A method and system for diagnosing the health status of wind turbine components

Abstract

The present invention proposes a method and system for diagnosing the health status of wind turbine components, which relates to the technical field of wind turbines. The method includes selecting input feature data and component reconstruction values ​​to calculate the initial health status of the components; calculating the adaptive threshold of the components through the initial health status; then determining the component energy efficiency level and the health status range corresponding to the component energy efficiency level through the constructed energy efficiency level judgment condition; based on the health status range, constructing a normalized function corresponding to the component energy efficiency, the normalized function is used to calculate the normalized health status of the component; based on the normalized health status, calculating the health status of the wind turbine component, the health status of the wind turbine component is used to diagnose the energy efficiency level of the wind turbine component. The present invention can ensure that the health status of the component is consistent with the energy efficiency level, so that when the component energy efficiency diagnosis level is the same, the value range of the health status corresponding to the component is also the same, thereby improving the diagnostic accuracy of the health status of the component.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbines, and in particular to a method and system for diagnosing the health status of wind turbine components. Background Art

[0002] With the increasing attention paid to energy consumption and the increasing requirements for energy efficiency, energy efficiency diagnosis of wind turbine components plays a key role in improving wind turbine performance and reducing energy consumption. This diagnostic technology monitors and analyzes the health status and performance indicators of each component of the wind turbine, identifies problematic components and provides repair or replacement suggestions. Common wind turbine component diagnosis includes blade diagnosis, bearing diagnosis, gearbox diagnosis, etc. Compared with methods that require the installation of sensors to monitor the health status of wind turbine components, such as status monitoring and audio monitoring of wind turbine status monitoring systems, the health status diagnosis method of components based only on Supervisory Control and Data Acquisition (SCADA) analysis is both economical and convenient, and has outstanding advantages.

[0003] The mainstream method for analyzing component status based on SCADA data is to use neural network methods, support vector regression methods, denoising autoencoder methods and their improved methods to evaluate the health of components. However, judging the energy efficiency health of components based only on health may not be accurate, because the correlation between comparison model selection, training features and target values ​​can better determine the size of health. To solve the above problems, the prior art also proposes to use an adaptive threshold method to evaluate the health of components based on health, but this method has a problem. When the energy efficiency diagnosis level of the same component is low, the assessed health of the component will be higher, which is prone to misdiagnosis of the health status of wind turbine components. Summary of the invention

[0004] In order to solve the problem in the above-mentioned prior art that the component health and energy efficiency level are inconsistent, resulting in low accuracy in diagnosing the health status of the components, the present invention proposes a health status diagnosis method and system for wind turbine components, which can ensure that the component health and energy efficiency level are consistent, so that when the component energy efficiency diagnosis level is the same, the value range of the health level corresponding to the component is also the same, thereby improving the diagnostic accuracy of the component health status.

[0005] In order to achieve the above technical effects, the technical solution of the present invention is as follows:

[0006] A method for diagnosing the health status of a wind turbine component comprises the following steps:

[0007] Inputting component data and operating condition data related to a wind turbine component as input feature data into a pre-trained denoising autoencoder, and outputting a component reconstruction value by the denoising autoencoder, wherein the wind turbine component includes a plurality of components;

[0008] constructing energy efficiency level judgment conditions based on the input feature data and the component reconstruction value;

[0009] Based on the energy efficiency level judgment condition, determine the energy efficiency level of the component and the health level range corresponding to the energy efficiency level of the component;

[0010] Based on the health range, constructing a normalized function corresponding to the component energy efficiency, wherein the normalized function is used to calculate the normalized health of the component;

[0011] Based on the normalized health, a wind turbine component health is calculated, and the wind turbine component health is used to diagnose a wind turbine component level representing a health status of the wind turbine component.

[0012] Preferably, the energy efficiency level judgment condition is constructed based on the input feature data and the component reconstruction value, including:

[0013] Calculating the initial health of the component using the input feature data and the component reconstruction value;

[0014] Calculating an adaptive threshold of a component using the initial health;

[0015] The energy efficiency level determination condition is set according to the adaptive threshold.

