PCA-SPE-CUSUM algorithm-based wind turbine generator early fault online detection method, program, equipment and storage medium

Abstract

The invention belongs to the technical field of wind turbine generator fault detection, and particularly relates to a PCA-SPE-CUSUM algorithm-based wind turbine generator early fault online detection method, program and equipment and a storage medium. According to the method, principal component analysis and a square prediction error cumulative sum algorithm are applied to monitoring data of the SCADA system of the wind turbine generator set, detection of early faults of the generator system of the wind turbine generator set can be achieved, and early warning can be given earlier than that of the SCADA system. According to the method, the problems that an SCADA data acquisition system is multiple in variable and large in data dimension are effectively solved through principal component analysis, the operation cost is reduced, and the fault detection efficiency is improved. The fault amplitude estimation model deduced based on the fault detection algorithm provided by the invention can realize amplitude estimation of the generator fault of the wind turbine generator, and is helpful for evaluating the severity of the fault.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind turbine fault detection, and particularly relates to an online detection method, program, device and storage medium for early faults of wind turbines based on the PCA-SPE-CUSUM algorithm. Background Art

[0002] With the continuous development of intelligent manufacturing and the transformation of the manufacturing industry, structural health monitoring and fault diagnosis have become particularly important. This includes aspects such as condition monitoring, fault diagnosis, trend prediction and life assessment. The reliability of various devices directly affects people's lives and interests. The occurrence of faults, especially in large devices and transportation means such as high-speed EMUs, large compressors, generator sets and large ships, may lead to serious accidents, causing huge economic losses and casualties.

[0003] Wind turbines are usually in harsh environments, commonly found on mountaintops, at sea and in remote areas. The operating conditions are extremely complex and the wind speed is unstable, which makes wind turbines prone to faults. In order to maximize the utilization of wind energy, these devices are often located in sparsely populated areas, which also means that it takes a long time for maintenance personnel to arrive after a fault is detected, resulting in high maintenance costs. At the same time, during maintenance, the shutdown of wind turbines will cause significant economic losses. At present, the maintenance of wind turbines mainly relies on regular inspections and after-fault repairs. These methods cannot detect early faults in a timely manner, resulting in an extended downtime when a fault occurs, increasing economic losses. At the same time, it may also lead to over-maintenance, wasting manpower and material resources.

[0004] Fault diagnosis methods are mainly divided into model-based methods and data-based methods. Model-based methods rely on the theoretical analysis of system behavior, and their effectiveness is affected by the accuracy of mathematical models and is prone to decline due to the uncertainty of operating conditions and environmental changes. In recent years, technological progress has made it possible to store a large amount of data, which can effectively describe the system state and can be used for state diagnosis based on historical process data. In contrast, data-based methods can be regarded as a black-box method. It does not rely on any system model or prior knowledge, but directly extracts key information from rich process data. Therefore, data-based fault diagnosis methods show broad application prospects.

[0005] The status detection data of wind turbines is mainly collected through the Supervisory Control And Data Acquisition (SCADA) system installed in the wind farm. At present, almost all operating wind farms are equipped with SCADA systems. However, the SCADA system only issues early warnings when the faults reach a certain severity level, and there may be risks of false positives and false negatives in its early warning mechanism. Therefore, it is necessary to deeply analyze the operation data collected by the SCADA system to better judge the operation status of wind turbines and early warning and monitor the early faults that occur. Summary of the Invention

[0006] The purpose of the present invention is to provide an online early fault detection method, program, device and storage medium for wind turbines based on the PCA-SPE-CUSUM algorithm, which can realize the health status monitoring, early fault detection and fault amplitude estimation of the generator system of wind turbines.

[0007] An online early fault detection method for wind turbines based on the PCA-SPE-CUSUM algorithm includes the following steps:

[0008] Step 1: Sample the wind turbine under the current working condition for a period of time, obtain the wind turbine status information at each sampling moment, including stator winding temperature, non-drive end bearing temperature, drive end bearing temperature, active power and nacelle temperature, and construct the status data set X of the wind turbine under the current working condition in ;

[0009] Step 2: Sample from the historical time series monitoring data set of the wind turbine in the normal working state, with the same sampling duration as in Step 1, obtain the wind turbine status information at each sampling moment, and construct the historical status data set X of the wind turbine in the normal working state 0 ;

