CN114626415A - Wind turbine generator set composite fault diagnosis method based on artificial intelligence - Google Patents

Wind turbine generator set composite fault diagnosis method based on artificial intelligence Download PDF

Info

Publication number
CN114626415A
CN114626415A CN202210245773.0A CN202210245773A CN114626415A CN 114626415 A CN114626415 A CN 114626415A CN 202210245773 A CN202210245773 A CN 202210245773A CN 114626415 A CN114626415 A CN 114626415A
Authority
CN
China
Prior art keywords
sample
fault
wind turbine
turbine generator
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202210245773.0A
Other languages
Chinese (zh)
Other versions
CN114626415B (en
Inventor
殷林飞
王恬
高放
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangxi University
Original Assignee
Guangxi University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangxi University filed Critical Guangxi University
Priority to CN202210245773.0A priority Critical patent/CN114626415B/en
Publication of CN114626415A publication Critical patent/CN114626415A/en
Application granted granted Critical
Publication of CN114626415B publication Critical patent/CN114626415B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12—Classification; Matching

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Mathematical Physics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Medical Informatics (AREA)
  • Signal Processing (AREA)
  • Control Of Eletrric Generators (AREA)

Abstract

The invention provides a wind turbine generator complex fault diagnosis method based on artificial intelligence, which comprises the following steps of firstly, acquiring a data sample in a wind turbine generator data acquisition and detection system by the artificial intelligence method, and processing the dirty data sample by an expectation maximization clustering method and a lazy matrix of a Gaussian mixture model; secondly, extracting features by using a convolutional neural network method, establishing a sample attribute space, using a generalized zero-order learning method to generate unknown fault samples, and performing pre-judgment classification on the samples by using a gating method; respectively completing diagnosis of a single fault sample and a composite fault sample through a multivariate classifier and similarity evaluation; and finally, finishing final diagnosis of the wind turbine generator fault through stack-based ensemble learning. The provided artificial intelligence method can solve the problem of large demand of a fault diagnosis model sample, realize the diagnosis of the compound fault of the wind turbine generator, simultaneously realize the comprehensive diagnosis of the fault of the wind turbine generator, and improve the diagnosis precision and the diagnosis speed of the compound fault of the wind turbine generator.

