阅读说明：本技术 基于代价敏感轻量级梯度提升机的风电机组故障检测方法 (Wind turbine generator fault detection method based on cost-sensitive lightweight gradient elevator ) 是由 唐明珠 彭巨 赵琪 陈荐 彭钟慧 彭书豪 于 2020-12-31 设计创作，主要内容包括：本申请公开了一种基于代价敏感轻量级梯度提升机的风电机组故障检测方法、装置、设备及计算机可读存储介质,方法包括：利用秩相关性分析算法从风电机组的数据集中筛选目标特征；利用目标特征对嵌入有误分类代价且以误分类代价最小化为目标的代价敏感轻量级梯度提升机故障检测模型进行训练；利用得到的最佳代价敏感轻量级梯度提升机故障检测模型及待测试集中的目标测试特征进行故障检测。本申请公开的上述技术方案,通过在代价敏感轻量级梯度提升机故障检测模型中嵌入误分类代价而提高对小类样本的关注,通过利用最佳代价敏感轻量级梯度提升机故障检测模型进行故障检测实现对数据不平衡和误分类代价不等问题的处理,提高风电机组故障检测的准确性。(The application discloses a wind turbine generator fault detection method, device and equipment based on a cost-sensitive lightweight gradient elevator and a computer-readable storage medium, wherein the method comprises the following steps: screening target characteristics from a data set of the wind turbine generator by using a rank correlation analysis algorithm; training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target to be minimized, by using target characteristics; and carrying out fault detection by using the obtained optimal cost-sensitive lightweight gradient elevator fault detection model and the target test characteristics to be tested and concentrated. According to the technical scheme, the attention to the subclass samples is improved by embedding the misclassification cost in the cost-sensitive lightweight gradient elevator fault detection model, the problems of data imbalance and unequal misclassification cost are solved by utilizing the optimal cost-sensitive lightweight gradient elevator fault detection model to perform fault detection, and the accuracy of the fault detection of the wind turbine generator is improved.)

1. A wind turbine generator fault detection method based on a cost-sensitive lightweight gradient elevator is characterized by comprising the following steps:

acquiring a data set of a wind turbine generator, and screening out target characteristics from the data set by using a rank correlation analysis algorithm;

training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target to be minimized, by using the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator;

and acquiring a to-be-tested set of the wind turbine generator, and performing fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and the target test characteristics in the to-be-tested set.

2. The wind turbine generator system fault detection method based on the cost-sensitive lightweight gradient elevator according to claim 1, wherein the loss function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

the objective function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

wherein, F (C)_F，C_N) Representing the cost of the fault class being classified as a normal class, F (C)_N，C_F) Representing the cost of the normal class being classified as a fault class, C_FIndicates a fault class, C_NDenotes normal class, P (c ═ F | x_i) Represents a sample x_iPosterior probability classified as fault class, P (c ═ N | x_i) Represents a sample x_iPosterior probability, y, divided into normal classes_iRepresents a sample x_iClass label of y_i1 denotes a fault sample, y_i0 denotes a normal sample, n denotes the number of samples, Ψ denotes a loss function, and Ω (f)_k) Representing a regularization term, F_t-1(x_i) Representing the last loss function, f_t(x_i) Representing the loss function at the current time.

3. The wind turbine generator system fault detection method based on the cost-sensitive lightweight gradient elevator according to claim 2, wherein the target feature is used for training a cost-sensitive lightweight gradient elevator fault detection model embedded with a misclassification cost and aiming at minimizing the misclassification cost to obtain an optimal cost-sensitive lightweight gradient elevator fault detection model, and the method comprises the following steps:

dividing the target features into a training set, a testing set and a verification set, and preprocessing the training set;

training the fault detection model of the cost-sensitive lightweight gradient elevator by using the preprocessed training set, adjusting the hyper-parameters of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using the verification set, primarily evaluating the capability of the model, evaluating the generalization capability of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using the test set, and calculating an evaluation index;

and judging whether the trained fault detection model of the cost-sensitive lightweight gradient elevator is optimal or not by using the evaluation index, if not, acquiring a training set again, and executing the step of preprocessing the training set, and if so, determining the optimally trained fault detection model of the cost-sensitive lightweight gradient elevator as the optimal fault detection model of the cost-sensitive lightweight gradient elevator.

4. The wind turbine generator system fault detection method based on the cost-sensitive lightweight gradient elevator as claimed in claim 1, wherein the step of screening out target features from the data set by using a rank correlation analysis algorithm comprises:

and calculating the Spearman rank correlation coefficient of each characteristic in the data set and the fault of the wind turbine generator by using a Spearman rank correlation coefficient method, and selecting the characteristic of the Spearman rank correlation coefficient in a threshold value range as the target characteristic.

5. The utility model provides a wind turbine generator system fault detection device based on sensitive lightweight gradient lifting machine of cost which characterized in that includes:

the screening module is used for acquiring a data set of the wind turbine generator and screening out target features from the data set by utilizing a rank correlation analysis algorithm;

the training module is used for training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost minimized as a target, by utilizing the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator;

and the detection module is used for acquiring a to-be-tested set of the wind turbine generator and carrying out fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and the target test characteristics in the to-be-tested set.

6. The wind turbine generator system fault detection device based on the cost-sensitive lightweight gradient elevator as claimed in claim 5, wherein the loss function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

the objective function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

wherein, F (C)_F，C_N) Representing the cost of the fault class being classified as a normal class, F (C)_N，C_F) Representing the cost of the normal class being classified as a fault class, C_FIndicates a fault class, C_NDenotes normal class, P (c ═ F | x_i) Represents a sample x_iPosterior probability classified as fault class, P (c ═ N | x_i) Represents a sample x_iPosterior probability, y, divided into normal classes_iRepresents a sample x_iClass label of y_i1 denotes a fault sample, y_i0 denotes a normal sample, n denotes the number of samples, Ψ denotes a loss function, and Ω (f)_k) Representing a regularization term, F_t-1(x_i) Representing the last loss function, f_t(x_i) Representing the loss function at the current time.

