotational speed feature vector and the second-scale rotational speed feature vector to obtain the rotational speed feature vector corresponding to each part.

[0112] In the embodiment of the present disclosure, Figure 5 As shown, the transmission logic characteristic matrix determination module 3 includes:

[0113] a transmission logic topology construction submodule 3-1, configured to construct a transmission logic matrix for each component of the transmission chain, wherein the eigenvalue of each non-diagonal position in the transmission logic matrix is used to indicate whether a transmission connection exists between the corresponding two components;

[0114] It should be noted that, considering that there are transmission-related characteristics between the various parts of the transmission chain, for example, the wind wheel drives the gearbox through the main shaft, and the gearbox drives the generator through the high-speed shaft. Therefore, if the transmission logic topology characteristics of each part of the transmission chain can be used to optimize the speed characteristic expression of each part, it will obviously be able to improve the accuracy of fault diagnosis. Therefore, in the technical solution of the present application, a transmission logic matrix of each part of the transmission chain is further constructed, wherein the characteristic values of each position on the non-diagonal position in the transmission logic matrix are used to indicate whether there is a transmission connection between the corresponding two components.

[0115] The transmission topology feature extraction submodule 3-2 is used to obtain a transmission logic feature matrix by passing the transmission logic matrix through a convolutional neural network model as a feature extractor; that is, the transmission logic matrix is subjected to feature mining by using a convolutional neural network model as a feature extractor that has excellent performance in extracting implicit correlation features, so as to extract the correlation features of each position in the transmission logic matrix, that is, the transmission logic topology features between the various parts of the transmission chain, thereby obtaining a transmission logic feature matrix.

[0116] The step of passing the transmission logic matrix through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix includes:

[0117] Using the convolution units of each layer of the convolutional neural network model to perform convolution processing based on a two-dimensional convolution kernel on the input data to obtain a convolution feature map;

[0118] Using the pooling units of each layer of the convolutional neural network model to perform pooling processing on the convolution feature map along the channel dimension to obtain a pooled feature map; and

[0119] Using the activation units of each layer of the convolutional neural network model to perform nonlinear activation on the feature values at each position in the pooling feature map to obtain an activation feature map;

[0120] Among them, the output of the last layer of the convolutional neural network model is the transmission logic feature matrix.

[0121] It should be noted that, in the transmission logic matrix, if there is a transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is one; if there is no transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is zero.

[0122] In the embodiment of the present disclosure, Figure 6 As shown, the characteristic distribution correction module 5 includes:

[0123] The characteristic vector correction factor calculation unit 5-1 is used to calculate the characteristic vector correction factor according to the formula Calculate the energy aggregation factor of the wavelet-like function family of each transmission topology global speed characteristic vector in the transmission topology global speed characteristic matrix respectively; where v i represents the eigenvalue of each position of the global speed eigenvector of each transmission topology, r represents the energy aggregation factor of the wavelet-like function family of the global speed eigenvector of each transmission topology, L is the length of the global speed eigenvector of each transmission topology, and log represents the logarithmic function value with base 2; that is, considering that for high-dimensional manifolds, since information representation tends to be concentrated on high-frequency components, information tends to be distributed on the edge of the manifold, the wavelet-like function family serves as a separable transformation for separating the edges on the feature distribution dimension, through which the latent state of the high-dimensional feature can be converted into frequency components, and the amount of information can be expressed in a wavelet-like energy manner.

[0124] The weighted correction unit 5-2 is used to use the energy aggregation factor of the wavelet-like function family of each transmission topology global speed characteristic vector as the weighted weight, and perform weighted optimization on each transmission topology global speed characteristic vector to obtain the corrected transmission topology global speed characteristic matrix.

[0125] It should be noted that, by using the energy aggregation factor of the wavelet-like function family of each transmission topology global speed feature vector in the multiple transmission topology global speed feature vectors obtained by expanding the transmission topology global speed feature matrix as a weighting coefficient to weight each transmission topology global speed feature vector, and then classifying the transmission topology global speed feature matrix through a classifier, it is possible to improve the degree of information aggregation of each transmission topology global speed feature vector in the transmission topology global speed feature matrix within its feature expression space, thereby improving the consistency of information expression between each transmission topology global speed feature vector, thereby avoiding inductive bias and improving classification accuracy. In this way, the transmission chain of the wind turbine generator set can be accurately monitored in real time, so as to automatically generate a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine.