Description

Wind turbine generator set composite fault diagnosis method based on artificial intelligence
Technical Field
The invention belongs to the field of fault diagnosis of wind turbine generators of power systems, relates to a composite fault diagnosis method based on artificial intelligence, and is suitable for composite fault diagnosis of the wind turbine generators of the power systems.
Background
At present, according to the strategic goals of carbon peak reaching and carbon neutralization in China, the installed capacity of wind power integration in China breaks through 25 hundred million kilowatts in 2060. The wind turbine generator is a complex mechanical system and works in severe environment for a long time, so that the wind turbine generator is easy to break down. The causes of the failure are environmental factors, high volatility and randomness of the wind, aging, saturation or thermal problems of the system components. In the actual industry, the wind turbine inevitably has compound faults, which is one of the difficulties in intelligent diagnosis of the faults of the wind turbine.
In the field of wind turbine fault diagnosis, machine learning-based methods are typically supervised or semi-supervised learning, and each compound fault type requires sufficient training data to train before a compound fault is successfully diagnosed. In the actual operation process of the wind turbine generator, the composite fault samples are far less than single fault samples, and unknown composite fault types which do not occur can also appear. In summary, complex fault samples are often difficult to collect, resulting in the absence of complex fault samples. The current machine learning composite fault diagnosis method has the limitations that: the composite fault type is regarded as a new single fault type, and the concurrency and coupling relation between the single fault and the composite fault is ignored; at the same time, a large number of labeled composite fault samples are required during the training process. Therefore, unknown compound fault types and compound fault samples are absent, and the two problems cause that the precision of the compound fault diagnosis model of the wind turbine generator at present is not high and the accuracy is low.
Disclosure of Invention
The invention provides a wind turbine generator set composite fault diagnosis method based on artificial intelligence, wherein the artificial intelligence method combines a lazy matrix, a convolutional neural network attribute space model, a generalized zero-order learning method and stack-based integrated learning and is used for diagnosing single faults and composite faults of a wind turbine generator set; the provided artificial intelligence method comprises the following steps in the using process:
step (1): in a wind turbine generator data acquisition and monitoring system, acquiring a wind turbine generator operation original data sample set D { (X, Y) | X ∈ X, and Y ∈ Y }; x is a feature vector; x is a feature vector set; y is a fault label; y is a failure tag set; a small amount of unusable data, namely dirty data samples, exists in the original data sample set; clustering the original data sample set by using an expectation maximization clustering method of a Gaussian mixture model, wherein the clustering results are a non-dirty data sample set FZ and a dirty data sample set Z; wherein the dimension of the feature vector in the non-dirty data sample is nFZThe dimension of the dirty feature vector in the dirty data sample is nZ(ii) a Availability of dirty data sample sets Z with a lazy matrix (n)FZ-nZ) Expanding dimension data into nFZDimension data, feature vector x of extended dirty data sampleFZComprises the following steps:
xFZ＝xFZ-ZLlazy (1)
wherein x isFZ-ZIs a feature vector available in dirty data samples of dimension (n)FZ-nZ)；LlazyIs (n)FZ-nZ) Line nFZA lazy matrix of columns;
step (2): converting the eigenvectors in the wind turbine generator data sample after the lazy matrix into the eigenvectors; converting the characteristic vector X into a characteristic matrix X 'by using a signal image conversion method, normalizing the numerical value of the characteristic matrix X', scaling the numerical value of the characteristic matrix X 'into a range of [0,1], so as to obtain a sample set D', D { (X ', Y) | X' ∈ X ', Y ∈ Y }, wherein X' is a characteristic matrix set;
and (3): learning a characteristic matrix set X' by using a convolutional neural network method to form a characteristic space and establishing a sample attribute space;
3.1) constructing a feature space, inputting a feature matrix X ', X ' belongs to X ', performing deep fault feature extraction through a convolutional neural network, and outputting a feature matrix X ', X ' belongs to X ', wherein X ' is an output feature matrix set; the convolutional neural network is composed of a convolutional layer, an active layer, a pooling layer and a full-connection layer, wherein the convolutional layer receives input and convolutes an input characteristic matrix, and nonlinear factors are introduced into the active layer;
output characteristic matrix u of the first convolutional layerlComprises the following steps:
ul＝xl-1*kl+bl (2)
wherein x isl-1Is the input feature matrix of the first convolution layer; "+" indicates convolution operation; k is a radical oflIs a kernel of layer i channel size; blIs the bias vector for layer l;
output characteristic matrix x of the l-th active layerlIs as follows;
xl＝σ(ul) (3)
where σ (·) is the activation function;
the pooling layer compresses the input feature matrix, the full connection is responsible for output, and the output feature matrix z of the first layer of pooling layerlComprises the following steps:
zl＝max(xl) (4)
where max (·) is a function of maximum pooling;
