CN109086793A - A kind of abnormality recognition method of wind-driven generator - Google Patents
A kind of abnormality recognition method of wind-driven generator Download PDFInfo
- Publication number
- CN109086793A CN109086793A CN201810675216.6A CN201810675216A CN109086793A CN 109086793 A CN109086793 A CN 109086793A CN 201810675216 A CN201810675216 A CN 201810675216A CN 109086793 A CN109086793 A CN 109086793A
- Authority
- CN
- China
- Prior art keywords
- blower
- attribute
- wind
- data set
- data
- 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
Links
Classifications
-
- 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
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
Abstract
A kind of abnormality recognition method of wind-driven generator of the invention, attribute abnormal in SCADA data is picked out, other attributes of reservation are clustered through k-means；Attribute is reduced through t-SNE into fixed dimension respectively；Valid data after dimensionality reduction are subjected to Pearson correlation coefficients and calculate sequence, are converted to input of the picture as CNN, realize that the normal anomaly of blower judges by classifying to picture two.For the abnormality recognition method when establishing blower anomalous identification model, input and output are related to all properties, and model can effectively identify multiclass attribute while a variety of exceptions occur, and model has versatility, are suitable for any wind field；Using the data preprocessing method of dimensionality reduction in class after first clustering, unnecessary interference is eliminated, accuracy is improved；Carrying out attribute abnormal identification using convolutional neural networks can inhibit noise jamming, accurate to find that blower attribute picture minor change, the ability and robustness of stronger distinguishing characteristics make it have better accuracy.
Description
Technical field
The invention belongs to fault diagnosis technology fields, and in particular to a kind of abnormality recognition method of wind-driven generator.
Background technique
With the Rapid Expansion of global wind power generation power sustainable growth and wind power plant, the continuous application of blower, unit machine
Cabin product increases with it, and hub positions are also higher, and the challenge of fan safe operation is increasing.Wind turbines are chronically exposed to greatly
In the extreme weathers such as wind, thunderstorm, hail, especially offshore wind turbine will be also very easy to occur mechanical by long-term sea water intrusion
The exceptions such as component or electrical system or failure.Blower different parts failure is all possible to that blower is caused to have a power failure, the study found that in wind
Machine maintenance area, even if some small failures can all bring expensive maintenance cost, the maintenance of blower is faced with numerous
Challenge.According to the study analysis shows, if to Wind turbines use anticipatory maintenance, by substantially reduce Wind turbines it is subsequent reparation and
Schedule maintenance cost, this has tremendous influence to the economic input aspect of extensive development Wind Power Generation Industry, therefore to wind turbine
The abnormality of group is analyzed, and is carried out fault pre-alarming and is of great significance to.In order to reduce blower failure and by
Failure bring economic loss, the condition monitoring and anomalous identification of wind-driven generator are in terms of reducing downtime and maintenance cost
It becomes more and more important.
Power of fan curve can provide the relationship between output power and wind speed at present, be blower anomalous identification and performance point
One of most common tool is analysed, researcher is estimated and monitored using different methods on the basis of wind power plant SCADA data
Power of fan curve.When blower be in work normally (i.e. active reactive control) state under, be blower based on SCADA parameter model
Another effective ways of anomalous identification.Compared with power curve monitoring technology, it can make full use of SCADA system to hide
It is pre- to have developed a variety of conditional parameters by using SCADA parameter method for digging for the exception of operation information identification fan assembly
Model is surveyed to detect the significant changes of blower behavior before failure occurs.But forefathers study when there are the problem of it is mainly as follows:
(1) when establishing blower anomalous identification model, it will usually artificially select several determinant attributes as the defeated of model
Enter, easily choose the attribute of mistake in this way, causes modeling to fail, or make model not general.(2) it when training pattern, only wraps
The sample that this state is controlled containing active reactive, because to wait for that the states such as wind, unit is standby are all included in non-for unit caused by natural cause
Normal operating conditions, and it is above-mentioned be only weather the non-blower of interference exception.(3) when identification is abnormal, it can only identify that single class is abnormal,
Multiclass exception cannot be identified simultaneously.In addition, residual error method is usually utilized to determine the state of blower, but the setting majority of threshold value is based on
Artificial experience often interferes the accuracy of anomalous identification.
Summary of the invention
In view of the deficiencies of the prior art, present invention implementation provides a kind of abnormality recognition method of wind-driven generator.
A kind of abnormality recognition method of wind-driven generator of the invention, comprising the following steps:
Step 1: including the blower data of a variety of blower attributes from wind field SCADA system acquisition multiple groups, constitute initial blower
Data set；
Step 2: the abnormal data in initial blower data set is deleted, to obtain complete blower data set, the exception number
According to the blower data for including: a large amount of missing blower attributes of certain group；Certain in blower data is daily, monthly or every year all invariable
Blower attribute；Processing rear fan attribute number is n；
Step 3: noise reduction process is carried out to complete blower data set using wavelet thresholding method；
Step 4: using the clustering method of k-means to the blower attributive classification in the entire blower data set after noise reduction,
Make attribute in class similar after cluster, between class attribute mutually from；
Step 5: dimension-reduction treatment is carried out to the blower data set after clustering processing using t-SNE algorithm；
Step 6: the calculating that the data after dimensionality reduction carry out Pearson correlation coefficients being analyzed, data according to related coefficient
Size is arranged, and blower data set organization is the image with identical size；
Step 7: the image form after Pearson correlation coefficients analysis being sent to convolutional neural networks, and carries out blower
Anomalous identification；
Step 8: after determining that blower is abnormal, pass criteria determines classification belonging to the blower attribute being abnormal；
Step 9: by the identification to blower attribute abnormal, verifying the validity of the wind-driven generator abnormality recognition method.