[0126] In the embodiment of the present disclosure, Figure 7 As shown, the system further includes an early warning device 7. When the classification result indicates that a certain part of the transmission chain has failed, a fault early warning reminder is automatically generated to notify the staff to perform timely maintenance, thereby ensuring the safe operation of the fan.

[0127] To sum up, the present embodiment proposes a transmission chain fault diagnosis system for a wind turbine generator set, which adopts deep learning-based artificial intelligence monitoring technology to extract the multi-scale neighborhood correlation features of each part of the transmission chain of the wind turbine generator set in time series, and further constructs a transmission logic topology matrix between the various components, and uses the transmission logic correlation features between the various components of the transmission chain to assist in the transmission chain fault diagnosis of the wind turbine generator set, thereby improving the accuracy of fault diagnosis. In this way, the transmission chain of the wind turbine generator set can be accurately monitored in real time by intelligent means, so as to automatically generate a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine.

[0128] Exemplary methods:

[0129] Figure 8 FIG. 1 is a flow chart of a method for diagnosing transmission chain faults of a wind turbine generator set according to one embodiment of the present application. Figure 8 As shown, the method includes:

[0130] Step 1: collecting the rotational speed values of each part of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period;

[0131] Step 2: determining a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of the various parts at a plurality of predetermined time points within a predetermined time period;

[0132] Step 3: constructing a transmission logic matrix of each part of the transmission chain, and determining a transmission logic characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix;

[0133] Step 4: The global speed characteristic matrix and the transmission logic characteristic matrix are passed through a graph neural network to obtain a transmission topology global speed characteristic matrix corresponding to the wind turbine to be diagnosed;

[0134] Step 5: performing characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix;

[0135] Step 6: Pass the corrected transmission topology global speed feature matrix through a classifier to obtain a classification result, which is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

[0136] Wherein, the step 5 specifically includes:

[0137] Step 5-1: Using the formula Calculate the energy aggregation factor of the wavelet-like function family of each transmission topology global speed characteristic vector in the transmission topology global speed characteristic matrix respectively; where v i represents the eigenvalue of each position of the global speed eigenvector of each transmission topology, r represents the energy aggregation factor of the wavelet-like function family of the global speed eigenvector of each transmission topology, L is the length of the global speed eigenvector of each transmission topology, and log represents the logarithmic function value with base 2;

[0138] Step 5-2: Using the energy aggregation factor of the wavelet-like function family of each transmission topology global speed feature vector as a weighted weight, perform weighted optimization on each transmission topology global speed feature vector to obtain the corrected transmission topology global speed feature matrix.

[0139] like Figure 9 As shown, the specific steps of this application include:

[0140] S110, obtaining rotational speed values of various parts of a transmission chain of a wind turbine generator set at a plurality of predetermined time points within a predetermined time period;

[0141] S120, arranging the speed values of the respective parts at a plurality of predetermined time points within a predetermined time period according to the time dimension as speed input vectors, and then performing the vector extraction through a multi-scale neighborhood feature extraction module to obtain a plurality of speed feature vectors;

[0142] S130, constructing a transmission logic matrix for each part of the transmission chain, wherein the eigenvalue of each non-diagonal position in the transmission logic matrix is used to indicate whether there is a transmission connection between the corresponding two parts;

[0143] S140, passing the transmission logic matrix through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix;

[0144] S150, performing two-dimensional arrangement on the plurality of speed feature vectors to obtain a global speed feature matrix;

[0145] S160, applying the global speed characteristic matrix and the transmission logic characteristic matrix to a graph neural network to obtain a transmission topology global speed characteristic matrix;

[0146] S170, performing characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix;

[0147] S180: Pass the corrected transmission topology global speed characteristic matrix through a classifier to obtain a classification result, where the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

