CN112611584B

CN112611584B - Fatigue failure detection method, device, equipment and medium for wind generating set

Info

Publication number: CN112611584B
Application number: CN202010421806.3A
Authority: CN
Inventors: 徐建波; 宋建军; 俞海国
Original assignee: Jiangsu Jinfeng Software Technology Co ltd; Qinghai Green Energy Data Co ltd; Beijing Goldwind Smart Energy Service Co Ltd
Current assignee: Jiangsu Jinfeng Software Technology Co ltd; Qinghai Green Energy Data Co ltd; Beijing Goldwind Smart Energy Service Co Ltd
Priority date: 2020-05-18
Filing date: 2020-05-18
Publication date: 2023-06-02
Anticipated expiration: 2040-05-18
Also published as: CN112611584A

Abstract

The application provides a fatigue failure detection method, device, equipment and medium for a wind generating set. The fatigue failure detection method comprises the following steps: acquiring historical fault information of a wind generating set to be predicted, and taking the historical fault information as the fault information to be predicted; predicting the fatigue failure condition of the wind generating set to be predicted according to the failure information to be predicted and the fatigue failure prediction model to obtain a first prediction result; acquiring historical transient data of the wind generating set to be predicted before a fault occurs, and taking the historical transient data as the historical transient data to be predicted; determining the aperiodic load characteristics of the wind generating set to be predicted under each aperiodic load stress born by the wind generating set; predicting fatigue failure conditions of the wind generating set to be predicted according to the aperiodic load characteristics to obtain a second prediction result; and obtaining a third prediction result according to the first prediction result and the second prediction result. The method and the device improve the identification accuracy of the wind generating set with higher fatigue failure risk.

Description

Fatigue failure detection method, device, equipment and medium for wind generating set

Technical Field

The application relates to the technical field of fatigue detection methods, in particular to a method, a device, equipment and a medium for detecting fatigue failure of a wind generating set.

Background

With the rapid development of wind power industry in recent years, the reliability of wind power generation equipment gradually becomes a focus of attention, the running environment of the wind power generation equipment is complex and severe, the fatigue damage problem of the wind power generation equipment is continuous, the damage condition occurs in the effective life period, and the maintenance cost of a wind power generator set is greatly increased.

The inventor finds that the cause of fatigue damage of wind power generation equipment mainly comprises the following points: 1. the equipment design production links have careless mistakes, such as polishing, electroplating and the like, which do not meet the requirements; 2. the equipment has a severe running environment, such as overlarge humidity, overhigh temperature, corrosive gas and liquid and the like; 3. the equipment is unbalanced in stress and abrasion during operation, so that the wind power generation equipment is scrapped in the effective life period for a long time.

Aiming at the problem of fatigue damage of wind power generation equipment, the research direction in the industry is mainly focused on fatigue estimation, and the health state of each part in a wind power generator set is simulated by utilizing software such as finite elements, so that an improvement scheme for each part is provided, but the method has lower recognition accuracy for the wind power generator set with fatigue failure and poorer real-time performance for monitoring the fatigue condition of each part in the wind power generator set.

Disclosure of Invention

Aiming at the defects of the existing mode, the application provides a fatigue failure detection method, device, equipment and medium of a wind generating set, which are used for solving the technical problems that the recognition accuracy of the wind generating set with fatigue failure is low and the real-time performance of monitoring the fatigue condition is poor in the prior art.

In a first aspect, an embodiment of the present application provides a method for detecting fatigue failure of a wind generating set, including:

acquiring historical fault information of a wind generating set to be predicted, and taking the historical fault information as the fault information to be predicted;

predicting the fatigue failure condition of the wind generating set to be predicted according to the failure information to be predicted and the fatigue failure prediction model to obtain a first prediction result;

acquiring historical transient data of the wind generating set to be predicted before a fault occurs, and taking the historical transient data as the historical transient data to be predicted;

determining the aperiodic load characteristics of the wind generating set to be predicted under each aperiodic load stress born according to the historical transient data to be predicted;

predicting fatigue failure conditions of the wind generating set to be predicted according to the aperiodic load characteristics to obtain a second prediction result;

And obtaining a third prediction result according to the first prediction result and the second prediction result.

In a second aspect, an embodiment of the present application provides a fatigue failure detection apparatus for a wind turbine generator system, including:

the first data acquisition module is used for acquiring historical fault information of the wind generating set to be predicted and taking the historical fault information as the fault information to be predicted;

the first prediction module is used for predicting the fatigue failure condition of the wind generating set to be predicted according to the failure information to be predicted and the fatigue failure prediction model to obtain a first prediction result;

the second data acquisition module is used for acquiring the calendar Shi Shuntai data of the wind generating set to be predicted before the fault occurs as the historical transient data to be predicted;

the load characteristic determining module is used for determining the aperiodic load characteristic of the wind generating set to be predicted under each aperiodic load stress born by the wind generating set to be predicted according to the historical transient data to be predicted;

the second prediction module is used for predicting the fatigue failure condition of the wind generating set to be predicted according to the aperiodic load characteristics to obtain a second prediction result;

and the third prediction module is used for obtaining a third prediction result according to the first prediction result and the second prediction result.

In a third aspect, an embodiment of the present application provides a fatigue failure detection apparatus for a wind turbine generator system, including:

a memory;

a processor electrically connected to the memory;

the memory stores a computer program that is executed by the processor to implement the fatigue failure detection method for a wind turbine generator set provided in the first aspect of the embodiment of the present application.

In a fourth aspect, embodiments of the present application provide a wind farm controller, comprising: the third aspect of the embodiment of the application provides fatigue failure detection equipment of a wind generating set.

In a fifth aspect, embodiments of the present application provide a computer readable storage medium storing a computer program, where the computer program when executed by a processor implements the fatigue failure detection method of a wind turbine generator set provided in the first aspect of the embodiments of the present application.

The technical scheme provided by the embodiment of the application has at least the following beneficial effects:

according to the embodiment of the application, the fatigue failure condition of the wind turbine generator set to be predicted is predicted by adopting two prediction modes based on the historical fault information and the historical transient data, and after the prediction results (the first prediction result and the second prediction result) obtained by the two prediction modes are comprehensively considered, the comprehensive prediction result (the third prediction result) can be obtained, the two prediction modes are mutually supplemented, so that the fatigue failure of the wind turbine generator set can be more accurately detected, the recognition accuracy of the wind turbine generator set with higher fatigue failure risk is improved, the real-time performance of detection and recognition is higher, and a more efficient detection scheme and a more referential detection result are provided for later operation and maintenance.

When the prediction is performed based on the historical fault information, the prediction is performed according to the fatigue failure prediction model, so that the prediction efficiency and accuracy can be improved, and the reliability of a prediction result can be improved; when predicting based on the historical transient data, the aperiodic load characteristics of the wind turbine generator system to be predicted under each aperiodic load stress can be determined according to the historical transient data, the characteristics can reflect the vibration characteristics of the wind turbine generator system in different vibration directions under the aperiodic load stress, and the fatigue condition of the wind turbine generator system can be accurately predicted based on the characteristics.

Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the application.

Drawings

The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

fig. 1 is a schematic flow chart of a fatigue failure detection method of a wind turbine generator system according to an embodiment of the present application;

FIG. 2 is a schematic diagram of a first distribution curve and a second distribution curve of a normal unit in an embodiment of the present application;

FIG. 3 is a graph showing a first distribution curve and a second distribution curve of a failure unit in an embodiment of the present application;

FIG. 4 is a schematic diagram of a distribution curve of accumulated fatigue values of a normal unit and a failed unit in an embodiment of the present application;

FIG. 5 is a schematic structural frame diagram of a fatigue failure detection device of a wind turbine generator system according to an embodiment of the present disclosure;

fig. 6 is a schematic structural frame diagram of a fatigue failure detection device of a wind turbine generator system according to an embodiment of the present application.

Detailed Description

Examples of embodiments of the present application are illustrated in the accompanying drawings, in which like or similar reference numerals refer to like or similar elements or elements having like or similar functionality throughout. Further, detailed descriptions of known techniques are omitted if they are not necessary for the illustrated features of the present application. The embodiments described below by referring to the drawings are exemplary only for the purpose of illustrating the present application and are not to be construed as limiting the present application.

It will be understood by those skilled in the art that all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs unless defined otherwise. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. The term "and/or" as used herein includes all or any and all combinations of one or more of the associated listed items.

Several terms which are referred to in this application are first introduced and explained:

permutation entropy (Permutation Entropy): the signal mutation detection method proposed for the spatial characteristics of the time series itself increases the value of the permutation entropy as the data becomes more chaotic.

The inventor of the application conducts research, and when analyzing and verifying the vibration distribution characteristics of the wind generating set at different rotating speeds to excavate the difference of signals, generally, the signals of the vibration of the generator bearings show typical random signal characteristics, when the bearings are in local faults, impact components which are characterized by the passing frequency of the bearings appear in the signals, the impact size depends on the range degree and the load size of the faults, and the stronger the faults, the larger the impact is, and the more severe the vibration of a transmission system is.

In order to represent the vibration change condition in a certain rotating speed interval, the concept of the arrangement entropy is introduced, and experiments prove that the method can well show the chaotic degree of vibration data.

The technical scheme of the present application and how the technical scheme of the present application solves the above technical problems are described in detail below with specific embodiments.

The embodiment of the application provides a fatigue failure detection method of a wind generating set, as shown in fig. 1, the method comprises the following steps:

s101, acquiring historical fault information of the wind generating set to be predicted, and taking the historical fault information as the fault information to be predicted.

Specifically, historical fault information of each component in the wind generating set to be predicted in a statistical period is obtained, wherein the components in the wind generating set comprise but are not limited to equipment capable of representing abnormal information such as a generator main bearing, a hub, a blade, a gear box, a yaw bearing, a pitch bearing, a tower barrel, a cabin and the like through vibration data, and the generator main bearing is taken as an example, and related faults comprise abnormal temperature, grease leakage, abnormal sound and the like; and judging whether the wind power generation equipment has faults or not, and taking the issued fault worksheet as the reference.

Optionally, the historical fault information includes a name (or a number) of the wind generating set, a fault type of each fault, a fault occurrence frequency of each fault, and a fatigue failure tag of the wind generating set, where the fault occurrence frequency of each fault can be obtained by statistics according to a historical occurrence time of each fault.

Alternatively, the time range of the statistical period may be set according to actual requirements, for example, the time range from a certain time of a certain month to a certain time of a certain year to a current time may be set as one statistical period, and the time range from a corresponding time when the wind power generation device starts to operate online to the current time may also be set as one statistical period.

The inventor of the present application conducted research and found that fatigue failure of wind power generation equipment is often caused by several non-periodic load stress actions, and that failure of wind power generation equipment in the past can be regarded as visual appearance of the fact that several non-periodic load stress actions act on the equipment. Therefore, the fatigue condition of the wind generating set can be detected based on the historical fault information of the wind generating set before fatigue failure, and further the recognition of the wind generating set with fatigue failure (hereinafter simply referred to as a failure set) is realized.

S102, predicting the fatigue failure condition of the wind generating set to be predicted according to the failure information to be predicted and the fatigue failure prediction model to obtain a first prediction result, and then executing step S106.

Optionally, the fatigue failure prediction model is trained by:

acquiring historical fault information of each wind generating set and fatigue failure data of each wind generating set, wherein the historical fault information and the fatigue failure data of each wind generating set are respectively used as sample fault information and sample fatigue failure data; inputting sample fault information into a fatigue failure prediction model, and obtaining a sample prediction result based on the output of the fatigue failure prediction model; determining loss of a sample prediction result according to sample fatigue failure data; and adjusting parameters of the fatigue failure prediction model according to the loss of the sample prediction result.

Alternatively, the information included in the historical fault information is as described above, and the obtained historical fault information of the generator in each wind generating set is shown in table 1, taking the generator as an example.

Table 1 historical fault information for generator

In table 1, wt1 to wt3 in the first row respectively represent wind power generation set 1 to wind power generation set 3, the meanings of the subsequent set names are analogized in order, and are not listed one by one in table 1; the numbers from the second row in the second column to the lower right corner area in table 1 represent the frequency (i.e. the number of times) of occurrence of a corresponding failure of the wind turbine generator set, for example, 23 in the second row in the second column represents the frequency of occurrence of a temperature abnormality of the generator set around the wind turbine generator set 1, and the other numbers are the same.

Alternatively, the fatigue failure data includes a record of whether the wind turbine generator set has failed in a statistical period, the record may be represented in the form of a tag, which may be represented by a letter, number, or other symbol; for example, a label of "1" indicates that the wind turbine generator system has fatigue failure within a statistical period, and a label of "0" indicates that the wind turbine generator system has not fatigue failure within a statistical period.

In one example, fatigue failure data for each wind turbine generator set is obtained as shown in Table 2.

Table 2 fatigue failure data for each wind turbine

Machine set name	Label (Label)
		wt1	1
wt2	1
		wt3	0
wt4	1
		wt5	0
……	……

In table 2, wt1 to wt5 represent wind turbine generator sets 1 to 5, respectively, and the meanings of the subsequent unit names are analogized in order, and are not listed in table 2.

Optionally, the fatigue failure prediction model is constructed based on a classification algorithm; the classification algorithm comprises any one algorithm of logic regression, a neural network, a support vector machine and a decision tree.

