Blade fault diagnosis method and diagnosis device of wind turbine generator and electronic equipment

Abstract

The invention provides a blade fault diagnosis method and device of a wind turbine generator and electronic equipment. The method comprises the steps that a blade vibration signal of the wind turbine generator is acquired; calculating a plurality of fault characteristic values of the blade vibration signal; determining a plurality of fault diagnosis results through a plurality of fault diagnosis models based on the plurality of fault characteristic values; determining a fused fault diagnosis result by fusing the multiple fault diagnosis results; and blade fault diagnosis is carried out based on the fused fault diagnosis result. According to the process, multiple fault feature values of blade vibration signals of the wind turbine generator are calculated, multiple fault diagnosis results are determined based on the multiple fault feature values and multiple fault diagnosis models, the multiple fault diagnosis results are fused to obtain a final diagnosis result for blade fault diagnosis, and the blade fault diagnosis accuracy is improved based on multi-dimensional fusion diagnosis. And the fault diagnosis accuracy is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of wind power generation, and more specifically, to a blade fault diagnosis method, a diagnosis device and an electronic device for a wind turbine. Background Art

[0002] At present, the blade damage of wind turbines is an important issue in the design, manufacture and operation of wind turbine blades. The wind turbines themselves are located in harsh climatic conditions and operate under alternating loads all day, which makes the wind turbine components more easily damaged. The rapid development of the wind power industry has put forward more stringent requirements for fault monitoring and diagnosis technology.

[0003] There are two main types of existing blade fault diagnosis for wind turbines. One is to combine the coherence analysis method with the nonlinear dynamics method to extract the fault features of the blade vibration signal, and then perform data preprocessing and statistical analysis on the extracted features, and use the XGBoost pre-trained model to determine the differences between the three blades to achieve fault diagnosis. The other is to use time domain, frequency domain, and time-frequency domain signal processing methods to achieve noise reduction and fault feature extraction for the blade vibration signal, or to use signal processing methods to reduce noise and combine with nonlinear dynamics methods to achieve fault feature extraction, and then perform data preprocessing and statistical analysis on the extracted features, and use the XGBoost pre-trained model to determine the differences between the three blades to achieve fault diagnosis.

[0004] The above two methods only obtain the characteristic value of the blade vibration signal from a single method and process the characteristic value. However, blades of different types and blades under various working conditions are quite different. In this case, the above two methods have poor generalization and high false alarm and missed alarm rates. Summary of the invention

[0005] To this end, the present disclosure provides a blade fault diagnosis method, a diagnosis device and an electronic device for a wind turbine generator set to reduce the false alarm rate and the missed alarm rate of blade fault diagnosis.

[0006] A first aspect of the present disclosure provides a blade fault diagnosis method for a wind turbine generator set, the blade fault diagnosis method comprising: acquiring a blade vibration signal of the wind turbine generator set; calculating a plurality of fault characteristic values ​​of the blade vibration signal; determining a plurality of fault diagnosis results through a plurality of fault diagnosis models based on the plurality of fault characteristic values; determining a fused fault diagnosis result by fusing the plurality of fault diagnosis results; and performing blade fault diagnosis based on the fused fault diagnosis result.

[0007] Optionally, the blade vibration signal includes a flapping vibration signal and a swing vibration signal; wherein, calculating multiple fault characteristic values ​​of the blade vibration signal includes: calculating a first fault characteristic value reflecting the coherence between the flapping vibration signal and the swing vibration signal based on the flapping vibration signal and the swing vibration signal; and calculating a second fault characteristic value reflecting the time-frequency domain characteristics of the flapping vibration signal or the swing vibration signal based on the flapping vibration signal or the swing vibration signal.

[0008] Optionally, based on the multiple fault characteristic values, multiple fault diagnosis results are determined through multiple fault diagnosis models, including: based on the first fault characteristic value, determining the coherence diagnosis result through a pre-trained coherence diagnosis model; based on the second fault characteristic value, determining the time-frequency domain diagnosis result through a pre-trained time-frequency domain diagnosis model.

[0009] Optionally, a fused fault diagnosis result is determined by fusing the multiple fault diagnosis results, including: determining the fused fault diagnosis result based on the coherence diagnosis result, the precision and recall of the coherence diagnosis model, the time-frequency domain diagnosis result, and the precision and recall of the time-frequency domain diagnosis model.

[0010] Optionally, the fusion fault diagnosis result is determined according to the coherence diagnosis result, the precision and recall of the coherence diagnosis model, the time-frequency domain diagnosis result, and the precision and recall of the time-frequency domain diagnosis model, including: calculating a first ratio of a first product to a first sum, wherein the first product is the product of the sum of the precision and recall of the coherence diagnosis model and the coherence diagnosis result, and the first sum is the sum of the precision and recall of the coherence diagnosis model and the precision and recall of the time-frequency domain diagnosis model; calculating a second ratio of a second product to the first sum, wherein the second product is the product of the sum of the precision and recall of the time-frequency domain diagnosis model and the time-frequency domain diagnosis result; and determining the fusion fault diagnosis result according to the sum of the first ratio and the second ratio.

[0011] Optionally, the coherence diagnostic model is trained through the following steps: selecting a coherence training feature from a preset coherence feature database, wherein the characteristic value distribution of the coherence training feature for each blade of the same wind turbine is abnormal; determining a first target training characteristic value based on the quartiles of the characteristic value of each blade of at least one wind turbine of the coherence training feature; and training the coherence diagnostic model using the first target training characteristic value.

[0012] Optionally, the first target characteristic value includes a first training characteristic value, a second training characteristic value and a third training characteristic value; wherein, the first target training characteristic value is determined according to the quartiles of the characteristic value of each blade of at least one wind turbine group according to the coherence training feature, including: sorting the characteristic values ​​of each blade of the same wind turbine group according to the coherence training feature; determining the quartiles of the characteristic value of each blade, wherein the quartiles include the first quartile, the second quartile and the third quartile; calculating the first difference between the first quartiles of each blade, and taking the ratio of the maximum first difference to the maximum first quartile as the first training characteristic value; calculating the second difference between the second quartiles of each blade, and taking the ratio of the maximum second difference to the maximum second quartile as the second training characteristic value; calculating the third difference between the third quartiles of each blade, and taking the ratio of the maximum third difference to the maximum third quartile as the third training characteristic value.

[0013] Optionally, the coherence feature database is established through the following steps: acquiring training flapping vibration signals and training swing vibration signals; screening the acquired training flapping vibration signals and training swing vibration signals according to the rotational speed data and yaw data of the wind turbine; calculating the coherence coefficient between the screened training flapping vibration signals and the training swing vibration signals, and determining multiple characteristic values ​​of the coherence feature based on the coherence coefficient; and establishing the coherence feature database according to the coherence feature and its corresponding multiple characteristic values.

[0014] Optionally, the time-frequency domain diagnostic model is trained through the following steps: selecting time-frequency domain training features from a preset time-frequency domain feature database, wherein the characteristic value distribution of the time-frequency domain training features for each blade of the same wind turbine is abnormal; determining a second target training feature value based on the mean and / or standard deviation of the characteristic value of each blade of at least one wind turbine using the time-frequency domain training features; and using the second target training feature value to train the time-frequency domain diagnostic model.

[0015] Optionally, the second target training feature value includes a fourth training feature value and a fifth training feature value; wherein, the second target training feature value is determined based on the mean and / or standard deviation of the feature value of each blade of at least one wind turbine group according to the time-frequency domain training feature, including: determining the mean and / or standard deviation of the feature value of each blade of the same wind turbine group according to the time-frequency domain training feature; calculating the fourth difference between the mean values ​​of the feature values ​​of each blade, and taking the ratio of the largest fourth difference to the largest mean value as the fourth training feature value; or calculating the fifth difference between the standard deviations of the feature values ​​of each blade, and taking the ratio of the largest fifth difference to the largest standard deviation as the fifth training feature value.