[0016] Preferably, the input feature data is X(x1, x2, ..., x i ), the component reconstruction value is X'(x'1,x'2,…,x' i ), based on X(x1,x2,…,x i ) and X'(x'1,x'2,…,x' i ), the expression for calculating the initial health is as follows:

[0017] R j =1-|x j -x′ j |

[0018] Among them, R j represents the initial health of the jth component, x j Represents the input feature data X(x1, x2, ..., x i ), where i represents a positive integer, 1≤j≤i, x' jrepresents the reconstruction value X'(x'1,x'2,…,x' i ) is the j-th input element in .

[0019] Preferably, based on the initial health, the adaptive threshold of the component is calculated by fitting the extreme value distribution function Pareto distribution, and the calculation expression of the adaptive threshold is as follows:

[0020]

[0021] Among them, T represents the adaptive threshold, P represents the initial threshold, and the value is the initial health R j The preset percentile of , σ represents the scale parameter, γ represents the shape parameter, and n represents the initial health R j The number of data samples, q represents the probability of extreme events, N t Indicates the number of samples in the sample data of initial health that are greater than the initial threshold P.

[0022] Preferably, the energy efficiency level judgment conditions include:

[0023] The proportion of the number of samples greater than the adaptive threshold T in the sample data of the initial health degree to the total sample data of the initial health degree is set to Pr;

[0024] If the ratio Pr∈Pr i , then the energy efficiency level of the component is judged to be L i , the energy efficiency level L i Corresponding health range R i for The expression of the energy efficiency level judgment condition is as follows:

[0025]

[0026] Among them, Pr i is the ratio range corresponding to the energy efficiency level, and its value is is the minimum ratio range corresponding to the energy efficiency level, is the maximum ratio range corresponding to the energy efficiency level, Energy efficiency level L i The corresponding minimum health range, Energy efficiency level L i The corresponding maximum health range.

[0027] Preferably, the determining, based on the energy efficiency level judgment condition, the component energy efficiency level and the health range corresponding to the component energy efficiency level comprises:

[0028] The proportion Pr of the number of samples greater than the adaptive threshold T in the sample data of the initial health to the total sample data of the initial health is calculated as follows:

[0029]

[0030] Among them, R j,k Represents the initial health R j (R j,1 ,R j,2 ,…,R j,n ), T j,k represents the adaptive threshold T(T j,1 ,T j,2 ,…,T j,n ), k∈(1,n);

[0031] Based on the statistical ratio Pr∈Pr i , query the energy efficiency level judgment condition, and obtain the energy efficiency level of the component as L i , the energy efficiency level L i Corresponding health range R i for

[0032] Preferably, the health range is based on The expression for constructing the normalized function corresponding to the energy efficiency of the components is as follows:

[0033]

[0034] Among them, R' j represents the normalized health of the jth component; max(R j )=max(R j,1 ,R j,2 ,…,R j,n ), min(R j )=min(R j,1 ,R j,2 ,…,R j,n ).

[0035] Preferably, the expression for calculating the health of wind turbine components based on the normalized health is as follows:

[0036]

[0037] Wherein, R represents the health of the wind turbine components, w k represents the weight, m ​​represents the number of sliding windows, R' k represents the normalized health of the kth component.

[0038] Preferably, the diagnosing the wind turbine component level representing the health status of the wind turbine component comprises:

[0039] According to the health level R of the wind turbine component, the health level range R to which the health level R of the wind turbine component belongs is queried in the energy efficiency level judgment condition. i ;

[0040] Based on the health range R to which the health level R of the wind turbine component belongs i , diagnose wind turbine component levels that characterize the health of wind turbine components.