Taking a logistic regression algorithm as an example, the construction of the fatigue failure prediction model is described as follows:

first, a linear boundary is constructed based on historical fault information, which can be represented by the following expression:

In expression (1), θ is a historical failure frequency weight parameter vector; θ _i The historical fault frequency weight parameter corresponding to the ith fault class can be obtained through calculation of maximum likelihood and gradient descent (refer to subsequent content); x is the historical failure frequency value, x _i The historical fault frequency value of the ith fault type is given, and m is the total number of fault categories; in expression (1), i is [0, m]Integers within the range.

Secondly, constructing a prediction function based on the sigmoid mapping function, wherein the prediction function is as follows:

in expression (2), h _θ (x) And g (z) is a sigmoid mapping function, and the meaning of the rest parameters can be referred to in the expression (1).

Then, a loss function is constructed as follows:

in expression (3), y is a label indicating whether fatigue failure occurs in the wind turbine generator set within a statistical period, COST (h _θ (x) Y) represents h _θ (x) And loss of y, the meaning of the remaining parameters can be referred to expressions (1) and (2).

When training the fatigue failure prediction model constructed according to the mode, substituting the historical failure frequency value, the initial historical failure frequency parameter vector and the total number of failure categories of each wind generating set into x, theta and m of an expression (2), and obtaining a prediction result h of the fatigue condition of each wind generating set through the expression (2) _θ (x) Calculating a prediction result h according to expression (3) _θ (x) And (3) adjusting the historical failure frequency weight parameter vector theta according to the obtained loss by a maximum likelihood method and a gradient descent method relative to the loss of the tag y until a preset loss function convergence condition is met.

When the historical fault frequency weight parameter vector theta is adjusted through a maximum likelihood method and a gradient descent method until a preset loss function convergence condition is met, firstly estimating an expression of the historical fault frequency weight parameter vector theta through the maximum likelihood method, and then realizing iterative operation of theta through the gradient descent method to obtain the final theta.

The core idea of maximum likelihood is to extrapolate the parameters that lead to the maximum result from the known results, while maximum likelihood estimation is the application of probability theory in statistics, which provides a method for evaluating model parameters given observation data, i.e. "model determined, parameter not determined", by several experimental observations, using a certain parameter of the experiment to maximize the probability of sample occurrence, called maximum likelihood trajectory.

Logistic regression is a supervised learning, labeled. Starting from the known result, the result parameter which can obtain the maximum probability is deduced, and the model can predict data more accurately as long as the parameter can be found.

Sigmoid function in the embodiment of the application

The value of (1) can be regarded as the posterior probability (determined by the sigmoid function properties) that the test tuple belongs to class "1" (i.e. the set tag above), and thus:

expression (4) is rewritable as:

p(y|X；θ)＝g(z) ^y (1-g(z)) ^1-y expression (5)

Expression (5) represents the posterior probability of the tuple class index y under the parameter θ, and X represents the input sample data.

Assuming that a sample is obtained at this time, the joint probability can be described as

The size of the model can reflect the loss cost of the model, and the larger the joint probability is, the closer the learning result is to the real situation is; the smaller the joint probability, the more the learning result deviates from the real situation.

The joint probability is subjected to logarithmization:

to this end, a loss cost function can be obtained:

expression (7) is the rewrite of expression (3), y _i Representing the predicted value of the ith sample, wherein the value range is 0 or 1; z _i Namely theta ^T X _i Watch (Table)Showing the value obtained by substituting the ith sample into expression (1); in expressions (6) and (7), i is [0, n]Integers within the range.

The basic principle of solving the parameter theta according to the gradient descent method is as follows: the negative direction of the gradient is the direction in which the loss function drops fastest, and the minimum value of the loss function is obtained through iteration, so that the parameter theta is obtained and is the final solution of logistic regression.

The change of each weight component is:

in expression (8), θ _j A value representing the j-th dimension of the weight parameter θ; η is learning rate and control steps.

Thus, the variable of the gradient descent update weight can be obtained:

in expressions (9) and (10), j represents the dimension of the weight parameter θ, θ _j A value representing the j-th dimension of the weight parameter θ _j : values following the expression representing the equal sign are assigned to θ _j ，x _ij Representing the value of the ith sample in the j-th dimension of the weighting parameter θ, other parameter meanings may be referred to the parameter meanings of each expression.

The neural network, the support vector machine and the decision tree in the embodiment of the application are all existing algorithms, and the application of the algorithms in the classification problem is existing, so that a person skilled in the art can understand how to apply the algorithms to realize classification in the embodiment of the application, thereby constructing a corresponding fatigue failure prediction model, and the algorithms are not developed and introduced one by one.

S103, acquiring historical transient data of the wind generating set to be predicted before the occurrence of the fault, and taking the historical transient data as the historical transient data to be predicted.

In particular, historical transient data of various components in the wind generating set to be predicted before a fault occurs are obtained, and as described above, the components of the wind generating set may be a generator main bearing, a hub, a blade, a gearbox, a yaw bearing, a pitch bearing, a tower, a nacelle, and the like.

Optionally, the historical transient data includes: rotational speed data, nacelle vibration data in a first direction, and nacelle vibration data in a second direction. Optionally, the historical transient data may also include wind speed data. The nacelle vibration data includes nacelle vibration acceleration.

The historical transient data in the embodiment of the application can adopt data with corresponding frequency, such as any one of second-level data, millisecond-level data and microsecond-level data, according to actual requirements, so that the accuracy of subsequent calculation is improved, and the higher the data frequency is, the more the accuracy of the subsequent calculation is improved.

The first direction and the second direction may be selected according to an actual vibration direction or according to actual requirements, in one example, based on an angle facing the impeller of the wind turbine, the first direction may be a left-to-right or right-to-left direction of the wind turbine, i.e. a radial direction of the nacelle, and the second direction may be a front-to-back or back-to-front direction of the wind turbine, i.e. an axial direction of the nacelle.

S104, determining the aperiodic load characteristics of the wind generating set to be predicted under each aperiodic load stress born according to the historical transient data to be predicted.

Optionally, determining a first permutation entropy value of the nacelle vibration data in the first direction under each rotational speed data and a second permutation entropy value of the nacelle vibration data in the second direction under each rotational speed data; and determining the aperiodic load characteristics of the wind generating set to be predicted under each aperiodic load stress born by the wind generating set according to the first permutation entropy value and the second permutation entropy value.

Optionally, before determining the permutation entropy value, the method may further include: and cleaning, filtering, sliding windowing and data grouping are sequentially carried out on the acquired historical transient data.

Optionally, during the cleaning process, the null value and the abnormal value in the historical transient data can be deleted, wherein whether a certain historical transient data is an abnormal value can be judged according to whether the data is in a preset normal value range or not, and can also be judged according to the difference value between the data and other data.