[0016] Optionally, the time-frequency domain feature database is established through the following steps, including: obtaining a training swing vibration signal or a training swing vibration signal; screening the obtained training swing vibration signal or training swing vibration signal according to the speed data and yaw data of the wind turbine generator set; calculating multiple eigenvalues ​​of frequency domain features according to the screened training swing vibration signal or training swing vibration signal; and establishing the time-frequency domain feature database according to the time-frequency domain features and their corresponding multiple eigenvalues.

[0017] Optionally, performing blade fault diagnosis based on the fused fault diagnosis result includes: in response to the fused fault diagnosis result satisfying a preset condition, determining that a blade fault occurs.

[0018] Optionally, the blade fault diagnosis method also includes: acquiring speed data and yaw data of the wind turbine corresponding to the blade vibration signal; filtering the blade vibration signal based on the speed data and yaw data to obtain a first filtered vibration signal; selecting a second filtered vibration signal that meets data quality detection conditions in the first filtered vibration signal; downsampling or zero-filling the second filtered vibration signal to obtain a target vibration signal, wherein calculating multiple fault characteristic values ​​of the blade vibration signal includes: calculating multiple fault characteristic values ​​of the target vibration signal.

[0019] A second aspect of the present disclosure provides a blade fault diagnosis device for a wind turbine generator set, the blade fault diagnosis device comprising: a signal acquisition unit, configured to acquire a blade vibration signal of the wind turbine generator set; a characteristic value determination unit, configured to calculate a plurality of fault characteristic values ​​of the blade vibration signal; a diagnosis result determination unit, configured to determine a plurality of fault diagnosis results through a plurality of fault diagnosis models based on the plurality of fault characteristic values; a fused fault diagnosis result determination unit, configured to determine a fused fault diagnosis result by fusing the plurality of fault diagnosis results; and a fault determination unit, configured to perform blade fault diagnosis based on the fused fault diagnosis result.

[0020] A third aspect of the present disclosure provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the blade fault diagnosis method of the wind turbine generator set as described above is implemented.

[0021] A fourth aspect of the present disclosure provides an electronic device, comprising: at least one processor; and at least one memory storing a computer program, wherein when the computer program is executed by the at least one processor, the blade fault diagnosis method of the wind turbine set as described above is implemented.

[0022] According to the blade fault diagnosis method, diagnostic device and electronic equipment of the wind turbine set disclosed in the present invention, by calculating multiple fault characteristic values ​​of the blade vibration signal of the wind turbine set, multiple fault diagnosis results are determined based on the multiple fault characteristic values ​​and multiple fault diagnosis models, and the multiple fault diagnosis results are fused to obtain the final diagnosis result for blade fault diagnosis, which can realize multi-dimensional fusion diagnosis, solve the problem of poor generalization of existing blade fault diagnosis methods, and improve the accuracy of blade fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 is a schematic flow chart illustrating a blade fault diagnosis method according to an exemplary embodiment of the present disclosure.

[0024] Figure 2 is a schematic flow chart illustrating a training method of a coherence diagnosis model according to an exemplary embodiment of the present disclosure.

[0025] Figure 3 is a schematic flow chart illustrating a training method of a time-frequency domain diagnostic model according to an exemplary embodiment of the present disclosure.

[0026] Figure 4 is a diagram illustrating an example application scenario of a blade fault diagnosis method according to an exemplary embodiment of the present disclosure.

[0027] Figure 5 is a schematic structural diagram showing a blade fault diagnosis device for a wind turbine according to an exemplary embodiment of the present disclosure.

[0028] Figure 6 is a block diagram illustrating an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] The application provides the following specific embodiments to help the reader obtain a comprehensive understanding of the method, device and / or system described herein. However, after understanding the disclosure of the application, various changes, modifications and equivalents of the method, device and / or system described herein will be clear. For example, the order of operations described herein is only an example, and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the application, except for the operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art can be omitted.

[0030] The features described herein can be implemented in different forms and should not be construed as being limited to the examples described herein. Rather, the examples described herein have been provided to illustrate only some of the many possible ways to implement the methods, devices, and / or systems described herein, which will be clear after understanding the disclosure of the present application.

[0031] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more.

[0032] Although terms such as "first", "second", and "third" may be used herein to describe various members, components, regions, layers, or portions, these members, components, regions, layers, or portions should not be limited by these terms. Instead, these terms are only used to distinguish one member, component, region, layer, or portion from another member, component, region, layer, or portion. Therefore, without departing from the teachings of the examples described herein, the first member, first component, first region, first layer, or first portion referred to in the examples may also be referred to as the second member, second component, second region, second layer, or second portion.

[0033] In the specification, when an element (such as a layer, a region, or a substrate) is described as being “on”, “connected to”, or “coupled to” another element, the element may be directly “on”, “connected to”, or “coupled to” another element, or one or more other elements may be present therebetween. Conversely, when an element is described as being “directly on”, “directly connected to”, or “directly coupled to” another element, there may be no other elements present therebetween.

[0034] The terms used herein are only used to describe various examples and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. The terms "comprise", "include" and "have" indicate the presence of the described features, quantities, operations, components, elements and / or combinations thereof, but do not exclude the presence or addition of one or more other features, quantities, operations, components, elements and / or combinations thereof.

[0035] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by a person of ordinary skill in the art to which the present disclosure belongs after understanding the present disclosure. Unless explicitly defined as such herein, terms (such as those defined in a general dictionary) should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and should not be interpreted in an idealized or overly formal manner.

[0036] Furthermore, in the description of examples, when it is considered that a detailed description of a well-known related structure or function would cause vague interpretation of the present disclosure, such a detailed description will be omitted.

[0037] As mentioned above, the existing wind turbine blade fault diagnosis method is easily affected by the quality of blade vibration data, and different types of blades and blades under various working conditions have certain differences, which may lead to a decrease in the diagnosis speed of blade faults, poor generalization and low accuracy, which is not conducive to ensuring the normal operation of wind turbines.

[0038] A blade fault diagnosis method for a wind turbine generator set, a blade fault diagnosis device for a wind turbine generator set, and an electronic device according to exemplary embodiments of the present disclosure will be described below with reference to the accompanying drawings to solve or alleviate at least one of the above problems.

[0039] The following will refer to Figure 1 A blade fault diagnosis method according to an exemplary embodiment of the present disclosure is described. The blade fault diagnosis method can be executed by a main control of a wind turbine generator set, can also be executed by other controllers arranged in the wind turbine generator set, or can be remotely executed by a controller of a wind farm and / or other networked controllers.

[0040] Reference Figure 1 In step S101, a blade vibration signal of a wind turbine generator system may be obtained.

[0041] Here, the blade vibration signal can be obtained by a vibration sensor set at the blade of the wind turbine. However, the method for obtaining the blade vibration signal is not limited to the above example. The blade vibration signal can also be obtained by an acceleration sensor set at the blade of the wind turbine. The present disclosure does not impose any special restrictions on this.

[0042] In step S102 , various fault characteristic values ​​of the blade vibration signal may be calculated.

[0043] According to an embodiment of the present disclosure, the blade vibration signal includes a flapping vibration signal and a shimmying vibration signal. In this case, a first fault characteristic value reflecting the coherence between the flapping vibration signal and the shimmying vibration signal can be calculated based on the flapping vibration signal and the shimmying vibration signal, and a second fault characteristic value reflecting the time-frequency domain characteristics of the flapping vibration signal or the shimmying vibration signal can be calculated based on the flapping vibration signal or the shimmying vibration signal.