[0041] The present invention also proposes a health status diagnosis system for wind turbine components, comprising:

[0042] A data selection module, used for inputting component data and operating condition data related to a wind turbine component as input feature data into a pre-trained denoising autoencoder, and outputting a component reconstruction value by the denoising autoencoder, wherein the wind turbine component includes a plurality of components;

[0043] An energy efficiency level judgment condition construction module, used to construct an energy efficiency level judgment condition according to the input feature data and the component reconstruction value;

[0044] An energy efficiency level determination module, used to determine the energy efficiency level of a component and a health range corresponding to the energy efficiency level of the component according to the energy efficiency level judgment condition;

[0045] A normalized health calculation module, used to construct a normalized function corresponding to the energy efficiency of the component according to the health range, and the normalized function is used to calculate the normalized health of the component;

[0046] The diagnosis module is used to calculate the health of the wind turbine component according to the normalized health, wherein the health of the wind turbine component is used to diagnose the wind turbine component level representing the health status of the wind turbine component.

[0047] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0048] The present invention proposes a method and system for diagnosing the health status of wind turbine components. Firstly, aiming at the concept that the value range of health status is the same for the same energy efficiency diagnosis level, it is proposed to determine the energy efficiency level of the component and the health status range corresponding to the energy efficiency level of the component by constructing energy efficiency level judgment conditions, so that the on-site operators can effectively identify the component or the energy efficiency status of the component according to the health status range, and then calculate the normalized health status of the component by constructing a normalized function corresponding to the energy efficiency of the component, and calculate the health status of the wind turbine component by the normalized health status, and further use the health status of the wind turbine component to diagnose the energy efficiency level of the wind turbine component, which can ensure that the health status of the component is consistent with the energy efficiency level, so that when the energy efficiency diagnosis level of the component is the same, the value range of the health status corresponding to the component is also the same, thereby improving the diagnostic accuracy of the health status of the component. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 A flow chart showing a method for diagnosing the health status of a wind turbine component provided in an embodiment of the present invention;

[0050] Figure 2 A flow chart showing conditions for establishing energy efficiency level determination provided in an embodiment of the present invention;

[0051] Figure 3 A graph showing the health of the generator bearing temperature and its corresponding adaptive threshold provided in an embodiment of the present invention;

[0052] Figure 4 A block diagram of a health status diagnosis system for wind turbine components provided in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0053] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0054] It is understandable to those skilled in the art that some well-known contents may be omitted in the drawings;

[0055] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be understood that the step numbers used in the text are only for the convenience of description and are not used as a limitation on the order of execution of the steps.

[0056] Example 1

[0057] like Figure 1As shown, this embodiment proposes a method for diagnosing the health status of wind turbine components, comprising the following steps:

[0058] S1. Inputting component data and operating condition data related to a wind turbine component as input feature data into a pre-trained denoising autoencoder, and outputting a component reconstruction value by the denoising autoencoder, wherein the wind turbine component includes a plurality of components;

[0059] In step S1, the denoising autoencoder is trained and tested, including:

[0060] Acquire input feature data of the denoising autoencoder, divide the input feature data into a training set and a test set, and train the denoising autoencoder using the training set to obtain a trained denoising autoencoder;

[0061] In the process of training the denoising autoencoder, firstly, component data and operating condition data related to wind turbine components are selected as input feature data of the denoising autoencoder to evaluate the health status of wind turbine components, the wind turbine component is a generator, the component data include the generator drive end bearing temperature, the generator non-drive end bearing temperature, the generator stator U winding temperature, the generator stator V winding temperature and the generator stator W winding temperature, the operating condition data include the ambient temperature, wind speed, active power, and the input feature data is dimensionless processed by the Min-max normalization method; secondly, the component data and operating condition data in the input feature data are selected as the training set when the wind turbine components are in normal working state; then the minimum reconstruction error of the denoising autoencoder is calculated by using the back propagation algorithm and the optimization convergence method Minimum reconstruction error The calculation expression is as follows:

[0062]

[0063] Among them, the minimum reconstruction error is the MSE error, Y represents the input feature data of the denoising autoencoder during the training process, and Y' represents the reconstructed data output by the denoising autoencoder after training;

[0064] The denoising autoencoder is tested using the test set to obtain an optimal denoising autoencoder. S2. Based on the input feature data and the component reconstruction value, an energy efficiency level judgment condition is constructed;