Optionally, in the filtering process, the historical transient data can be filtered based on a preset filtering rule, and the filtering rule can be set according to actual requirements; in one example, if the filtering rule is set as follows: when the absolute value of the cabin vibration acceleration is between 0 and 0.45, deleting the cabin vibration acceleration with the absolute value not conforming to the rule during filtering, and only keeping the cabin vibration acceleration with the absolute value between 0 and 0.45; in another example, if the filtering rule is set as follows: if the wind speed value is not lower than 3m/s, deleting the wind speed data lower than 3m/s during filtering, and only retaining the wind speed data higher than or equal to 3 m/s.

Optionally, after the cleaning and filtering are finished, the residual data volume can be detected, whether the residual data volume is less than 30% of the original data volume before the data cleaning and data filtering are performed is judged, if the residual data volume is greater than or equal to 30% of the original data volume, the subsequent steps are continuously performed, and if the residual data volume is less than 30% of the original data volume, the cleaning or filtering rule can be reset, so that the residual data volume after the data cleaning and data filtering are performed again is greater than or equal to 30% of the original data volume, the subsequent calculation is ensured to have a larger data volume basis, and the calculation result is more accurate.

Optionally, in the sliding windowing process, the historical transient data is divided into a plurality of data segments according to a time range, for example, the historical transient data in one month is used as data of one data segment, that is, each data segment shares one month of data volume, and the sliding step length of each data segment is one week.

Optionally, in the data grouping process, the historical transient data may be grouped based on the rotational speed data, specifically, the rotational speed data is reserved for one decimal, the rotational speed data with the same numerical value after the reserved one decimal is grouped into one group, the nacelle vibration data in the first direction corresponding to the same group of rotational speed data is calculated by a first permutation entropy value, and the nacelle vibration data in the second direction corresponding to the same group of rotational speed data is calculated by a second permutation entropy value.

In one example, the rotational speed data, the nacelle vibration data in the first direction, and the nacelle vibration data in the second direction are shown in the first, third, and fourth columns of table 1, the data retaining one decimal place for the rotational speed data is shown in the second column of table 1, the x-direction in table 1 represents the first direction, and the y-direction represents the second direction.

Table 3 rotational speed data and nacelle vibration data

The rotation speed data after one decimal retention and the corresponding cabin vibration data in table 3 are grouped as shown in table 4, the rotation speed data with the same numerical value in table 4 and the cabin vibration acceleration data corresponding to the rotation speed data are grouped, and the arrangement entropy value is calculated for the cabin vibration speed in the x direction and the y direction of each group.

Table 4 rotational speed data and nacelle vibration data

Optionally, determining the aperiodic load characteristic of the wind turbine generator system to be predicted under each aperiodic load stress according to the first permutation entropy value and the second permutation entropy value includes:

determining a first distribution curve of the first permutation entropy value along with the change of the rotating speed according to the first permutation entropy value and the corresponding rotating speed data; determining a second distribution curve of the second permutation entropy value along with the change of the rotating speed according to the second permutation entropy value and the corresponding rotating speed data; determining the non-periodic load characteristic according to the number of the crossing points of the first distribution curve and the second distribution curve.

The inventors of the present application have studied and found that the effect of non-periodic load stress causes vibration of wind power generation equipment, and fatigue failure may be caused.

Meanwhile, when analyzing a large number of wind turbine generator set cases, the inventor of the application compares the arrangement entropy values of cabin vibration data of a normal turbine generator set (namely, a wind turbine generator set which does not have fatigue failure) and a failure turbine generator set (or called an abnormal turbine generator set), namely, a wind turbine generator set which has fatigue failure, in the same wind power plant. Alternatively, the first direction may be the x-direction and the second direction may be the y-direction. A distribution curve (i.e., a first distribution curve) of the permutation entropy value (hereinafter referred to as a first permutation entropy value) of the nacelle vibration data of the normal unit in the first direction is shown as a solid line in fig. 2; a distribution curve (i.e., a second distribution curve) of the permutation entropy value (hereinafter referred to as a second permutation entropy value) of the nacelle vibration data of the normal unit in the second direction, as shown by a dotted line in fig. 2; the first and second profiles of the failing unit are shown in solid and dashed lines, respectively, in fig. 3.

As can be seen from fig. 2 and 3, in each rotational speed data segment, the first permutation entropy value in the x direction and the second permutation entropy value in the y direction are both substantially linearly changed, and when the x direction is the left-right direction of the wind turbine generator set and the y direction is the front-rear direction of the wind turbine generator set, the first permutation entropy value is slightly higher than the second permutation entropy value; for the normal unit shown in fig. 2, there are substantially no intersections between the first distribution curve and the second distribution curve, and for the failure unit shown in fig. 3, in some rotation speed intervals (rotation speed intervals are not fixed), the first permutation entropy value and the second permutation entropy value are irregular in appearance, and there are more intersections between the first distribution curve and the second distribution curve, such as the intersections in the circles in fig. 3.

Thus, for a wind power plant whose component vibrations fluctuate with the rotational speed to be predicted, the characteristic can be characterized by the intersection of a first distribution curve of first permutation entropy values and a second distribution curve of second permutation entropy values based on nacelle vibration data in the x-direction and the y-direction, and thus the present application determines the aperiodic load characteristic of the aperiodic load stress inducing vibrations from the intersection of the first distribution curve and the second distribution curve.

In an alternative embodiment, the intersection points of the first distribution curve and the second distribution curve may be interNum as non-periodic load features; in another alternative embodiment, the intersection points intersnum may be multiplied by a certain coefficient according to the actual requirement, and the result is taken as the aperiodic load characteristic.

S105, predicting the fatigue failure condition of the wind generating set to be predicted according to the aperiodic load characteristics to obtain a second prediction result.

Optionally, predicting the fatigue failure condition of the wind generating set to be predicted according to the aperiodic load characteristic includes:

determining fatigue values of the wind generating set to be predicted under each aperiodic load stress according to the aperiodic load characteristics; determining an accumulated fatigue value of the wind generating set to be predicted under each aperiodic load stress according to fatigue values of the wind generating set to be predicted under each aperiodic load stress; determining characteristic parameters of accumulated fatigue values of the wind generating set to be predicted under each aperiodic load stress along with the change of the aperiodic load stress; and comparing the characteristic parameters with characteristic parameter thresholds, and predicting the fatigue failure condition of the wind generating set to be predicted according to the comparison result.

In an alternative embodiment, the accumulated fatigue value of the wind power generation equipment along with time can be calculated according to the aperiodic load characteristic and combined with the Miner-Palmgren fatigue theory; specifically, according to the aperiodic load characteristics, the fatigue value under each aperiodic load stress can be obtained by using the dynamics principle in the following way:

in expression (11), S represents a fatigue value under a certain aperiodic load stress, d represents an aperiodic load characteristic under the aperiodic load stress, and α represents a material constant in terms of dynamics.