[0044] Here, the first fault characteristic value reflecting the coherence between the flapping vibration signal and the swing vibration signal can be calculated using the transfer function. Specifically, the autopower spectrum density of the flapping vibration signal and the swing vibration signal of the blade can be calculated first, and the calculation method is as follows:

[0045] G xx =X * ·X

[0046] Among them, G xx is the auto-power spectral density, X *represents the conjugate of the Fourier transform of the signal x(t), X represents the Fourier transform of the signal x(t), and x(t) is a swing vibration signal or a shimmying vibration signal.

[0047] Secondly, the cross power spectral density of the flapping vibration signal and the pendulum vibration signal can be calculated as follows:

[0048] G xy =X * ·Y

[0049] Among them, G xy is the cross power spectral density, X and Y represent the Fourier transform of signals x(t) and y(t), respectively, x(t) is one of the waving vibration signal and the swing vibration signal, and y(t) is the other of the waving vibration signal and the swing vibration signal.

[0050] Next, the transfer function can be calculated based on the auto-power spectrum density and cross-power spectrum density of the flapping vibration signal and the swing vibration signal of the blade. For example, the first transfer function H of the system with the swing vibration signal as the impact signal and the flapping vibration signal as the response signal can be calculated. 1 , and the second transfer function H of the system with the swing vibration signal as the impact signal and the shimmy vibration signal as the response signal can be calculated 2 , the calculation formula is as follows:

[0051]

[0052]

[0053] Among them, G1 xy , G2xy, G3 xy , … represent the cross power spectrum density after equally dividing the time domain numbers of the swing vibration signal and the waving vibration signal, G1yx, G2yx, G3yx, … represent the cross power spectrum density after equally dividing the time domain numbers of the waving vibration signal and the swing vibration signal.

[0054] Then, the coherence coefficient γ reflecting the similarity between the first transfer function and the second transfer function can be calculated. 2 , coherence coefficient γ 2 is the sequence data in the frequency domain, where the calculation formula of the coherence coefficient is as follows:

[0055]

[0056] Finally, based on the coherence coefficient γ 2 The first fault characteristic value is calculated.

[0057] Table 1 below gives an example of a method for calculating a characteristic value (ie, a first fault characteristic value). As shown in Table 1, the corresponding first fault characteristic value can be calculated using a sequence of coherence coefficients according to the calculation method in Table 1.

[0058] Table 1

[0059]

[0060]

[0061]

[0062] As shown in Table 1, the corresponding eigenvalues ​​can be calculated according to the eigenvalue calculation method in Table 1. For example, according to the eigenvalue calculation method described in sequence number 1, the square sum of the sequence of coherence coefficients can be calculated, and the calculation result is used as the first fault eigenvalue. However, the exemplary embodiments of the present disclosure are not limited to this, and in addition to the calculation method recorded in Table 1, other methods can be used to calculate the eigenvalue, and the present disclosure does not make any special limitations.

[0063] Alternatively, the second fault characteristic value reflecting the time-frequency domain characteristics of the waving vibration signal may be calculated according to the characteristic value calculation method of Table 1 based on the time-frequency domain sequence of the waving vibration signal, or the second fault characteristic value reflecting the time-frequency domain characteristics of the waving vibration signal may be calculated according to the characteristic value calculation method of Table 1 based on the time-frequency domain sequence of the waving vibration signal. For example, the time-domain fault characteristic value may be calculated according to the characteristic value calculation method described in sequence number 13 in Table 1 based on the acquired time-domain sequence of the waving vibration signal, that is, the mean of the time-domain sequence is calculated, the number of values ​​in the time-domain sequence that are greater than the mean is determined, and the determined number is used as the time-domain fault characteristic value of the waving vibration signal.

[0064] In this way, by using the coherence characteristics of the flapping vibration signal and the swing vibration signal in the blade vibration signal and extracting the time-frequency domain characteristics of the flapping vibration signal and the swing vibration signal to reflect the blade fault information, the deficiency that a single feature cannot accurately reflect the blade fault can be avoided, and the accuracy of blade fault detection can be improved.

[0065] According to an embodiment of the present disclosure, before calculating multiple fault characteristic values ​​of the blade vibration signal, the blade vibration signal can be processed to obtain a target vibration signal (i.e., an effective vibration signal), and multiple fault characteristic values ​​of the target vibration signal can be calculated as multiple fault characteristic values ​​of the blade vibration signal.

[0066] Specifically, first, the rotation speed data and yaw data of the wind turbine corresponding to the blade vibration signal may be acquired, and then, the blade vibration signal may be screened based on the rotation speed data and the yaw data to obtain a first screened vibration signal.

[0067] Here, the corresponding speed data and yaw data can be determined according to the acquired blade vibration signal. If the corresponding speed data is less than the preset speed threshold, or if there is corresponding yaw data, the blade vibration signal can be deleted to obtain the first screened vibration signal. In this way, by preliminarily screening the blade vibration signal, the quality problem of the vibration signal caused by too low speed or yaw can be eliminated, and the data quality of the blade vibration signal can be improved to ensure the accuracy of fault diagnosis.

[0068] Next, a second screening vibration signal that meets the data quality detection condition may be selected from the first screening vibration signals.

[0069] Here, the data quality detection condition is related to the characteristic value of the blade vibration signal. The method of calculating the characteristic value includes but is not limited to calculating the mean deviation, the ratio of the number of points with the same data value to the total data volume, the ratio of the difference between the number of positive and negative data points to the total number of data points, the positive impact of vibration data, the drift anomaly threshold (zero crossing rate), the interruption of vibration data, the sliding window mean difference, the effective value, the kurtosis value, the proportion of null values ​​and non-real values ​​of vibration data, etc. When the characteristic value of the first screened vibration signal calculated based on the above method meets the data quality detection condition, the first screened vibration signal can be used as the second screened vibration signal, and the first screened vibration signal that does not meet the data quality detection condition is deleted. The second screened vibration signal obtained by performing a second screening on the blade vibration signal (i.e., performing data quality detection on the blade vibration signal) further ensures the data quality of the blade vibration signal, which can improve the accuracy of subsequent fault diagnosis.

[0070] Finally, the second filtered vibration signal may be downsampled or zero-filled to obtain a target vibration signal.

[0071] Specifically, the second screening vibration signal can be processed according to the length of the second screening vibration signal. For example, if the length of the second screening vibration signal exceeds 65536, the second screening vibration signal can be downsampled, and if the length of the second screening vibration signal is less than 65536, the second screening vibration signal can be padded with zeros to satisfy the signal length of 65536 and the sampling frequency of 50Hz. In this way, by processing the blade vibration signal, the difference in blade vibration signals of different types of blades or under different working conditions is reduced, thereby ensuring the accuracy of fault diagnosis.

[0072] In step S103, multiple fault diagnosis results may be determined based on multiple fault characteristic values ​​through multiple fault diagnosis models.

[0073] Specifically, the coherence diagnosis result can be determined based on the first fault characteristic value through a pre-trained coherence diagnosis model; and the time-frequency domain diagnosis result can be determined based on the second fault characteristic value through a pre-trained time-frequency domain diagnosis model.

[0074] Here, the pre-trained coherence diagnosis model and the pre-trained time-frequency domain diagnosis model are both data-driven models, which can be pre-trained based on the existing fault case library. The coherence diagnosis model and the time-frequency domain diagnosis model have no parameter update during blade fault diagnosis, and only judge the fault feature value. The coherence diagnosis result determined by the coherence diagnosis model and the time-frequency domain diagnosis result determined by the time-frequency domain diagnosis model have a value range of 0 to 1, where the diagnosis result can be understood as the possibility of a fault, 0 represents the absence of a fault at all, and 1 represents the existence of a fault.