[0065] In step S2, the energy efficiency level judgment condition is constructed based on the input feature data and the component reconstruction value, see Figure 2 ,include:

[0066] S21. Calculating the initial health of a component using the input feature data and the component reconstruction value;

[0067] In S21, the input feature data is selected as X(x1, x2, ..., x i ), the component reconstruction value is X'(x'1,x'2,…,x' i ),

[0068] The expression for calculating the initial health of a component is as follows:

[0069] R j =1-|x j -x′ j |

[0070] Among them, R j represents the initial health of the jth component, x j Represents the input feature data X(x1, x2, ..., x i ), where i represents a positive integer, 1≤j≤i, x' j represents the reconstructed value X'(x'1, x'2, ..., x' i ) in the jth input element;

[0071] S22. Based on the initial health, calculating the adaptive threshold of the component;

[0072] In step S22, since the working effects of various components of the wind turbine may change with time and environment, if the threshold is set manually, the user needs to have sufficient data usage experience, and this setting method is very random. Therefore, an adaptive threshold method based on extreme value theory (Drift streaming Peaks-Over_threshold, DSPOT) is proposed as a method for judging whether the characteristics of wind turbine components are in an abnormal state. DSPOT calculates the extreme value under a specified probability by fitting the extreme value distribution, and this extreme value is judged as an abnormal threshold. The specific process is as follows:

[0073] First, the probability value q and the number of sliding windows m are set; then based on the initial health R of the wind turbine components under normal operating conditions j (R j,1 ,R j,2 ,…,R j,n ),

[0074] The initial health degree is calculated by fitting the extreme value distribution function Pareto distribution to calculate the adaptive threshold value T (T j,1 ,T j,2 ,…,T j,n ), the calculation expression of the adaptive threshold is as follows:

[0075]

[0076] Among them, T represents the adaptive threshold, P represents the initial threshold, and the value is the initial health R j , σ represents the scale parameter, γ represents the shape parameter, and n represents the initial health R j The number of data samples, q represents the probability of extreme events, N t The number of samples in the sample data representing the initial health that are greater than the initial threshold P. The parameters σ and γ are calculated using the maximum likelihood estimation and GrimShaw strategy; see Figure 3 , Figure 3 Displays the health of the generator bearing temperature and its corresponding adaptive threshold within a day.

[0077] Since data may change over time, the local mean of samples in the sliding window is used to update the adaptive threshold model.

[0078] If the new sample data value is greater than T, it is judged as an outlier, the model distribution is not updated, and the window average value is not updated;

[0079] If the new sample data value is less than or equal to T, it is judged as an outlier, the model distribution is updated, and the window average is updated;

[0080] If the new sample data value is greater than P, the model distribution is updated and the window continues to slide forward for calculation;

[0081] If the new sample data value is less than or equal to P, the model distribution is not updated and the window continues to slide forward for calculation.

[0082] S23. According to the adaptive threshold, set the energy efficiency level judgment condition.

[0083] In step S23, the energy efficiency level judgment conditions include:

[0084] The proportion of the number of samples greater than the adaptive threshold T in the sample data of the initial health degree to the total sample data of the initial health degree is set to Pr;

[0085] If the ratio Pr∈Pr i , then the energy efficiency level of the component is judged to be L i , the energy efficiency level L i Corresponding health range R i for The expression of the energy efficiency level judgment condition is as follows:

[0086]

[0087] Among them, Pri is the ratio range corresponding to the energy efficiency level, and its value is is the minimum ratio range corresponding to the energy efficiency level, is the maximum ratio range corresponding to the energy efficiency level, Energy efficiency level L i The corresponding minimum health range, Energy efficiency level L i The corresponding maximum health range.