7. The wind turbine generator system fault detection device based on the cost-sensitive lightweight gradient elevator as claimed in claim 6, wherein the training module comprises:

the dividing unit is used for dividing the target features into a training set, a test set and a verification set and preprocessing the training set;

the training unit is used for training the fault detection model of the cost-sensitive lightweight gradient elevator by utilizing the preprocessed training set, verifying the fault detection model of the trained cost-sensitive lightweight gradient elevator by utilizing the verification set and calculating an evaluation index;

and the judging unit is used for judging whether the trained fault detection model of the cost-sensitive lightweight gradient elevator reaches the optimum value by using the evaluation index, if not, acquiring the training set again, executing the step of preprocessing the training set, and if so, determining the optimally trained fault detection model of the cost-sensitive lightweight gradient elevator as the optimum fault detection model of the cost-sensitive lightweight gradient elevator.

8. The wind turbine generator system fault detection device based on the cost-sensitive lightweight gradient elevator as claimed in claim 5, wherein the screening module comprises:

and the selecting unit is used for calculating the Spearman rank correlation coefficient of each characteristic in the data set and the fault of the wind turbine generator by using a Spearman rank correlation coefficient method, and selecting the characteristic of the Spearman rank correlation coefficient in a threshold value range as the target characteristic.

9. The utility model provides a wind turbine generator system fault detection equipment based on sensitive lightweight gradient lifting machine of cost which characterized in that includes:

a memory for storing a computer program;

a processor for implementing the steps of the wind turbine generator system fault detection method based on the cost-sensitive lightweight gradient elevator according to any one of claims 1 to 4 when executing the computer program.

10. A computer-readable storage medium, characterized in that a computer program is stored in the computer-readable storage medium, and when executed by a processor, the computer program implements the steps of the wind turbine generator system fault detection method based on the cost-sensitive lightweight gradient elevator according to any one of claims 1 to 4.

Technical Field

The application relates to the technical field of wind turbine generator fault detection, in particular to a wind turbine generator fault detection method, device and equipment based on a cost-sensitive lightweight gradient elevator and a computer-readable storage medium.

Background

In the wind turbine, the wind turbine gearbox fault is the reason causing the longest downtime of the wind turbine and the largest economic loss, and the gearbox fault directly affects the overall performance of the equipment, so that the fault detection and the rapid fault identification are carried out on the wind turbine gearbox component, and the method has important significance for reducing the operation and maintenance cost of the wind turbine and improving the production efficiency of the whole wind field.

At present, a wind turbine generator fault diagnosis method is generally based on machine learning fault detection, namely existing data are analyzed and processed, a fault diagnosis model is established, and fault diagnosis is realized by using the model, while the existing machine learning algorithm is used for solving the problem of wind turbine generator fault detection, a fault sample and a normal sample are assumed to be in balanced distribution, but the existing machine learning-based wind turbine generator fault diagnosis method does not consider the problem of data imbalance and the loss caused by fault misreport and fault omission, and directly uses a kini coefficient, information gain and the like as optimization targets, and the wind turbine generator has short fault occurrence time and more normal states in the actual operation process, so that the problems of lower fault detection accuracy and poor effect exist in the wind turbine generator fault detection by using the existing machine learning algorithm.

In summary, how to improve the accuracy of detecting the fault of the wind turbine generator is a technical problem to be solved urgently by technical personnel in the field at present.

Disclosure of Invention

In view of this, an object of the present application is to provide a wind turbine generator fault detection method, device, and apparatus based on a cost-sensitive lightweight gradient elevator, and a computer-readable storage medium, which are used to improve the accuracy of wind turbine generator fault detection.

In order to achieve the above purpose, the present application provides the following technical solutions:

a wind turbine generator fault detection method based on a cost-sensitive lightweight gradient elevator comprises the following steps:

acquiring a data set of a wind turbine generator, and screening out target characteristics from the data set by using a rank correlation analysis algorithm;

training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target to be minimized, by using the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator;

and acquiring a to-be-tested set of the wind turbine generator, and performing fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and the target test characteristics in the to-be-tested set.

Preferably, the loss function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

the objective function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

wherein, F (C)_F,C_N) Indicating the cost of the fault class being classified as a normal class,F(C_N,C_F) Representing the cost of the normal class being classified as a fault class, C_FIndicates a fault class, C_NDenotes normal class, P (c ═ F | x_i) Represents a sample x_iPosterior probability classified as fault class, P (c ═ N | x_i) Represents a sample x_iPosterior probability, y, divided into normal classes_iRepresents a sample x_iClass label of y_i1 denotes a fault sample, y_i0 denotes a normal sample, n denotes the number of samples, Ψ denotes a loss function, and Ω (f)_k) Representing a regularization term, F_t-1(x_i) Representing the last loss function, f_t(x_i) Representing the loss function at the current time.

Preferably, the training of the fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target to be minimized, by using the target features to obtain the optimal fault detection model of the cost-sensitive lightweight gradient elevator, includes:

dividing the target features into a training set, a testing set and a verification set, and preprocessing the training set;

training the fault detection model of the cost-sensitive lightweight gradient elevator by using the preprocessed training set, adjusting the hyper-parameters of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using the verification set, primarily evaluating the capability of the model, evaluating the generalization capability of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using the test set, and calculating an evaluation index;

and judging whether the trained fault detection model of the cost-sensitive lightweight gradient elevator is optimal or not by using the evaluation index, if not, acquiring a training set again, and executing the step of preprocessing the training set, and if so, determining the optimally trained fault detection model of the cost-sensitive lightweight gradient elevator as the optimal fault detection model of the cost-sensitive lightweight gradient elevator.