3.2) constructing a sample attribute space, using the single fault sample characteristics of the wind turbine generator extracted by the convolutional neural network as a base vector, wherein the single fault sample attribute space set of the wind turbine generator is represented as A0＝{a1,a2,...,ai,...,apAnd f, wherein the single fault sample attributes of the wind turbine generator are p in total, and the single fault sample attribute a of the wind turbine generator is ai＝[αi,1,αi,2,...,ai,j,...,αi,m]，αi,jThe j-dimension attribute of the ith type of wind turbine generator single fault sample in the fault attribute space is defined, and the fault attribute space has m dimensions; establishing a wind turbine generator set composite fault sample attribute space set represented as A through a characteristic relation1＝{A12,A13,...,A1i,...,A1kWind, windMotor group composite fault sample attribute set A1iIn order to simultaneously have i types of single fault sample attributes of the wind turbine generator, the compound fault sample attribute set of the wind turbine generator simultaneously has k types of single fault sample attributes of the wind turbine generator at most;
and (4): the generalized zero-time learning method is used for generating unknown fault samples;
4.1) using the failure sample attribute set A ═ A0∪A1The known fault sample set S is:
S＝{(xs,as,ys)|xs∈X″,as∈A,ys∈Ys} (5)
wherein x issRepresenting a known fault signature matrix; a issAttributes representing known fault samples; y issIndicating a known fault label; y issRepresenting a set of known fault tags; the set U of unknown fault sample attributes and unknown fault labels is:
U＝{(au,yu)|au∈A,yu∈Yu} (6)
wherein, auAttributes representing unknown fault samples; y isuIndicating an unknown fault label; y isuRepresenting a set of unknown fault labels;
4.2) generating unknown fault samples by adopting a condition-based variational automatic coding method in the generalized zero-order learning method; the conditional variational automatic coding consists of an encoder and a decoder, x', asAnd auIs used as input to the encoder, which will apply x ", asAnd auThe combination of (a) is converted into a latent variable z, subject to a gaussian distribution of mean μ and variance σ; the encoder outputs a latent variable z; z, asAnd auIs used as the input of the decoder to obtain the decoder output reconstruction feature matrix
Figure BDA0003544402690000021
Figure BDA0003544402690000022
Reconstructing a characteristic matrix set for the known fault sample and the unknown fault sample to obtain a reconstructed characteristic matrix setKnown fault sample set
Figure BDA0003544402690000031
Figure BDA0003544402690000032
Representing the reconstructed known fault feature matrix and unknown fault sample set
Figure BDA0003544402690000033
Figure BDA0003544402690000034
Representing the reconstructed unknown fault feature matrix;
and (5): a gating method is utilized to prejudge the sample as a single fault sample or a composite fault sample;
5.1) a gating method based on a binary classification principle is used for prejudging whether a sample belongs to a single fault sample or a composite fault sample, wherein a sample in a normal state is regarded as the single fault sample; gate control method calculates reconstruction loss function L through square of 2-normrec；
Reconstruction loss function
Figure BDA0003544402690000035
Comprises the following steps:
Figure BDA0003544402690000036
wherein x isi"represents the i-th feature matrix before sample reconstruction;
Figure BDA0003544402690000037
representing an ith feature matrix after sample reconstruction; r represents the total number of feature matrices; i | · | purple wind2Represents a 2 norm;
the minimum optimization function L (-) is:
Figure BDA0003544402690000038
wherein KL (·) represents a Kulbeckleffler divergence function; n (0, I) is a normal distribution; | | is OR operation;
5.2) for a test sample t, calculating the minimum optimized value L of the test sample ttBy comparing the minimum optimum value LtMaximum L of minimum optimization in training samplemaxThe method is characterized in that whether the test sample t belongs to a single fault sample or a composite fault sample is judged in advance, the diagnosis speed is improved, and the method is divided into the following two conditions:
if L ist＜LmaxPre-judging that the test sample t belongs to a single fault sample, and inputting the test sample t into a trained multivariate Softmax classifier; if L ist≥LmaxPre-judging that the test sample t belongs to a composite fault sample, and inputting the test sample t into an attribute space of the composite fault sample;
and (6): respectively carrying out fault diagnosis on the single fault sample and the composite fault sample;
6.1) Single failure sample: training a multivariate Softmax classifier to diagnose;
6.2) composite fault sample: in composite fault sample attribute space A1In (1), diagnosis is performed by similarity evaluation of 2-norm, objective function Emin(. is):
Figure BDA0003544402690000039
wherein,
Figure BDA00035444026900000310
representing the attribute corresponding to the ith feature matrix of the test sample t; a iscRepresenting the c-th composite fault sample attribute;
Figure BDA00035444026900000311
is a regular coefficient;
Figure BDA00035444026900000312
representing the ith characteristic matrix after the test sample t is reconstructed;
and (7): and (3) using stack-based integrated learning for final fault diagnosis of the wind turbine generator, using the traditional modal decomposition model and the model as a base model, training the model on the k-1 fold by using a k-fold verification method, verifying on the k-fold, and using the trained base model for integration to realize final fault diagnosis.
Compared with the prior art, the invention has the following advantages and effects:
(1) dirty data samples are processed through an expectation maximization clustering method and a lazy matrix of a Gaussian mixture model to form non-dirty data samples, and higher data quality is provided for training and testing of subsequent models.
(2) By the generalized zero-time learning method, the unknown fault sample is generated by utilizing the known fault sample and the fault attribute, the problem of large demand of the current fault diagnosis model sample is solved, the diagnosis of the unknown fault type is realized, the problem that the model is easy to misdiagnose or miss-diagnose is solved, and the diagnosis precision of the model is improved.
(3) The gating method is utilized to prejudge the sample to be a single fault sample or a composite fault sample, so that comprehensive diagnosis of different fault types is efficiently realized, and the diagnosis speed of the model is increased.
(4) Compared with the existing fault diagnosis model, the method utilizes stack-based integrated learning, combines the traditional modal decomposition and the artificial intelligence model, and further improves the diagnosis precision of the model.
Drawings
FIG. 1 is a flow chart of the wind turbine complex fault diagnosis of the method of the present invention.
FIG. 2 is a schematic diagram of a generalized zero-order learning generative model of the method of the present invention.
FIG. 3 is a schematic diagram of a convolutional neural network learning forming sample feature space of the method of the present invention.
FIG. 4 is a schematic diagram of the gate control method of the present invention.
FIG. 5 is a schematic diagram of stack-based ensemble learning for the method of the present invention.
Detailed Description
The invention provides a wind turbine generator set composite fault diagnosis method based on artificial intelligence, which is described in detail in the following steps in combination with the accompanying drawings:
FIG. 1 is a flow chart of the wind turbine complex fault diagnosis of the method of the present invention. Firstly, collecting original data samples, and processing dirty data samples in the original data samples through an expectation maximization clustering method of a Gaussian mixture model and a lazy matrix to form non-dirty data samples. The method comprises the following steps that non-dirty data samples are used as input and are respectively applied to two models, one is that original data samples are used as time series data, and a modal decomposition method is selected to realize fault diagnosis; and the other method converts the characteristics of the original data sample into a characteristic matrix which is used as the input of a convolutional neural network, and forms a characteristic space through the learning of the convolutional neural network to obtain the final output characteristics. Then, constructing a generalized zero-order learning generation model to generate unknown fault samples, and carrying out prejudgment through a gating method to classify the unknown fault samples into single fault samples and composite fault samples; completing the establishment of a sample attribute space on the basis of the feature space; the single fault sample is subjected to single fault diagnosis through a multivariate Softmax classifier, and the composite fault is subjected to composite fault diagnosis through sample space and similarity evaluation. And finally, combining the two models by using a stack-based ensemble learning method to realize final fault diagnosis.
FIG. 2 is a schematic diagram of a generalized zero-order learning generative model of the method of the invention. First, a fault attribute and a known fault sample are obtained. The generator is then trained using known fault samples and fault attributes, and finally, unknown fault samples are generated with unknown fault attributes.
FIG. 3 is a schematic diagram of a convolutional neural network learning forming sample feature space of the method of the present invention. First, the input layer is composed of sensing nodes, accepting a feature matrix. The calculation flow then alternates between convolution and sub-sampling: the first hidden layer is convoluted and is composed of feature maps, and each feature map is composed of neurons; the second hidden layer realizes sampling and local averaging; the third hidden layer is convolved a second time, which operates in a similar way as the first convolutional layer; the fourth hidden layer performs a second sub-sampling and local average calculation, which operates in a similar manner to the first sampling; the fifth hidden layer realizes the final stage of convolution, and finally, the full connection layer obtains output characteristics for constructing a characteristic space.
FIG. 4 is a schematic diagram of the gate control method of the present invention. Firstly, the condition variational mode automatic coding consists of an encoder and a decoder, and the encoder outputs a latent variable z in a training stage; z, asAnd auIs used as input to the decoder and the output of the decoder is obtained by a latent variable z obeying a gaussian distribution
Figure BDA0003544402690000041
Then, a gating method formed by binary classification is utilized to carry out prejudgment through training; and finally, selecting a test sample to be judged as a single fault sample or a composite fault sample in the test stage.
FIG. 5 is a schematic diagram of stack-based ensemble learning for the method of the present invention. Firstly, training a model on a k-1 fold by using a k-fold verification method, and verifying on a k-th fold; then, integrating the trained base models to obtain integrated models; and finally, operating the integrated model on the test set to obtain a final fault diagnosis result.