In the abnormality recognition method of wind-driven generator of the invention, small echo described in step 3 is db5, and Decomposition order is
5。
In the abnormality recognition method of wind-driven generator of the invention, the step 4 is specifically included:
Step 4-1: the element in blower data set is subjected to normalization according to the following formula:
Wherein, ViIndicate that arbitrary element, max (A) represent the element maximum value of the blower attribute, min (A) represents the blower
The element minimum value of attribute, V are the values after the arbitrary element normalizing of the blower attribute；
Step 4-2: maximum cluster numbers are according to formulaIt determines, wherein kmaxFor maximum cluster numbers, n is blower attribute
Number；
Step 4-3: calculating by step 4-2, can arrive blower Attribute transposition for 2Class determines cluster numbers by following formula
Mesh:
Wherein, SC indicates silhouette coefficient, aiIndicate an element XiBeing averaged between element every other in same classification
Distance, aiFor quantifying inner integrated degree；biFor quantifying the separating degree between cluster, in above-mentioned element XiExcept select one
Classify b, calculates element XiAnd the average distance b in classification b between all elementsi, every other classification is traversed, is found recently
Average distance bi, SC is between -1 to+1, and value is bigger, and expression Clustering Effect is better, chooses corresponding poly- when silhouette coefficient maximum
Class number.
In the abnormality recognition method of wind-driven generator of the invention, the step 5 is specifically included:
Step 5-1: the blower data set after cluster is divided into multiple blower sample matrix day by data acquisition, to blower
The data of sample matrix are normalized, and are transformed to the master sample matrix X={ x that mean value is 0, variance is 11,x2,…
xn}；
Step 5-2: master sample matrix X={ x is defined1,x2,…xnIn condition similitude expression formula between two attributes such as
Under:
Wherein, σiFor constant, represent with xiCentered on the variance of Gaussian Profile put, it is different because of data point difference；
xi, xj, xkIt indicates 3 attributes in master sample matrix, calculates the similarity of two attributes according to the following formula:
Step 5-3: puzzlement degree Perp, the number of iterations T, learning rate η, momentum α (t) are rule of thumb set, are randomly provided
Blower data set after dimensionality reduction is Y={ y1,y2,…,yn, according to the following formula in the blower data set after calculating dimensionality reduction between two attributes
Similarity:
The mode of binary search is used in the case where given Perp and finds suitable σ according to the following formula, and gradient formula is such as
Under:
Wherein, C represents loss function, constantly adjustment Perp, the number of iterations T, until find out above formula most level off to 0 when it is corresponding
Perp and T value；
Blower data set after dimensionality reduction is calculated according to above formula；
By blower Attribute transposition it is K class after step 5-4:K-means cluster, reduces every class at 3 after t-SNE algorithm dimensionality reduction
It ties up, the blower data set after dimensionality reduction is that Y=K × 3 is tieed up, and can be used as effective row input of convolutional neural networks.
In the abnormality recognition method of wind-driven generator of the invention, the step 6 is specifically included:
Step 6-1: Pearson correlation coefficients calculating is carried out to the K group data after sequence；
Wherein cov (M, N) is represented takes out different classification M and classification N progress from K classification in blower data set Y
Covariance coefficient calculates, σMRepresent the standard deviation of classification M, σNRepresent the standard deviation of classification N.
Step 6-2: the Pearson correlation coefficients completed will be calculated and carry out comparative sorting two-by-two according to size, being combined into has
The image of continuity.
In the abnormality recognition method of wind-driven generator of the invention, the step 7 is specifically included:
Step 7-1: normalization layer is introduced, the input of convolutional neural networks is standardized as identical size；
Step 7-2: introducing convolutional layer, generates several characteristic patterns by non-linear two operations of convolution sum and mentions to carry out feature
It takes；
Step 7-3: introducing pond layer, and the size for reducing characteristic pattern is operated by pondization；
Step 7-4: it after the characteristic present for obtaining higher level, is converted into 1-D vector and is fed to classification layer；
Step 7-5: classification layer using S type power function as activation primitive, to blower have it is without exception judge, wherein
Output 0 represents blower exception, and 1 to represent blower normal.
In the abnormality recognition method of wind-driven generator of the invention, the step 8 is specifically included:
Step 8-1: calculating the mean value of 3 attributes in the blower data set of the health status after dimensionality reduction as unit of day,
It is denoted as a, b, c respectively；
Step 8-2: classical formulas inference is improved:
And then it releases:
Step 8-3: m is set1=a+b+c；It is calculated daily by trainingWithMost value, look for m1Minimum value be denoted asLook for m2Maximum value be denoted asAnd then obtain the normal shape of blowing machine
The range of m is under stateIf m exceeds this range, such attribute abnormal of blower is determined.
A kind of abnormality recognition method of wind-driven generator of the invention compared with prior art, at least has below beneficial to effect
Fruit:
1, when establishing blower anomalous identification model, input and output are related to all properties, and model can effectively identify multiclass category
Property occur a variety of exceptions simultaneously, the model of foundation has versatility, is applicable to any wind field；
2, when training pattern, unit caused by weather conditions is waited for into the weather such as wind, unit is standby interference rather than blower is abnormal
Situations such as be included in normal condition, these can be to avoid erroneous judgement；
3, after first clustering in class dimensionality reduction data preprocessing method, the use most succinct feature representation state of blower goes
In addition to unnecessary interference, facilitate the raising of accuracy；
4, the input sample of convolutional neural networks model is provided by complete blower image segments, wherein comprising comprehensive different
Normal information；Feature is to noise-sensitive simultaneously, and especially for abnormal unconspicuous situation, convolutional neural networks model can inhibit
The unfavorable interference of noise, the accurate variation for finding that blower attribute picture is small, the ability and robustness of stronger distinguishing characteristics
Make it have better accuracy.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the abnormality recognition method of wind-driven generator of the invention；
Fig. 2 a is the abnormal data schematic diagram that the sample point data value of embodiment of the present invention largely lacks,
Fig. 2 b is the invariable schematic diagram of the collected blower attribute value of embodiment of the present invention；
Fig. 3 is the structure chart of the convolutional neural networks of embodiment of the present invention；
Fig. 4 is the output characteristic pattern of every layer of the convolutional neural networks of embodiment of the present invention；
Fig. 5 is that five groups of output figures of individual event exception occur for the first generic attribute of embodiment of the present invention；
Fig. 6 is that multinomial abnormal accuracy rate statistical chart occurs for the 6th generic attribute of embodiment of the present invention；
Fig. 7 is that multinomial five groups of abnormal output figures occur for the 6th generic attribute of embodiment of the present invention；
Fig. 8 is that multinomial abnormal accuracy rate statistical chart occurs for the 6th generic attribute of embodiment of the present invention；
Fig. 9 is that five groups of output figures of multiclass exception occur for the multiclass attribute of embodiment of the present invention；
Figure 10 is five groups of output figures of BPNN of embodiment of the present invention.