Preferably, the screening out the target feature from the data set by using a rank correlation analysis algorithm includes:

and calculating the Spearman rank correlation coefficient of each characteristic in the data set and the fault of the wind turbine generator by using a Spearman rank correlation coefficient method, and selecting the characteristic of the Spearman rank correlation coefficient in a threshold value range as the target characteristic.

A wind turbine generator system fault detection device based on a cost-sensitive lightweight gradient elevator comprises:

the screening module is used for acquiring a data set of the wind turbine generator and screening out target features from the data set by utilizing a rank correlation analysis algorithm;

the training module is used for training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost minimized as a target, by utilizing the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator;

and the detection module is used for acquiring a to-be-tested set of the wind turbine generator and carrying out fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and the target test characteristics in the to-be-tested set.

Preferably, the loss function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

the objective function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

wherein, F (C)_F,C_N) Representing the cost of the fault class being classified as a normal class, F (C)_N,C_F) Representing the cost of the normal class being classified as a fault class, C_FIndicates a fault class, C_NDenotes normal class, P (c ═ F | x_i) Represents a sample x_iPosterior probability classified as fault class，P(c＝N|x_i) Represents a sample x_iPosterior probability, y, divided into normal classes_iRepresents a sample x_iClass label of y_i1 denotes a fault sample, y_i0 denotes a normal sample, n denotes the number of samples, Ψ denotes a loss function, and Ω (f)_k) Representing a regularization term, F_t-1(x_i) Represents the last loss, f_t(x_i) Representing the loss function at the current time.

Preferably, the training module comprises:

the dividing unit is used for dividing the target features into a training set, a test set and a verification set and preprocessing the training set;

the training unit is used for training the fault detection model of the cost-sensitive lightweight gradient elevator by utilizing the preprocessed training set, verifying the fault detection model of the trained cost-sensitive lightweight gradient elevator by utilizing the verification set and calculating an evaluation index;

and the judging unit is used for judging whether the trained fault detection model of the cost-sensitive lightweight gradient elevator reaches the optimum value by using the evaluation index, if not, acquiring the training set again, executing the step of preprocessing the training set, and if so, determining the optimally trained fault detection model of the cost-sensitive lightweight gradient elevator as the optimum fault detection model of the cost-sensitive lightweight gradient elevator.

Preferably, the screening module comprises:

and the selecting unit is used for calculating the Spearman rank correlation coefficient of each characteristic in the data set and the fault of the wind turbine generator by using a Spearman rank correlation coefficient method, and selecting the characteristic of the Spearman rank correlation coefficient in a threshold value range as the target characteristic.

A wind turbine generator system fault detection device based on a cost-sensitive lightweight gradient elevator comprises:

a memory for storing a computer program;

and the processor is used for realizing the steps of the wind turbine generator fault detection method based on the cost-sensitive lightweight gradient elevator in any one of the above manners when the computer program is executed.

A computer readable storage medium, in which a computer program is stored, which, when being executed by a processor, implements the steps of the method for detecting a fault of a wind turbine generator based on a cost-sensitive lightweight gradient hoist according to any one of the above.

The application provides a wind turbine generator fault detection method, device and equipment based on a cost-sensitive lightweight gradient elevator and a computer-readable storage medium, wherein the method comprises the following steps: acquiring a data set of a wind turbine generator, and screening target characteristics from the data set by using a rank correlation analysis algorithm; training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target to be minimized, by using the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator; and acquiring a to-be-tested set of the wind turbine generator, and performing fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and target test characteristics in the to-be-tested set.

The technical scheme disclosed by the application improves the attention to the subclass samples by embedding the misclassification cost in the fault detection model of the cost-sensitive lightweight gradient elevator, so as to improve the classification effect of unbalanced data in the fault detection of the wind turbine generator, train the fault detection model of the cost-sensitive lightweight gradient elevator with the minimized misclassification cost by utilizing the target characteristics screened from the acquired data set by the rank correlation analysis algorithm to obtain the optimal fault detection model of the cost-sensitive lightweight gradient elevator, and the fault detection is carried out by utilizing the fault detection model of the optimal cost sensitive lightweight gradient elevator, so that better fault detection is realized, the problems of unbalanced data and unequal misclassification cost are solved, the false alarm rate and the missing report rate are reduced, and the accuracy of the fault detection of the wind turbine generator is improved.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

Fig. 1 is a flowchart of a wind turbine generator fault detection method based on a cost-sensitive lightweight gradient elevator according to an embodiment of the present disclosure;

FIG. 2 is a flowchart of offline modeling and online detection of a fault detection model of a cost-sensitive lightweight gradient elevator provided in an embodiment of the present application;

fig. 3 is a flowchart of feature selection based on Spearman rank correlation coefficients according to an embodiment of the present application;

FIG. 4 is a distribution diagram of a failure detection and false-positive rate of a gearbox of a wind turbine generator system under different algorithms provided by an embodiment of the present application;

FIG. 5 is a wind turbine generator gearbox fault detection false alarm rate distribution diagram under different algorithms provided by the embodiment of the present application;

FIG. 6 is a MCC index distribution graph of three different sets of fault data provided by an embodiment of the present application;

fig. 7 is a schematic structural diagram of a wind turbine generator fault detection device based on a cost-sensitive lightweight gradient elevator according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of a wind turbine generator fault detection device based on a cost-sensitive lightweight gradient elevator according to an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

Referring to fig. 1, which shows a flowchart of a wind turbine generator fault detection method based on a cost-sensitive lightweight gradient elevator provided in an embodiment of the present application, a wind turbine generator fault detection method based on a cost-sensitive lightweight gradient elevator provided in an embodiment of the present application may include:

s11: and acquiring a data set of the wind turbine generator, and screening out target characteristics from the data set by using a rank correlation analysis algorithm.