[0148] like Figure 10 FIG. 1 is a schematic diagram of the structure of a transmission chain fault diagnosis method for a wind turbine generator set according to an embodiment of the present application. Figure 10 As shown, in the architecture of the wind turbine transmission chain fault diagnosis method of the embodiment of the present application, first, the speed values of each component of the wind turbine transmission chain at multiple predetermined time points within a predetermined time period are obtained. The speed values of each component at multiple predetermined time points within the predetermined time period are arranged according to the time dimension as speed input vectors, which are then passed through a multi-scale neighborhood feature extraction module to obtain multiple speed feature vectors. Then, the multiple speed feature vectors are arranged in two dimensions to obtain a global speed feature matrix. Simultaneously, a transmission logic matrix is constructed for each component of the transmission chain, and the transmission logic matrix is passed through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix. Next, the global speed feature matrix and the transmission logic feature matrix are passed through a graph neural network to obtain a transmission topology global speed feature matrix. The transmission topology global speed feature matrix is then subjected to feature distribution correction to obtain a corrected transmission topology global speed feature matrix. Finally, the corrected transmission topology global speed feature matrix is passed through a classifier to obtain a classification result, which is used to indicate whether the wind turbine transmission chain has a fault.

[0149] In a specific embodiment of the present application, the speed values of each part at multiple predetermined time points within a predetermined time period are arranged as speed input vectors according to the time dimension and then passed through a multi-scale neighborhood feature extraction module to obtain multiple speed feature vectors, including: passing the speed input vector through a first one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale speed feature vector, wherein the first one-dimensional convolution layer uses a one-dimensional convolution kernel with a first length to perform one-dimensional convolution processing on the speed input vector; passing the speed input vector through a second one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale speed feature vector, wherein the second one-dimensional convolution layer uses a one-dimensional convolution kernel with a second length to perform one-dimensional convolution processing on the speed input vector; and cascading the first-scale speed feature vector and the second-scale speed feature vector to obtain the speed feature vector corresponding to each part.

[0150] In a specific embodiment of the present application, the step of passing the speed input vector through the first one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale speed feature vector includes: using the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the speed input vector using the following formula to obtain the first-scale neighborhood speed feature vector;

[0151] Wherein, the formula is:

[0152]

[0153] Where a is the width of the first convolution kernel in the x direction, F is the first convolution kernel parameter vector, G is the local vector matrix operated with the first convolution kernel function, W is the size of the first convolution kernel, and X is the speed input vector.

[0154] In a specific embodiment of the present application, the step of passing the speed input vector through the second one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale speed feature vector includes: using the second convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the speed input vector using the following formula to obtain the second-scale neighborhood speed feature vector;

[0155] Wherein, the formula is:

[0156]

[0157] Where b is the width of the second convolution kernel in the x direction, f is the second convolution kernel parameter vector, g is the local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents the speed input vector.

[0158] In a specific embodiment of the present application, in the transmission logic matrix, if there is a transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is one; if there is no transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is zero.

[0159] In a specific embodiment of the present application, the transmission logic matrix is passed through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix, including: each layer of the convolutional neural network model performs the following on the input data in the forward pass of the layer: using the convolution units of each layer of the convolutional neural network model to perform convolution processing based on a two-dimensional convolution kernel on the input data to obtain a convolution feature map; using the pooling units of each layer of the convolutional neural network model to perform pooling processing on the convolution feature map along the channel dimension to obtain a pooling feature map; and using the activation units of each layer of the convolutional neural network model to perform nonlinear activation on the eigenvalues of each position in the pooling feature map to obtain an activation feature map; wherein, the output of the last layer of the convolutional neural network model is the transmission logic feature matrix.

[0160] In a specific embodiment of the present application, performing characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix includes: calculating the wavelet-like function family energy aggregation factor of each transmission topology global speed characteristic vector in a plurality of transmission topology global speed characteristic vectors obtained by expanding the transmission topology global speed characteristic matrix using the following formula;

[0161] Wherein, the formula is:

[0162]

[0163] where v i Representing the eigenvalues of each position of each transmission topology global speed feature vector, r represents the wavelet-like function family energy aggregation factor of each transmission topology global speed feature vector, L is the length of each transmission topology global speed feature vector, and log represents the logarithmic function value with base 2; and, using the wavelet-like function family energy aggregation factor of each transmission topology global speed feature vector in the multiple transmission topology global speed feature vectors obtained by expanding the transmission topology global speed feature matrix as a weighted weight, weighted optimization is performed on each transmission topology global speed feature vector in the multiple transmission topology global speed feature vectors obtained by expanding the transmission topology global speed feature matrix to obtain the corrected transmission topology global speed feature matrix.