Specific embodiment
It elaborates with reference to the accompanying drawing to one embodiment of the present invention.
Attribute abnormal in collected wind field SCADA data is picked out, other blower attributes of reservation are poly- through k-means
Class, silhouette coefficient is used as judging the mark of Clustering Effect, to determine clusters number and each generic attribute；It then, will be of all categories
Attribute is reduced through t-SNE into fixed dimension respectively, and new attribute is the set of attribute after dimensionality reduction of all categories, is convolutional neural networks mould
Type provides the valid data of anomalous identification, and convolutional neural networks is made to play maximum effect；Finally, the valid data after dimensionality reduction are drawn
It is divided into the input that square matrix is converted to picture as convolutional neural networks, realizes that the normal anomaly of blower is sentenced by classifying to picture two
Disconnected, the validity of proposed method will be proved by three tests.
As shown in Figure 1, a kind of abnormality recognition method of wind-driven generator, includes the following steps:
Step 1: including the blower data of a variety of blower attributes from wind field SCADA system acquisition multiple groups, constitute initial blower
Data set；When it is implemented, the one group of wind data of acquisition in every 30 seconds；
Step 2: the abnormal data in initial blower data set is deleted, to obtain complete blower data set, the exception number
According to the blower data for including: a large amount of missing blower attributes of certain group；Certain in blower data is daily, monthly or every year all invariable
Blower attribute；Processing rear fan attribute number is n；
Fig. 2 a is the abnormal data schematic diagram that sample point data value largely lacks, i.e. the wind of a large amount of missing blower attributes of certain group
Machine data, wherein missing point is the 0-505 sampled point；Fig. 2 b is that the collected blower attribute value of embodiment of the present invention is permanent
Fixed constant schematic diagram, i.e. daily, the monthly or every year all invariable blower attribute of certain in blower data are invariable
Schematic diagram.
Step 3: noise reduction process being carried out to complete blower data set using wavelet thresholding method, to inhibit useless in signal
Part, enhances part useful in signal, and the small echo is db5, Decomposition order 5；
Step 4: blower attribute is divided by K class using the clustering method of k-means, makes attribute in class similar after cluster,
Between class attribute mutually from；When it is implemented, blower attribute is divided into 7 classes in present embodiment.The step 4 specifically includes:
Step 4-1: it since the dimension of blower attribute is different, is not easy to cluster, so needing the member in blower data set
Element carries out normalization according to the following formula:
Wherein, ViIndicate that arbitrary element, max (A) represent the element maximum value of the blower attribute, min (A) represents the blower
The element minimum value of attribute, V are the values after the arbitrary element normalizing of the blower attribute；
Step 4-2: maximum cluster numbers are according to formulaIt determines, wherein kmaxFor maximum cluster numbers, n is blower attribute
Number；
Step 4-3: calculating by step 4-2, can arrive blower Attribute transposition for 2Class determines cluster numbers by following formula
Mesh:
Wherein, SC indicates silhouette coefficient, aiIndicate an element XiBeing averaged between element every other in same classification
Distance, aiFor quantifying inner integrated degree；biFor quantifying the separating degree between cluster, in above-mentioned element XiExcept select one
Classify b, calculates element XiAnd the average distance b in classification b between all elementsi, every other classification is traversed, is found recently
Average distance bi, SC is between -1 to+1, and value is bigger, and expression Clustering Effect is better, chooses corresponding poly- when silhouette coefficient maximum
Class number K.
Step 5: dimension-reduction treatment being carried out to the blower data set after clustering processing using t-SNE algorithm, by every generic attribute point
The higher attribute of similar degree in the class Jiang not be replaced more to represent power and persuasion property with fixed low-dimensional attribute at 3 dimensions.The step
5 specifically include:
Step 5-1: the blower data set after cluster is divided into multiple blower sample matrix day by data acquisition, to blower
The data of sample matrix are normalized, and are transformed to the master sample matrix X={ x that mean value is 0, variance is 11,x2,…
xn}；
Step 5-2: master sample matrix X={ x is defined1,x2,…xnIn condition similitude expression formula between two attributes such as
Under:
Wherein, σiFor constant, represent with xiCentered on the variance of Gaussian Profile put, it is different because of data point difference；
xi, xj, xkIt indicates 3 attributes in master sample matrix, calculates the similarity of two attributes according to the following formula:
Step 5-3: puzzlement degree Perp, the number of iterations T, learning rate η, momentum α (t) are rule of thumb set, are randomly provided
Blower data set after dimensionality reduction is Y={ y1,y2,…,yn, according to the following formula in the blower data set after calculating dimensionality reduction between two attributes
Similarity:
The mode of binary search is used in the case where given Perp and finds suitable σ according to the following formula, and gradient formula is such as
Under:
Wherein, C represents loss function, constantly adjustment Perp, the number of iterations T, until find out above formula most level off to 0 when it is corresponding
Perp and T value；
Blower data set after dimensionality reduction is calculated according to above formula；
By blower Attribute transposition it is 7 classes after step 5-4:K-means cluster, reduces every class at 3 after t-SNE algorithm dimensionality reduction
It ties up, the blower data set after dimensionality reduction is that Y=K × 3 is tieed up, and can be used as effective row input of convolutional neural networks.
Step 6: Pearson correlation coefficients analysis being carried out to 7 class blower attributes after dimensionality reduction, by this 7 generic attribute according to Pierre
Gloomy related coefficient size is arranged, and the coherent image with correlation is formed.The step 6 specifically includes:
Step 6-1: Pearson correlation coefficients calculating is carried out to the K group data after sequence, Pearson correlation coefficients formula is such as
Under:
Wherein cov (M, N) is represented takes out different classification M and classification N progress from K classification in blower data set Y
Covariance coefficient calculates, σMRepresent the standard deviation of classification M, σNRepresent the standard deviation of classification N.