[0088] S3. Based on the energy efficiency level judgment condition, determine the energy efficiency level of the component and the health range corresponding to the energy efficiency level of the component;

[0089] In step S3, the determining of the component energy efficiency level and the health range corresponding to the component energy efficiency level based on the energy efficiency level judgment condition includes:

[0090] The proportion Pr of the number of samples greater than the adaptive threshold T in the sample data of the initial health to the total sample data of the initial health is calculated as follows:

[0091]

[0092] Among them, R j,k Represents the initial health R j (R j,1 ,R j,2 ,…,R j,n ), T j,k represents the adaptive threshold T(T j,1 ,T j,2 ,…,T j,n ), k∈(1,n);

[0093] Calculate the total health of the components within the preset time series r j as follows:

[0094]

[0095] Based on the statistical ratio Pr∈Pr i , query the energy efficiency level judgment condition, and obtain the energy efficiency level of the component as L i , the energy efficiency level L i Corresponding health range R i for

[0096] S4. Based on the health range, construct a normalized function corresponding to the component energy efficiency, wherein the normalized function is used to calculate the normalized health of the component;

[0097] In step S4, the health range is based on The expression for constructing the normalized function corresponding to the energy efficiency of the components is as follows:

[0098]

[0099] Among them, R' j represents the normalized health of the jth component; max(R j )=max(R j,1 ,R j,2 ,…,R j,n ), min(R j )=min(R j,1 ,R j,2 ,…,R j,n );

[0100] Using the normalization function, calculate the normalized health R' of the jth component j , using the normalized health R' j Normalize the total health r j .

[0101] S5. Based on the normalized health, calculate the health of the wind turbine component, wherein the health of the wind turbine component is used to diagnose a wind turbine component level representing a health status of the wind turbine component.

[0102] In step S5, the expression for calculating the health of wind turbine components is as follows:

[0103]

[0104] Wherein, R represents the health of the wind turbine components, w k represents the weight, m ​​represents the number of sliding windows, R' k Represents the normalized health of the kth component; the corresponding weight is calculated according to the frequency of each component in the fault record

[0105] The diagnostic characterizes the health status of the wind turbine component at the wind turbine component level, including:

[0106] According to the health level R of the wind turbine component, the health level range R to which the health level R of the wind turbine component belongs is queried in the energy efficiency level judgment condition. i ;

[0107] Based on the health range R to which the health level R of the wind turbine component belongs i , diagnose wind turbine component levels that characterize the health of wind turbine components.

[0108] In this embodiment, firstly, aiming at the same energy efficiency diagnosis level and the same value range of health degree, it is proposed to determine the energy efficiency level of components and the health degree range corresponding to the energy efficiency level of the components by constructing energy efficiency level judgment conditions, so that the operators at the site can effectively identify the energy efficiency status of components or components according to the health degree range, and then calculate the normalized health degree of the components by constructing a normalized function corresponding to the energy efficiency of the components, and calculate the health degree of wind turbine components by the normalized health degree, and further use the health degree of the wind turbine components to diagnose the energy efficiency level of the wind turbine components, which can ensure that the health degree of the components is consistent with the energy efficiency level, so that when the energy efficiency diagnosis level of the components is the same, the value range of the health degree corresponding to the components is also the same, thereby improving the diagnostic accuracy of the health status of the components.

[0109] Example 2

[0110] This embodiment takes the MY1.5 wind turbine in a certain wind farm as an example to further explain the health status diagnosis method of the components of the MY1.5 wind turbine.

[0111] SCADA data of component data and operating condition data related to MY1.5 wind turbine components with a time resolution of 1 minute in a certain year. The component data includes the generator drive end bearing temperature, the generator non-drive end bearing temperature, the generator stator U winding temperature, the generator stator V winding temperature and the generator stator W winding temperature, and the operating condition information includes the ambient temperature, wind speed and active power.

[0112] In the first step, component data and operating condition data related to the wind turbine components are input into the pre-trained denoising autoencoder as input feature data of the pre-trained denoising autoencoder, and the denoising autoencoder outputs component reconstruction values;

[0113] In the first step, assume that the input feature data is X(x1, x2, ..., x8), the trained denoising autoencoder is H_DAE, and the trained denoising autoencoder outputs the component reconstruction value X′(x1', x2', ..., x8').