Claims (1)

1. A wind turbine generator set composite fault diagnosis method based on artificial intelligence is characterized in that the artificial intelligence method combines a lazy matrix, a convolutional neural network attribute space model, a generalized zero-order learning method and stack-based integrated learning and is used for diagnosing single faults and composite faults of a wind turbine generator set; the provided artificial intelligence method comprises the following steps in the using process:
step (1): in a wind turbine generator data acquisition and monitoring system, acquiring a wind turbine generator operation original data sample set D { (X, Y) | X ∈ X, Y ∈ Y }; x is a feature vector; x is a feature vector set; y is a fault label; y is a failure tag set; a small amount of unavailable data, namely dirty data samples, exists in the original data sample set; clustering the original data sample set by using an expectation maximization clustering method of a Gaussian mixture model, wherein the clustering result is non-dirty dataA sample set FZ and a dirty data sample set Z; wherein the dimension of the feature vector in the non-dirty data sample is nFZThe dimension of the dirty feature vector in the dirty data sample is nZ(ii) a Availability of dirty data sample sets Z with lazy matrices (n)FZ-nZ) Dimension data expansion as nFZDimension data, feature vector x of extended dirty data sampleFZComprises the following steps:
xFZ＝xFZ-ZLlazy (1)
wherein x isFZ-ZIs a feature vector available in dirty data samples, with dimension (n)FZ-nZ)；LlazyIs (n)FZ-nZ) Line nFZA lazy matrix of columns;
step (2): converting the eigenvectors in the wind turbine generator data sample after the lazy matrix into the eigenvectors; converting the feature vector X into a feature matrix X 'by using a signal image conversion method, normalizing the numerical value of the feature matrix X', scaling the numerical value of the feature matrix X 'into a range of [0,1], so as to obtain a sample set D', D { (X ', Y) | X' belongs to X ', Y belongs to Y }, wherein X' is a feature matrix set;
and (3): learning a characteristic matrix set X' by using a convolutional neural network method to form a characteristic space and establishing a sample attribute space;
3.1) constructing a feature space, inputting a feature matrix X ', X ' belongs to X ', performing deep fault feature extraction through a convolutional neural network, and outputting a feature matrix X ', X ' belongs to X ', wherein X ' is an output feature matrix set; the convolutional neural network is composed of a convolutional layer, an active layer, a pooling layer and a full-connection layer, wherein the convolutional layer receives input and convolutes an input characteristic matrix, and nonlinear factors are introduced into the active layer;
output characteristic matrix u of the first convolutional layerlComprises the following steps:
ul＝xl-1*kl+bl (2)
wherein x isl-1Is the input feature matrix of the first convolution layer; "+" indicates convolution operation; k is a radical oflIs the first layer of channel is largeA small kernel; blIs the bias vector for layer l;
output characteristic matrix x of the l-th active layerlIs as follows;
xl＝σ(ul) (3)
where σ (·) is the activation function;
the pooling layer compresses the input feature matrix, the full connection is responsible for output, and the output feature matrix z of the first layer of pooling layerlComprises the following steps:
zl＝max(xl) (4)
where max (·) is a function of maximum pooling;
3.2) constructing a sample attribute space, using the single fault sample characteristics of the wind turbine generator extracted by the convolutional neural network as a base vector, wherein the single fault sample attribute space set of the wind turbine generator is represented as A0＝{a1,a2,...,ai,...,apAnd f, wherein the single fault sample attributes of the wind turbine generator are p in total, and the single fault sample attribute a of the wind turbine generator is ai＝[αi,1,αi,2,...,ai,j,...,αi,m]，αi,jThe j-dimension attribute of the ith type of wind turbine generator single fault sample in the fault attribute space is defined, and the fault attribute space has m dimensions; establishing a wind turbine generator set composite fault sample attribute space set represented as A through a characteristic relation1＝{A12,A13,...,A1i,...,A1kWind turbine generator set composite fault sample attribute set A1iIn order to simultaneously have i types of single fault sample attributes of the wind turbine generator, the compound fault sample attribute set of the wind turbine generator simultaneously has k types of single fault sample attributes of the wind turbine generator at most;
and (4): the generalized zero-order learning method is used for generating unknown fault samples;
4.1) Using the failure sample Attribute set A ═ A0∪A1The known fault sample set S is:
S＝{(xs,as,ys)|xs∈X″,as∈A,ys∈Ys} (5)