Step 6-2: the Pearson correlation coefficients completed will be calculated and carry out comparative sorting two-by-two according to size, being combined into has
The image of continuity.When it is implemented, blower data set is brought into Pearson correlation coefficients formula according to classification, 7 classes are obtained
Correlation coefficient charts between data are ranked up according to the size of incidence coefficient, this 7 class data is ranked up, image is spliced into.
Step 7: the image form after Pearson correlation coefficients analysis being sent to convolutional neural networks, and carries out blower
Anomalous identification, the step 7 specifically includes:
Step 7-1: normalization layer is introduced, the input of convolutional neural networks is standardized as identical size.
When it is implemented, finding the image maximum value for being input to normalization layer and minimum value and its corresponding position；Then make
Required size is normalized to Downsapling method；Finally replace maximum value and minimum value；
Step 7-2: introducing convolutional layer, generates several characteristic patterns by non-linear two operations of convolution sum and mentions to carry out feature
It takes.
When it is implemented, each characteristic pattern is the special characteristic expression for inputting blower attribute image in some regions, convolution
Operation can pass through yj=∑ikij*xiIt indicates.Wherein, * represents convolution operation；yjIt is j-th of characteristic pattern of output；kijIt is that can instruct
Experienced convolution kernel (also referred to as filter)；xiIt is i-th of input；
Step 7-3: introducing pond layer, and the size for reducing characteristic pattern is operated by pondization.
When it is implemented, input is divided into several non-overlap rectangular areas with same size, maximum pondization behaviour first
Obtain the maximum value in rectangular area；Average pondization operation obtains the average value in rectangular area.
Step 7-4: it after the characteristic present for obtaining higher level, is converted into 1-D vector and is fed to classification layer.
Step 7-5: classification layer using S type power function as activation primitive, to blower have it is without exception judge, wherein
Output 0 represents blower exception, and 1 to represent blower normal.
Step 8: after determining that blower is abnormal, pass criteria determines classification belonging to the blower attribute being abnormal；Institute
Step 8 is stated to specifically include:
Step 8-1: calculating the mean value of 3 attributes in the blower data set of the health status after dimensionality reduction as unit of day,
It is denoted as a, b, c respectively；
Step 8-2: classical formulas inference is improved:
And then it releases:
Step 8-3: m is set1=a+b+c；It is calculated daily by trainingWithMost value, look for m1Minimum value be denoted asLook for m2Maximum value be denoted asAnd then obtain the normal shape of blowing machine
The range of m is under stateIf m exceeds this range, such attribute abnormal of blower is determined.
Step 9: by the identification to blower attribute abnormal, verifying the validity of the wind-driven generator abnormality recognition method.
Fig. 3 is the structure chart of the convolutional neural networks of embodiment of the present invention, including convolutional layer, pond layer, convolutional layer, pond
The classification layer changing layer and being fully connected；
The convolutional neural networks model structure setting principle of foundation, X=21 in this experiment, each input picture are standardizing
Layer is normalized to 21 × 21 size, and input picture each in this way represents the attributive character in blower 10min, other settings such as table
Shown in 1.By testing to obtain relatively optimal model repeatedly, specifically to learn details as shown in table 2.
The model structure for the convolutional neural networks that table 1 proposes
Table 2 learns details
Wherein, the quantity 3 of second layer convolution kernels is to optimize performance by many experiments to carry out selection.By taking Fig. 3 as an example,
Test sample as shown in the figure is finally correctly classified, and normalized blower image is shown in Far Left, later from left to right according to
Secondary is C1, S1, C2, S2.Fig. 4 is every layer of output characteristic pattern of embodiment of the present invention.
Below by the identification of different classes of blower attribute being abnormal, to verify wind-driven generator of the invention
The validity of abnormality recognition method.
(1) it by the first generic attribute (k=1) occurring the identification of individual event exception, verifies the abnormal of the wind-driven generator and knows
The validity of other method；
Training sample is the blower attribute picture that 20,000 sizes are 21 × 21, wherein including a variety of exceptions, test sample is
The blower attribute picture that 100 sizes are 21 × 21, wherein only including a kind of this exception of gear-box rear bearing temperature overheating.
Fan condition is judged based on CNN model, to avoid coincidence, randomly selects test sample, the test sample of selection
Intersect under two states.The five groups of tests carried out, wherein normal sample is 48, exceptional sample 52.
The actual value marked with grey i.e. our given labels, five groups of colored predicted values are the knots of every group of prediction of model
Fruit, if output, which is more than or equal to 0.5, is defaulted as 1, output is defaulted as 0 less than 0.5, judges fan condition with this criterion.Judge blower
The accuracy of state recognition, as a result as shown in table 3 and Fig. 5.
The statistics of 3 five groups of accuracys rate of table
Table 3 has recorded TA, FA, TH, FH value of each group of test, and Fig. 6 is accuracy rate Q1, Q2, Q of every group of test.Five groups
Test reaches 90.8% to the average value of the blower correct recognition rata of health status, and every group is attained by 85% or more
Accuracy rate, though the blower recognition accuracy to abnormality is largely slightly below normal condition, and every group is attained by 83%
Or more accuracy rate, the average value of comprehensive accuracy rate is 90%.When K=1, δ1=-0.018；δ2=2.52 × 10-3, first
The m value of generic attribute is fallen in outside range, remaining is normal, and blower first kind attribute abnormal is verified by test of many times.
(2) it by the way that multinomial abnormal identification occurs to the 6th generic attribute (k=6), verifies the abnormal of the wind-driven generator and knows
The validity of other method；
Identical as the first kind attribute abnormal verifying selection principle of test sample, Fig. 7 show five groups of tests of progress, and
And normal anomaly samples pictures quantity is 50.Wherein first and second, four group of intensity of anomaly is more apparent.Table 4 be each group of test TA,
The number statistical of FA, TH, FH, Fig. 8 are the Statistical Comparison figure of accuracy rate Q1, Q2, Q.