In consideration of the fact that the existing wind turbine generator fault detection method based on machine learning assumes balanced distribution of fault samples and normal samples, and does not consider the data imbalance problem and the loss caused by fault false alarm and fault false alarm, and the wind turbine generator has short fault occurrence time and more normal states in the actual operation process, the problem that the detection accuracy is lower, and the false alarm rate are higher when the existing method is adopted to detect the wind turbine generator fault is solved.

Specifically, a Data set of the wind turbine generator may be obtained from an SCADA (Supervisory Control And Data Acquisition) database of the wind turbine generator, where the obtained Data set includes a plurality of gearbox Data characteristics, such as an average wind speed of 30s, an oil temperature inside the gearbox, an oil temperature of the gearbox, And a winding temperature of the generator set. After the data set is acquired, data preprocessing such as normalization may be performed on the data in the data set.

Considering that the wind turbine generator has massive big data, the characteristics have correlation, the characteristic relationship between the data has correlation and redundant characteristics, and the like, therefore, the characteristic selection can be performed on the massive big data of the wind turbine generator to screen out target characteristics related to fault detection from the data set, specifically, the correlation between each characteristic in the data set and the fault of the wind turbine generator can be calculated by using a rank correlation analysis algorithm, and the characteristic of which the correlation is not lower than a threshold value is selected as the target characteristic according to the calculated correlation.

S12: and training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target, by using the target characteristics to obtain the optimal fault detection model of the cost-sensitive lightweight gradient elevator.

Considering that fewer wind turbine generator fault samples and more normal samples exist, and a general classification algorithm is not suitable for unbalanced data, scholars propose a cost sensitive method, namely, a misclassification cost is introduced in attribute splitting to replace indexes such as information gain and a kini coefficient, the minimum classification cost is sought, and attention to subclass samples is improved, wherein the misclassification cost is expressed in a cost matrix mode, and the method is specifically shown in table 1:

TABLE 1 cost matrix

Wherein, C_FIndicates a fault class, C_NIndicates normal class, F (C)_F,C_F) Representing the cost of the fault class being correctly classified as a fault class, F (C)_F,C_N) Representing the cost of the fault class being classified as a normal class, F (C)_N,C_F) Representing the cost of the normal class being classified as a fault class, F (C)_N,C_N) Indicating the cost of the normal class being classified as a normal class.

If we know the misclassification cost matrix C, then when the true class is j and the prediction class is i, if i is j, then the prediction is correct, and the optimal prediction structure of one sample x should be the class that minimizes the expected total samples: r (C)_I|x)＝∑P(C_j|x)F(C_j,C_i) Wherein, P (C)_j| x) represents the division of sample x into C_jThe posterior probability of a class. Given a training set S of N samples, where S { (x)_i,y_i)},i＝1,…,N，x_i(x_iE X) denotes X in k-dimension vector space X ═ X₁,x₂,…,x_k}，y_ie.Y is x {0,1}_iClass label of y_i1 indicates a few class samples, i.e. failure samples. General formula F (C)_F,C_N)>F(C_F,C_F),F(C_N,C_F)>F(C_N,C_N). The cost sensitive nature is that even though there is a greater likelihood of assigning a sample x to a certain class, it is necessary to partition x into the class that minimizes the cost.

A lightweight Gradient hoist (LightGBM) is a Decision Tree algorithm-based distributed Gradient hoist (GBDT) framework proposed in 2017. The GBDT algorithm can process discretization information data, but only utilizes first derivative information when optimizing a loss function, and residual errors of an n-1 tree are needed when an nth tree is trained, so that parallelization operation is difficult to realize. The XGboost algorithm is characterized in that second-order derivatives are introduced to perform Taylor expansion on a loss function, L2 regularization of parameters and the like to integrally evaluate the complexity of a model, parallel calculation is supported, and the accuracy of the algorithm is improved. On the basis of the former, a decision tree algorithm based on Hisgram is provided, a leaf growth strategy with depth limitation is utilized, and multithreading optimization is adopted, so that the lightweight gradient elevator has low memory occupancy rate, can process large-scale data, and is more efficient and higher in precision.

Given a supervised learning data setThe aim of the lightweight gradient elevator is to find a mapping relationTo approximate the function f (x) such that the desired value of the loss function Ψ (y, f (x)) is minimized:

the regression tree may be represented in another form, namely w_q(x)Q ∈ {1,2, …, J }, generation JTable leaf node number, q represents decision rule of tree, w represents sample weight, objective function Obj^(t)Can be expressed as:

wherein, Ω (f)_k) Representing a regularization term.

Conventional GBDTs use the steepest descent method, which only considers the gradient of the loss function. And in the lightweight gradient elevator, the Newton method is used for quickly approximating the target function, and the target function Obj is simplified^(t)After that, it is possible to obtain:

wherein, g_i、h_iRepresenting first-order and second-order loss functions, respectively, i.e.

By means of I_jTo represent the sample set of leaf j, the first order loss function can be as follows:

given the structure q (x) of the tree, the optimal weight for each leaf nodeAndcan be obtained by quadratic programming：

The gain calculation formula is:

compared with the GBM algorithm, the lightweight Gradient hoist algorithm is more efficient in processing high-dimensional large data because Exclusive Feature Bundling (EFB) and Gradient-based One-Side Sampling (GOSS) are used. The GOSS method introduces a data example with a constant multiplier and small gradient, and can sample data with the same distribution and essence as the original data from a large amount of data, thereby reducing the data amount, ensuring the classification precision and improving the classification speed. In high dimensional space, data is sparsely encoded, while in sparse feature space, non-0 values rarely occur simultaneously. The EFB method is used for feature sampling, and two features are bound to form a new feature, so that the size of data is further reduced. In addition, the traditional gradient lifting method uses an exhaustion method to find segmented features and thresholds, and the lightweight gradient lifter uses a Histogram-based method (Histogram optimization) to find segmented features and thresholds with suboptimal solutions, so that the calculation time is reduced. Specifically, a certain feature of the data is discretized into a histogram, statistics are accumulated in the histogram according to the discretized value as an index, after the data is traversed once, the histogram accumulates needed statistics, and then the optimal segmentation point is searched in a traversing mode according to the discrete value of the histogram. And the lightweight gradient elevator is grown by a gradient-first (leaf-wise) method, so that more loss strategies can be reduced.