[0010] Step 3: Preprocess the data sets X in and X 0 , eliminate the data in the time period with missing values and the data in the time period with zero active power; for the samples in the preprocessed data sets X in and X 0 , calculate the differences between the stator winding temperature, non-drive end bearing temperature, drive end bearing temperature and nacelle temperature, and construct the results into data sets X test and X h ;

[0011] Step 4: Perform decentralization and standardization processing on the data sets X test and X h to obtain the data sets and

[0012] Step 5: Use the principal component analysis method to project the data set onto a new mapping space and construct a real-time data calculation result matrix T test , and determine the principal component space and the remaining principal component space of matrix T test ;

[0013] Step 6: Calculate the remaining principal components of the historical data calculation result matrix T, calculate the squared prediction error and its cumulative sum for each row in the historical data calculation result matrix T, and determine the fault threshold H; based on the remaining principal components of the real-time data calculation result matrix T test , calculate the squared prediction error and its cumulative sum for each row in the real-time data calculation result matrix T test , and judge whether the wind turbine has a fault according to the fault threshold.

[0014] Furthermore, the state data set X of the wind turbine under the current working condition in step 1 in is:

[0015]

[0016] where, represents the state information of the wind turbine at the sampling moment under the current working condition , including the stator winding temperature T 1 , the non-drive end bearing temperature T 2 , the drive end bearing temperature T 3 , the active power P 0 , and the nacelle temperature T 0 ; N 1 is the number of samples in the data set X in ;

[0017] Furthermore, the normal working historical state data set X of the wind turbine in step 2 0 is:

[0018]

[0019] where, represents the state information of the wind turbine at the sampling moment in the normal working state , N 2 is the number of samples in the data set X 0 , N is the sampling duration.

[0020] Furthermore, the data sets X test and X h in step 3 are:

[0021]

[0022] Among them, N 3 and N 4 are the number of samples in the preprocessed dataset X in and X 0 respectively.

[0023] Furthermore, the specific steps of step 5 are as follows:

[0024] Step 5.1: Based on the dataset Construct the covariance matrix S;

[0025]

[0026] Step 5.2: Perform eigenvalue decomposition on the covariance matrix S to obtain eigenvalues (λ 1 , λ 2 ,..., λ m ) and the eigenvector matrix P;

[0027] Step 5.3: Based on the eigenvector matrix P and the dataset Construct the real-time data calculation result matrix T test ;

[0028]

[0029] Step 5.4: Take the first l columns in the real-time data calculation result matrix T test as the principal components, and the remaining m - l columns as the remaining principal components; among them, the calculation method of l is:

[0030] Calculate the contribution rate CPV(λ d ) of each eigenvalue λ d , d = 1,..., m, and select the largest first l contribution rates and the sum of the first l contribution rates is greater than 90%;

[0031]

[0032] Furthermore, the specific steps of step 6 are as follows:

[0033] Step 6.1: Based on the remaining principal components of the historical data calculation result matrix T, calculate the squared prediction error of each row in the historical data calculation result matrix T Obtain the mean μ h ;

[0034]

[0035] Among them, is the squared prediction error of the a-th row in the historical data calculation result matrix T, a = 1, 2,..., N 4; It is the element in the a-th row and k-th column of the historical data calculation result matrix T;

[0036] Step 6.2: Let G h (1)=0, and use the recurrence formula to calculate the cumulative sum G h of the squared prediction errors of each row in the historical data calculation result matrix T h : Take the maximum value of G

[0037]

[0038] H = max{G h (a), a ∈ [1, N 4}

[0039] where G h (a) is the cumulative sum of the squared prediction errors of the a-th row in the historical data calculation result matrix T; δ is the allowable deviation, δ = 5%μ h ;

[0040] Step 6.3: Based on the remaining principal components of the real-time data calculation result matrix T test , calculate the squared prediction error of each row in the real-time data calculation result matrix T test

[0041]

[0042] where, is the squared prediction error of the a-th test row in the real-time data calculation result matrix T test , a test = 1, 2,..., N 3 ; is the element in the a-th test row and k-th column of the real-time data calculation result matrix T test ;

[0043] Step 6.4: Let G test (1)=0, and use the recurrence formula to calculate the cumulative sum G test of the squared prediction errors of each row in the real-time data calculation result matrix T test (a test ):

[0044]

[0045] If until a test = N 3 , G test (a test) are all less than the fault threshold H, it is determined that the wind turbine has not failed;

[0046] If a test = b, G test (b) is greater than or equal to the fault threshold H, it is determined that the wind turbine has failed.