4 five groups of accuracy rate statistics of table
Five groups of tests reach 95.2% to the average value of the blower correct recognition rata of health status, and every group is attained by
84% or more accuracy rate is still up to 95.2% to the average value of the blower correct recognition rata of abnormality, but still reaches
Average value is 87.3%, and every group be attained by 84% or more accuracy rate, the average value of comprehensive accuracy rate is 95.4%,
Higher than the 90% of step 7.Compared to step 7, step 8 shows in the apparent situation of intensity of anomaly, the standard of the model anomalous identification
True rate will be promoted.When K=6, δ1=-0.028；δ2=1.12 × 10-3, the m value of the 6th generic attribute falls in outside range, remaining is just
Often, the 6th generic sexual abnormality of blower, is verified by test of many times.
(3) by the way that multinomial abnormal identification occurs to multiclass attribute, having for the wind-driven generator abnormality recognition method is verified
Effect property.
For avoid sample only have a certain generic sexual abnormality unicity be not enough to specification exception identification accuracy, step 9
Verifying identifies while multiclass exception occurs.Five groups of tests of Fig. 9, the Exception Type of every group of test are different from, and all simultaneously
Exception comprising plurality of classes.Table 5 counts the accuracy rate of each group and whole fan condition identification.
5 groups of accuracy rate statistics of table
As shown in Table 5, five groups of tests reach 95.6% to the average value of the blower correct recognition rata of health status, and every
Group is attained by 90% or more accuracy rate, is up to 96% to the average value of the blower correct recognition rata of abnormality, and
Every group be attained by 84% or more accuracy rate, the average value of comprehensive accuracy rate is 95.8%, the results showed that, this method is to more
Higher accuracy rate is still held in the identification of class exception, and because abnormal more apparent, accuracy rate is high for rear three groups of tests.The m of multiclass attribute
Value is abnormal.
Table 6 be it is proposed that method compared with BPNN anomalous identification effect, two methods using identical data into
Row is tested, wherein process of data preprocessing of the BPNN without dimensionality reduction in class after first clustering, and the anomalous identification result based on BPNN is such as
Shown in Figure 10 and table 7.It is proposed that method it is as shown in table 7 compared with the effect of BPNN anomalous identification.
6 groups of accuracy rate statistics of table
7 each method anomalous identification effect of table statistics
In conjunction with three cases, one demonstrate it is proposed that method to the accuracy of blower anomalous identification；Secondly showing
Intensity of anomaly is more obvious, anomalous identification accuracy rate higher feature.
The foregoing is merely presently preferred embodiments of the present invention, the thought being not intended to limit the invention, all of the invention
Within spirit and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (7)
1. a kind of abnormality recognition method of wind-driven generator, which comprises the following steps:
Step 1: including the blower data of a variety of blower attributes from wind field SCADA system acquisition multiple groups, constitute initial blower data
Collection；
Step 2: the abnormal data in initial blower data set is deleted, to obtain complete blower data set, the abnormal data packet
It includes: the blower data of a large amount of missing blower attributes of certain group；Daily, the monthly or every year all invariable wind of certain in blower data
Machine attribute, processing rear fan attribute number are n；
Step 3: noise reduction process is carried out to complete blower data set using wavelet thresholding method；
Step 4: using the clustering method of k-means to the blower attributive classification in the entire blower data set after noise reduction, cluster
After make attribute in class similar, between class attribute mutually from；
Step 5: dimension-reduction treatment is carried out to the blower data set after clustering processing using t-SNE algorithm；
Step 6: the calculating that the data after dimensionality reduction carry out Pearson correlation coefficients being analyzed, data according to the size of related coefficient
It is arranged, blower data set organization is the image with identical size；
Step 7: the image form after Pearson correlation coefficients analysis being sent to convolutional neural networks, and carries out the different of blower
Common sense is other；
Step 8: after determining that blower is abnormal, pass criteria determines classification belonging to the blower attribute being abnormal；
Step 9: by the identification to blower attribute abnormal, verifying the validity of the wind-driven generator abnormality recognition method.
2. the abnormality recognition method of wind-driven generator as described in claim 1, which is characterized in that small echo described in step 3 is
Db5, Decomposition order 5.
3. the abnormality recognition method of wind-driven generator as described in claim 1, which is characterized in that the step 4 specifically includes:
Step 4-1: the element in blower data set is subjected to normalization according to the following formula:
Wherein, ViIndicate that arbitrary element, max (A) represent the element maximum value of the blower attribute, min (A) represents the blower attribute
Element minimum value, V is the value after the arbitrary element normalizing of the blower attribute；
Step 4-2: maximum cluster numbers are according to formulaIt determines, wherein kmaxFor maximum cluster numbers, n is of blower attribute
Number；
Step 4-3: calculating by step 4-2, can arrive blower Attribute transposition for 2Class determines clusters number by following formula:
Wherein, SC indicates silhouette coefficient, aiIndicate an element XiWith the average departure between element every other in same classification
From aiFor quantifying inner integrated degree；biFor quantifying the separating degree between cluster, in above-mentioned element XiExcept select one point
Class b calculates element XiAnd the average distance b in classification b between all elementsi, every other classification is traversed, is found nearest
Average distance bi, SC is between -1 to+1, and value is bigger, and expression Clustering Effect is better, chooses corresponding cluster when silhouette coefficient maximum
Number K.