[0075] Furthermore, the calculation method of the first fault characteristic value may correspond to the pre-trained coherence diagnosis model, and the calculation method of the second fault characteristic value may correspond to the pre-trained time-frequency domain diagnosis model. For example, if the features required for fault diagnosis according to the pre-trained coherence diagnosis model are the features corresponding to sequence number 1 in Table 1 (the sum of squares of the sequence) and the features corresponding to sequence number 2 (the sum of the absolute values ​​of the continuous changes of the sequence x), the sum of squares of the sequence may be calculated according to the sequence of the coherence coefficients of the waving vibration signal and the swing vibration signal to obtain the characteristic value corresponding to sequence number 1, and the sum of the absolute values ​​of the continuous changes of the sequence of the coherence coefficients may be calculated to obtain the characteristic value corresponding to sequence number 2. The characteristic value corresponding to sequence number 1 and the characteristic value corresponding to sequence number 2 are input into the coherence diagnosis model to obtain the coherence diagnosis result. In this way, according to the features required by the pre-trained coherence diagnosis model and the time-frequency domain diagnosis model for fault diagnosis, the corresponding first fault characteristic value and the second fault characteristic value may be determined.

[0076] In this way, selecting features that are prone to failure and calculating the corresponding first fault feature value and second fault feature value can reduce the amount of calculation of the fault features and improve the accuracy of fault diagnosis.

[0077] In step S104, a fused fault diagnosis result may be determined by fusing multiple fault diagnosis results, wherein the multiple fault diagnosis results include coherence diagnosis results and time-frequency domain diagnosis results.

[0078] According to the embodiments of the present disclosure, the fused fault diagnosis result can be determined based on the coherence diagnosis result, the precision and recall of the coherence diagnosis model, the time-frequency domain diagnosis result, and the precision and recall of the time-frequency domain diagnosis model. By multi-dimensionally fusing the coherence diagnosis result and the time-frequency domain diagnosis result, the problem of poor generalization, high false alarm rate and high missed alarm rate of the single diagnosis method of the fault diagnosis system can be effectively solved.

[0079] Specifically, first, a first ratio of the first product to the first sum can be calculated, wherein the first product is the product of the sum of the precision and recall of the coherence diagnosis model and the coherence diagnosis result, and the first sum is the sum of the precision and recall of the coherence diagnosis model and the precision and recall of the time-frequency domain diagnosis model; then, a second ratio of the second product to the first sum can be calculated, wherein the second product is the product of the sum of the precision and recall of the time-frequency domain diagnosis model and the time-frequency domain diagnosis result; finally, the fusion fault diagnosis result can be determined according to the sum of the first ratio and the second ratio.

[0080] Here, the precision and recall of the coherence diagnosis model and the precision and recall of the time-frequency domain diagnosis model can be calculated separately in various ways, and the present disclosure does not impose any restrictions on this. Optionally, the fault diagnosis result parameters of the coherence diagnosis model are calculated based on the coherence diagnosis results, the precision and recall of the coherence diagnosis model, and the fault result parameters of the time-frequency domain diagnosis model are calculated based on the time-frequency domain diagnosis results, the precision and recall of the time-frequency domain diagnosis model, and the fault diagnosis result parameters of the time-frequency domain diagnosis model are calculated, and the fault diagnosis result parameters of the two diagnosis models are added to obtain a fused fault diagnosis result. Specifically, the fused fault diagnosis result can be calculated by the following formula:

[0081]

[0082] Among them, result is the fusion fault diagnosis result, i takes the value of 0 and 1, representing the coherence diagnosis model and the time-frequency domain diagnosis model respectively, prop i is the coherence diagnosis result or the time-frequency domain diagnosis result, thre i is the accuracy of the coherence diagnosis model or the time-frequency domain diagnosis model, recall i is the recall rate of the coherence diagnosis model or the time-frequency domain diagnosis model. In this way, the diagnosis results of the coherence diagnosis model and the time-frequency domain diagnosis model are fused through the above formula to obtain a multi-dimensional fault diagnosis result based on coherence and time-frequency domain, solving the problem of poor generalization, high false alarm rate and missed alarm rate of a single system diagnosis method.

[0083] In step S105, blade fault diagnosis may be performed based on the fused fault diagnosis result.

[0084] Here, in response to the fusion fault diagnosis result satisfying the preset conditions, it is determined that a blade fault occurs. For example, the blade of the wind turbine can be diagnosed for faults according to the numerical value of the fusion fault diagnosis result. When the numerical value of the fusion fault diagnosis result is greater than or equal to 0.5, it is determined that a blade fault occurs. The blade fault may be a crack in the blade. However, the present disclosure is not limited thereto. The blade fault may also be sand holes on the blade surface, blade cracking, blade icing, blade surface corrosion, etc., and the present disclosure does not place any particular restrictions on this.

[0085] As described above, by determining multiple fault characteristic values ​​of the blade vibration signal, using the multiple fault characteristic values ​​to obtain the corresponding diagnosis results of multiple fault diagnosis models, and fusing the multiple diagnosis results to determine whether the blade is faulty, the multi-dimensional fusion of the diagnosis results of the fault diagnosis model is achieved, making the fault diagnosis results more accurate.

[0086] Figure 2 is a schematic flow chart illustrating a training method of a coherence diagnosis model according to an exemplary embodiment of the present disclosure.

[0087] Reference Figure 2 In step S201, a coherence training feature can be selected from a preset coherence feature database, and the selected coherence training feature has an abnormal distribution of feature values ​​for each blade of the same wind turbine. Here, a feature with an abnormal distribution of feature values ​​is selected from the preset coherence feature database as a training feature of the coherence diagnosis model, wherein the feature with an abnormal distribution of feature values ​​is a feature that is prone to failure. In this way, the accuracy of the coherence diagnosis model can be improved by using the feature values ​​of features that are prone to failure for training.

[0088] Specifically, a characteristic value change curve of each coherence feature on each blade of the same wind turbine in the same time period can be plotted. For example, a characteristic value change curve of each blade of the 52 features in Table 1 is plotted within 30 seconds for a certain wind turbine. The characteristic value change curve of each blade is compared with the characteristic value change curve of normal operation on the wind turbine blade, and the coherence features with abnormal characteristic value distribution in the characteristic value change curve (i.e., having a large difference from the normal characteristic value change curve (for example, the difference is greater than a preset threshold)) are used as coherence training features.

[0089] In addition, according to an embodiment of the present disclosure, a coherence feature database can be established by the following method. First, a training waving vibration signal and a training swing vibration signal can be obtained. Here, the training waving vibration signal and the training swing vibration signal can be historical vibration signals of the wind turbine during historical operation. However, the training waving vibration signal and the training swing vibration signal can also be real-time vibration signals in real-time operation of the wind turbine, which is not limited by the embodiment of the present disclosure.

[0090] Then, the obtained training flapping vibration signals and training shimmy vibration signals can be screened according to the speed data and yaw data of the wind turbine. Here, as described above, the training flapping vibration signals and training shimmy vibration signals with a speed less than a preset speed threshold or with yaw can be deleted, thereby obtaining vibration signals with higher data quality.

[0091] Next, the coherence coefficient between the screened training swing vibration signal and the training shimmy vibration signal can be calculated, and multiple characteristic values ​​of the coherence feature can be determined based on the calculated coherence coefficient. Here, the first transfer function can be calculated with the training shimmy vibration signal as the impact signal and the training swing vibration signal as the response signal; the second transfer function can be calculated with the training swing vibration signal as the impact signal and the training shimmy vibration signal as the response signal; the coherence coefficient can be determined based on the ratio of the first transfer function to the second transfer function; and the characteristic value of the coherence feature can be determined based on the coherence coefficient.