wherein x issRepresenting a known fault signature matrix; a issAttributes representing known fault samples; y issIndicating a known fault label; y issRepresenting a set of known fault tags;
the set U of unknown fault sample attributes and unknown fault labels is:
U＝{(au,yu)|au∈A,yu∈Yu} (6)
wherein, auAttributes representing unknown fault samples; y isuRepresenting an unknown fault label; y isuRepresenting a set of unknown fault labels;
4.2) generating unknown fault samples by adopting a condition-based variational automatic coding method in the generalized zero-order learning method; the conditional variational automatic coding consists of an encoder and a decoder, x', asAnd auIs used as input to the encoder, which will apply x ", asAnd auThe combination of (a) is converted into a latent variable z, subject to a gaussian distribution of mean μ and variance σ; the latent variable z is output by the encoder; z, asAnd auIs used as the input of the decoder to obtain the decoder output reconstruction feature matrix
Figure FDA0003544402680000021
Figure FDA0003544402680000022
Figure FDA0003544402680000023
Reconstructing a characteristic matrix set for the known fault sample and the unknown fault sample to obtain a reconstructed known fault sample set
Figure FDA0003544402680000024
Figure FDA0003544402680000025
Representing a reconstructed known fault signature matrix, unknownSample set of faults
Figure FDA0003544402680000026
Figure FDA0003544402680000027
Representing the reconstructed unknown fault characteristic matrix;
and (5): a gating method is utilized to prejudge the sample as a single fault sample or a composite fault sample;
5.1) a gating method based on a binary classification principle is used for prejudging whether a sample belongs to a single fault sample or a composite fault sample, wherein a sample in a normal state is regarded as the single fault sample; the gate control method calculates the reconstruction loss function L through the square of 2-normrec；
Reconstructing a loss function
Figure FDA0003544402680000028
Comprises the following steps:
Figure FDA0003544402680000029
wherein x isi"represents the i-th feature matrix before sample reconstruction;
Figure FDA00035444026800000210
representing an ith feature matrix after sample reconstruction; r represents the total number of feature matrices; i | · | purple wind2Represents a 2 norm;
the minimum optimization function L (-) is:
Figure FDA0003544402680000031
wherein KL (·) represents a Kulbeckleffler divergence function; n (0, I) is a normal distribution; | | is OR operation;
5.2) for a test sample t, calculating the minimum optimized value L of the test sample ttBy comparing the minimum optimum value LtMaximum L of minimum optimization in training samplemaxThe method is characterized in that whether the test sample t belongs to a single fault sample or a composite fault sample is judged in advance, the diagnosis speed is improved, and the method is divided into the following two conditions:
if L ist＜LmaxPre-judging that the test sample t belongs to a single fault sample, and inputting the test sample t into a trained multivariate Softmax classifier;
if L ist≥LmaxPre-judging that the test sample t belongs to a composite fault sample, and inputting the test sample t into an attribute space of the composite fault sample;
and (6): respectively carrying out fault diagnosis on the single fault sample and the composite fault sample;
6.1) Single failure sample: training a multivariate Softmax classifier to diagnose;
6.2) composite fault samples: in composite fault sample attribute space A1In (1), diagnosis is performed by similarity evaluation of 2-norm, objective function Emin(. is):
Figure FDA0003544402680000032
wherein,
Figure FDA0003544402680000033
representing the attribute corresponding to the ith feature matrix of the test sample t; a is acRepresenting the c-th composite fault sample attribute;
Figure FDA0003544402680000034
is a regular coefficient;
Figure FDA0003544402680000035
representing the ith characteristic matrix after the test sample t is reconstructed;
and (7): and (3) using stack-based integrated learning for final fault diagnosis of the wind turbine generator, using the traditional mode decomposition model and the model as a base model, training the model on a k-1 fold by using a k-fold verification method, verifying on a k-fold, and using the trained base model for integration to realize final fault diagnosis.
CN202210245773.0A 2022-03-14 2022-03-14 Wind turbine generator composite fault diagnosis method based on artificial intelligence Active CN114626415B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210245773.0A CN114626415B (en) 2022-03-14 2022-03-14 Wind turbine generator composite fault diagnosis method based on artificial intelligence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210245773.0A CN114626415B (en) 2022-03-14 2022-03-14 Wind turbine generator composite fault diagnosis method based on artificial intelligence