4. the abnormality recognition method of wind-driven generator as described in claim 1, which is characterized in that the step 5 specifically includes:
Step 5-1: the blower data set after cluster is divided into multiple blower sample matrix day by data acquisition, to blower sample
The data of matrix are normalized, and are transformed to the master sample matrix X={ x that mean value is 0, variance is 11,x2,…xn}；
Step 5-2: master sample matrix X={ x is defined1,x2,…xnIn condition similitude expression formula between two attributes it is as follows:
Wherein, σiFor constant, represent with xiCentered on the variance of Gaussian Profile put, it is different because of data point difference；xi, xj,
xkIt indicates 3 attributes in master sample matrix, calculates the similarity of two attributes according to the following formula:
Step 5-3: puzzlement degree Perp, the number of iterations T, learning rate η, momentum α (t) are rule of thumb set, dimensionality reduction is randomly provided
Blower data set afterwards is Y={ y1,y2,…,yn, the phase in the blower data set after calculating dimensionality reduction according to the following formula between two attributes
Like degree:
The mode of binary search is used in the case where given Perp and finds suitable σ according to the following formula, and gradient formula is as follows:
Wherein, C represents loss function, constantly adjustment Perp, the number of iterations T, until find out above formula most level off to 0 when it is corresponding
The value of Perp and T；
Blower data set after dimensionality reduction is calculated according to above formula；
By blower Attribute transposition it is K class after step 5-4:K-means cluster, reduces every class at 3 dimensions after t-SNE algorithm dimensionality reduction,
Blower data set after dimensionality reduction is that Y=K × 3 is tieed up, and can be used as effective row input of convolutional neural networks.
5. the abnormality recognition method of wind-driven generator as claimed in claim 4, which is characterized in that the step 6 specifically includes:
Step 6-1: Pearson correlation coefficients calculating is carried out to the K group data after sequence；
Wherein cov (M, N) representative takes out different classification M and classification N from K classification in blower data set Y and carries out association side
Poor coefficient calculates, σMRepresent the standard deviation of classification M, σNRepresent the standard deviation of classification N.
Step 6-2: will calculate the Pearson correlation coefficients completed and carry out comparative sorting two-by-two according to size, be combined into have and link up
The image of property.
6. the abnormality recognition method of wind-driven generator as described in claim 1, which is characterized in that the step 7 specifically includes:
Step 7-1: normalization layer is introduced, the input of convolutional neural networks is standardized as identical size；
Step 7-2: introducing convolutional layer, generates several characteristic patterns by non-linear two operations of convolution sum to carry out feature extraction；
Step 7-3: introducing pond layer, and the size for reducing characteristic pattern is operated by pondization；
Step 7-4: it after the characteristic present for obtaining higher level, is converted into 1-D vector and is fed to classification layer；
Step 7-5: classification layer using S type power function as activation primitive, to blower have it is without exception judge, wherein exporting
0 represents blower exception, and 1 to represent blower normal.
7. the abnormality recognition method of wind-driven generator as described in claim 1, which is characterized in that the step 8 specifically includes:
Step 8-1: the mean value of 3 attributes in the blower data set of the health status after dimensionality reduction is calculated as unit of day, respectively
It is denoted as a, b, c；
Step 8-2: classical formulas inference is improved:
And then it releases:
Step 8-3: m is set1=a+b+c；It is logical
Training is crossed to calculate dailyAnd m2maxMost value, look for m1Minimum value be denoted asLook for m2Maximum value be denoted asAnd then the range for obtaining m under blowing machine normal condition isIf m exceeds this model
It encloses, then determines such attribute abnormal of blower.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810675216.6A CN109086793B (en) | 2018-06-27 | 2018-06-27 | Abnormity identification method for wind driven generator |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810675216.6A CN109086793B (en) | 2018-06-27 | 2018-06-27 | Abnormity identification method for wind driven generator |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109086793A true CN109086793A (en) | 2018-12-25 |
| CN109086793B CN109086793B (en) | 2021-11-16 |
Family
ID=64839830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810675216.6A Active CN109086793B (en) | 2018-06-27 | 2018-06-27 | Abnormity identification method for wind driven generator |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109086793B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409740A (en) * | 2018-10-25 | 2019-03-01 | 江苏方天电力技术有限公司 | A method of the wind-power electricity generation quality of data is checked based on Pearson correlation coefficients |
| CN109978039A (en) * | 2019-03-19 | 2019-07-05 | 南京邮电大学 | A kind of lower fan blade icing prediction technique based on unbalanced dataset situation |
| CN110161408A (en) * | 2019-05-31 | 2019-08-23 | 江铃汽车股份有限公司 | The recognition methods of gathering chain motor status and system |
| CN110262460A (en) * | 2019-07-01 | 2019-09-20 | 山东浪潮人工智能研究院有限公司 | A kind of combination Clustering carries out the concrete piston failure prediction method of feature extraction |
| CN110287233A (en) * | 2019-06-18 | 2019-09-27 | 华北电力大学 | A kind of system exception method for early warning based on deep learning neural network |
| CN110610212A (en) * | 2019-09-20 | 2019-12-24 | 云南电网有限责任公司电力科学研究院 | Fault classification method and fault classification device for transformer of power distribution network |
| CN110674864A (en) * | 2019-09-20 | 2020-01-10 | 国网上海市电力公司 | Wind power abnormal data identification method with synchronous phasor measurement device |