[0047] Furthermore, in step 6, if a test = b, G test (b) is greater than or equal to the fault threshold, it is determined that the wind turbine has failed, and continue to calculate Gn = G test (N 3 ) through the recurrence formula, and then calculate the fault amplitude based on the cumulative sum Gn of the squared prediction errors of the wind turbine under the current working condition Judge the severity of the wind turbine fault;

[0048]

[0049] Among them, the superscript " * " means removing noise; is the element in the a-th row and j-th column of the matrix X h after removing noise, a = 1,..., N 4 , j = 1, 2, 3; is the element in the a-th row and the i-th column of the matrix after removing noise; v aj is the element in the a-th row and j-th column of the noise signal matrix; p jc is the element in the j-th row and c-th column of the eigenvector matrix P; is the mean of all elements in the j-th column of the matrix X h after removing noise; is the standard deviation of all elements in the j-th column of the matrix X h after removing noise;

[0050] If it is considered that the wind turbine is in the early fault stage. As time goes by, the severity of the fault will gradually increase, and it is necessary to monitor the development of the fault and perform regular maintenance;

[0051] If it is considered that the wind turbine has a relatively serious fault, and it should be shut down for repair or component replacement immediately.

[0052] A computer device / system, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the steps of the above-mentioned online early fault detection method for wind turbines based on the PCA - SPE - CUSUM algorithm.

[0053] A computer-readable storage medium has a computer program / instructions stored thereon. When the computer program / instructions are executed by a processor, the steps of the above-mentioned online early fault detection method for wind turbine generator sets based on the PCA-SPE-CUSUM algorithm are implemented.

[0054] A computer program product includes a computer program / instructions. When the computer program / instructions are executed by a processor, the steps of the above-mentioned online early fault detection method for wind turbine generator sets based on the PCA-SPE-CUSUM algorithm are implemented.

[0055] The beneficial effects of the present invention are as follows:

[0056] The present invention applies principal component analysis and the cumulative sum of squared prediction errors algorithm to the monitoring data of the SCADA system of wind turbine generator sets, which can realize the detection of early faults in the generator system of wind turbine generator sets and can issue an early warning earlier than the SCADA system. The present invention effectively solves the problems of many variables and large data dimensions in the SCADA data acquisition system by using principal component analysis, reduces the operation cost, and improves the fault detection efficiency. The fault amplitude estimation model derived from the fault detection algorithm proposed by the present invention can realize the amplitude estimation of the generator fault of the wind turbine generator set, which helps to evaluate the severity of the fault. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 It is the overall architecture diagram of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0058] The following further describes the present invention with reference to the drawings.

[0059] The present invention provides an online early fault detection method for wind turbine generator sets based on the PCA-SPE-CUSUM algorithm, which can realize the online detection of early faults in the generator system of wind turbine generator sets; at the same time, the present invention also establishes an amplitude estimation model based on the PCA-SPE-CUSUM fault diagnosis algorithm, which can estimate the severity of faults in the generator system of wind turbine generator sets.

[0060] Step 1: Sample the wind turbine generator set under the current working condition for a period of time, obtain the state information of the wind turbine generator set at each sampling moment, and construct a data set X in ;

[0061]

[0062]

[0063] Among them, represents the state information of the wind turbine generator set at the sampling moment under the current working condition including the stator winding temperature T 1 and the non-drive end bearing temperature T2 , the temperature T of the drive-end bearing 3 , the active power P 0 and the nacelle temperature T 0 ; N 1 is the number of samples in the dataset X in ;

[0064] Step 2: Sample from the historical time-series monitoring dataset of the wind turbine under normal operating conditions. The sampling duration is the same as that in Step 1 to obtain the wind turbine status information at each sampling moment and construct the dataset X 0 ;

[0065]

[0066] wherein represents the status information of the wind turbine at the sampling moment under normal operating conditions , N 2 is the number of samples in the dataset X 0 ; N is the sampling duration;

[0067] Step 3: Preprocess the datasets X in and X 0 by removing the data within the time period with missing values and the data within the time period where the corresponding active power is zero; for the samples in the preprocessed datasets X in and X 0 , calculate the differences between the stator winding temperature, non-drive-end bearing temperature, drive-end bearing temperature and nacelle temperature, and construct the results into the datasets X test and X h ;