[0092] Finally, a coherence feature database can be established based on the coherence features and their corresponding multiple feature values.

[0093] In step S202, a first target training feature value may be determined based on the quartiles of the feature value of each blade of at least one wind turbine generator set for the coherence training feature. Specifically, because the distribution of blade features of different wind turbines and different models is quite different, the ratio of the maximum difference of the quartiles of the three blades of each feature to the maximum quartile may be calculated to generate a new feature value, i.e., the first target training feature value, for the training of the final coherence diagnosis model.

[0094] According to an embodiment of the present disclosure, the first target feature value includes a first training feature value, a second training feature value, and a third training feature value. Specifically, the first target training feature value can be determined by the following method. First, the feature values ​​of the coherence training feature for each blade of the same wind turbine can be sorted. Here, the feature values ​​of the coherence training feature on each blade of the wind turbine can be sorted, wherein the sorting method can be sorting from large to small, or from small to large, and the present disclosure does not limit this.

[0095] Next, the quartiles of the eigenvalues ​​of each blade may be determined, wherein the quartiles include the first quartile, the second quartile, and the third quartile. Specifically, the coherence training features have a eigenvalue sequence (i.e., a sorted eigenvalue sequence) on each blade of the same wind turbine. The first quartile, the second quartile, and the third quartile in the eigenvalue sequence (i.e., the sorted eigenvalue sequence) on each blade are determined.

[0096] Then, the first difference between the first quartiles of each leaf may be calculated, and the ratio of the largest first difference to the largest first quartile may be used as the first training feature value.

[0097] Specifically, the first difference between the first quartiles of the three blades of the same wind turbine can be calculated, that is, the first difference between the first quartile of the first blade and the first quartile of the second blade, the first difference between the first quartile of the second blade and the first quartile of the third blade, and the first difference between the first quartile of the first blade and the first quartile of the third blade. The maximum value among the absolute values ​​of the first differences is selected, and its ratio to the largest first quartile of the three blades is used as the first training feature value.

[0098] Alternatively, the second difference between the second quartiles of each leaf may be calculated, and the ratio of the largest second difference to the largest second quartile may be used as the second training feature value.

[0099] Specifically, the second difference between the second quartiles of the three blades of the same wind turbine can be calculated, that is, the second difference between the second quartile of the first blade and the second quartile of the second blade, the second difference between the second quartile of the second blade and the second quartile of the third blade, and the second difference between the second quartile of the first blade and the second quartile of the third blade. The maximum value among the absolute values ​​of the second differences is selected, and its ratio to the largest second quartile of the three blades is used as the second training feature value.

[0100] Alternatively, the third difference between the third quartiles of each leaf can be calculated, and the ratio of the largest third difference to the largest third quartile is used as the third training feature value. In this way, the first training feature value, the second training feature value and the third training feature value are used as new feature values ​​for training the coherence diagnosis model.

[0101] Here, the quartiles of the eigenvalues ​​of each leaf are used to describe the distribution of the eigenvalues ​​of the feature, thereby excluding abnormal eigenvalues. Further, new training eigenvalues ​​are determined by the ratio of the difference between the quartiles corresponding to the three leaves and the maximum quartile, in order to ensure the data quality during the training of the coherence diagnosis model.

[0102] In step S203, the coherence diagnosis model may be trained using the first target training feature value. In this way, the coherence diagnosis model may be trained using the XGBoost model and the first target training feature value.

[0103] Specifically, the Xgboost model uses the classification and regression decision tree (CART) as the base learner and outputs the sum of the results of all learners as the integrated model. The basic structure of CART is a binary tree, using the Gini gain as the partitioning criterion, and has the characteristics of automatically handling missing values, automatic pruning, and being friendly to class imbalanced data.

[0104] set up Where D represents a dataset with n samples and m features. Furthermore, D is the first target training feature value dataset, x i is the input space, y i is the output space, is a set of real numbers. The output of the boosting model containing K trees is:

[0105]

[0106] in, is the loss function, represents the function space of the classification and regression tree, and T is the number of leaves of the decision tree. k It consists of two parts: tree structure q and leaf weight ω.

[0107] Each base learner f k It is generated by minimizing the following objective function:

[0108]

[0109] The first part l is the measure of the predicted value and the true value y i The second part Ω is the regularization term that controls the complexity of the tree model.

[0110] Since the objective function of the boosted tree model cannot be optimized using traditional optimization methods in Euclidean space, the gradient boosted tree improves the boosted tree and gradually trains each tree model in an additive manner. Each round of training fits the residual of the previous model's prediction results. The prediction result of the t-th round model is:

[0111]

[0112] The objective function of the t-th round of model training can be written as:

[0113]

[0114] Using the second-order Taylor expansion to approximate the above formula, the objective function of the tth round can be rewritten as:

[0115]

[0116] in, and are the first and second order partial derivatives of the loss function, respectively. is a regularization term that limits the complexity of the tree, and λ and γ are hyperparameters.

[0117] Let I j ={i|q(g i)=j} represents the sample set contained in leaf j. After removing the constant term in the objective function, it is as follows:

[0118]

[0119]

[0120]

[0121] The above formula is used as an evaluation function to evaluate the tree structure q(x) to obtain the optimal structure for classification problems. Xgboost generates a tree structure through a greedy algorithm. Starting from each node, iteratively adds branches to expand the tree structure. By calculating the gain of each split method, the split method with the largest gain is selected as the partition node. The gain calculation method is as follows:

[0122]

[0123] Among them, IL and IR are the sample sets of the left and right nodes after division, I = IL ∪ IR represents the sample set contained in the parent node, and γ is the complexity introduced by adding the node.

[0124] Figure 3 is a schematic flow chart illustrating a training method of a time-frequency domain diagnostic model according to an exemplary embodiment of the present disclosure.

[0125] Reference Figure 3 In step S301, time-frequency domain training features can be selected from a preset time-frequency domain feature database, and the selected time-frequency domain training features have abnormal distribution of characteristic values ​​for each blade of the same wind turbine. Here, features with abnormal distribution of characteristic values ​​are selected from the preset time-frequency domain feature database as training features of the time-frequency domain diagnostic model, wherein the features with abnormal distribution of characteristic values ​​are features that are prone to failure. In this way, the accuracy of the time-frequency domain diagnostic model can be improved by using the characteristic values ​​of features that are prone to failure for training.

[0126] Specifically, a characteristic value change curve of each coherence feature on each blade of the same wind turbine in the same time period can be plotted. For example, a characteristic value change curve of each blade of the 52 features in Table 1 is plotted within one year for a certain wind turbine. The characteristic value change curve of each blade is compared with the characteristic value change curve of normal operation on the wind turbine blade, and the coherence features with abnormal characteristic value distribution in the characteristic value change curve (i.e., having a large difference from the normal characteristic value change curve (for example, the difference is greater than a preset threshold)) are used as time-frequency domain training features.

[0127] In addition, according to an embodiment of the present disclosure, a time-frequency domain feature database can be established by the following method. First, a training waving vibration signal or a training swing vibration signal can be obtained. Then, the obtained training waving vibration signal or training swing vibration signal can be screened according to the speed data and yaw data of the wind turbine. Next, a plurality of eigenvalues ​​of frequency domain features can be calculated based on the screened training waving vibration signal or training swing vibration signal. Here, for the frequency spectrum and envelope spectrum of the training waving vibration signal or the training swing vibration signal, the eigenvalues ​​of the time-frequency domain can be calculated according to the eigenvalue calculation method shown in Table 1. Finally, a time-frequency domain feature database can be established based on the time-frequency domain features and their corresponding plurality of eigenvalues.