Considering that the wind turbine generator has fewer fault samples and more normal samples, the method improves the algorithm of the lightweight gradient elevator, introduces a Cost adjusting function to replace an information gain rate in a weight formula of the algorithm to form a Cost-sensitive lightweight gradient elevator (CS-LightGBM) algorithm, pays attention to the fault sample in each iteration update, improves the classification effect of unbalanced data, specifically, constructs a Cost-sensitive lightweight gradient elevator fault detection model embedded with misclassification Cost and taking the misclassification Cost as a target, trains the constructed Cost-sensitive lightweight gradient elevator fault detection model embedded with the misclassification Cost and taking the misclassification Cost as the target by using the target characteristics obtained in the step S11 to obtain an optimal Cost-sensitive lightweight gradient elevator fault detection model corresponding to the minimum misclassification Cost, so as to effectively improve the recognition rate of the wind turbine generator fault sample.

S13: and acquiring a to-be-tested set of the wind turbine generator, and performing fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and target test characteristics in the to-be-tested set.

After the optimal cost-sensitive lightweight gradient elevator fault detection model is obtained, a to-be-tested set of the wind turbine generator system can be obtained, target test characteristics (specifically, the target test characteristics can be obtained by a method similar to the method for obtaining the target characteristics) in the to-be-tested set are input into the optimal cost-sensitive lightweight gradient elevator fault detection model to obtain an output fault prediction value y, and an evaluation index can be simultaneously calculated, wherein if y is 1, a fault occurs, and if y is 0, a normal condition is obtained, and the failure rate and the false alarm rate can be used as the evaluation index of fault detection.

By combining the processes, the fault detection model of the optimal cost sensitive lightweight gradient elevator provided by the application is used for detecting the faults of the wind turbine generator, so that the loss caused by fault false report and fault missing report when the unbalanced data of the wind turbine generator is detected can be solved, the fault detection efficiency of the wind turbine generator can be improved, and the real-time performance of the fault detection of the wind turbine generator is ensured.

The technical scheme disclosed by the application improves the attention to the subclass samples by embedding the misclassification cost in the fault detection model of the cost-sensitive lightweight gradient elevator, so as to improve the classification effect of unbalanced data in the fault detection of the wind turbine generator, train the fault detection model of the cost-sensitive lightweight gradient elevator with the minimized misclassification cost by utilizing the target characteristics screened from the acquired data set by the rank correlation analysis algorithm to obtain the optimal fault detection model of the cost-sensitive lightweight gradient elevator, and the fault detection is carried out by utilizing the fault detection model of the optimal cost sensitive lightweight gradient elevator, so that better fault detection is realized, the problems of unbalanced data and unequal misclassification cost are solved, the false alarm rate and the missing report rate are reduced, and the accuracy of the fault detection of the wind turbine generator is improved.

According to the wind turbine generator fault detection method based on the cost-sensitive lightweight gradient elevator provided by the embodiment of the application, a loss function of a fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

the objective function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

wherein, F (C)_F,C_N) Representing the cost of the fault class being classified as a normal class, F (C)_N,C_F) Representing the cost of the normal class being classified as a fault class, C_FIndicates a fault class, C_NDenotes normal class, P (c ═ F | x_i) Represents a sample x_iPosterior probability classified as fault class, P (c ═ N | x_i) Represents a sample x_iPosterior probability, y, divided into normal classes_iRepresents a sample x_iClass label of y_i1 denotes a fault sample, y_i0 denotes a normal sample, n denotes the number of samples, Ψ denotes a loss function, and Ω (f)_k) Representing a regularization term, F_t-1(x_i) Representing the last loss function, f_t(x_i) Representing the loss function at the current time.

For the binary problem, the common logic loss function of the lightweight gradient elevator is a logarithmic loss function, and the expression is as follows:

wherein, P (x)_i) Expressing the posterior probability, and in the application, embedding a log loss function of a fault detection model of the cost-sensitive light-weight gradient elevator with misclassification cost and minimizing the misclassification cost as a target, and dividing P (x)_i) The substitutions are as follows:

wherein the content of the first and second substances,δ(x_i) And η are both intermediate variables, which are expressed by taking the two variables for convenient calculation, the logic loss function of the fault detection model of the cost-sensitive lightweight gradient elevator can be simplified as follows:

wherein P (c ═ F | x)_i) Represents a sample x_iPosterior probability classified as fault class, P (c ═ N | x_i) Represents a sample x_iPosterior probability classified as normal, apparently P (c ═ F | x)_i)＝1-P(c＝N|x_i)，y_iRepresents a sample x_iClass label of y_i1 denotes a fault sample, y_i0 denotes a normal sample.

According toThe objective function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

where Ψ denotes a loss function, i.e., a logical loss function CSloglos (x) corresponding to the above-mentioned failure detection model of the cost-sensitive lightweight gradient elevator_i,y_i)，Ω(f_k) Representing a regularization term, F_t-1(x_i) Represents the loss of the previous step, f_t(x_i) Indicating the current loss. According to the two-stage Taylor expansion, the above objective function can be written as:

wherein x is_iFirst order loss function g of_iAnd a second order loss function h_iRespectively as follows:

g_i(x_i)＝2δ[y_i-P(x_i)]

h_i(x_i)＝-4δ²p(x_i)[1-P(x_i)]

given the structure of the tree, the optimal weight for each leaf nodeThe following were used:

λ is the model complexity, wherein the pseudo code of the cost sensitive lightweight gradient elevator is shown in table 2:

TABLE 2 pseudo code for cost sensitive lightweight gradient elevators

According to the expression of the loss function of the fault detection model of the cost-sensitive lightweight gradient elevator, the misclassification cost F (C) is obtained_F,C_N)、F(C_N,C_F) Embedding the fault detection model loss function into the cost-sensitive lightweight gradient elevator, and obtaining the target function Obj of the fault detection model of the cost-sensitive lightweight gradient elevator by combining the loss function embedded with the misclassification cost^(t)The method and the device improve the attention to the fault sample, achieve better detection conveniently by minimizing the misclassification cost, and reduce the false alarm rate and the missing report rate of fault detection, so that the wind turbine generator fault detection provided by the application has good engineering application value.