According to Miner-Pallgren fatigue theory, the fatigue damage of wind power generation equipment generated under each aperiodic load stress can be linearly overlapped in the following manner to obtain accumulated fatigue values under each aperiodic load stress:

in expression (12), F represents an accumulated fatigue value of the wind power plant under each aperiodic load stress, and when f=1, fatigue failure of the wind power plant occurs; n represents the number of aperiodic load stresses, i represents the i-th aperiodic load stress of the n aperiodic load stresses, and S (i) represents the fatigue value at the i-th aperiodic load stress.

Over time, the accumulated fatigue value F of the wind power generation equipment is continuously increased, a series of data of accumulated fatigue values F can be obtained, when the inventor of the application analyzes a large number of wind power generation set cases, the accumulated fatigue values F of a normal set and a failure set of the same wind power plant are compared as shown in a graph 4, a curve wt1 in the graph 4 is a distribution curve of the accumulated fatigue values of the failure set, a curve wt2, a curve wt3 and a curve wt4 are respectively a distribution curve of the accumulated fatigue values of each of 3 normal sets, and an abscissa in the graph 4 is the generating frequency of non-periodic load, namely the number of non-periodic load stresses; as can be seen from fig. 4, the magnitude of the change in the accumulated fatigue value of the failed unit is higher than that of the normal unit.

Therefore, the fatigue failure condition of each component in the wind generating set to be predicted can be predicted based on the change trend of the accumulated fatigue value.

Optionally, determining a characteristic parameter of the cumulative fatigue value of the wind generating set to be predicted under each aperiodic load stress as a function of the aperiodic load stress includes:

performing linear fitting on the accumulated fatigue values of the wind generating set to be predicted under each aperiodic load stress to obtain an accumulated fatigue value linear equation of the wind generating set to be predicted; and extracting the slope value of the accumulated fatigue value linear equation as a characteristic parameter.

In an alternative embodiment, a linear regression algorithm is adopted to linearly fit the accumulated fatigue values of the wind turbine generator set to be predicted at each non-periodic load action moment, so as to obtain an accumulated fatigue value linear equation of the wind turbine generator set to be predicted.

Optionally, the characteristic parameter threshold is determined by:

acquiring historical transient data of each wind generating set before a fault occurs, and taking the historical transient data as sample historical transient data; determining the aperiodic load characteristics of each wind generating set under each stressed load according to the sample historical transient data; determining accumulated fatigue values of each wind generating set under each aperiodic load stress according to the aperiodic load characteristics; determining an accumulated fatigue value distribution curve of the accumulated fatigue value of each wind generating set along with the aperiodic load stress variation; and determining a characteristic parameter threshold according to the distribution parameters of the accumulated fatigue value distribution curve.

Optionally, determining the aperiodic load characteristic of each wind generating set under each load stress born by the wind generating set according to the sample historical transient data, and its principle and optional implementation are similar to the principle of the previous step S104 and the optional implementation of step S104, for example, by determining the arrangement entropy of the nacelle vibration data in the sample historical transient data, the aperiodic load characteristic of each wind generating set under each load stress born by the wind generating set is further determined, and the relevant content of step S104 may be referred to specifically.

Optionally, determining the accumulated fatigue value of each wind generating set at each non-periodic load acting moment according to the non-periodic load characteristics, including:

for each wind generating set, determining the fatigue value of the wind generating set under each aperiodic load stress according to the aperiodic load characteristics of the wind generating set under each aperiodic load stress born by the wind generating set; and determining the accumulated fatigue value of the wind generating set under each aperiodic load stress according to the fatigue value of the wind generating set under each aperiodic load stress.

In an alternative embodiment, the manner of calculating the fatigue value under each aperiodic load stress according to the aperiodic load characteristic may refer to the relevant content of the foregoing expression (11), which is not described herein.

In an alternative embodiment, the accumulated fatigue value of the wind turbine generator set under each aperiodic load stress is determined according to the fatigue value of the wind turbine generator set under each aperiodic load stress, and the specific manner may refer to the related content of the foregoing expression (12), which is not described herein.

In an alternative embodiment, a cumulative fatigue value profile of each wind park as a function of the aperiodic load is determined, the profile of which can be referred to in fig. 4.

In an alternative embodiment, determining the characteristic parameter threshold from the distribution parameters of the accumulated fatigue value distribution curve includes:

for each wind generating set, performing linear fitting on the accumulated fatigue value of the wind generating set at each aperiodic load acting moment to obtain an accumulated fatigue value linear equation of the wind generating set; extracting the slope value of an accumulated fatigue value linear equation of each wind generating set; deleting the slope value of the accumulated fatigue value linear equation with the slope value of the first 20% to prevent the influence on the accurate estimation of the subsequent normal distribution parameters due to the overlarge slope value corresponding to the failure unit; estimating normal distribution parameters of the slope values of the rest wind generating sets according to a maximum likelihood estimation method (corresponding slope values of the normal sets usually show normal distribution according to research), wherein the normal distribution parameters comprise a mean value mu and a standard difference sigma; the characteristic parameter threshold value may be set to μ+4σ (other values may also be set according to actual conditions) according to the determined normal distribution parameter.

In one example, after comparing the characteristic parameter of the wind turbine to be predicted with the characteristic parameter threshold, if the comparison result is that the characteristic parameter is greater than the characteristic parameter threshold, the obtained second prediction result is that the risk of fatigue failure of the wind turbine to be predicted is greater; if the comparison result is that the characteristic parameter is smaller than or equal to the characteristic parameter threshold value, the second prediction result is that the risk of fatigue failure of the wind generating set to be predicted is small.

Optionally, when predicting the fatigue failure condition of the wind turbine generator set to be predicted according to the aperiodic load characteristic, the accumulated fatigue value F of the wind turbine generator set to be predicted may be calculated according to the aperiodic load characteristic, and the specific calculation mode may refer to the foregoing embodiment, and the accumulated fatigue value F of the wind turbine generator set to be predicted is used as input, and abnormal recognition is performed by using an outlier detection algorithm (such as an isolated forest algorithm), so as to predict the risk of fatigue failure of the wind turbine generator set to be predicted.

Optionally, when predicting the fatigue failure condition of the wind turbine generator set to be predicted according to the aperiodic load characteristic, the accumulated fatigue value F of the wind turbine generator set to be predicted at each time point or under each aperiodic load stress can be used as input, and abnormal recognition is performed according to any one of algorithms such as a logistic regression algorithm, a support vector machine, a neural network and the like, so that the risk of fatigue failure of the wind turbine generator set to be predicted is predicted.

The outlier detection algorithm, the logistic regression algorithm, the support vector machine, the neural network and other algorithms in the embodiment of the application are all existing algorithms, and a person skilled in the art can understand how to apply these algorithms to realize the anomaly identification of the accumulated fatigue value F of the wind turbine generator set to be predicted in the embodiment of the application, so that the algorithms are not developed and introduced one by one.