[0128] In step S302, the second target training feature value can be determined based on the mean and / or standard deviation of the feature value of each blade of at least one wind turbine according to the time-frequency domain training feature. Specifically, because the distribution of blade features of different wind turbines and different models is quite different, the ratio of the maximum difference of the mean and standard deviation of the three blades of each feature to the maximum mean or standard deviation can be calculated to generate a new feature value, that is, the second target training feature value, for the training of the final time-frequency domain diagnostic model.

[0129] According to an embodiment of the present disclosure, the second target training feature value includes a fourth training feature value and a fifth training feature value. Specifically, the second target training feature value can be determined by the following method. First, the mean value and / or standard deviation of the feature value of the time-frequency domain training feature for each blade of the same wind turbine can be determined. Specifically, the mean value or standard deviation of the feature value pairs of the time-frequency domain training feature for the three blades of the same wind turbine can be calculated.

[0130] Then, the fourth difference between the mean values ​​of the eigenvalues ​​of each leaf can be calculated, and the ratio of the largest fourth difference to the largest mean value can be used as the fourth training eigenvalue; or the fifth difference between the standard deviations of the eigenvalues ​​of each leaf can be calculated, and the ratio of the largest fifth difference to the largest standard deviation can be used as the fifth training eigenvalue.

[0131] Here, firstly, the eigenvalue of a certain feature on the three blades of the same generator set is calculated, and the corresponding average value is calculated. Then, the fourth difference between the average of the eigenvalues ​​of the first blade and the average of the eigenvalues ​​of the second blade is calculated, the fourth difference between the average of the eigenvalues ​​of the second blade and the average of the eigenvalues ​​of the third blade is calculated, and the fourth difference between the average of the eigenvalues ​​of the first blade and the average of the eigenvalues ​​of the third blade is calculated. The maximum value in the fourth difference is selected, and its ratio to the maximum value in the average of the eigenvalues ​​of the three blades is used as the fourth training feature.

[0132] Alternatively, the characteristic values ​​of a certain characteristic on three blades of the same generator set can be calculated, and the corresponding standard deviations can be calculated, and then the fifth difference between the standard deviation of the characteristic value of the first blade and the standard deviation of the characteristic value of the second blade is calculated, the fifth difference between the standard deviation of the characteristic value of the second blade and the standard deviation of the characteristic value of the third blade is calculated, and the fifth difference between the standard deviation of the characteristic value of the first blade and the standard deviation of the characteristic value of the third blade is calculated, and the maximum value of the fifth difference is selected, and the ratio of the maximum value of the fifth difference to the standard deviation of the characteristic values ​​of the three blades is used as the fifth training feature. In this way, the fourth training characteristic value and the fifth training characteristic value are used as new characteristic values ​​for training the time-frequency domain diagnosis model.

[0133] Here, by selecting the mean value or standard deviation of the characteristic value of each leaf, the stability of the data during the training process of the time-frequency domain diagnosis model can be guaranteed.

[0134] In step S303, the time-frequency domain diagnosis model may be trained using the second target training feature value. Specifically, the time-frequency domain diagnosis model may be trained using the XGBoost model and the second target training feature value.

[0135] According to an embodiment of the present disclosure, by calculating two different blade fault characteristic values ​​of the blade vibration signal of a wind turbine, namely, a fault characteristic value related to coherence and a fault characteristic value related to the time-frequency domain, the two blade fault characteristic values ​​are respectively input into corresponding fault diagnosis models, namely, a coherence diagnosis model and a time-frequency domain diagnosis model, to obtain two fault diagnosis results, and the two fault diagnosis results are fused to determine whether there is a fault in the blade, thereby reducing the false alarm rate and missed alarm rate of fault diagnosis through multi-dimensional fusion diagnosis.

[0136] In addition, according to the embodiments of the present disclosure, the following will refer to Figure 4 An example application scenario of a blade fault diagnosis method according to an exemplary embodiment of the present disclosure is described.

[0137] First, the blade vibration signal data of the wind turbine is obtained, wherein the data includes blade vibration signals of multiple units (for example, but not limited to, 233 units) in multiple (for example, but not limited to, 14) wind farms in a certain period of time. Then, the vibration signals with low rotation speed or yaw in the blade vibration signals are eliminated. The data quality of the eliminated blade vibration signals is detected, that is, characteristic indexes are calculated based on the blade vibration signals, such as calculating and determining the mean deviation of the correlation diagnosis model, the ratio of the number of points with the same data value to the total data volume, the ratio of the difference between the number of positive and negative data points to the total number of data points, etc. The blade vibration signals that pass the data quality test are screened out, and the data that fail the data quality test are saved in the database.

[0138] Next, the fault features that need to be calculated are determined according to the correlation diagnosis model and the time-frequency domain diagnosis model, the fault features are calculated using the screened blade vibration signals to obtain the fault feature values, and the fault feature values ​​are input into the correlation diagnosis model and the time-frequency domain diagnosis model to obtain the correlation diagnosis results and the time-frequency domain diagnosis results.

[0139] Here, a coherence feature database and a time-frequency domain feature database can be established based on the training blade vibration signals, i.e., the training flapping vibration signals and the training swing vibration signals. Corresponding coherence training features and time-frequency domain training features are selected from the coherence feature database and the time-frequency domain feature database, and new training features are determined based on the coherence training features and the time-frequency domain training features, i.e., the first target training feature value for training the coherence diagnostic model and the second target training feature value for training the time-frequency domain diagnostic model.

[0140] Then, the fusion fault diagnosis result is calculated according to the coherence diagnosis result, the precision and recall of the coherence diagnosis model, the time-frequency domain diagnosis result, and the precision and recall of the time-frequency domain diagnosis model.

[0141] The above describes a method for diagnosing blade faults of a wind turbine according to an exemplary embodiment of the present disclosure. Figure 5 A blade fault diagnosis device for a wind turbine according to an exemplary embodiment of the present disclosure is described.

[0142] like Figure 5 As shown, the blade fault diagnosis device 500 includes: a signal acquisition unit 501, a characteristic value determination unit 502, a diagnosis result determination unit 503, a fusion fault diagnosis result determination unit 504, and a fault determination unit 505.

[0143] The signal acquisition unit 501 is configured to acquire the blade vibration signal of the wind turbine; the characteristic value determination unit 502 is configured to calculate multiple fault characteristic values ​​of the blade vibration signal; the diagnosis result determination unit 503 is configured to determine multiple fault diagnosis results through multiple fault diagnosis models based on the multiple fault characteristic values; the fused fault diagnosis result determination unit 504 is configured to determine the fused fault diagnosis result by fusing the multiple fault diagnosis results; the fault determination unit 505 is configured to perform blade fault diagnosis based on the fused fault diagnosis result.

[0144] As an example, the blade vibration signal includes a flapping vibration signal and a shimmying vibration signal. The characteristic value determining unit 502 may be configured to calculate a first fault characteristic value reflecting the coherence between the flapping vibration signal and the shimmying vibration signal based on the flapping vibration signal and the shimmying vibration signal; and calculate a second fault characteristic value reflecting the time-frequency domain characteristics of the flapping vibration signal or the shimmying vibration signal based on the flapping vibration signal or the shimmying vibration signal.

[0145] Optionally, the diagnosis result determination unit 503 may be configured to determine the coherence diagnosis result based on the first fault characteristic value by a pre-trained coherence diagnosis model; and to determine the time-frequency domain diagnosis result based on the second fault characteristic value by a pre-trained time-frequency domain diagnosis model.