The wind turbine generator fault detection method based on the cost-sensitive lightweight gradient elevator provided by the embodiment of the application trains a cost-sensitive lightweight gradient elevator fault detection model embedded with misclassification cost and taking the misclassification cost as a target by using target characteristics to obtain an optimal cost-sensitive lightweight gradient elevator fault detection model, and can include:

dividing target characteristics into a training set, a test set and a verification set, and preprocessing the training set;

training a fault detection model of the cost-sensitive lightweight gradient elevator by using a preprocessed training set, adjusting hyper-parameters of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using a verification set, primarily evaluating the capability of the model, evaluating the generalization capability of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using a test set, and calculating an evaluation index;

and judging whether the trained fault detection model of the cost-sensitive lightweight gradient elevator reaches the optimum by utilizing the evaluation index, if not, acquiring the training set again, and executing the step of preprocessing the training set, and if so, determining the trained fault detection model of the cost-sensitive lightweight gradient elevator reaching the optimum as the optimum fault detection model of the cost-sensitive lightweight gradient elevator.

Referring to fig. 2, a flowchart of offline modeling and online detection of a fault detection model of a cost-sensitive lightweight gradient elevator provided in the embodiment of the present application is shown, where the process of offline modeling (i.e., training) of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

step 1: dividing target characteristics into a training set, a testing set and a verification set, and preprocessing data in the training set; the preprocessing mentioned here includes deletion of missing values, normalization, and the like.

Step 2: training a fault detection model of the cost-sensitive lightweight gradient elevator by using a preprocessed training set, adjusting hyper-parameters of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using a verification set, primarily evaluating the capability of the model, evaluating the generalization capability of the trained fault detection model of the cost-sensitive lightweight gradient elevator by using a test set, and calculating an evaluation index;

step 3: judging whether the trained fault detection model of the cost-sensitive lightweight gradient elevator reaches the optimum according to the calculated evaluation index, if not, executing steps of acquiring a training set again according to target characteristics, executing preprocessing on the training set, training the fault detection model of the cost-sensitive lightweight gradient elevator by using the preprocessed training set and the like, and if so, determining the optimally trained fault detection model of the cost-sensitive lightweight gradient elevator as the optimum fault detection model of the cost-sensitive lightweight gradient elevator.

In the process, the training set is used for data fitting of the fault detection model of the cost-sensitive lightweight gradient elevator, the test set is used for evaluating the generalization capability of the obtained fault detection model of the cost-sensitive lightweight gradient elevator, and the verification set is used for adjusting the hyper-parameters of the fault detection model of the cost-sensitive lightweight gradient elevator and primarily evaluating the capability of the model, so that the optimal fault detection model of the cost-sensitive lightweight gradient elevator can be obtained finally.

After obtaining the optimal cost-sensitive lightweight gradient elevator fault detection model, the online fault detection shown in the right half of fig. 2 can be performed, and the main steps are as follows:

step 1: processing the test set in a manner similar to that used in the training phase, i.e., pre-processing the test set;

step 2: performing fault detection according to the fault detection model of the optimal cost-sensitive lightweight gradient elevator to obtain an output fault prediction value y, wherein if y is 1, a fault occurs, and if y is 0, the normal working condition is obtained;

step 3: and calculating an evaluation index.

The wind turbine generator fault detection method based on the cost-sensitive lightweight gradient elevator provided by the embodiment of the application screens out target characteristics from a data set by using a rank correlation analysis algorithm, and can include the following steps:

and calculating the Spearman rank correlation coefficient of each characteristic in the data set and the fault of the wind turbine generator by using a Spearman rank correlation coefficient method, and selecting the characteristic of the Spearman rank correlation coefficient in a threshold value range as a target characteristic.

In the application, a Spearman rank correlation coefficient method can be used for screening out target features from a data set, specifically, a Spearman rank correlation coefficient method can be used for calculating the Spearman rank correlation coefficient of each feature in the data set and the fault of a wind turbine generator, and the feature of the Spearman rank correlation coefficient in a threshold value range is selected as the target feature.

Wherein the Spearman rank correlation coefficient is used to measure the linear or non-linear relationship of variables. Given x, y two discrete features, M data samples, the Spearman rank correlation coefficient can be calculated by the following equation:

wherein the content of the first and second substances,the above expression is alsoThe following can be written:

where cov represents the standard deviation and σ represents the covariance. Spearman rank correlation coefficient r_sThe value range is [ -1,1 [)]R is_sWhen the value is 1, x and y are strictly in positive correlation, and r is_sWhen the formula is-1, x and y are strictly negative, and r is_sWhen 0, the two features are independent of each other.

In addition, when Spearman rank correlation coefficients are used to measure the correlation between feature vectors, data loss may be caused if the index with a higher correlation coefficient is directly deleted. Therefore, after selecting a feature with the highest information coefficient in the original data set, the features with high linear correlation with other fault features may be grouped into a set of feature sets according to a threshold until the fault features in the original data set are completely eliminated or selected, which can ensure that redundancy between the fault features is reduced and information of different features is retained, and the feature selection method is specifically shown in fig. 3, which shows a flowchart for performing feature selection based on Spearman rank correlation coefficients provided by an embodiment of the present application.