S106, obtaining a third prediction result according to the first prediction result and the second prediction result.

Optionally, weights are respectively set for the first prediction result and the second prediction result; and combining the first prediction result and the second prediction result based on the weight to obtain a third prediction result.

In an alternative embodiment, the third prediction result may be determined by:

l (x) =αh (x) + (1- α) f (x) expression (13)

In the expression (13), h (x) represents a first prediction result, h (x) represents that the risk of fatigue failure of the wind turbine generator set to be predicted is large when h (x) is 1, and h (x) represents that the risk of fatigue failure of the wind turbine generator set to be predicted is small when h (x) is 0; f (x) represents a second prediction result, f (x) represents that the risk of fatigue failure of the wind turbine generator set to be predicted is high when the f (x) is 1, and f (x) represents that the risk of fatigue failure of the wind turbine generator set to be predicted is low when the f (x) is 0; alpha is a weight of h (x), which can be determined according to actual requirements or empirical values, and in one alternative embodiment the weight has a value in the range of [0,1], and in another alternative embodiment the weight has a value in the range of [0, 0.5); l (x) represents the third prediction result, and the value of L (x) is between 0 and 1.

Optionally, the method for detecting fatigue failure of a wind generating set provided in the embodiment of the present application further includes, on the basis of the steps S101 to S106, the steps of: and comparing the third prediction result with a result threshold value, and sending out fatigue failure early warning of the wind generating set to be predicted when the third prediction result is larger than the result threshold value.

In an alternative embodiment, the result threshold of the present embodiment may be set to a value between 0 and 1.

The third prediction result comprehensively considers two prediction results obtained by two prediction modes, so that the fatigue condition of the wind turbine generator set to be predicted can be reflected more accurately, and more accurate prediction and early warning of fatigue failure are facilitated.

Based on the same inventive concept, the fatigue failure detection device of the wind generating set provided in the embodiment of the present application, as shown in fig. 5, includes: a first data acquisition module 501, a first prediction module 502, a second data acquisition module 503, a load feature determination module 504, a second prediction module 505, and a third prediction module 506.

The first data obtaining module 501 is configured to obtain historical fault information of a wind turbine generator set to be predicted, as the fault information to be predicted.

The first prediction module 502 is configured to predict a fatigue failure condition of the wind turbine generator set to be predicted according to the failure information to be predicted and the fatigue failure prediction model, so as to obtain a first prediction result.

The second data obtaining module 503 is configured to obtain historical transient data of the wind turbine generator set to be predicted before the fault occurs, as the historical transient data to be predicted.

The load characteristic determining module 504 is configured to determine, according to the historical transient data to be predicted, an aperiodic load characteristic of the wind turbine to be predicted under each aperiodic load stress to which the wind turbine to be predicted is subjected.

And the second prediction module 505 is configured to predict a fatigue failure condition of the wind turbine generator set to be predicted according to the aperiodic load characteristic, so as to obtain a second prediction result.

And a third prediction module 506, configured to obtain a third prediction result according to the first prediction result and the second prediction result.

Optionally, the fatigue failure detection device 500 for a wind generating set provided in the implementation of the present application further includes: and an early warning module.

The early warning module is used for: and comparing the third prediction result with a result threshold value, and sending out fatigue failure early warning of the wind generating set to be predicted when the third prediction result is larger than the result threshold value.

Optionally, the fatigue failure detection device 500 for a wind generating set provided in the implementation of the present application further includes: and a model training module.

The model training module is used for training the fatigue failure prediction model by the following modes:

Optionally, the load feature determination module 504 is specifically configured to: determining a first arrangement entropy value of the cabin vibration data in the first direction under each rotating speed data and a second arrangement entropy value of the cabin vibration data in the second direction under each rotating speed data; and determining the aperiodic load characteristics of the wind generating set to be predicted under each aperiodic load stress born according to the first permutation entropy value and the second permutation entropy value.

Optionally, the load feature determination module 504 is specifically configured to: determining a first distribution curve of the first permutation entropy value along with the change of the rotating speed according to the first permutation entropy value and the corresponding rotating speed data; determining a second distribution curve of the second arrangement entropy value along with the change of the rotating speed according to the second arrangement entropy value and the corresponding rotating speed data; the aperiodic load characteristic is determined based on the number of intersections of the first distribution curve and the second distribution curve.

Optionally, the second prediction module 505 is specifically configured to: determining fatigue values of the wind generating set to be predicted under the action of each aperiodic load according to the aperiodic load characteristics; according to the fatigue value of the wind generating set to be predicted under the action of each aperiodic load, determining the accumulated fatigue value of the wind generating set to be predicted at the action moment of each aperiodic load; determining characteristic parameters of accumulated fatigue values of the wind generating set to be predicted at each non-periodic load action moment along with the non-periodic load change; and comparing the characteristic parameters with characteristic parameter thresholds, and predicting fatigue failure conditions of the wind generating set to be predicted according to the comparison results.

Optionally, the second prediction module 505 is specifically configured to: performing linear fitting on the accumulated fatigue values of the wind generating set to be predicted at each aperiodic load acting moment to obtain an accumulated fatigue value linear equation of the wind generating set to be predicted; and extracting the slope value of the accumulated fatigue value linear equation as a characteristic parameter.

Optionally, the second prediction module 505 is specifically configured to determine the feature parameter threshold by:

acquiring historical transient data of each wind generating set before a fault occurs, and taking the historical transient data as sample historical transient data; determining the aperiodic load characteristics of each wind generating set under each load stress born by each wind generating set according to the sample historical transient data; determining accumulated fatigue values of each wind driven generator set at each aperiodic load acting moment according to the aperiodic load characteristics; determining an accumulated fatigue value distribution curve of the accumulated fatigue value of each wind generating set along with the non-periodic load change; and determining a characteristic parameter threshold according to the distribution parameters of the accumulated fatigue value distribution curve.

Optionally, the third prediction module 506 is specifically configured to: respectively setting weights for the first prediction result and the second prediction result; and combining the first prediction result and the second prediction result based on the weight to obtain a third prediction result.

The fatigue failure detection device 500 of the wind turbine generator system according to the present embodiment may execute any of the fatigue failure detection methods of the wind turbine generator system provided in the embodiments of the present application, and the implementation principle is similar, and details not shown in the present embodiment may refer to the foregoing method embodiments, which are not repeated herein.

Based on the same inventive concept, embodiments of the present application provide a wind farm controller, including: fatigue failure detection equipment of the wind generating set; the apparatus includes: the device comprises a memory and a processor, wherein the memory is electrically connected with the processor.

The memory stores a computer program that is executed by the processor to implement any of the fatigue failure detection methods for wind turbine generator systems provided in the embodiments of the present application.