[0146] Optionally, the fusion fault diagnosis result determination unit 504 may be configured to determine the fusion fault diagnosis result based on the coherence diagnosis result, the precision and recall of the coherence diagnosis model, the time-frequency domain diagnosis result, and the precision and recall of the time-frequency domain diagnosis model.

[0147] Optionally, the fusion fault diagnosis result determination unit 504 can also be configured to: calculate a first ratio of a first product to a first sum, wherein the first product is the product of the sum of the precision and recall of the coherence diagnosis model and the coherence diagnosis result, and the first sum is the sum of the precision and recall of the coherence diagnosis model and the precision and recall of the time-frequency domain diagnosis model; calculate a second ratio of a second product to the first sum, wherein the second product is the product of the sum of the precision and recall of the time-frequency domain diagnosis model and the time-frequency domain diagnosis result; determine the fusion fault diagnosis result according to the sum of the first ratio and the second ratio.

[0148] Optionally, the blade fault diagnosis device 500 may also include a coherence diagnosis model training unit (not shown), which may be configured to select a coherence training feature from a preset coherence feature database, wherein the characteristic value distribution of the coherence training feature for each blade of the same wind turbine is abnormal; determine a first target training feature value based on the quartiles of the characteristic value of the coherence training feature for each blade of at least one wind turbine; and train the coherence diagnosis model using the first target training feature value.

[0149] Optionally, the first target characteristic value includes a first training characteristic value, a second training characteristic value, and a third training characteristic value. The coherence diagnosis model training unit can also be configured to sort the characteristic values ​​of the coherence training characteristics for each blade of the same wind turbine; determine the quartiles of the characteristic value of each blade, wherein the quartiles include the first quartile, the second quartile, and the third quartile; calculate the first difference between the first quartiles of each blade, and use the ratio of the largest first difference to the largest first quartile as the first training characteristic value; calculate the second difference between the second quartiles of each blade, and use the ratio of the largest second difference to the largest second quartile as the second training characteristic value; calculate the third difference between the third quartiles of each blade, and use the ratio of the largest third difference to the largest third quartile as the third training characteristic value.

[0150] Optionally, the blade fault diagnosis device 500 may further include a coherence feature database establishment unit (not shown), which may be configured to obtain training flapping vibration signals and training swing vibration signals; filter the obtained training flapping vibration signals and training swing vibration signals according to the speed data and yaw data of the wind turbine; calculate the coherence coefficient between the filtered training flapping vibration signals and the training swing vibration signals, and determine multiple characteristic values ​​of the coherence feature based on the coherence coefficient; and establish a coherence feature database according to the coherence feature and its corresponding multiple characteristic values.

[0151] Optionally, the blade fault diagnosis device 500 may also include a time-frequency domain diagnostic model training unit (not shown), which may be configured to select time-frequency domain training features from a preset time-frequency domain feature database, wherein the characteristic value distribution of the time-frequency domain training features for each blade of the same wind turbine is abnormal; determine a second target training feature value based on the mean and / or standard deviation of the characteristic values ​​of the time-frequency domain training features for each blade of at least one wind turbine; and train the time-frequency domain diagnostic model using the second target training feature value.

[0152] Optionally, the second target training feature value includes a fourth training feature value and a fifth training feature value. The time-frequency domain diagnostic model training unit may also be configured to determine the mean value and / or standard deviation of the feature value of each blade of the same wind turbine generator set for the time-frequency domain training feature; calculate the fourth difference between the mean values ​​of the feature values ​​of each blade, and use the ratio of the largest fourth difference to the largest mean value as the fourth training feature value; or calculate the fifth difference between the standard deviations of the feature values ​​of each blade, and use the ratio of the largest fifth difference to the largest standard deviation as the fifth training feature value.

[0153] Alternatively, the fault determination unit 505 may be configured to determine that a blade fault occurs in response to the fused fault diagnosis result satisfying a preset condition.

[0154] Optionally, the blade fault diagnosis device 500 may also include a data processing unit (not shown), which may be configured to obtain speed data and yaw data of the wind turbine corresponding to the blade vibration signal; filter the blade vibration signal based on the speed data and yaw data to obtain a first filtered vibration signal; select a second filtered vibration signal that meets the data quality detection condition in the first filtered vibration signal; downsample or zero-fill the second filtered vibration signal to obtain a target vibration signal, wherein multiple fault characteristic values ​​of the blade vibration signal are calculated, including: calculating multiple fault characteristic values ​​of the target vibration signal.

[0155] Regarding the device in the above embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0156] Figure 6 is a block diagram showing an electronic device according to an embodiment of the present disclosure. Figure 6 The electronic device 600 includes: a processor 601; a memory 602, wherein the memory 602 stores computer executable instructions, wherein when the computer executable instructions are executed by the processor 601, the processor 601 is prompted to execute a blade fault diagnosis method for a wind turbine according to an exemplary embodiment of the present disclosure.

[0157] As an example, the electronic device 600 can be a PC computer, a tablet device, a personal digital assistant, a smart phone or other device capable of executing the above-mentioned instruction set. Here, the electronic device 600 is not necessarily a single electronic device, but can also be any device or circuit capable of executing the above-mentioned instruction (or instruction set) individually or jointly. The electronic device 600 can also be a part of an integrated control system or a system manager, or can be configured as a portable electronic device interconnected with an interface locally or remotely (for example, via wireless transmission). In addition, the electronic device 600 can also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.). All components of the electronic device 600 can be connected to each other via a bus and / or a network.

[0158] Processor 601 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, processor 601 may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0159] The processor 601 may execute instructions or codes stored in the memory 602, which may also store data. Instructions and data may also be sent and received over a network via a network interface device, which may employ any known transmission protocol.

[0160] The memory 602 may be integrated with the processor 601, for example, by placing RAM or flash memory within an integrated circuit microprocessor or the like. In addition, the memory 602 may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory 602 and the processor 601 may be operatively coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor 601 can read files stored in the memory 602.

[0161] According to an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor, the processor is prompted to execute a blade fault diagnosis method for a wind turbine according to an exemplary embodiment of the present disclosure.

[0162] Specifically, the blade fault diagnosis method of the wind turbine according to the embodiment of the present disclosure can be written as a computer program and stored on a computer-readable storage medium. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer programs and any associated data, data files and data structures in a non-transitory manner and provide computer programs and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. In one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0163] According to the blade fault diagnosis method, diagnostic device and electronic equipment of the wind turbine set disclosed in the present invention, by calculating multiple fault characteristic values ​​of the blade vibration signal of the wind turbine set, multiple fault diagnosis results are determined based on the multiple fault characteristic values ​​and multiple fault diagnosis models, and the multiple fault diagnosis results are fused to obtain the final diagnosis result for blade fault diagnosis, which can realize multi-dimensional fusion diagnosis, solve the problem of poor generalization of existing blade fault diagnosis methods, and improve the accuracy of blade fault diagnosis.

[0164] The specific implementation methods of the present disclosure have been described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments may be modified and varied without departing from the principles and spirit of the present disclosure, the scope of which is defined by the claims and their equivalents. These modifications and variations should also be within the scope of protection of the claims of the present disclosure.

Claims

1. A method for diagnosing blade faults of a wind turbine generator set, characterized in that: The blade fault diagnosis method comprises: Obtain blade vibration signals of wind turbines; Calculating multiple fault characteristic values ​​of the blade vibration signal; Based on the multiple fault characteristic values, determining multiple fault diagnosis results through multiple fault diagnosis models; Determine a fused fault diagnosis result by fusing the multiple fault diagnosis results; Blade fault diagnosis is performed based on the fused fault diagnosis result.