It should be noted that a Kendall rank correlation coefficient method may also be used to screen out target features in a data set.

In order to verify the effectiveness of the cost-sensitive lightweight gradient elevator in the wind turbine generator gearbox fault detection method by using other integrated algorithm methods, a comparison experiment is set, and the experiment steps are as follows:

step 1: collecting data from an SCADA database of the wind turbine generator and preprocessing the data;

step 2: performing feature selection on the extracted features by using a correlation analysis method;

step 3: dividing an existing data set into a training set, a testing set and a verification set, and establishing a cost-sensitive lightweight gradient elevator model;

step 4: carrying out online detection according to the established cost-sensitive lightweight gradient elevator model;

step 5: and evaluating the fault detection method of the cost-sensitive lightweight gradient elevator, and calculating the fault missing report rate and the false report rate.

1) Data extraction

A certain 1.5MW fan of a certain wind power plant in Shandong is taken as a research object, 3-year gearbox data of the wind power plant is extracted from SCADA data, the sampling interval is 2s, 17 state parameters are selected according to expert experience, and part of original data is shown in a table 3:

table 32018 year, 2 month and 27 day wind turbine generator set partial original data set

Data sets containing the fault of the oil temperature of the gearbox, the fault of the filtering pressure of the oil of the gearbox and the fault of the oil level of the gearbox are selected from the normal working condition data of the SCADA and are respectively recorded as a data set 1, a data set 2 and a data set 3, and the data sets are shown in a table 4:

table 4 data description

2) Feature selection

The health of the gearbox is evaluated by means of gearbox bearing temperature information, and when selecting the state parameter, the parameter that has a greater impact on this parameter is mainly selected. The correlation strength of each state parameter with the gearbox bearing temperature was calculated according to the Spearman rank correlation coefficient analysis method, as shown in table 5:

TABLE 5 wind turbine generator gearbox shaft temperature Spearman rank correlation analysis results

As can be seen from the correlation analysis results in Table 5, the temperature correlation between the state parameters and the gearbox bearing is greatly different, and in order to avoid the influence of irrelevant and weakly relevant state parameters on the gearbox fault detection, the characteristic that the correlation coefficient is between +/-0.50 and +/-0.95 is selected, as shown in the bold part of Table 5.

3) Gear case fault detection evaluation index

And recording 4 states corresponding to a normal state, a gear box oil temperature overrun, a gear box oil filtering pressure fault and a gear box oil level fault as P ═ 0,1,2 and 3 respectively, dividing the data set into 4 parts, combining the 3 faults and the normal state in sequence, combining the four groups of data sets, and performing fault diagnosis through a LightGBMCost algorithm to obtain four groups of classification indexes. In order to measure the classification condition of the unbalanced data, a Matthews Correlation Coefficient (MCC) is introduced to evaluate a fault detection model. Meanwhile, the False Alarm Rate (FAR) and the Missing Detection Rate (MDR) are used as the fault detection evaluation indexes. The binary problem mixing matrix is shown in the following table:

TABLE 6 hybrid matrix

Wherein TP is the number of the fault samples predicted to be fault samples, TN is the number of the normal samples predicted to be normal samples, FP is the number of the normal samples predicted to be fault samples, and FN is the number of the normal samples predicted to be normal samples. The indexes under the two classifications are as follows:

4) experimental results and discussion

The experimental data is obtained by selecting a three-year SCADA data set from a certain wind power plant for experiment, and the effectiveness of the proposed cost-sensitive lightweight gradient elevator on the fault detection of the fan gearbox is verified through experiments. In order to further prove the superiority of the method, three advanced fault diagnosis methods are compared and researched, wherein the three advanced fault diagnosis methods comprise cost-sensitive AdaBoost (Adacost), cost-sensitive GBDT (GBDTcost) and cost-sensitive XGBoost (XGBboost) and cost-sensitive lightweight gradient hoisting machine (LightGBMboost). By using different evaluation standards in three different data sets, the test comparison false-positive rate and the test comparison false-positive rate of the proposed algorithm, the adaboost algorithm, the GBDTcost algorithm and the XGBboost algorithm under different fault conditions are respectively shown in fig. 4 and fig. 5, wherein fig. 4 shows a wind turbine generator gearbox fault detection false-positive rate distribution diagram under different algorithms provided by the embodiment of the application, fig. 5 shows a wind turbine generator gearbox fault detection false-positive rate distribution diagram under different algorithms provided by the embodiment of the application, as can be seen from fig. 4 and fig. 5, the cost-sensitive lightweight gradient elevator method is lower than the other three algorithms in the FAR and MDR aspects, and the XGBboost index is generally superior to the adaboost and GBDTcost methods. When analyzing the failure data set 2, the FAR index of the LightGBMcost method is only 1.43%, and the MDR index is only 1.01%. The method has good fault detection performance. The traditional cost-sensitive Boost method has the problems of high missing report rate and false report rate in the fault detection process, and the missing report rate and the false report rate of the LightGBMcmos method are lower than those of the other three methods. To avoid overfitting the model, the model was evaluated using a five-fold cross-validation method. The smaller the FAR and MDR, the better the performance.

In addition, fig. 6 shows MCC index profiles of three different fault data sets provided by the embodiment of the present application, and the MCC index can also be used in case of sample imbalance. The closer the MCC index is to 1, the better the process performance. As can be seen from fig. 6, the MCC index of the cost sensitive lightweight gradient elevator method in data set 2 is as high as 99.61%, and the MCC indices of the remaining data sets are all higher than those of the other three methods.