Those skilled in the art will appreciate that the electronic devices provided by the embodiments of the present application may be specially designed and constructed for the required purposes, or may comprise known devices in general purpose computers. These devices have computer programs stored therein that are selectively activated or reconfigured. Such a computer program may be stored in a device (e.g., computer) readable medium or in any type of medium suitable for storing electronic instructions and coupled to a bus, respectively.

The present application provides, in an alternative embodiment, a fatigue failure detection apparatus of a wind turbine generator system, as shown in fig. 6, including: the memory 601 and the processor 602, the memory 601 and the processor 602 being electrically connected, such as by a bus 603.

Optionally, the memory 601 is used for storing application program codes for executing the embodiments of the present application, and the execution is controlled by the processor 602. The processor 602 is configured to execute application program codes stored in the memory 601, so as to implement any one of the fatigue failure detection methods of the wind turbine generator set provided in the embodiments of the present application.

The Memory 601 may be, but is not limited to, a ROM (Read-Only Memory) or other type of static storage device that can store static information and instructions, a RAM (Random Access Memory ) or other type of dynamic storage device that can store information and instructions, an EEPROM (Electrically Erasable Programmable Read Only Memory, electrically erasable programmable Read-Only Memory), a CD-ROM (Compact Disc Read-Only Memory) or other optical disk storage, optical disk storage (including compact disc, laser disc, optical disc, digital versatile disc, blu-ray disc, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer.

The processor 602 may be a CPU (Central Processing Unit ), general purpose processor, DSP (Digital Signal Processor, data signal processor), ASIC (Application Specific Integrated Circuit ), FPGA (Field-Programmable Gate Array, field programmable gate array) or other programmable logic device, transistor logic device, hardware component, or any combination thereof. Which may implement or perform the various exemplary logic blocks, modules, and circuits described in connection with this application disclosure. The processor 602 may also be a combination that performs computing functions, such as including one or more microprocessors, a combination of a DSP and a microprocessor, and the like.

Bus 603 may include a path that communicates information between the components. The bus may be a PCI (Peripheral Component Interconnect, peripheral component interconnect standard) bus or an EISA (Extended Industry Standard Architecture ) bus. The buses may be divided into address buses, data buses, control buses, etc. For ease of illustration, only one thick line is shown in fig. 6, but not only one bus or one type of bus.

Optionally, the fatigue failure detection device 600 of the wind turbine may further comprise a transceiver 604. The transceiver 604 may be used for both reception and transmission of signals. The transceiver 604 may allow the electronic device 600 to communicate wirelessly or by wire with other devices to exchange data. It should be noted that, in practical application, the transceiver 604 is not limited to one.

Optionally, the fatigue failure detection device 600 of the wind power generation set may further comprise an input unit 605. The input unit 605 may be used to receive input digital, character, image and/or sound information or to generate key signal inputs related to user settings and function control of the fatigue failure detection device 600 of the wind turbine generator system. The input unit 605 may include, but is not limited to, one or more of a touch screen, a physical keypad, function keys (such as volume control keys, switch keys, etc.), a track ball, a mouse, a joystick, a camera, a microphone, etc.

Optionally, the fatigue failure detection device 600 of the wind turbine may further comprise an output unit 606. An output unit 606 may be used to output or present information processed by the processor 602. The output unit 806 may include, but is not limited to, one or more of a display device, a speaker, a vibration device, and the like.

While FIG. 6 illustrates a fatigue failure detection apparatus 600 for a wind turbine generator set having various devices, it should be understood that not all of the illustrated devices are required to be implemented or provided. More or fewer devices may be implemented or provided instead.

The fatigue failure detection apparatus 600 of the wind turbine generator set provided in the embodiment of the present application has the same inventive concept as that of the foregoing embodiments, and details not shown in the fatigue failure detection apparatus may refer to the foregoing embodiments, and will not be described herein.

Based on the same inventive concept, the embodiments of the present application provide a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements any one of the fatigue failure detection methods of a wind turbine generator set provided in the embodiments of the present application.

The computer readable medium includes, but is not limited to, any type of disk including floppy disks, hard disks, optical disks, CD-ROMs, and magneto-optical disks, ROM, RAM, EPROM (Erasable Programmable Read-Only Memory), EEPROMs, flash Memory, magnetic cards, or optical cards. That is, a readable medium includes any medium that stores or transmits information in a form readable by a device (e.g., a computer).

The embodiment of the application provides a fatigue failure detection method for a wind turbine generator system, wherein the fatigue failure detection method is applicable to any wind turbine generator system. And will not be described in detail herein.

By applying the embodiment of the application, at least the following beneficial effects can be realized:

1) According to the embodiment of the application, the fatigue failure condition of the wind turbine to be predicted is predicted by adopting two prediction modes based on the historical fault information and the historical transient data, and after the prediction results (the first prediction result and the second prediction result) obtained by the two prediction modes are comprehensively considered, the obtained comprehensive prediction result (the third prediction result) is comprehensively considered, and the two prediction modes are mutually complementary, so that the fatigue failure of the wind turbine can be more accurately detected, the recognition accuracy of the wind turbine with higher fatigue failure risk is improved, the real-time performance of detection and recognition is higher, and a more efficient detection scheme and a more referential detection result are provided for later operation and maintenance.

2) When the prediction is performed based on the historical fault information, the prediction is performed according to the fatigue failure prediction model, the fatigue failure prediction model can be trained based on the historical fault information and the fatigue failure data of each wind generating set, the advantages of statistics and machine learning algorithms are combined, the trained fatigue failure prediction model is used for prediction, the prediction accuracy rate can be improved, and the reliability of a prediction result is improved.

3) When the wind power generation set is predicted based on the historical transient data, the aperiodic load characteristics of the wind power generation set to be predicted under each aperiodic load stress can be determined according to the historical transient data, the characteristics can reflect the vibration characteristics of the wind power generation set in different vibration directions under the aperiodic load stress, and the accumulated fatigue value of the wind power generation set can be obtained based on the characteristics and the Miner-Palmgren fatigue theory, so that the fatigue condition of the wind power generation set can be accurately predicted.

Those of skill in the art will appreciate that the various operations, methods, steps in the flow, steps, means, arrangements may be interchanged, altered, combined, or deleted in the application. Further, other steps, acts, schemes, and arrangements of various operations, methods, flows that have been discussed in this application may also be alternated, altered, rearranged, split, combined, or deleted. Further, steps, measures, schemes in the prior art with various operations, methods, flows disclosed in the present application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

In the description of the present application, it should be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or number of technical features indicated. Thus, a feature defining "a first" or "a second" may include one or more of the feature explicitly or implicitly. In the description of the present application, unless otherwise indicated, the meaning of "a plurality" is two or more.

It should be understood that, although the steps in the flowcharts of the figures are shown in order as indicated by the arrows, these steps are not necessarily performed in order as indicated by the arrows. The steps are not strictly limited in order and may be performed in other orders, unless explicitly stated herein. Moreover, at least some of the steps in the flowcharts of the figures may include a plurality of sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, the order of their execution not necessarily being sequential, but may be performed in turn or alternately with other steps or at least a portion of the other steps or stages.

The foregoing is only a partial embodiment of the present application, and it should be noted that, for a person skilled in the art, several improvements and modifications can be made without departing from the principle of the present application, and these improvements and modifications should also be considered as the protection scope of the present application.

Claims

1. The fatigue failure detection method of the wind generating set is characterized by comprising the following steps of:

acquiring historical transient data of the wind generating set to be predicted before a fault occurs as the historical transient data to be predicted;

determining the aperiodic load characteristics of the wind generating set to be predicted under each aperiodic load stress born by the wind generating set to be predicted according to the historical transient data to be predicted; the aperiodic load characteristic represents the vibration characteristics of the wind power generation equipment in different vibration directions under the aperiodic load stress;

predicting the fatigue failure condition of the wind generating set to be predicted according to the aperiodic load characteristic to obtain a second prediction result;

2. The fatigue failure detection method according to claim 1, wherein the fatigue failure prediction model is trained by:

acquiring historical fault information of each wind generating set and fatigue failure data of each wind generating set, wherein the historical fault information and the fatigue failure data of each wind generating set are respectively used as sample fault information and sample fatigue failure data;

Inputting the sample fault information into the fatigue failure prediction model, and obtaining a sample prediction result based on the output of the fatigue failure prediction model;

determining a loss of the sample prediction result according to the sample fatigue failure data;

and adjusting parameters of the fatigue failure prediction model according to the loss of the sample prediction result.

3. The fatigue failure detection method according to claim 2, wherein the fatigue failure prediction model is constructed based on a classification algorithm;

the classification algorithm comprises any one algorithm of logistic regression, a neural network, a support vector machine and a decision tree.

4. The fatigue failure detection method according to claim 1, wherein the historical transient data includes: rotational speed data, nacelle vibration data in a first direction, and nacelle vibration data in a second direction;

and determining, according to the historical transient data, an aperiodic load characteristic of the wind generating set to be predicted under each aperiodic load stress borne by the wind generating set to be predicted, including:

determining a first permutation entropy value of the cabin vibration data in the first direction under each of the rotational speed data and a second permutation entropy value of the cabin vibration data in the second direction under each of the rotational speed data;

And determining the aperiodic load characteristics of the wind generating set to be predicted under each aperiodic load stress according to the first permutation entropy value and the second permutation entropy value.

5. The method for fatigue failure detection according to claim 4, wherein determining the aperiodic load characteristic of the wind power generation set to be predicted under each aperiodic load stress according to the first permutation entropy value and the second permutation entropy value comprises:

determining a first distribution curve of the first permutation entropy value along with the change of the rotating speed according to the first permutation entropy value and the corresponding rotating speed data;

determining a second distribution curve of the second permutation entropy value along with the change of the rotating speed according to the second permutation entropy value and the corresponding rotating speed data;

the number of intersections of the first distribution curve and the second distribution curve is taken as the aperiodic load characteristic.

6. The fatigue failure detection method according to claim 1, wherein the predicting the fatigue failure condition of the wind turbine to be predicted according to the aperiodic load characteristic includes:

determining fatigue values of the wind generating set to be predicted under each aperiodic load stress according to the aperiodic load characteristics;

Determining an accumulated fatigue value of the wind generating set to be predicted under each aperiodic load stress according to the fatigue value of the wind generating set to be predicted under each aperiodic load stress;

determining characteristic parameters of the accumulated fatigue value of the wind generating set to be predicted under each aperiodic load stress along with the change of the aperiodic load stress;

and comparing the characteristic parameters with characteristic parameter thresholds, and predicting the fatigue failure condition of the wind generating set to be predicted according to the comparison result.

7. The method for detecting fatigue failure according to claim 6, wherein the determining a characteristic parameter of a cumulative fatigue value of the wind turbine generator set to be predicted under each of the aperiodic load stresses as a function of the aperiodic load stress includes:

performing linear fitting on the accumulated fatigue values of the wind generating set to be predicted under the aperiodic load stress to obtain an accumulated fatigue value linear equation of the wind generating set to be predicted;

and extracting a slope value of the accumulated fatigue value linear equation as the characteristic parameter.

8. The fatigue failure detection method according to claim 6, wherein the characteristic parameter threshold value is determined by:

Acquiring historical transient data of each wind generating set before a fault occurs, and taking the historical transient data as sample historical transient data;

determining the aperiodic load characteristics of each wind generating set under each aperiodic load stress born by each wind generating set according to the sample historical transient data;

determining accumulated fatigue values of each wind generating set under each aperiodic load stress according to the aperiodic load characteristics;

determining a cumulative fatigue value distribution curve of the cumulative fatigue value of each wind generating set along with the aperiodic load stress variation;

and determining the characteristic parameter threshold according to the distribution parameters of the accumulated fatigue value distribution curve.

9. The method of claim 1, wherein the obtaining a third predicted result from the first predicted result and the second predicted result comprises:

weights are respectively set for the first prediction result and the second prediction result;

and combining the first prediction result and the second prediction result based on the weight to obtain a third prediction result.

10. The fatigue failure detection method according to claim 1, further comprising:

And comparing the third prediction result with a result threshold value, and sending out fatigue failure early warning of the wind generating set to be predicted when the third prediction result is larger than the result threshold value.

11. A fatigue failure detection device for a wind turbine generator system, comprising:

the second data acquisition module is used for acquiring historical transient data of the wind generating set to be predicted before the occurrence of the fault and taking the historical transient data as the historical transient data to be predicted;

the load characteristic determining module is used for determining the aperiodic load characteristic of the wind generating set to be predicted under each aperiodic load stress born by the wind generating set to be predicted according to the historical transient data to be predicted; the aperiodic load characteristic represents the vibration characteristics of the wind power generation equipment in different vibration directions under the aperiodic load stress;

The second prediction module is used for predicting the fatigue failure condition of the wind generating set to be predicted according to the aperiodic load characteristic to obtain a second prediction result;

12. The fatigue failure detection apparatus according to claim 11, further comprising:

and the early warning module is used for comparing the third prediction result with a result threshold value, and sending out fatigue failure early warning of the wind generating set to be predicted when the third prediction result is larger than the result threshold value.

13. A fatigue failure detection apparatus for a wind turbine, comprising:

a memory;

a processor electrically connected to the memory;

the memory stores a computer program that is executed by the processor to implement the fatigue failure detection method of a wind turbine generator set according to any of claims 1-10.

14. A wind farm controller, comprising: a fatigue failure detection apparatus for a wind turbine according to claim 13.

15. A computer-readable storage medium, characterized in that a computer program is stored, which, when being executed by a processor, implements a fatigue failure detection method of a wind park according to any of claims 1-10.