2. The blade fault diagnosis method according to claim 1, characterized in that: The blade vibration signal includes a flapping vibration signal and a shimmying vibration signal; The calculation of multiple fault characteristic values ​​of the blade vibration signal includes: calculating, based on the flapping vibration signal and the shimmying vibration signal, a first fault characteristic value reflecting the coherence between the flapping vibration signal and the shimmying vibration signal; Based on the flapping vibration signal or the shimmying vibration signal, a second fault characteristic value reflecting time-frequency domain characteristics of the flapping vibration signal or the shimmying vibration signal is calculated.

3. The blade fault diagnosis method according to claim 2, characterized in that: Based on the multiple fault characteristic values, multiple fault diagnosis results are determined through multiple fault diagnosis models, including: Based on the first fault characteristic value, determining a coherence diagnosis result through a pre-trained coherence diagnosis model; Based on the second fault characteristic value, a time-frequency domain diagnosis result is determined by a pre-trained time-frequency domain diagnosis model.

4. The blade fault diagnosis method according to claim 3, characterized in that: The fused fault diagnosis result is determined by fusing the multiple fault diagnosis results, including: The fusion fault diagnosis result is determined according to the coherence diagnosis result, the precision and recall of the coherence diagnosis model, the time-frequency domain diagnosis result, and the precision and recall of the time-frequency domain diagnosis model.

5. The blade fault diagnosis method according to claim 4, characterized in that: Determining the fusion fault diagnosis result according to the coherence diagnosis result, the precision and recall of the coherence diagnosis model, the time-frequency domain diagnosis result, and the precision and recall of the time-frequency domain diagnosis model includes: Calculate a first ratio of a first product to a first sum, wherein the first product is the product of the sum of the precision and the recall of the coherence diagnosis model and the coherence diagnosis result, and the first sum is the sum of the precision and the recall of the coherence diagnosis model and the precision and the recall of the time-frequency domain diagnosis model; Calculating a second ratio of a second product to the first sum, wherein the second product is the product of the sum of the precision and the recall of the time-frequency domain diagnosis model and the time-frequency domain diagnosis result; The fusion fault diagnosis result is determined according to the sum of the first ratio and the second ratio.

6. The blade fault diagnosis method according to claim 3, characterized in that: The coherence diagnosis model is trained by the following steps: Selecting a coherence training feature from a preset coherence feature database, wherein the coherence training feature has an abnormal distribution of feature values ​​for each blade of the same wind turbine; Determining a first target training feature value according to the quartile of the feature value of each blade of at least one wind turbine generator set of the coherence training feature; The coherence diagnosis model is trained using the first target training feature value.

7. The blade fault diagnosis method according to claim 6, characterized in that: The first target feature value includes a first training feature value, a second training feature value and a third training feature value; Wherein, determining the first target training feature value according to the quartile of the feature value of each blade of at least one wind turbine generator set of the coherence training feature includes: Sort the feature values ​​of the coherence training features for each blade of the same wind turbine generator set; Determine the quartiles of the characteristic value of each leaf, wherein the quartiles include a first quartile, a second quartile, and a third quartile; Calculate the first difference between the first quartiles of each leaf, and use the ratio of the largest first difference to the largest first quartile as the first training feature value; Calculate the second difference between the second quartiles of each leaf, and take the ratio of the maximum second difference to the maximum second quartile as the second training feature value; The third difference between the third quartiles of each leaf is calculated, and the ratio of the maximum third difference to the maximum third quartile is used as the third training feature value.

8. The fault diagnosis method according to claim 6, characterized in that: The coherence feature database is established by the following steps: Get Training waving vibration signal and training shimmy vibration signal; The obtained training flapping vibration signal and training shimmy vibration signal are screened according to the rotation speed data and yaw data of the wind turbine; calculating a coherence coefficient between the screened training waving vibration signal and the training shimmy vibration signal, and determining a plurality of characteristic values ​​of the coherence feature based on the coherence coefficient; The coherence feature database is established according to the coherence features and their corresponding multiple feature values.

9. The blade fault diagnosis method according to claim 3, characterized in that: The time-frequency domain diagnostic model is trained by the following steps: Selecting a time-frequency domain training feature from a preset time-frequency domain feature database, wherein the time-frequency domain training feature has an abnormal distribution of feature values ​​for each blade of the same wind turbine generator set; Determine a second target training feature value according to the mean and / or standard deviation of the feature value of each blade of at least one wind turbine generator set for the time-frequency domain training feature; The time-frequency domain diagnostic model is trained using the second target training feature value.

10. The blade fault diagnosis method according to claim 9, characterized in that: The second target training feature value includes a fourth training feature value and a fifth training feature value; Wherein, determining the second target training feature value according to the mean and / or standard deviation of the feature value of each blade of at least one wind turbine generator set according to the time-frequency domain training feature includes: Determine the average value and / or standard deviation of the characteristic value of each blade of the same wind turbine generator set for the time-frequency domain training feature; Calculate the fourth difference between the average values ​​of the characteristic values ​​of each leaf, and use the ratio of the largest fourth difference to the largest average value as the fourth training characteristic value; or The fifth difference between the standard deviations of the characteristic values ​​of each leaf is calculated, and the ratio of the largest fifth difference to the largest standard deviation is used as the fifth training characteristic value.

11. The blade fault diagnosis method according to claim 9, characterized in that: The time-frequency domain feature database is established by the following steps, including: Acquire a training waving vibration signal or a training shimmying vibration signal; The obtained training flapping vibration signal or training shimmy vibration signal is screened according to the rotation speed data and yaw data of the wind turbine; Calculating multiple characteristic values ​​of frequency domain characteristics according to the screened training swing vibration signal or the training swing vibration signal; The time-frequency domain feature database is established according to the time-frequency domain features and their corresponding multiple feature values.

12. The blade fault diagnosis method according to claim 1, characterized in that: Performing blade fault diagnosis based on the fused fault diagnosis result includes: In response to the fused fault diagnosis result satisfying a preset condition, it is determined that a blade fault occurs.

13. The blade fault diagnosis method according to claim 1, characterized in that: The blade fault diagnosis method further comprises: Acquire rotation speed data and yaw data of the wind turbine generator set corresponding to the blade vibration signal; Filtering the blade vibration signal based on the rotation speed data and the yaw data to obtain a first filtered vibration signal; Selecting a second screening vibration signal that meets a data quality detection condition from among the first screening vibration signals; downsampling or zero-padding the second filtered vibration signal to obtain a target vibration signal, The step of calculating a plurality of fault characteristic values ​​of the blade vibration signal includes: calculating a plurality of fault characteristic values ​​of the target vibration signal.

14. A blade fault diagnosis device for a wind turbine generator set, characterized in that: The blade fault diagnosis device comprises: A signal acquisition unit, configured to acquire a blade vibration signal of a wind turbine; a characteristic value determination unit, configured to calculate a plurality of fault characteristic values ​​of the blade vibration signal; a diagnosis result determination unit configured to determine a plurality of fault diagnosis results through a plurality of fault diagnosis models based on the plurality of fault characteristic values; a fused fault diagnosis result determining unit, configured to determine a fused fault diagnosis result by fusing the multiple fault diagnosis results; The fault determination unit is configured to perform blade fault diagnosis based on the fused fault diagnosis result.

15. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the blade fault diagnosis method for a wind turbine set according to any one of claims 1 to 13 is implemented.

16. An electronic device, characterized in that: include: at least one processor; At least one memory stores a computer program, and when the computer program is executed by the at least one processor, the blade fault diagnosis method of the wind turbine set according to any one of claims 1 to 13 is implemented.

Images