By combining the processes, the fault rate of the wind turbine gearbox is increased and faults are easy to occur when the wind turbine gearbox works in a severe operation condition for a long time, the accuracy of diagnosis of the wind turbine gearbox is often influenced by a plurality of factors such as severe environment and extreme weather, in order to improve the accuracy of fault detection and analyze and compare the defects of a traditional algorithm, a wind turbine gearbox fault detection method based on a cost-sensitive lightweight gradient elevator is provided, the method can (1) analyze the fault characteristics of the wind turbine gearbox and extract fault characteristic variables and fault characteristic indexes of the wind turbine gearbox, and the correlation among the characteristic indexes is used for improving the fault diagnosis performance (2) the method based on the cost-sensitive lightweight gradient elevator is provided and applied to actual fault diagnosis of the wind turbine and compared with the traditional sensitive method.

Experimental results show that the method is superior to the traditional cost-sensitive Boost classification method in the aspects of FDR, MDR and MCC indexes. The method provided by the application combines the advantages of mechanism and data modeling, determines a proper diagnosis data source based on the mechanism, selects the lightweight gradient elevator method based on the data to design the cost-sensitive lightweight gradient elevator algorithm, and has the advantages of clear physical significance, simple structure, easiness in engineering realization and the like, and the application prospect is good.

The embodiment of the present application further provides a wind turbine generator system fault detection device based on the cost-sensitive lightweight gradient elevator, refer to fig. 7, which shows a schematic structural diagram of the wind turbine generator system fault detection device based on the cost-sensitive lightweight gradient elevator provided in the embodiment of the present application, and the schematic structural diagram may include:

the screening module 71 is used for acquiring a data set of the wind turbine generator and screening out target features from the data set by using a rank correlation analysis algorithm;

the training module 72 is used for training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost minimized as a target, by using the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator;

and the detection module 73 is used for acquiring a to-be-tested set of the wind turbine generator and performing fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and target test characteristics of the to-be-tested set.

According to the wind turbine generator fault detection device based on the cost-sensitive lightweight gradient elevator, the loss function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

the objective function of the fault detection model of the cost-sensitive lightweight gradient elevator is as follows:

wherein, F (C)_F,C_N) Representing the cost of the fault class being classified as a normal class, F (C)_N,C_F) Representing the cost of the normal class being classified as a fault class, C_FIndicates a fault class, C_NDenotes normal class, P (c ═ F | x_i) Represents a sample x_iPosterior probability classified as fault class, P (c ═ N | x_i) Represents a sample x_iPosterior probability, y, divided into normal classes_iRepresents a sample x_iClass label of y_i1 denotes a fault sample, y_i0 denotes a normal sample, n denotes the number of samples, Ψ denotes a loss function, and Ω (f)_k) Representing a regularization term, F_t-1(x_i) Representing the last loss function, f_t(x_i) Representing the loss function at the current time.

According to the wind turbine generator fault detection device based on the cost-sensitive lightweight gradient elevator, the training module 72 can include:

the dividing unit is used for dividing the target characteristics into a training set, a test set and a verification set and preprocessing the training set;

the training unit is used for training the fault detection model of the cost-sensitive lightweight gradient elevator by using the preprocessed training set, adjusting the hyper-parameters of the fault detection model of the trained cost-sensitive lightweight gradient elevator by using the verification set, primarily evaluating the capability of the model, evaluating the generalization capability of the fault detection model of the trained cost-sensitive lightweight gradient elevator by using the test set, and calculating the evaluation index;

and the judging unit is used for judging whether the trained fault detection model of the cost-sensitive lightweight gradient elevator reaches the optimum value by utilizing the evaluation index, if not, acquiring the training set again, executing the step of preprocessing the training set, and if so, determining the trained fault detection model of the cost-sensitive lightweight gradient elevator reaching the optimum value as the optimum fault detection model of the cost-sensitive lightweight gradient elevator.

The wind turbine generator system fault detection device based on the cost-sensitive lightweight gradient elevator provided by the embodiment of the application has the advantages that the screening module 71 can comprise:

and the selecting unit is used for calculating the Spearman rank correlation coefficient of each characteristic in the data set and the fault of the wind turbine generator by using a Spearman rank correlation coefficient method, and selecting the characteristic of the Spearman rank correlation coefficient in a threshold value range as a target characteristic.

The embodiment of the present application further provides a wind turbine generator fault detection device based on a cost-sensitive lightweight gradient elevator, refer to fig. 8, which shows a schematic structural diagram of a wind turbine generator fault detection device based on a cost-sensitive lightweight gradient elevator, provided by the embodiment of the present application, and the schematic structural diagram may include:

a memory 81 for storing a computer program;

the processor 82, when executing the computer program stored in the memory 81, may implement the following steps:

acquiring a data set of a wind turbine generator, and screening target characteristics from the data set by using a rank correlation analysis algorithm; training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target to be minimized, by using the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator; and acquiring a to-be-tested set of the wind turbine generator, and performing fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and target test characteristics in the to-be-tested set.

An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored, and when executed by a processor, the computer program can implement the following steps:

acquiring a data set of a wind turbine generator, and screening target characteristics from the data set by using a rank correlation analysis algorithm; training a fault detection model of the cost-sensitive lightweight gradient elevator, which is embedded with a misclassification cost and takes the misclassification cost as a target to be minimized, by using the target characteristics to obtain an optimal fault detection model of the cost-sensitive lightweight gradient elevator; and acquiring a to-be-tested set of the wind turbine generator, and performing fault detection by using the optimal cost-sensitive lightweight gradient elevator fault detection model and target test characteristics in the to-be-tested set.

The computer-readable storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.

For the descriptions of the wind turbine generator fault detection device and equipment based on the cost-sensitive lightweight gradient elevator and the relevant parts in the computer-readable storage medium provided in the embodiment of the present application, reference may be made to the detailed descriptions of the corresponding parts in the wind turbine generator fault detection method based on the cost-sensitive lightweight gradient elevator provided in the embodiment of the present application, and details are not repeated here.

It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include elements inherent in the list. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element. In addition, parts of the above technical solutions provided in the embodiments of the present application, which are consistent with the implementation principles of corresponding technical solutions in the prior art, are not described in detail so as to avoid redundant description.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.