[0114] The second step is to construct energy efficiency level judgment conditions based on the input feature data and the component reconstruction value; the specific steps of the second step are as follows:

[0115] Firstly, the initial health of the component is calculated using the input feature data and the component reconstruction value;

[0116] In this step, the initial health of the component is calculated using the input feature data and the component reconstruction value in the training set as follows:

[0117] Rtrain =1-|Y-Y'|

[0118] Using the input feature data and the component reconstruction value in the test set, the initial health of the component is calculated as follows:

[0119] R test =1-|XX′|

[0120] One day of online SCADA data is collected according to the time dimension of 1 minute, and data with the same characteristics as the training data set is extracted as the input feature data X(x1,x2,…,x8) of the test. The denoising autoencoder H_DAE obtained through training obtains the reconstruction value X'(x1',x2',…,x8') of the parts.

[0121] Then, for the initial health of any component, R test,k (k∈[1,5]), setting the outlier probability to 3% and the number of sliding windows to 5, we get the adaptive threshold T.

[0122] Next, according to the adaptive threshold, an energy efficiency level judgment condition is constructed;

[0123]

[0124] The third step is to determine the energy efficiency level of the component and the health range corresponding to the energy efficiency level of the component based on the energy efficiency level judgment condition;

[0125] In the third step, the step of determining the component energy efficiency level and the health range corresponding to the component energy efficiency level includes:

[0126] (1) The number of samples in the statistical test data set that are greater than the adaptive threshold accounts for the proportion of the total sample data in the test data set, which is Pr = 5.37%;

[0127] (2) Calculate the health of components within a day

[0128] R 发电机驱动端轴承温度 =0.973

[0129] (3) Determine the grade of components:

[0130] Pr = 5.37%, the proportion range is Pr 2 ∈(5%,15%], then the component level is L 2 , the health range is R 2 ∈(0.75,0.85];

[0131] Step 4: Based on the health range R 2∈(0.75,0.85], the normalized function corresponding to the energy efficiency of the components is constructed as follows:

[0132]

[0133] Calculate the normalized health R' of the component using the normalized function 发电机驱动端轴承温度 ;

[0134] Step 5: Set the weight to w k =0.2, (k∈[1,5]). Except for the generator drive end bearing temperature health of 0.84, the health of the remaining components is 0.92, 0.83, 0.89, and 0.7 respectively. Based on the normalized health, the weighted calculation of the health of the wind turbine components is:

[0135]

[0136] Using the health degree R=0.836 of the wind turbine component, the energy efficiency level judgment condition is queried. The health degree range of R=0.836 is R 2 ∈(0.75,0.85], the energy efficiency level of the wind turbine components is diagnosed as average.

[0137] In this embodiment, firstly, aiming at the same energy efficiency diagnosis level and the same value range of health degree, it is proposed to determine the energy efficiency level of components and the health degree range corresponding to the energy efficiency level of the components by constructing energy efficiency level judgment conditions, so that the operators at the site can effectively identify the energy efficiency status of components or components according to the health degree range, and then calculate the normalized health degree of the components by constructing a normalized function corresponding to the energy efficiency of the components, and calculate the health degree of wind turbine components by the normalized health degree, and further use the health degree of the wind turbine components to diagnose the energy efficiency level of the wind turbine components, which can ensure that the health degree of the components is consistent with the energy efficiency level, so that when the energy efficiency diagnosis level of the components is the same, the value range of the health degree corresponding to the components is also the same, thereby improving the diagnostic accuracy of the health status of the components.

[0138] Example 3

[0139] See also Figure 4 This embodiment provides a health status diagnosis system for wind turbine components, including:

[0140] A data selection module, used for inputting component data and operating condition data related to a wind turbine component as input feature data into a pre-trained denoising autoencoder, and outputting a component reconstruction value by the denoising autoencoder, wherein the wind turbine component includes a plurality of components;

[0141] An energy efficiency level judgment condition construction module, used to construct an energy efficiency level judgment condition according to the input feature data and the component reconstruction value;

[0142] An energy efficiency level determination module, used to determine the energy efficiency level of a component and a health range corresponding to the energy efficiency level of the component according to the energy efficiency level judgment condition;

[0143] A normalized health calculation module, used to construct a normalized function corresponding to the energy efficiency of the component according to the health range, and the normalized function is used to calculate the normalized health of the component;

[0144] The diagnosis module is used to calculate the health of the wind turbine component according to the normalized health, wherein the health of the wind turbine component is used to diagnose the wind turbine component level representing the health status of the wind turbine component.