[0068]

[0069] wherein, N 3 and N 4 are respectively the number of samples in the preprocessed datasets X in and X 0 ;

[0070] Step 4: Perform de-centralization and standardization on the datasets X test and X h to obtain the datasets and

[0071] De-centralization means subtracting the sample mean of each variable from the observed value corresponding to the variable in the dataset; standardization means dividing the observed value corresponding to each variable in the de-centralized dataset by the sample standard deviation of the variable;

[0072] Step 5: Use the principal component analysis method to project the data set and onto a new mapping space, and determine the principal component space and the remaining principal component space;

[0073] Step 5.1: Based on the data set Construct the covariance matrix S;

[0074]

[0075] Step 5.2: Perform eigenvalue decomposition on the covariance matrix S to obtain the eigenvalues (λ 1 , λ 2 ,..., λ m ) and the eigenvector matrix P;

[0076] Step 5.3: Based on the eigenvector matrix P and the data set Construct the historical data calculation result matrix T, and based on the eigenvector matrix P and the data set Construct the real-time data calculation result matrix T test ; T is a matrix with N 4 rows and m columns, and T test is a matrix with N 3 rows and m columns;

[0077]

[0078] Step 5.4: Take the first l columns in the historical data calculation result matrix T as the principal components, and the remaining m - l columns as the remaining principal components; take the first l columns in the real-time data calculation result matrix T test as the principal components, and the remaining m - l columns as the remaining principal components;

[0079] The calculation method of l is: calculate the contribution rate CPV(λ d ) of each eigenvalue λ d , d = 1,..., m, and select the largest first l contribution rates and the sum of the first l contribution rates is greater than 90%;

[0080]

[0081] Step 6: Based on the remaining principal components of the historical data calculation result matrix T, calculate the squared prediction error of each row in the historical data calculation result matrix T and its cumulative sum G h (a), and determine the fault threshold H; based on the remaining principal components of the real-time data calculation result matrix T test , calculate the squared prediction error of each row in the real-time data calculation result matrix T test and its cumulative sum G and its cumulative sum G test (atest ) According to G test (a test ) and the fault threshold H, determine whether the wind turbine has a fault;

[0082] Step 6.1: Calculate the remaining principal components of the result matrix T based on historical data, and calculate the squared prediction error of each row in the result matrix T of historical data calculation Obtain the mean value μ h ;

[0083]

[0084] Among them, is the squared prediction error of the a-th row in the result matrix T of historical data calculation, a = 1, 2,..., N 4 ; is the element of the a-th row and the k-th column in the result matrix T of historical data calculation;

[0085] Step 6.2: Let G h (1) = 0, and use the recurrence formula to calculate the cumulative sum G h (a) of the squared prediction error of each row in the result matrix T of historical data calculation: Take the maximum value of G h (a) as the fault threshold H of the wind turbine;

[0086]

[0087] H = max{G h (a), a ∈ [1, N 4}

[0088] Among them, G h (a) is the cumulative sum of the squared prediction error of the a-th row in the result matrix T of historical data calculation; δ is the allowable deviation, μ f is the mean value of the cumulative sum of the squared prediction error when the wind turbine is in the fault state. Here, the fault should be detected when it exceeds the early fault stage. Therefore, set μ f = 110%μ h ; that is, δ = 5%μ h ;

[0089] Step 6.3: Calculate the remaining principal components of the result matrix T based on real-time data, and calculate the squared prediction error of each row in the result matrix T test of real-time data calculation test in the result matrix T

[0090]

[0091] Among them, is the square prediction error of the a-th row in the real-time data calculation result matrix T test in the test a-th row, where a test = 1, 2,..., N 3 ; is the element in the a-th row and k-th column of the real-time data calculation result matrix T test in the test a-th row and k-th column;

[0092] Step 6.4: Let G test (1) = 0, and calculate the cumulative sum G test (a test ) of the square prediction errors of each row in the real-time data calculation result matrix T test using the recurrence formula:

[0093]

[0094] If until a test = N 3 , G test (a test ) is less than the fault threshold H, it is determined that the wind turbine has not failed;