| CN110751213A (en) * | 2019-10-21 | 2020-02-04 | 东北电力大学 | Method for identifying and supplementing abnormal wind speed data of wind measuring tower |
| CN111796576A (en) * | 2020-06-16 | 2020-10-20 | 北京工业大学 | Process monitoring visualization method based on dual-core t-distribution random neighbor embedding |
| CN112000831A (en) * | 2020-08-13 | 2020-11-27 | 贵州电网有限责任公司 | Abnormal data identification optimization method based on transformer substation graph transformation |
| CN112697268A (en) * | 2020-12-04 | 2021-04-23 | 佳木斯电机股份有限公司 | Motor anomaly detection integrated algorithm based on T-SNE |
| CN113723502A (en) * | 2021-08-27 | 2021-11-30 | 西安热工研究院有限公司 | Wind generating set frequency converter network side abnormity identification method and system based on current waveform |
| CN115035988A (en) * | 2022-08-15 | 2022-09-09 | 武汉明炀大数据科技有限公司 | Medical image processing method, system, equipment and medium based on cloud computing |
| CN116089843A (en) * | 2023-04-10 | 2023-05-09 | 国网浙江省电力有限公司 | New energy settlement data anomaly detection and early warning method based on data clustering |
| CN112000831B (en) * | 2020-08-13 | 2024-04-19 | 贵州电网有限责任公司 | Abnormal data identification optimization method based on substation graph transformation |
Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104765964A (en) * | 2015-04-15 | 2015-07-08 | 北京空间飞行器总体设计部 | Space environment sensitive parameter screening method |
| CN105699080A (en) * | 2015-12-18 | 2016-06-22 | 华北电力大学（保定） | Wind turbine generator set bearing fault feature extraction method based on vibration data |
| CN105719002A (en) * | 2016-01-18 | 2016-06-29 | 重庆大学 | Wind turbine generator state parameter abnormity identification method based on combination prediction |
| CN105841961A (en) * | 2016-03-29 | 2016-08-10 | 中国石油大学(华东) | Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network |
| CN106014858A (en) * | 2016-07-21 | 2016-10-12 | 浙江运达风电股份有限公司 | Automatic calibration method and device for air aligning errors of wind generation set |
| CN106779200A (en) * | 2016-12-07 | 2017-05-31 | 东北大学 | Based on the Wind turbines trend prediction method for carrying out similarity in the historical data |
| CN107132478A (en) * | 2017-04-13 | 2017-09-05 | 南京师范大学 | A kind of wind-driven generator method for diagnosing faults measured based on polarization maintaining optical fibre with neural network classification |
| CN107144569A (en) * | 2017-04-27 | 2017-09-08 | 西安交通大学 | The fan blade surface defect diagnostic method split based on selective search |
| CN107153842A (en) * | 2017-04-27 | 2017-09-12 | 西安交通大学 | The fan blade diaphragm damage detecting method split based on edge |
| CN107301433A (en) * | 2017-07-14 | 2017-10-27 | 南京华苏科技有限公司 | Net based on clustering and discriminant model about car discrimination method and system |
| CN107355342A (en) * | 2017-06-30 | 2017-11-17 | 北京金风科创风电设备有限公司 | The abnormal recognition methods of wind generating set pitch control and device |
| CN107403194A (en) * | 2017-07-26 | 2017-11-28 | 广州慧扬健康科技有限公司 | Cutaneum carcinoma image recognition visualization model based on t SNE |
| CN107560849A (en) * | 2017-08-04 | 2018-01-09 | 华北电力大学 | A kind of Wind turbines Method for Bearing Fault Diagnosis of multichannel depth convolutional neural networks |
| CN107689059A (en) * | 2017-06-30 | 2018-02-13 | 北京金风科创风电设备有限公司 | The abnormal recognition methods of wind generating set pitch control and device |
| CN107704953A (en) * | 2017-09-20 | 2018-02-16 | 河海大学 | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests |
-
2018
- 2018-06-27 CN CN201810675216.6A patent/CN109086793B/en active Active
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104765964A (en) * | 2015-04-15 | 2015-07-08 | 北京空间飞行器总体设计部 | Space environment sensitive parameter screening method |
| CN105699080A (en) * | 2015-12-18 | 2016-06-22 | 华北电力大学（保定） | Wind turbine generator set bearing fault feature extraction method based on vibration data |
| CN105719002A (en) * | 2016-01-18 | 2016-06-29 | 重庆大学 | Wind turbine generator state parameter abnormity identification method based on combination prediction |
| CN105841961A (en) * | 2016-03-29 | 2016-08-10 | 中国石油大学(华东) | Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network |
| CN106014858A (en) * | 2016-07-21 | 2016-10-12 | 浙江运达风电股份有限公司 | Automatic calibration method and device for air aligning errors of wind generation set |
| CN106779200A (en) * | 2016-12-07 | 2017-05-31 | 东北大学 | Based on the Wind turbines trend prediction method for carrying out similarity in the historical data |
| CN107132478A (en) * | 2017-04-13 | 2017-09-05 | 南京师范大学 | A kind of wind-driven generator method for diagnosing faults measured based on polarization maintaining optical fibre with neural network classification |
| CN107144569A (en) * | 2017-04-27 | 2017-09-08 | 西安交通大学 | The fan blade surface defect diagnostic method split based on selective search |
| CN107153842A (en) * | 2017-04-27 | 2017-09-12 | 西安交通大学 | The fan blade diaphragm damage detecting method split based on edge |
| CN107355342A (en) * | 2017-06-30 | 2017-11-17 | 北京金风科创风电设备有限公司 | The abnormal recognition methods of wind generating set pitch control and device |
| CN107689059A (en) * | 2017-06-30 | 2018-02-13 | 北京金风科创风电设备有限公司 | The abnormal recognition methods of wind generating set pitch control and device |
| CN107301433A (en) * | 2017-07-14 | 2017-10-27 | 南京华苏科技有限公司 | Net based on clustering and discriminant model about car discrimination method and system |
| CN107403194A (en) * | 2017-07-26 | 2017-11-28 | 广州慧扬健康科技有限公司 | Cutaneum carcinoma image recognition visualization model based on t SNE |
| CN107560849A (en) * | 2017-08-04 | 2018-01-09 | 华北电力大学 | A kind of Wind turbines Method for Bearing Fault Diagnosis of multichannel depth convolutional neural networks |
| CN107704953A (en) * | 2017-09-20 | 2018-02-16 | 河海大学 | The short-term wind-electricity power probability density Forecasting Methodology of EWT quantile estimate forests |
Non-Patent Citations (2)
| Title |
|---|
| 崔杨 等: "双馈感应风电机组异常脱网及其无功需求分析", 《电网技术》 * |
| 杨茂 等: "基于风特征分析的风电机组异常数据识别算法", 《中国电机工程学报》 * |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109409740B (en) * | 2018-10-25 | 2021-08-10 | 江苏方天电力技术有限公司 | Method for checking quality of wind power generation data based on Pearson correlation coefficient |
| CN109409740A (en) * | 2018-10-25 | 2019-03-01 | 江苏方天电力技术有限公司 | A method of the wind-power electricity generation quality of data is checked based on Pearson correlation coefficients |
| CN109978039B (en) * | 2019-03-19 | 2020-10-16 | 南京邮电大学 | Fan blade icing prediction method based on unbalanced data set |
| CN109978039A (en) * | 2019-03-19 | 2019-07-05 | 南京邮电大学 | A kind of lower fan blade icing prediction technique based on unbalanced dataset situation |
| CN110161408A (en) * | 2019-05-31 | 2019-08-23 | 江铃汽车股份有限公司 | The recognition methods of gathering chain motor status and system |
| CN110287233A (en) * | 2019-06-18 | 2019-09-27 | 华北电力大学 | A kind of system exception method for early warning based on deep learning neural network |
| CN110262460B (en) * | 2019-07-01 | 2020-07-03 | 浪潮集团有限公司 | Concrete piston fault prediction method for extracting features by combining clustering idea |
| CN110262460A (en) * | 2019-07-01 | 2019-09-20 | 山东浪潮人工智能研究院有限公司 | A kind of combination Clustering carries out the concrete piston failure prediction method of feature extraction |
| CN110674864A (en) * | 2019-09-20 | 2020-01-10 | 国网上海市电力公司 | Wind power abnormal data identification method with synchronous phasor measurement device |
| CN110610212A (en) * | 2019-09-20 | 2019-12-24 | 云南电网有限责任公司电力科学研究院 | Fault classification method and fault classification device for transformer of power distribution network |
| CN110674864B (en) * | 2019-09-20 | 2024-03-15 | 国网上海市电力公司 | Wind power abnormal data identification method comprising synchronous phasor measurement device |
| CN110751213A (en) * | 2019-10-21 | 2020-02-04 | 东北电力大学 | Method for identifying and supplementing abnormal wind speed data of wind measuring tower |
| CN111796576A (en) * | 2020-06-16 | 2020-10-20 | 北京工业大学 | Process monitoring visualization method based on dual-core t-distribution random neighbor embedding |
| CN111796576B (en) * | 2020-06-16 | 2023-03-31 | 北京工业大学 | Process monitoring visualization method based on dual-core t-distribution random neighbor embedding |
| WO2021253550A1 (en) * | 2020-06-16 | 2021-12-23 | 北京工业大学 | Process monitoring visualization method based on bi-kernel t-distributed stochastic neighbor embedding |
| CN112000831A (en) * | 2020-08-13 | 2020-11-27 | 贵州电网有限责任公司 | Abnormal data identification optimization method based on transformer substation graph transformation |
| CN112000831B (en) * | 2020-08-13 | 2024-04-19 | 贵州电网有限责任公司 | Abnormal data identification optimization method based on substation graph transformation |
| CN112697268A (en) * | 2020-12-04 | 2021-04-23 | 佳木斯电机股份有限公司 | Motor anomaly detection integrated algorithm based on T-SNE |
| CN112697268B (en) * | 2020-12-04 | 2022-06-24 | 佳木斯电机股份有限公司 | Motor anomaly detection integrated algorithm based on T-SNE |
| CN113723502A (en) * | 2021-08-27 | 2021-11-30 | 西安热工研究院有限公司 | Wind generating set frequency converter network side abnormity identification method and system based on current waveform |
| CN115035988B (en) * | 2022-08-15 | 2022-12-20 | 武汉明炀大数据科技有限公司 | Medical image processing method, system, equipment and medium based on cloud computing |
| CN115035988A (en) * | 2022-08-15 | 2022-09-09 | 武汉明炀大数据科技有限公司 | Medical image processing method, system, equipment and medium based on cloud computing |
| CN116089843A (en) * | 2023-04-10 | 2023-05-09 | 国网浙江省电力有限公司 | New energy settlement data anomaly detection and early warning method based on data clustering |
| CN116089843B (en) * | 2023-04-10 | 2023-09-08 | 国网浙江省电力有限公司 | New energy settlement data anomaly detection and early warning method based on data clustering |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109086793B (en) | 2021-11-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086793A (en) | A kind of abnormality recognition method of wind-driven generator | |
| CN110132598B (en) | Fault noise diagnosis algorithm for rolling bearing of rotating equipment | |
| Liu et al. | Time series classification with multivariate convolutional neural network | |
| CN111237134B (en) | Offshore double-fed wind driven generator fault diagnosis method based on GRA-LSTM-stacking model | |
| CN102112933B (en) | Error detection method and system | |
| CN107436597B (en) | A kind of chemical process fault detection method based on sparse filtering and logistic regression | |
| CN101907681B (en) | Analog circuit dynamic online failure diagnosing method based on GSD-SVDD | |
| CN105631596B (en) | Equipment fault diagnosis method based on multi-dimensional piecewise fitting | |
| CN105677791B (en) | For analyzing the method and system of the operation data of wind power generating set | |
| CN111368885B (en) | Gas circuit fault diagnosis method for aircraft engine | |
| CN106355030A (en) | Fault detection method based on analytic hierarchy process and weighted vote decision fusion | |
| CN110362045B (en) | Marine doubly-fed wind turbine generator fault discrimination method considering marine meteorological factors | |
| CN107238508B (en) | A kind of equipment state diagnostic method and device | |
| CN109491338A (en) | A kind of relevant method for diagnosing faults of multimode procedure quality based on sparse GMM | |
| Charbonnier et al. | A weighted dissimilarity index to isolate faults during alarm floods | |
| Wang et al. | Fault detection and diagnosis for multiple faults of VAV terminals using self-adaptive model and layered random forest | |
| CN105930792A (en) | Human action classification method based on video local feature dictionary | |
| CN111126820A (en) | Electricity stealing prevention method and system | |
| Calvo-Bascones et al. | A collaborative network of digital twins for anomaly detection applications of complex systems. Snitch Digital Twin concept | |
| Saleh et al. | Toward a taxonomy and computational models of abnormalities in images | |
| CN112860183A (en) | Multisource distillation-migration mechanical fault intelligent diagnosis method based on high-order moment matching | |
| CN113884807B (en) | Power distribution network fault prediction method based on random forest and multi-layer architecture clustering | |
| CN115017978A (en) | Fault classification method based on weighted probability neural network | |
| Yang et al. | Convolutional autoencoder-based sensor fault classification | |
| Yeh et al. | Application of auto-encoder for time series classification with class imbalance |
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 |