Wind turbine bearing lubrication system fault diagnosis system and method

Abstract

Disclosed is a fault diagnosis system and method for a bearing lubrication system of a wind turbine generator set, which passes wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period through a first convolutional neural network as a filter and performs correction to obtain a corrected linkage feature vector, and performs deep correlation feature extraction on the speed values ​​of the bearings at multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector, then calculates the responsiveness estimate of the multi-scale speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix, and finally passes the classification feature matrix through a classifier to obtain a classification result for indicating whether the fault risk of the bearing lubrication system exceeds a predetermined threshold. In this way, the bearing lubrication system can be given a fault warning, and the control strategy of the bearing lubrication system can be adjusted in a timely manner.

Description

Technical Field

[0001] The present application relates to the technical field of intelligent fault diagnosis, and more specifically, to a bearing lubrication system fault diagnosis system and method for a wind turbine generator set. Background Art

[0002] Wind turbines use wind turbines to convert wind power into electricity. The core component that connects the wind turbine and the generator is the bearing. As a rotating component, the bearing has high lubrication requirements and needs to be regularly filled with grease. At present, automatic grease pumps are mainly used to automatically lubricate the bearings, but this simple lubrication method is not very reliable, especially in winter when the temperature is low. It cannot effectively grease the bearings, which affects the lubrication of the bearing components, accelerates the wear of the bearing components, and reduces the reliability of the wind turbine.

[0003] Several major fan manufacturers in my country have experienced grease blockage problems. Grease is blocked in pipes and valve bodies, causing poor bearing lubrication and even failure. Therefore, it is very important to monitor the operating status of the bearing online to provide early warning of failures in the bearing lubrication system, so that the control strategy of the bearing lubrication system can be adjusted in time, for example, by increasing the temperature of the grease distributor.

[0004] Therefore, a fault diagnosis scheme for the bearing lubrication system of a wind turbine generator system is desired. Summary of the invention

[0005] In order to solve the above technical problems, the present application is proposed. An embodiment of the present application provides a bearing lubrication system fault diagnosis system and method for a wind turbine generator set, which passes the wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period through a first convolutional neural network as a filter and performs correction to obtain a corrected linkage feature vector, and performs deep correlation feature extraction on the speed values ​​of the bearings at multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector, and then calculates the responsiveness estimate of the multi-scale speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix, and finally passes the classification feature matrix through a classifier to obtain a classification result for indicating whether the failure risk of the bearing lubrication system exceeds a predetermined threshold. In this way, the bearing lubrication system can be given a fault warning, and the control strategy of the bearing lubrication system can be adjusted in time.

[0006] According to one aspect of the present application, a bearing lubrication system fault diagnosis system for a wind turbine generator set is provided, comprising:

[0007] An external excitation data acquisition module is used to obtain wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period;

[0008] an external excitation data structuring module, for arranging the wind speed values ​​and the pitch angles at a plurality of predetermined time points within the predetermined time period into a wind speed input vector and a pitch angle input vector according to the time dimension, respectively, and calculating the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage association matrix;

[0009] An external stimulus linkage feature extraction module, used for passing the linkage association matrix through a first convolutional neural network as a filter to obtain a linkage feature vector;

[0010] A feature correction module, used for correcting the feature value of each position in the linkage feature vector based on the second norm of the linkage feature vector to obtain a corrected linkage feature vector;

[0011] A bearing speed acquisition module, used to obtain the speed values ​​of the bearing at the plurality of predetermined time points;

[0012] A bearing speed feature extraction module, used for obtaining a multi-scale speed feature vector by passing the speed values ​​of the bearing at the plurality of predetermined time points through a multi-scale neighborhood feature extraction module;

[0013] a responsiveness estimation module, configured to calculate a responsiveness estimation of the multi-scale rotational speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix; and

[0014] The fault diagnosis result generating module is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0015] In the above-mentioned fault diagnosis system for the bearing lubrication system of the wind turbine generator set,

[0016] According to another aspect of the present application, a method for diagnosing a fault in a bearing lubrication system of a wind turbine generator set is provided, comprising:

[0017] Obtain wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period;

[0018] After arranging the wind speed values ​​and the pitch angles at a plurality of predetermined time points within the predetermined time period into a wind speed input vector and a pitch angle input vector according to the time dimension, respectively, calculating the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage correlation matrix;

[0019] Passing the linkage association matrix through a first convolutional neural network as a filter to obtain a linkage feature vector;

[0020] Based on the binary norm of the linkage feature vector, correcting the feature value of each position in the linkage feature vector to obtain a corrected linkage feature vector;

[0021] Acquire the rotational speed values ​​of the bearing at the plurality of predetermined time points;

[0022] The speed values ​​of the bearings at the plurality of predetermined time points are passed through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector;

[0023] Calculating a responsiveness estimate of the multi-scale rotational speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix; and

[0024] The classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0025] Compared with the prior art, the present application provides a fault diagnosis system and method for the bearing lubrication system of a wind turbine generator set, which passes the wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period through a first convolutional neural network as a filter and performs correction to obtain a corrected linkage feature vector, and performs deep correlation feature extraction on the speed values ​​of the bearings at multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector, and then calculates the responsiveness estimate of the multi-scale speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix, and finally passes the classification feature matrix through a classifier to obtain a classification result for indicating whether the failure risk of the bearing lubrication system exceeds a predetermined threshold. In this way, the bearing lubrication system can be given a fault warning, and the control strategy of the bearing lubrication system can be adjusted in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0027] Figure 1 The diagram illustrates an application scenario of a bearing lubrication system fault diagnosis system for a wind turbine generator set according to an embodiment of the present application.