[0095] If when a test = b, G test (b) is greater than or equal to the fault threshold H, it is determined that the wind turbine has failed. Continue to calculate Gn = G test (N 3 ) using the recurrence formula, and then execute Step 7 to calculate the fault amplitude and determine the severity of the wind turbine fault;

[0096] Step 7: Calculate the fault amplitude based on the cumulative sum Gn of the square prediction errors of the wind turbine under the current working condition

[0097] For a single fault, when a test = b, G test (b) being greater than or equal to the fault threshold H means that within the sampling duration N, the fault starts at the b-th sampling point of the j-th variable, and the last N - b observed values are affected by the fault:

[0098]

[0099]

[0100] where the superscript " * " indicates noise removal; is the element in the a-th row and j-th column of the matrix X h after noise removal, where a = 1,..., N 4, where \(j = 1, 2, 3\); is a matrix the element in the \(a\)-th row and \(i\)-th column after removing noise; \(v\) aj the element in the \(a\)-th row and \(j\)-th column of the noise signal matrix; \(p\) jc the element in the \(j\)-th row and \(c\)-th column of the eigenvector matrix \(P\); is the matrix \(X\) h the mean value of all elements in the \(j\)-th column after removing noise; is the matrix \(X\) h the standard deviation of all elements in the \(j\)-th column after removing noise;

[0101] If it is considered that the wind turbine is in the early fault stage. As time goes by, the severity of the fault will gradually increase, and it is necessary to monitor the development of the fault and perform regular maintenance;

[0102] If it is considered that the wind turbine has a relatively serious fault, and it should be shut down for repair or component replacement immediately.

[0103] The present invention applies principal component analysis and the squared prediction error cumulative sum algorithm to the monitoring data of the wind turbine SCADA system, which can realize the detection of early faults in the generator system of the wind turbine and can issue an early warning earlier than the SCADA system. The present invention effectively solves the problems of many variables and large data dimensions in the SCADA data acquisition system by using principal component analysis, reduces the operation cost, and improves the fault detection efficiency. The fault amplitude estimation model derived from the fault detection algorithm proposed by the present invention can realize the amplitude estimation of the generator fault of the wind turbine, which is helpful for evaluating the severity of the fault.

[0104] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for online detection of early faults of wind turbines based on PCA-SPE-CUSUM algorithm, characterized in that: The following steps are involved: Step 1: Sample the wind turbine under the current working condition within a period of time, obtain the wind turbine status information at each sampling moment, including stator winding temperature, non-drive end bearing temperature, drive end bearing temperature, active power and cabin temperature, and construct the wind turbine status data set X under the current working condition. in ; Step 2: Sampling from the historical time series monitoring data set of the wind turbine in normal working state, the sampling duration is consistent with step 1, the state information of the wind turbine at each sampling moment is obtained, and the normal working state historical data set X0 of the wind turbine is constructed; Step 3: For dataset X in Preprocess with X0 to remove the data in the time period containing missing values ​​and the data in the time period corresponding to zero active power; For the preprocessed dataset X in The difference between the stator winding temperature, the non-drive end bearing temperature, the drive end bearing temperature and the cabin temperature is calculated with the samples in X0, and the results are constructed as the data set X test With X h ; Step 4: For dataset X test With X h Decentralize and standardize to obtain a data set and Step 5: Use principal component analysis to analyze the data set Project to the new mapping space and construct the real-time data calculation result matrix T test , determine the matrix T test The principal component space and the remaining principal component space; Step 6: Based on the remaining principal components of the historical data calculation result matrix T, calculate the square prediction error and its cumulative sum of each row in the historical data calculation result matrix T, and determine the fault threshold H; Calculate the result matrix T based on real-time data test The remaining principal components of the real-time data are calculated and the result matrix T is calculated. test The square prediction error and its cumulative sum of each row in are used to determine whether the wind turbine has a fault according to the fault threshold.

2. The method for online detection of early faults of wind turbine generator systems based on the PCA-SPE-CUSUM algorithm according to claim 1, characterized in that: In step 1, the state data set X of the wind turbine generator set under the current working condition in for: in, Indicates that the wind turbine is under the current working condition The state information at the sampling time includes the stator winding temperature T1, the non-drive end bearing temperature T2, the drive end bearing temperature T3, the active power P0 and the cabin temperature T0; N1 is the data set X in The number of samples in .

3. The method for online detection of early faults of wind turbine generator systems based on PCA-SPE-CUSUM algorithm according to claim 2, characterized in that: The wind turbine normal working status data set X0 in step 2 is: in, Indicates that the wind turbine is in normal working condition The state information at the sampling time, N2 is the number of samples in the data set X0, N is the sampling duration.

4. The method for online detection of early faults of wind turbine generator sets based on PCA-SPE-CUSUM algorithm according to claim 3 is characterized in that: In step 3, the dataset X test With X h for: Among them, N3 and N4 are the preprocessed data sets X in with the number of samples in X0.

5. The method for online detection of early faults of wind turbine generator sets based on PCA-SPE-CUSUM algorithm according to claim 4 is characterized in that: The step 5 is specifically as follows: Step 5.1: Based on the dataset Construct the covariance matrix S; Step 5.2: Perform eigenvalue decomposition on the covariance matrix S and obtain the eigenvalues ​​(λ1,λ2,...,λ m ) and the eigenvector matrix P; Step 5.3: Based on the eigenvector matrix P and the data set Construct real-time data calculation result matrix T test ; Step 5.4: Calculate the real-time data result matrix T test The first l columns are used as principal components, and the remaining ml columns are used as residual principal components; where l is calculated as: Calculate each eigenvalue λ d Contribution rate CPV(λ d ), d = 1, ..., m, select the largest first l contribution rates and the sum of the first l contribution rates is greater than 90%; 6. The method for online detection of early faults of wind turbine generator systems based on PCA-SPE-CUSUM algorithm according to claim 5, characterized in that: The step 6 is specifically as follows: Step 6.1: Based on the remaining principal components of the historical data calculation result matrix T, calculate the squared forecast error of each row in the historical data calculation result matrix T Get The mean μ h ; in, The squared forecast error of the ath row in the historical data calculation result matrix T, a = 1, 2, ..., N4; The element in the ath row and the kth column of the historical data calculation result matrix T; Step 6.2: Let G h (1) = 0, and the recursive formula is used to calculate the cumulative sum G of the squared forecast errors of each row in the historical data calculation result matrix T h (a): Take G h The maximum value of (a) is taken as the fault threshold H of the wind turbine; H=max{G h (a),a∈[1,N4]} Among them, G h (a) is the cumulative sum of the squared forecast errors of the ath row in the historical data calculation result matrix T; δ is the allowable deviation, δ = 5% μ h ; Step 6.3: Calculate the result matrix T based on real-time data test The remaining principal components of the real-time data are calculated and the result matrix T is calculated. test The squared prediction error for each row in in, Calculate the result matrix T for real-time data test Middle a test The squared prediction error of the row, a test =1,2,...,N3; Calculate the result matrix T for real-time data test Middle a test The element in the row and column k; Step 6.4: Let G test (1) = 0, and the real-time data calculation result matrix T is calculated using the recursive formula test The cumulative sum of squared prediction errors for each row in G test (a test ): If until a test =N3,G test (a test ) are both less than the fault threshold H, it is determined that the wind turbine has no fault; If a test = b, G test (b) If the value is greater than or equal to the fault threshold H, it is determined that the wind turbine generator set has a fault.

7. The method for online detection of early faults of wind turbine generator systems based on PCA-SPE-CUSUM algorithm according to claim 6, characterized in that: In step 6.4, if a test = b, G test (b) If it is greater than or equal to the fault threshold, it is determined that the wind turbine has a fault, and Gn=G is calculated by the recursive formula. test (N3) After that, the fault amplitude is calculated based on the cumulative sum Gn of the square prediction error of the wind turbine under the current working condition. Determine the severity of wind turbine failure; Among them, the superscript "*" indicates noise removal; is the matrix X h The element in the ath row and jth column after removing the noise, a=1,...,N4, j=1,2,3; For the matrix The element in the ath row and the ith column after removing the noise; v aj is the element in the ath row and jth column of the noise signal matrix; p jc is the element in the jth row and cth column of the eigenvector matrix P; is the matrix X h The mean of all elements in the jth column after removing noise; is the matrix X h The standard deviation of all elements in the jth column after removing noise; like It is considered that the wind turbine is in the early stage of failure. As time goes by, the severity of the failure will gradually increase, and it is necessary to monitor the development of the failure and perform regular maintenance. like It is considered that the wind turbine has a serious fault and should be shut down immediately for repair or component replacement.

8. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images