[0145] In this embodiment, firstly, aiming at the same energy efficiency diagnosis level and the same value range of health degree, it is proposed to determine the energy efficiency level of components and the health degree range corresponding to the energy efficiency level of the components by constructing energy efficiency level judgment conditions, so that the operators at the site can effectively identify the energy efficiency status of components or components according to the health degree range, and then calculate the normalized health degree of the components by constructing a normalized function corresponding to the energy efficiency of the components, and calculate the health degree of wind turbine components by the normalized health degree, and further use the health degree of the wind turbine components to diagnose the energy efficiency level of the wind turbine components, which can ensure that the health degree of the components is consistent with the energy efficiency level, so that when the energy efficiency diagnosis level of the components is the same, the value range of the health degree corresponding to the components is also the same, thereby improving the diagnostic accuracy of the health status of the components.

[0146] Obviously, the above embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for diagnosing the health status of wind turbine components, characterized in that: The following steps are involved: Inputting component data and operating condition data related to a wind turbine component as input feature data into a pre-trained denoising autoencoder, and outputting a component reconstruction value by the denoising autoencoder, wherein the wind turbine component includes a plurality of components; Based on the input feature data and the component reconstruction value, an energy efficiency level judgment condition is constructed, including: Calculating the initial health of the component using the input feature data and the component reconstruction value; Based on the initial health, calculating an adaptive threshold value of the component; According to the adaptive threshold, setting the energy efficiency level judgment condition; The input feature data is X(x1, x2, ..., x i ), the component reconstruction value is X′(x′1, x′2, …, x′ i ), based on X(x1, x2, …, x i ) and X′(x′1, x′2, …, x′ i ), the expression for calculating the initial health is as follows: R j =1-|x j -x′ j | Among them, R j represents the initial health of the jth component, x j Represents the input feature data X(x1, x2, ..., x i ), where i represents a positive integer, 1≤j≤i, x′ j represents the reconstruction value X′(x′1, x′2, …, x′ i ) in the jth input element; Based on the initial health, the adaptive threshold of the component is calculated by using the fitted extreme value distribution function Pareto distribution. The calculation expression of the adaptive threshold is as follows: Among them, T represents the adaptive threshold, P represents the initial threshold, and the value is the initial health R j The preset percentile of , σ represents the scale parameter, γ represents the shape parameter, and n represents the initial health R j The number of data samples, q represents the probability of extreme events, N t The number of samples in the sample data representing the initial health that are greater than the initial threshold value P; Based on the energy efficiency level judgment condition, determine the energy efficiency level of the component and the health level range corresponding to the energy efficiency level of the component; Based on the health range, constructing a normalized function corresponding to the component energy efficiency, wherein the normalized function is used to calculate the normalized health of the component; Based on the normalized health, a wind turbine component health is calculated, and the wind turbine component health is used to diagnose a wind turbine component level representing a health status of the wind turbine component.

2. The method for diagnosing the health status of wind turbine components according to claim 1, characterized in that: The energy efficiency level judgment conditions include: The proportion of the number of samples greater than the adaptive threshold T in the sample data of the initial health degree to the total sample data of the initial health degree is set to Pr; If the ratio Pr∈Pr i , then the energy efficiency level of the component is judged to be L i , the energy efficiency level L i Corresponding health range R i for The expression of the energy efficiency level judgment condition is as follows: Among them, Pr i is the ratio range corresponding to the energy efficiency level, and its value is is the minimum ratio range corresponding to the energy efficiency level, is the maximum ratio range corresponding to the energy efficiency level, Energy efficiency level L i The corresponding minimum health range, Energy efficiency level L i The corresponding maximum health range.

3. The method for diagnosing the health status of wind turbine components according to claim 2, characterized in that: The determining, based on the energy efficiency level judgment condition, the component energy efficiency level and the health level range corresponding to the component energy efficiency level includes: The proportion Pr of the number of samples greater than the adaptive threshold T in the sample data of the initial health to the total sample data of the initial health is calculated as follows: Among them, R j,k Represents the initial health R j (R j,1 ,R j,2 ,…,R j,n ), T j,k represents the adaptive threshold T(T j,1 ,T j,2 ,…,T j,n ), k∈(1,n); Based on the statistical ratio Pr∈Pr i , query the energy efficiency level judgment condition, and obtain the energy efficiency level of the component as L i , the energy efficiency level L i Corresponding health range R i for 4. The method for diagnosing the health status of wind turbine components according to claim 3, characterized in that: Based on the health range The expression for constructing the normalized function corresponding to the energy efficiency of the components is as follows: Among them, R' j represents the normalized health of the jth component; max(R j )=max(R j,1 ,R j,2 ,…,R j,n ), min(R j )=min(R j,1 ,R j,2 ,…,R j,n ).

5. The method for diagnosing the health status of wind turbine components according to claim 4, characterized in that: The expression for calculating the health of wind turbine components based on the normalized health is as follows: Wherein, R represents the health of the wind turbine components, w k represents the weight, m ​​represents the number of sliding windows, R' k represents the normalized health of the kth component.

6. The method for diagnosing the health status of wind turbine components according to claim 5, characterized in that: The diagnostic characterizes the health status of the wind turbine component at the wind turbine component level, including: According to the health level R of the wind turbine component, the health level range R to which the health level R of the wind turbine component belongs is queried in the energy efficiency level judgment condition. i ; Based on the health range R to which the health level R of the wind turbine component belongs i , diagnose wind turbine component levels that characterize the health of wind turbine components.

7. A health status diagnosis system for wind turbine components, characterized in that: include: A data selection module, used for inputting component data and operating condition data related to a wind turbine component as input feature data into a pre-trained denoising autoencoder, and outputting a component reconstruction value by the denoising autoencoder, wherein the wind turbine component includes a plurality of components; The energy efficiency level judgment condition construction module is used to construct the energy efficiency level judgment condition according to the input characteristic data and the component reconstruction value, including: Calculating the initial health of the component using the input feature data and the component reconstruction value; Based on the initial health, calculating an adaptive threshold value of the component; According to the adaptive threshold, setting the energy efficiency level judgment condition; The input feature data is X(x1, x2, ..., x i ), the component reconstruction value is X'(x'1,x'2,…,x' i ), based on X(x1,x2,…,x i ) and X'(x'1,x'2,…,x' i ), the expression for calculating the initial health is as follows: R j =1-|x j -x′ j | Among them, R j represents the initial health of the jth component, x j Represents the input feature data X(x1, x2, ..., x i ), where i represents a positive integer, 1≤j≤i, x' j represents the reconstructed value X'(x'1, x'2, ..., x' i ) in the jth input element; Based on the initial health, the adaptive threshold of the component is calculated by using the fitted extreme value distribution function Pareto distribution. The calculation expression of the adaptive threshold is as follows: Among them, T represents the adaptive threshold, P represents the initial threshold, and the value is the initial health R j The preset percentile of , σ represents the scale parameter, γ represents the shape parameter, and n represents the initial health R j The number of data samples, q represents the probability of extreme events, N t The number of samples in the sample data representing the initial health that are greater than the initial threshold value P; An energy efficiency level determination module, used to determine the energy efficiency level of a component and a health range corresponding to the energy efficiency level of the component according to the energy efficiency level judgment condition; A normalized health calculation module, used to construct a normalized function corresponding to the energy efficiency of the component according to the health range, and the normalized function is used to calculate the normalized health of the component; The diagnosis module is used to calculate the health of the wind turbine component according to the normalized health, wherein the health of the wind turbine component is used to diagnose the wind turbine component level representing the health status of the wind turbine component.

Images