[0028] Figure 2 The diagram illustrates a block diagram of a bearing lubrication system fault diagnosis system for a wind turbine generator set according to an embodiment of the present application.

[0029] Figure 3 The diagram illustrates a block diagram of the feature correction module in the bearing lubrication system fault diagnosis system of a wind turbine generator set according to an embodiment of the present application.

[0030] Figure 4 The figure illustrates a block diagram of the bearing speed feature extraction module in the bearing lubrication system fault diagnosis system of a wind turbine generator set according to an embodiment of the present application.

[0031] Figure 5 The figure illustrates a flow chart of a method for diagnosing faults in a bearing lubrication system of a wind turbine generator set according to an embodiment of the present application.

[0032] Figure 6 The figure shows a schematic diagram of the architecture of a method for diagnosing faults in a bearing lubrication system of a wind turbine generator set according to an embodiment of the present application. DETAILED DESCRIPTION

[0033] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0034] Scenario Overview

[0035] As described above, considering that under normal circumstances, the coordinated configuration of wind speed and pitch angle will cause the bearing to operate at a predetermined speed, if one wants to perform online monitoring of the operating state of the bearing to provide fault warning for the bearing lubrication system, one can evaluate whether the ratio of the energy provided by the coordination of the wind speed and pitch angle converted into the bearing speed meets the predetermined requirements based on the responsiveness estimation between the coordinated driving characteristics of the wind speed and the pitch angle and the speed characteristics of the bearing, that is, whether the performance of the bearing lubrication system meets the predetermined requirements.

[0036] Specifically, in the technical solution of the present application, first, various sensors, such as an anemometer and a pitch angle meter, are used to collect wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period. Then, considering that under normal circumstances, the wind speed and the pitch angle will be configured in a coordinated manner so that the bearing runs at a predetermined speed, in order to more fully extract the coordinated driving characteristics of the wind speed and the pitch angle, a correlation matrix between the data of the two is further constructed. That is, the wind speed values ​​and pitch angles at multiple predetermined time points within the predetermined time period are arranged as wind speed input vectors and pitch angle input vectors according to the time dimension to integrate the wind speed and pitch angle information at each time point, and then the product between the transposed vector of the wind speed input vector and the pitch angle input vector is calculated to obtain a linkage correlation matrix.

[0037] Furthermore, since the convolutional neural network model has excellent performance in extracting implicit correlation features between data, in the technical solution of the present application, a first convolutional neural network is used as a filter to process the linkage correlation matrix to extract the implicit correlation feature information of each position in the linkage correlation matrix, that is, the collaborative driving characteristics of the wind speed and the pitch angle, thereby obtaining a linkage feature vector.

[0038] In particular, in the technical solution of the present application, when the linkage association matrix passes through the first convolutional neural network as a filter to obtain a linkage feature vector, since the convolution kernel of the first convolutional neural network acts as a filter to perform forward pixel-level propagation based on downsampling on the association features between the matrix values ​​of the linkage association matrix, the correlation of the eigenvalues ​​at each position of the linkage feature vector may become worse than the correlation between the values ​​at each position of the linkage association matrix, thereby affecting the feature expression effect of the linkage feature vector on the linkage association matrix.

[0039] Therefore, the linkage feature vector is forward propagated and guided to be corrected, namely:

[0040]

[0041] ∑ is the autocovariance matrix of vector V, that is, the value of each position of the matrix is ​​the variance between the eigenvalues ​​of every two positions of the linkage eigenvector V, and μ and σ are the global mean and variance of the linkage eigenvector V respectively.

[0042] That is, in view of the characteristic of filter-based convolutional neural network that performs forward pixel-level propagation of features based on downsampling, the long-range dependency between the eigenvalues ​​of the linkage feature vector V and the linkage association matrix in the width and height dimensions is effectively modeled through feature engineering guided by learnable normal sampling offset, and the correlation of the eigenvalues ​​of the linkage feature vector V is further repaired based on local and non-local neighborhoods to enhance the feature expression effect of the linkage feature vector on the linkage association matrix, thereby improving the classification accuracy.

[0043] Furthermore, in order to accurately evaluate whether the ratio of the energy provided by the combination of the wind speed and the pitch angle to the bearing speed meets the predetermined requirements, it is also necessary to extract the speed dynamic characteristics of the bearing speed value in the time dimension. That is, first, the speed values ​​of the bearing at the plurality of predetermined time points are obtained through a speed sensor. And, the speed values ​​of the bearing at the plurality of predetermined time points are arranged as a bearing speed input vector.

[0044] Then, considering that the bearing speed will present different mode states in different time periods, for example, the bearing speed will be fast at one time and slow at another time in different time periods, therefore, if you want to perform deep correlation feature extraction for the bearing speed in different time spans, further use the convolution layer of the multi-scale neighborhood feature extraction module with one-dimensional convolution kernels of different scales to perform one-dimensional convolution encoding on the bearing speed input vector respectively, and then cascade the obtained feature vectors corresponding to the two one-dimensional convolution kernels of different scales to obtain the multi-scale speed feature vector. It is worth mentioning that in this way, the multi-scale neighborhood correlation of the bearing speed features of different time spans is extracted, so that the output features include both smoothed features and original input features, avoiding information loss and improving the accuracy of subsequent classification. In other examples of the present application, the multi-scale neighborhood feature extraction module may also include a larger number of one-dimensional convolution layers, which use one-dimensional convolution kernels of different lengths to extract correlation features in the neighborhood of the bearing speed features of different scales, which is not limited to the present application.

[0045] It should be understood that, since the characteristic scales between the cooperative driving characteristics of the wind speed and the pitch angle and the rotational speed characteristics of the bearing are different, and since under normal circumstances, the cooperative configuration of the wind speed and the pitch angle will cause the bearing to operate at a predetermined rotational speed, the cooperative driving characteristics of the wind speed and the pitch angle can be regarded as responsive characteristics to the rotational speed characteristics of the bearing in the high-dimensional feature space. Therefore, in the technical solution of the present application, it is further evaluated based on the responsiveness estimation between the cooperative driving characteristics of the wind speed and the pitch angle and the rotational speed characteristics of the bearing to evaluate whether the ratio of the energy provided by the coordination of the wind speed and the pitch angle converted into the rotational speed of the bearing meets the predetermined requirements, that is, whether the performance of the bearing lubrication system meets the predetermined requirements. In this way, it is possible to further accurately judge whether the failure risk of the bearing lubrication system exceeds the predetermined threshold, so as to provide a fault warning for the bearing lubrication system, so as to timely adjust the control strategy of the bearing lubrication system, for example, to increase the temperature of the grease dispenser.

[0046] Based on this, the present application proposes a fault diagnosis system for a bearing lubrication system of a wind turbine generator set, which includes: an external excitation data acquisition module, used to obtain wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period; an external excitation data structuring module, used to arrange the wind speed values ​​and pitch angles at multiple predetermined time points within the predetermined time period into wind speed input vectors and pitch angle input vectors according to the time dimension, and then calculate the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage association matrix; an external excitation linkage feature extraction module, used to pass the linkage association matrix through a first convolutional neural network as a filter to obtain a linkage feature vector; a feature correction module, used to obtain a linkage feature vector based on the linkage feature vector The bi-norm of the linkage feature vector is used to correct the feature values ​​at each position in the linkage feature vector to obtain the corrected linkage feature vector; a bearing speed acquisition module is used to obtain the speed value of the bearing at the multiple predetermined time points; a bearing speed feature extraction module is used to pass the speed value of the bearing at the multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector; a responsiveness estimation module is used to calculate the responsiveness estimate of the multi-scale speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix; and a fault diagnosis result generation module is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0047] Figure 1 The figure shows an application scenario diagram of the bearing lubrication system fault diagnosis system of a wind turbine generator set according to an embodiment of the present application. Figure 1 As shown, in this application scenario, various sensors, such as an anemometer (e.g., Figure 1 C1) and a pitch angle measuring instrument (e.g., Figure 1 C2) in the diagram) collects data from wind turbines (e.g. Figure 1 F) wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period, and wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period, as well as ... Figure 1 C3) as shown in the figure obtains the rotation speed values ​​of the bearings of the wind turbine generator set at the plurality of predetermined time points; then, the obtained wind speed value, pitch angle and rotation speed value of the bearing are input to a server (for example, as shown in the figure) in which a fault diagnosis system for the bearing lubrication system of the wind turbine generator set is deployed. Figure 1 In S) shown in the figure, the server processes the wind speed value, pitch angle and bearing speed value using the bearing lubrication system fault diagnosis algorithm of the wind turbine generator set to generate a classification result indicating whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0048] 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.

[0049] Exemplary Systems

[0050] Figure 2 The figure shows a block diagram of a bearing lubrication system fault diagnosis system for a wind turbine generator set according to an embodiment of the present application. Figure 2 As shown, according to the embodiment of the present application, the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set includes: an external excitation data acquisition module 110, which is used to obtain the wind speed value and the pitch angle at multiple predetermined time points within a predetermined time period; an external excitation data structuring module 120, which is used to arrange the wind speed value and the pitch angle at multiple predetermined time points within the predetermined time period into a wind speed input vector and a pitch angle input vector according to the time dimension, and then calculate the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage association matrix; an external excitation linkage feature extraction module 130, which is used to pass the linkage association matrix through a first convolutional neural network as a filter to obtain a linkage feature vector; a feature correction module 140, which is used to obtain a linkage feature vector based on the linkage feature vector The second norm of the quantity is used to correct the eigenvalues ​​of each position in the linkage feature vector to obtain the corrected linkage feature vector; a bearing speed acquisition module 150 is used to obtain the speed value of the bearing at the multiple predetermined time points; a bearing speed feature extraction module 160 is used to pass the speed value of the bearing at the multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector; a responsiveness estimation module 170 is used to calculate the responsiveness estimate of the multi-scale speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix; and a fault diagnosis result generation module 180 is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0051] Specifically, in the embodiment of the present application, the external excitation data acquisition module 110 is used to obtain the wind speed value and the pitch angle at multiple predetermined time points within a predetermined time period. It should be understood that the coordinated configuration of the wind speed and the pitch angle will cause the bearing to operate at a predetermined speed. Therefore, if the operating state of the bearing is to be monitored online to provide a fault warning for the bearing lubrication system, it is possible to evaluate whether the ratio of the energy provided by the coordination of the wind speed and the pitch angle to the bearing speed meets the predetermined requirements based on the responsiveness estimation between the coordinated driving characteristics of the wind speed and the pitch angle and the speed characteristics of the bearing, that is, whether the performance of the bearing lubrication system meets the predetermined requirements.

[0052] Therefore, in the technical solution of the present application, firstly, various sensors, such as an anemometer and a pitch angle meter, are used to respectively collect wind speed values ​​and pitch angles at a plurality of predetermined time points within a predetermined time period.

[0053] Specifically, in the embodiment of the present application, the external excitation data structuring module 120 is used to arrange the wind speed values ​​and the pitch angles at multiple predetermined time points within the predetermined time period into wind speed input vectors and pitch angle input vectors according to the time dimension, and then calculate the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage association matrix. As mentioned above, under normal circumstances, the wind speed and the pitch angle will be configured in a coordinated manner so that the bearing runs at a predetermined speed. Therefore, in order to more fully extract the coordinated driving characteristics of the wind speed and the pitch angle, a correlation matrix between the data of the two is further constructed.

[0054] That is, after collecting the wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period, the wind speed values ​​and pitch angles at multiple predetermined time points within the predetermined time period are arranged as wind speed input vectors and pitch angle input vectors according to the time dimension to integrate the wind speed and pitch angle information at each time point, and then the product between the transposed vector of the wind speed input vector and the pitch angle input vector is calculated to obtain a linkage association matrix.

[0055] Specifically, in the embodiment of the present application, the external excitation linkage feature extraction module 130 is used to pass the linkage association matrix through the first convolutional neural network as a filter to obtain a linkage feature vector. Since the convolutional neural network model has excellent performance in extracting implicit association features between data, therefore, in the technical solution of the present application, the first convolutional neural network as a filter is used to process the linkage association matrix to extract the implicit association feature information of each position in the linkage association matrix, that is, the collaborative driving characteristics of the wind speed and the pitch angle, so as to obtain a linkage feature vector.

[0056] Furthermore, in the technical solution of the present application, each layer of the first convolutional neural network serving as a filter performs the following steps in the forward pass of the layer: performing convolution processing on the input data to obtain a convolutional feature map; performing mean pooling based on a local feature matrix on the convolutional feature map to obtain a pooled feature map; and performing nonlinear activation on the pooled feature map to obtain an activated feature map; wherein the output of the last layer of the first convolutional neural network serving as a filter is the linkage feature vector, and the input of the first layer of the first convolutional neural network serving as a filter is the linkage association matrix.

[0057] That is, the linkage association matrix is ​​first input into the first layer of the first convolutional neural network serving as a filter, and then each layer of the first convolutional neural network serving as a filter performs convolution processing, mean pooling processing, and nonlinear activation processing on the input data in the forward pass of the layer, and finally the output of the last layer of the first convolutional neural network serving as a filter is the linkage feature vector.

[0058] Specifically, in the embodiment of the present application, the feature correction module 140 is used to correct the eigenvalues ​​of each position in the linkage feature vector based on the second norm of the linkage feature vector to obtain the corrected linkage feature vector. In particular, in the embodiment of the present application, since the eigenvalues ​​of each position of the linkage feature vector are obtained by global mean pooling of each feature matrix of the linkage association matrix, it is expected to improve the expression ability of the linkage feature vector to the linkage association matrix regardless of the dimension of the feature matrix.

[0059] More specifically, in the embodiments of the present application, Figure 3 FIG. 1 is a block diagram of the feature correction module in the fault diagnosis system of the bearing lubrication system of the wind turbine generator set according to an embodiment of the present application, as shown in FIG. Figure 3 As shown, the feature correction module 140 includes: an autocovariance matrix construction unit 210, which is used to calculate the variance between the eigenvalues ​​of each two positions in the linkage feature vector to obtain an autocovariance matrix; an autocorrelation unit 220, which is used to calculate the product between the linkage feature vector and the autocovariance matrix to obtain an autocorrelation feature vector; a first position difference unit 230, which is used to calculate the position difference between the linkage feature vector and the autocorrelation feature vector to obtain a decorrelation linkage feature vector; a first exponential operation unit 240, which is used to calculate the natural exponential function value with the eigenvalues ​​of each position in the decorrelation linkage feature vector as the power to obtain an exponential decorrelation. linkage feature vector; a second position-based difference unit 250, used to calculate the position-based difference between the linkage feature vector and the second norm of the linkage feature vector to obtain a linkage feature vector represented by a de-low rank method; a second exponential operation unit 260, used to calculate the natural exponential function value raised to the power of the eigenvalues ​​of each position in the linkage feature vector represented by a de-low rank method to obtain an exponential de-low rank method of representing the linkage feature vector; and a fusion correction unit 270, used to fuse the exponential decorrelation linkage feature vector and the exponential de-low rank method of representing the linkage feature vector based on the global mean and variance of the eigenvalue set of all positions in the linkage feature vector to obtain the corrected linkage feature vector.

[0060] In particular, in the technical solution of the present application, when the linkage association matrix passes through the first convolutional neural network as a filter to obtain a linkage feature vector, since the convolution kernel of the first convolutional neural network acts as a filter to perform forward pixel-level propagation based on downsampling on the association features between the matrix values ​​of the linkage association matrix, the correlation of the eigenvalues ​​at each position of the linkage feature vector may become worse than the correlation between the values ​​at each position of the linkage association matrix, thereby affecting the feature expression effect of the linkage feature vector on the linkage association matrix.

[0061] Therefore, the linkage feature vector is forward propagated and guided by correlation correction, that is:

[0062]

[0063] ∑ is the autocovariance matrix of vector V, that is, the value of each position of the matrix is ​​the variance between the eigenvalues ​​of every two positions of the linkage eigenvector V, and μ and σ are the global mean and variance of the linkage eigenvector V respectively.

[0064] That is, in view of the characteristic of filter-based convolutional neural network that performs forward pixel-level propagation of features based on downsampling, the long-range dependency between the eigenvalues ​​of the linkage feature vector V and the linkage association matrix in the width and height dimensions is effectively modeled through feature engineering guided by learnable normal sampling offset, and the correlation of the eigenvalues ​​of the linkage feature vector V is further repaired based on local and non-local neighborhoods to enhance the feature expression effect of the linkage feature vector on the linkage association matrix, thereby improving the classification accuracy.

[0065] Further, in an embodiment of the present application, the exponential decorrelation linkage feature vector and the exponential de-low-rank representation linkage feature vector are fused based on the global mean and variance of the eigenvalue set at all positions in the linkage feature vector using the following formula to obtain the corrected linkage feature vector; wherein the formula is:

[0066]

[0067] Among them, V1 represents the exponential decorrelation linkage eigenvector, V2 represents the exponential de-low-rank representation linkage eigenvector, V′ represents the corrected linkage eigenvector, μ and σ respectively represent the global mean and variance of the eigenvalue set of all positions in the linkage eigenvector, and ⊕ represents the summation by position.

[0068] Specifically, in the embodiment of the present application, the bearing speed acquisition module 150 is used to obtain the speed values ​​of the bearing at the multiple predetermined time points. It should be understood that in order to accurately evaluate whether the ratio of the energy provided by the combination of the wind speed and the pitch angle converted into the bearing speed meets the predetermined requirements, it is also necessary to extract the speed dynamic characteristics of the bearing speed value in the time dimension. That is, first, the speed values ​​of the bearing at the multiple predetermined time points are obtained by the speed sensor. And the speed values ​​of the bearing at the multiple predetermined time points are arranged as bearing speed input vectors.

[0069] Specifically, in the embodiment of the present application, the bearing speed feature extraction module 160 is used to pass the speed values ​​of the bearings at the plurality of predetermined time points through the multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector. Considering that the bearing speed may present different mode states in different time periods, for example, the bearing speed may be fast at one time and slow at another time in different time periods, if it is desired to perform deep correlation feature extraction for the bearing speeds in different time spans, the convolution layer with one-dimensional convolution kernels of different scales of the multi-scale neighborhood feature extraction module is further used to perform one-dimensional convolution encoding on the bearing speed input vector respectively, and then the obtained feature vectors corresponding to the two one-dimensional convolution kernels of different scales are cascaded to obtain the multi-scale speed feature vector.

[0070] It is worth mentioning that in this way, the multi-scale neighborhood association of the bearing speed features of different time spans is extracted, so that the output features include both the smoothed features and the original input features, avoiding information loss and improving the accuracy of subsequent classification. In other examples of the present application, the multi-scale neighborhood feature extraction module may also include a larger number of one-dimensional convolution layers, which use one-dimensional convolution kernels of different lengths to extract neighborhood association features of different scales of the bearing speed features, which is not limited to the present application.

[0071] More specifically, in the embodiments of the present application, Figure 4 FIG. 1 is a block diagram of the bearing speed feature extraction module in the bearing lubrication system fault diagnosis system of the wind turbine generator set according to an embodiment of the present application, such as Figure 4As shown, the bearing speed feature extraction module 160 includes: a bearing speed value quantization unit 310, which is used to arrange the speed values ​​of the bearings at the multiple predetermined time points into a bearing speed input vector; a first scale feature extraction unit 320, which is used to input the bearing speed input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale tension-related feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; a second scale feature extraction unit 330, which is used to input the bearing speed input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale tension-related feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and a cascade unit 340, which is used to cascade the first neighborhood scale tension-related feature vector and the second neighborhood scale tension-related feature vector to obtain the multi-scale speed feature vector.

[0072] Further, the first convolution layer of the multi-scale neighborhood feature extraction module is used to perform one-dimensional convolution encoding on the bearing speed input vector using the following formula to obtain the first neighborhood scale tension-related feature vector; wherein the formula is:

[0073]

[0074] Among them, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the convolution kernel function, w is the size of the first convolution kernel, and X represents the bearing speed input vector.

[0075] Furthermore, the second convolution layer of the multi-scale neighborhood feature extraction module is used to perform one-dimensional convolution encoding on the bearing speed input vector using the following formula to obtain the second neighborhood scale tension-related feature vector; wherein the formula is:

[0076]

[0077] Among them, b is the width of the first convolution kernel in the x direction, F(b) is the first convolution kernel parameter vector, G(xb) is the local vector matrix operated with the convolution kernel function, m is the size of the first convolution kernel, and X represents the bearing speed input vector.

[0078] Specifically, in the embodiment of the present application, the responsiveness estimation module 170 is used to calculate the responsiveness estimation of the multi-scale rotational speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix. It should be understood that, since the characteristic scales between the coordinated driving characteristics of the wind speed and the pitch angle and the rotational speed characteristics of the bearing are different, and since under normal circumstances, the coordinated configuration of the wind speed and the pitch angle will cause the bearing to operate at a predetermined speed, the coordinated driving characteristics of the wind speed and the pitch angle can be regarded as responsive characteristics to the rotational speed characteristics of the bearing in the high-dimensional feature space.

[0079] Accordingly, the responsiveness estimate of the multi-scale rotational speed feature vector relative to the corrected linkage feature vector is calculated by the following formula to obtain the classification feature matrix; wherein the formula is:

[0080] V s =M*V l

[0081] Where V s represents the multi-scale speed feature vector, V l represents the linkage feature vector after correction, and M represents the classification feature matrix.

[0082] It should be understood that in the technical solution of the present application, the ratio of the energy provided by the coordination of the wind speed and the pitch angle and converted into the bearing speed is further evaluated based on the responsiveness estimation between the coordinated driving characteristics of the wind speed and the pitch angle and the speed characteristics of the bearing to evaluate whether it meets the predetermined requirements, that is, whether the performance of the bearing lubrication system meets the predetermined requirements. In this way, it is possible to further accurately judge whether the failure risk of the bearing lubrication system exceeds the predetermined threshold, so as to provide a failure warning for the bearing lubrication system, so as to timely adjust the control strategy of the bearing lubrication system, for example, to increase the temperature of the grease dispenser.

[0083] Specifically, in the embodiment of the present application, the fault diagnosis result generating module 180 is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold. That is, the classification feature matrix is ​​input into a classification function to obtain a classification function value, wherein the classification function value is the classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0084] More specifically, the classifier processes the classification feature matrix according to the following formula to generate a classification result, wherein the formula is: softmax{(W n ,B n):…:(W1,B1)|Project(F)}, where Project(F) represents projecting the classification feature matrix into a vector, W1 to W n is the weight matrix of each fully connected layer, B1 to B n Represents the bias matrix of each fully connected layer.

[0085] In summary, the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set based on the embodiment of the present application is explained, which passes the wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period through the first convolutional neural network as a filter and performs correction to obtain a corrected linkage feature vector, and the speed values ​​of the bearings at multiple predetermined time points are subjected to deep correlation feature extraction through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector, and then, the responsiveness estimate of the multi-scale speed feature vector relative to the corrected linkage feature vector is calculated to obtain a classification feature matrix, and finally, the classification feature matrix is ​​passed through a classifier to obtain a classification result for indicating whether the failure risk of the bearing lubrication system exceeds a predetermined threshold. In this way, the bearing lubrication system can be given a fault warning, and the control strategy of the bearing lubrication system can be adjusted in a timely manner.

[0086] As described above, the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set according to the embodiment of the present application can be implemented in various terminal devices, such as a server for bearing lubrication system fault diagnosis of the wind turbine generator set. In one example, the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set according to the embodiment of the present application can be integrated into the terminal device as a software module and / or a hardware module. For example, the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set can also be one of the many hardware modules of the terminal device.

[0087] Alternatively, in another example, the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set and the terminal device may also be separate devices, and the bearing lubrication system fault diagnosis system 100 of the wind turbine generator set may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0088] Exemplary Methods

[0089] Figure 5 The figure shows a flow chart of a method for diagnosing a fault in a bearing lubrication system of a wind turbine generator set according to an embodiment of the present application. Figure 5As shown, according to the embodiment of the present application, the fault diagnosis method of the bearing lubrication system of the wind turbine generator set includes: S110, obtaining the wind speed value and the pitch angle at multiple predetermined time points within a predetermined time period; S120, arranging the wind speed value and the pitch angle at multiple predetermined time points within the predetermined time period into a wind speed input vector and a pitch angle input vector according to the time dimension, respectively, and calculating the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage association matrix; S130, passing the linkage association matrix through a first convolutional neural network as a filter to obtain a linkage feature vector; S140, based on the second convolutional neural network of the linkage feature vector norm, correcting the eigenvalues ​​at each position in the linkage feature vector to obtain a corrected linkage feature vector; S150, obtaining the speed values ​​of the bearings at the multiple predetermined time points; S160, passing the speed values ​​of the bearings at the multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector; S170, calculating the responsiveness estimate of the multi-scale speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix; and, S180, passing the classification feature matrix through a classifier to obtain a classification result, which is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0090] Figure 6 FIG. 1 is a schematic diagram of the structure of a method for diagnosing a fault in a bearing lubrication system of a wind turbine generator set according to an embodiment of the present application. Figure 6 As shown, in the network architecture of the bearing lubrication system fault diagnosis method of the wind turbine generator set, first, the wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period are obtained; then, the wind speed values ​​and pitch angles at multiple predetermined time points within the predetermined time period are arranged as wind speed input vectors and pitch angle input vectors according to the time dimension, respectively, and the product between the transposed vector of the wind speed input vector and the pitch angle input vector is calculated to obtain a linkage association matrix; then, the linkage association matrix is ​​passed through a first convolutional neural network as a filter to obtain a linkage feature vector; then, based on the binary norm of the linkage feature vector , correcting the characteristic values ​​of each position in the linkage characteristic vector to obtain a corrected linkage characteristic vector; then, obtaining the speed values ​​of the bearing at the multiple predetermined time points; then, passing the speed values ​​of the bearing at the multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed characteristic vector; then, calculating the responsiveness estimate of the multi-scale speed characteristic vector relative to the corrected linkage characteristic vector to obtain a classification characteristic matrix; and, finally, passing the classification characteristic matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

[0091] In one embodiment of the present application, in the above-mentioned method for diagnosing faults of the bearing lubrication system of a wind turbine generator set, the linkage association matrix is ​​passed through a first convolutional neural network as a filter to obtain a linkage feature vector, comprising: each layer of the first convolutional neural network as a filter performs the following steps in the forward pass of the layer: convolution processing on the input data to obtain a convolution feature map; mean pooling based on a local feature matrix on the convolution feature map to obtain a pooled feature map; and nonlinear activation on the pooled feature map to obtain an activation feature map; wherein the output of the last layer of the first convolutional neural network as a filter is the linkage feature vector, and the input of the first layer of the first convolutional neural network as a filter is the linkage association matrix.

[0092] In one embodiment of the present application, in the above-mentioned method for diagnosing faults in the bearing lubrication system of a wind turbine generator set, the eigenvalues ​​at each position in the linkage eigenvector are corrected based on the second norm of the linkage eigenvector to obtain a corrected linkage eigenvector, including: calculating the variance between the eigenvalues ​​at every two positions in the linkage eigenvector to obtain an autocovariance matrix; calculating the product between the linkage eigenvector and the autocovariance matrix to obtain an autocorrelation eigenvector; calculating the positional difference between the linkage eigenvector and the autocorrelation eigenvector to obtain a decorrelation linkage eigenvector; calculating the decorrelation linkage eigenvector based on the decorrelation linkage eigenvector. The eigenvalue of each position in the vector is raised to a natural exponential function value to obtain an exponential decorrelation linkage eigenvector; the positional difference between the linkage eigenvector and the second norm of the linkage eigenvector is calculated to obtain a de-low-rank representation linkage eigenvector; the natural exponential function value of the eigenvalue of each position in the de-low-rank representation linkage eigenvector is calculated to obtain an exponential de-low-rank representation linkage eigenvector; and the exponential decorrelation linkage eigenvector and the exponential de-low-rank representation linkage eigenvector are fused based on the global mean and variance of the eigenvalue set of all positions in the linkage eigenvector to obtain the corrected linkage eigenvector.

[0093] In one embodiment of the present application, in the above-mentioned bearing lubrication system fault diagnosis method of the wind turbine generator set, the exponential decorrelation linkage feature vector and the exponential removal of low-rank representation linkage feature vector are fused based on the global mean and variance of the eigenvalue set at all positions in the linkage feature vector to obtain the corrected linkage feature vector, including: based on the global mean and variance of the eigenvalue set at all positions in the linkage feature vector, the exponential decorrelation linkage feature vector and the exponential removal of low-rank representation linkage feature vector are fused according to the following formula to obtain the corrected linkage feature vector; wherein the formula is:

[0094]

[0095] Wherein, V1 represents the exponential decorrelation linkage eigenvector, V2 represents the exponential de-low-rank linkage eigenvector, V′ represents the corrected linkage eigenvector, μ and σ represent the global mean and variance of the eigenvalue set of all positions in the linkage eigenvector, respectively. Indicates sum by position.

[0096] In one embodiment of the present application, in the above-mentioned method for diagnosing faults in a bearing lubrication system of a wind turbine generator set, the step of passing the speed values ​​of the bearings at the multiple predetermined time points through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector includes: arranging the speed values ​​of the bearings at the multiple predetermined time points as a bearing speed input vector; inputting the bearing speed input vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale tension-associated feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; inputting the bearing speed input vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale tension-associated feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and cascading the first neighborhood scale tension-associated feature vector and the second neighborhood scale tension-associated feature vector to obtain the multi-scale speed feature vector.

[0097] In one embodiment of the present application, in the above-mentioned method for diagnosing faults in a bearing lubrication system of a wind turbine generator set, the step of inputting the bearing speed input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale tension-associated feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length, comprises: using the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the bearing speed input vector using the following formula to obtain the first neighborhood scale tension-associated feature vector; wherein the formula is:

[0098]

[0099] Among them, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the convolution kernel function, w is the size of the first convolution kernel, and X represents the bearing speed input vector.

[0100] In one embodiment of the present application, in the above-mentioned method for diagnosing faults in a bearing lubrication system of a wind turbine generator set, the inputting the bearing speed input vector into the second convolutional layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale tension-associated feature vector, wherein the second convolutional layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length, comprises: using the second convolutional layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the bearing speed input vector using the following formula to obtain the second neighborhood scale tension-associated feature vector; wherein the formula is:

[0101]

[0102] Among them, b is the width of the first convolution kernel in the x direction, F(b) is the first convolution kernel parameter vector, G(xb) is the local vector matrix operated with the convolution kernel function, m is the size of the first convolution kernel, and X represents the bearing speed input vector.

[0103] In one embodiment of the present application, in the above-mentioned method for diagnosing faults in the bearing lubrication system of a wind turbine generator set, the calculating of the responsiveness estimation of the multi-scale speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix comprises: calculating the responsiveness estimation of the multi-scale speed feature vector relative to the corrected linkage feature vector using the following formula to obtain the classification feature matrix; wherein the formula is:

[0104] V s =M*V l

[0105] Where V s represents the multi-scale speed feature vector, V l represents the linkage feature vector after correction, and M represents the classification feature matrix.

[0106] In one embodiment of the present application, in the above-mentioned method for diagnosing faults in a bearing lubrication system of a wind turbine generator set, the classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold, including: the classifier processes the classification feature matrix using the following formula to generate a classification result, wherein the formula is: softmax{(W n ,B n ):…:(W1,B1)|Project(F)}, where Project(F) represents projecting the classification feature matrix into a vector, W1 to W n is the weight matrix of each fully connected layer, and B1 to B represent the bias matrix of each fully connected layer.

[0107] Here, those skilled in the art will appreciate that the specific functions and operations of the above-mentioned method for diagnosing a fault in a bearing lubrication system of a wind turbine generator set have been described in detail in the above reference. Figures 1 to 4 The bearing lubrication system fault diagnosis system of the wind turbine generator set has been introduced in detail, and therefore, its repeated description will be omitted.

Claims

1. A bearing lubrication system fault diagnosis system for a wind turbine generator set, characterized in that: include: An external excitation data acquisition module is used to obtain wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period; an external excitation data structuring module, for arranging the wind speed values ​​and the pitch angles at a plurality of predetermined time points within the predetermined time period into a wind speed input vector and a pitch angle input vector according to the time dimension, respectively, and calculating the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage association matrix; An external stimulus linkage feature extraction module, used for passing the linkage association matrix through a first convolutional neural network as a filter to obtain a linkage feature vector; A feature correction module, used for correcting the feature value of each position in the linkage feature vector based on the second norm of the linkage feature vector to obtain a corrected linkage feature vector; A bearing speed acquisition module, used to obtain the speed values ​​of the bearing at the plurality of predetermined time points; A bearing speed feature extraction module, used for obtaining a multi-scale speed feature vector by passing the speed values ​​of the bearing at the plurality of predetermined time points through a multi-scale neighborhood feature extraction module; A responsiveness estimation module, used for calculating a responsiveness estimation of the multi-scale rotational speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix; as well as The fault diagnosis result generating module is used to pass the classification feature matrix through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

2. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 1, characterized in that: The external excitation linkage feature extraction module is further used for: each layer of the first convolutional neural network as a filter is respectively performed in the forward transfer of the layer: Perform convolution on the input data to obtain a convolution feature map; Performing mean pooling on the convolution feature map based on a local feature matrix to obtain a pooled feature map; as well as Performing nonlinear activation on the pooled feature map to obtain an activated feature map; Among them, the output of the last layer of the first convolutional neural network serving as a filter is the linkage feature vector, and the input of the first layer of the first convolutional neural network serving as a filter is the linkage association matrix.

3. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 2, characterized in that: The feature correction module comprises: An autocovariance matrix construction unit is used to calculate the variance between the eigenvalues ​​of every two positions in the linkage eigenvector to obtain an autocovariance matrix; An autocorrelation unit, used for calculating the product between the linkage eigenvector and the autocovariance matrix to obtain an autocorrelation eigenvector; A first position difference unit, used for calculating the position difference of the linkage feature vector and the autocorrelation feature vector to obtain a decorrelation linkage feature vector; A first exponential operation unit, used for calculating a natural exponential function value with the eigenvalue of each position in the decorrelation linkage eigenvector raised to a power to obtain an exponential decorrelation linkage eigenvector; A second position difference unit is used to calculate the position difference between the linkage feature vector and the bi-norm of the linkage feature vector to obtain a linkage feature vector without low rank representation; A second exponential operation unit is used to calculate a natural exponential function value with the eigenvalue of each position in the de-low-rank representation linkage eigenvector raised to power to obtain an exponential de-low-rank representation linkage eigenvector; and A fusion correction unit is used to fuse the exponential decorrelation linkage feature vector and the exponential de-low-rank representation linkage feature vector based on the global mean and variance of the eigenvalue set at all positions in the linkage feature vector to obtain the corrected linkage feature vector.

4. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 3, characterized in that: The fusion correction unit is further used to: fuse the exponential decorrelation linkage feature vector and the exponential de-low-rank representation linkage feature vector based on the global mean and variance of the eigenvalue set at all positions in the linkage feature vector using the following formula to obtain the corrected linkage feature vector; Wherein, the formula is: Wherein, V1 represents the exponential decorrelation linkage eigenvector, V2 represents the exponential de-low-rank linkage eigenvector, V′ represents the corrected linkage eigenvector, μ and σ represent the global mean and variance of the eigenvalue set of all positions in the linkage eigenvector, respectively. Indicates sum by position.

5. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 4, characterized in that: The bearing speed feature extraction module comprises: a bearing speed value vectorization unit, used for arranging the bearing speed values ​​at the plurality of predetermined time points into a bearing speed input vector; A first scale feature extraction unit, configured to input the bearing speed input vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first neighborhood scale tension-related feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel of a first length; a second scale feature extraction unit, configured to input the bearing speed input vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second neighborhood scale tension associated feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel of a second length, and the first length is different from the second length; and A cascading unit is used to cascade the first neighborhood scale tension-related feature vector and the second neighborhood scale tension-related feature vector to obtain the multi-scale rotational speed feature vector.

6. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 5, characterized in that: The first scale feature extraction unit is further used to: use the first convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the bearing speed input vector according to the following formula to obtain the first neighborhood scale tension-related feature vector; Wherein, the formula is: Among them, a is the width of the first convolution kernel in the x direction, F(a) is the first convolution kernel parameter vector, G(xa) is the local vector matrix operated with the convolution kernel function, w is the size of the first convolution kernel, and X represents the bearing speed input vector.

7. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 6, characterized in that: The second scale feature extraction unit is further used to: use the second convolution layer of the multi-scale neighborhood feature extraction module to perform one-dimensional convolution encoding on the bearing speed input vector according to the following formula to obtain the second neighborhood scale tension-related feature vector; Wherein, the formula is: Among them, b is the width of the first convolution kernel in the x direction, F(b) is the first convolution kernel parameter vector, G(xb) is the local vector matrix operated with the convolution kernel function, m is the size of the first convolution kernel, and X represents the bearing speed input vector.

8. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 7, characterized in that: The responsiveness estimation module is further used to: calculate the responsiveness estimation of the multi-scale rotational speed feature vector relative to the corrected linkage feature vector using the following formula to obtain the classification feature matrix; Wherein, the formula is: In s =M*V l Where V s represents the multi-scale speed feature vector, V l represents the corrected linkage feature vector, and M represents the classification feature matrix.

9. The bearing lubrication system fault diagnosis system of a wind turbine generator set according to claim 8, characterized in that: The fault diagnosis result generating module is further used for: the classifier processes the classification feature matrix according to the following formula to generate a classification result, wherein the formula is: softmax{(W n , B n ):...:(W1,B1)|Project(F)}, where Project(F) represents the projection of the classification feature matrix into a vector, W1 to W n is the weight matrix of each fully connected layer, B1 to B n Represents the bias matrix of each fully connected layer.

10. A method for diagnosing faults in a bearing lubrication system of a wind turbine generator set, characterized in that: include: Obtain wind speed values ​​and pitch angles at multiple predetermined time points within a predetermined time period; After arranging the wind speed values ​​and the pitch angles at a plurality of predetermined time points within the predetermined time period into a wind speed input vector and a pitch angle input vector according to the time dimension, respectively, calculating the product between the transposed vector of the wind speed input vector and the pitch angle input vector to obtain a linkage correlation matrix; Passing the linkage association matrix through a first convolutional neural network as a filter to obtain a linkage feature vector; Based on the binary norm of the linkage feature vector, correcting the feature value of each position in the linkage feature vector to obtain a corrected linkage feature vector; Obtaining the rotational speed values ​​of the bearing at the plurality of predetermined time points; The speed values ​​of the bearings at the plurality of predetermined time points are passed through a multi-scale neighborhood feature extraction module to obtain a multi-scale speed feature vector; Calculating a responsiveness estimate of the multi-scale rotational speed feature vector relative to the corrected linkage feature vector to obtain a classification feature matrix; as well as The classification feature matrix is ​​passed through a classifier to obtain a classification result, and the classification result is used to indicate whether the failure risk of the bearing lubrication system exceeds a predetermined threshold.

Images