[0164] In a specific embodiment of the present application, the step of passing the corrected transmission topology global speed characteristic matrix through a classifier to obtain a classification result includes: using the classifier to process the corrected transmission topology global speed characteristic matrix using the following formula to obtain a classification result;

[0165] Wherein, the formula is: O=softmax{(W c , B c)|Project(M)}, where Project(M) represents the projection of the corrected transmission topology global speed characteristic matrix into a vector, W c is the weight matrix of the fully connected layer, B c Represents the bias vector of the fully connected layer.

[0166] To sum up, the transmission chain fault diagnosis method of a wind turbine generator set proposed in this embodiment can accurately perform real-time intelligent monitoring of the transmission chain of the wind turbine generator set, so as to automatically generate a fault warning when a fault is detected in a certain part of the transmission chain, thereby ensuring the safe operation of the wind turbine and improving the accuracy of fault diagnosis.

[0167] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0168] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed out of the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application belong.

[0169] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A transmission chain fault diagnosis system for a wind turbine generator set, characterized in that: include: A data acquisition module, used to collect the rotational speed values of various parts of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period; A global speed characteristic matrix determination module is used to determine the global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of the various parts at multiple predetermined time points within a predetermined time period; a transmission logic characteristic matrix determination module, configured to construct a transmission logic matrix of each part of the transmission chain, and determine a transmission logic characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix; A graph data encoding module, configured to pass the global speed characteristic matrix and the transmission logic characteristic matrix through a graph neural network to obtain a transmission topology global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed; a characteristic distribution correction module, configured to perform characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix; The fault diagnosis result generating module is used to pass the corrected transmission topology global speed characteristic matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

2. The transmission chain fault diagnosis system according to claim 1, characterized in that: The global speed characteristic matrix determination module includes: A component speed feature extraction and encoding submodule is used to arrange the speed values of each component at multiple predetermined time points within a predetermined time period according to the time dimension as a speed input vector and then pass it through a multi-scale neighborhood feature extraction module to obtain a speed feature vector corresponding to each component; A matrixing submodule, configured to perform a two-dimensional arrangement on the plurality of speed characteristic vectors to obtain a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed; The component speed feature extraction and encoding submodule includes: a first-scale speed feature extraction unit, configured to pass the speed input vector through a first one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale speed feature vector, wherein the first one-dimensional convolution layer performs one-dimensional convolution processing on the speed input vector using a one-dimensional convolution kernel having a first length; a second-scale speed feature extraction unit, configured to pass the speed input vector through a second one-dimensional convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale speed feature vector, wherein the second one-dimensional convolution layer performs one-dimensional convolution processing on the speed input vector using a one-dimensional convolution kernel having a second length; The multi-scale rotational speed feature aggregation unit is configured to concatenate the first-scale rotational speed feature vector and the second-scale rotational speed feature vector to obtain the rotational speed feature vector corresponding to each part.

3. The transmission chain fault diagnosis system according to claim 2, characterized in that: The first scale speed feature extraction unit uses the first convolution layer of the multi-scale neighborhood feature extraction module to calculate the speed of rotation using the formula Perform one-dimensional convolution encoding on the speed input vector to obtain the first-scale neighborhood speed feature vector; wherein a is the width of the first convolution kernel in the x direction, F is the first convolution kernel parameter vector, G is the local vector matrix operated with the first convolution kernel function, w is the size of the first convolution kernel, and X is the speed input vector.

4. The transmission chain fault diagnosis system according to claim 2, characterized in that: The second scale speed feature extraction unit uses the second convolution layer of the multi-scale neighborhood feature extraction module to calculate the speed feature of the multi-scale neighborhood feature extraction module according to the formula One-dimensional convolution encoding is performed on the speed input vector to obtain the second-scale neighborhood speed feature vector; wherein b is the width of the second convolution kernel in the x direction, f is the second convolution kernel parameter vector, g is the local vector matrix operated with the convolution kernel function, m is the size of the second convolution kernel, and X represents the speed input vector.

5. The transmission chain fault diagnosis system according to claim 1, characterized in that: The transmission logic characteristic matrix determination module includes: a transmission logic topology construction submodule, configured to construct a transmission logic matrix for each part of the transmission chain, wherein the eigenvalue of each non-diagonal position in the transmission logic matrix is used to indicate whether a transmission connection exists between the corresponding two parts; a transmission topology feature extraction submodule, configured to pass the transmission logic matrix through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix; The step of passing the transmission logic matrix through a convolutional neural network model as a feature extractor to obtain a transmission logic feature matrix includes: Using the convolution units of each layer of the convolutional neural network model to perform convolution processing based on a two-dimensional convolution kernel on the input data to obtain a convolution feature map; Using the pooling units of each layer of the convolutional neural network model to perform pooling processing on the convolution feature map along the channel dimension to obtain a pooled feature map; and Using the activation units of each layer of the convolutional neural network model to perform nonlinear activation on the feature values at each position in the pooling feature map to obtain an activation feature map; Among them, the output of the last layer of the convolutional neural network model is the transmission logic feature matrix.

6. The transmission chain fault diagnosis system according to claim 5, characterized in that: In the transmission logic matrix, if there is a transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is one; if there is no transmission connection relationship between the two components, the eigenvalue of the corresponding position in the transmission logic matrix is zero.

7. The transmission chain fault diagnosis system according to claim 1, characterized in that: The feature distribution correction module includes: Eigenvector correction factor calculation unit, used to calculate the Calculate the energy aggregation factor of the wavelet-like function family of each transmission topology global speed characteristic vector in the transmission topology global speed characteristic matrix respectively; where v i represents the eigenvalue of each position of the global speed eigenvector of each transmission topology, r represents the energy aggregation factor of the wavelet-like function family of the global speed eigenvector of each transmission topology, L is the length of the global speed eigenvector of each transmission topology, and log represents the logarithmic function value with base 2; A weighted correction unit is used to use the energy aggregation factor of the wavelet-like function family of each transmission topology global speed characteristic vector as a weighted weight, and perform weighted optimization on each transmission topology global speed characteristic vector to obtain the corrected transmission topology global speed characteristic matrix.

8. The transmission chain fault diagnosis system according to claim 1, characterized in that: The fault diagnosis result generation module uses the classifier to calculate the fault diagnosis result using the formula O=softmax{(W c , B c )|Project(M)}, the corrected transmission topology global speed feature matrix is processed to obtain a classification result; wherein, Project(M) represents the projection of the corrected transmission topology global speed feature matrix into a vector, W c is the weight matrix of the fully connected layer, B c Represents the bias vector of the fully connected layer.

9. The transmission chain fault diagnosis system according to claim 1, characterized in that: The system also includes an early warning device; The early warning device is used to automatically generate a fault early warning reminder when the classification result shows that a certain part of the transmission chain has failed.

10. A method for diagnosing transmission chain faults of a wind turbine generator set, characterized in that: The method comprises: Collecting the rotational speed values of various parts of the transmission chain of the wind turbine generator set to be diagnosed at multiple predetermined time points within a predetermined time period; Determining a global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the speed values of the various parts at a plurality of predetermined time points within a predetermined time period; Constructing a transmission logic matrix of each part of the transmission chain, and determining a transmission logic characteristic matrix corresponding to the wind turbine generator set to be diagnosed based on the transmission logic matrix; The global speed characteristic matrix and the transmission logic characteristic matrix are passed through a graph neural network to obtain a transmission topology global speed characteristic matrix corresponding to the wind turbine generator set to be diagnosed; Performing characteristic distribution correction on the transmission topology global speed characteristic matrix to obtain a corrected transmission topology global speed characteristic matrix; The corrected transmission topology global speed characteristic matrix is passed through a classifier to obtain a classification result, and the classification result is used to indicate whether there is a fault in the transmission chain of the wind turbine generator set.

Images