Publications (2)

Publication Number Publication Date
CN114626415A true CN114626415A (en) 2022-06-14
CN114626415B CN114626415B (en) 2024-08-09

Family

ID=81902643

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210245773.0A Active CN114626415B (en) 2022-03-14 2022-03-14 Wind turbine generator composite fault diagnosis method based on artificial intelligence

Country Status (1)

Country Link
CN (1) CN114626415B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108875818A (en) * 2018-06-06 2018-11-23 西安交通大学 Based on variation from code machine and confrontation network integration zero sample image classification method
CN112393934A (en) * 2020-11-20 2021-02-23 湖南工业大学 Wind turbine generator fault diagnosis method based on sparse self-coding and extreme learning machine
CN113609569A (en) * 2021-07-01 2021-11-05 湖州师范学院 Discriminant generalized zero-sample learning fault diagnosis method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108875818A (en) * 2018-06-06 2018-11-23 西安交通大学 Based on variation from code machine and confrontation network integration zero sample image classification method
CN112393934A (en) * 2020-11-20 2021-02-23 湖南工业大学 Wind turbine generator fault diagnosis method based on sparse self-coding and extreme learning machine
CN113609569A (en) * 2021-07-01 2021-11-05 湖州师范学院 Discriminant generalized zero-sample learning fault diagnosis method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
MINGZHU TANG 等: "Review and Perspectives of Machine Learning Methods for Wind Turbine Fault Diagnosis", 《FRONTIERS IN ENERGY RESEARCH》, vol. 9, 15 November 2021 (2021-11-15), pages 1 - 15 *
徐健: "基于卷积神经网络的风电机组轴承故障诊断方法研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》, no. 12, 15 December 2020 (2020-12-15), pages 042 - 43 *

Also Published As

Publication number Publication date
CN114626415B (en) 2024-08-09

Similar Documents

Publication Publication Date Title
CN110823574B (en) Fault diagnosis method based on semi-supervised learning deep countermeasure network
CN114048568B (en) Rotary machine fault diagnosis method based on multisource migration fusion shrinkage framework
WO2019214268A1 (en) Photovoltaic array fault diagnosis method based on composite information
CN112257530B (en) Rolling bearing fault diagnosis method based on blind signal separation and support vector machine
CN113609569B (en) Distinguishing type generalized zero sample learning fault diagnosis method
CN114818774A (en) Intelligent gearbox fault diagnosis method based on multi-channel self-calibration convolutional neural network
CN108509701B (en) Rotary machine fault direct intelligent diagnosis method based on vibration signal
CN113008559B (en) Bearing fault diagnosis method and system based on sparse self-encoder and Softmax
CN111079861A (en) Power distribution network voltage abnormity diagnosis method based on image rapid processing technology
CN107451760A (en) Based on when the limited Boltzmann machine of window sliding Fault Diagnosis of Roller Bearings
CN102262736A (en) Method for classifying and identifying spatial target images
CN112017204A (en) Tool state image classification method based on edge marker graph neural network
CN111401126A (en) Combustion condition monitoring method and device based on flame image and deep learning
CN115358259A (en) Self-learning-based unsupervised cross-working-condition bearing fault diagnosis method
CN114169377A (en) G-MSCNN-based fault diagnosis method for rolling bearing in noisy environment
CN113432874A (en) Rolling bearing fault diagnosis method based on 2DPCA-CWT and CNN
CN115452376A (en) Bearing fault diagnosis method based on improved lightweight deep convolution neural network
CN114564987A (en) Rotary machine fault diagnosis method and system based on graph data
CN117592543A (en) Aeroengine rolling bearing fault diagnosis method based on self-supervision learning
CN114626415A (en) Wind turbine generator set composite fault diagnosis method based on artificial intelligence
CN114383846B (en) Bearing composite fault diagnosis method based on fault label information vector
CN116296396A (en) Rolling bearing fault diagnosis method based on mixed attention mechanism residual error network
Ye et al. Monitoring and diagnosis of multi-channel profile data based on uncorrelated multilinear discriminant analysis
CN114185321B (en) Electric actuator fault diagnosis method for improving multi-classification twin support vector machine
Xu et al. A graph convolutional shrinkage network-based fault diagnosis method for industrial process

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant