CN105930861B - A kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm - Google Patents
A kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm Download PDFInfo
- Publication number
- CN105930861B CN105930861B CN201610227946.0A CN201610227946A CN105930861B CN 105930861 B CN105930861 B CN 105930861B CN 201610227946 A CN201610227946 A CN 201610227946A CN 105930861 B CN105930861 B CN 105930861B
- Authority
- CN
- China
- Prior art keywords
- transformer
- weight
- basis function
- sample
- neural network
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000003745 diagnosis Methods 0.000 title claims abstract description 53
- 238000012549 training Methods 0.000 claims abstract description 55
- 238000012360 testing method Methods 0.000 claims abstract description 12
- 238000013528 artificial neural network Methods 0.000 claims description 28
- 239000007789 gas Substances 0.000 claims description 20
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims description 17
- 230000000694 effects Effects 0.000 claims description 12
- HSFWRNGVRCDJHI-UHFFFAOYSA-N alpha-acetylene Natural products C#C HSFWRNGVRCDJHI-UHFFFAOYSA-N 0.000 claims description 10
- OTMSDBZUPAUEDD-UHFFFAOYSA-N Ethane Chemical compound CC OTMSDBZUPAUEDD-UHFFFAOYSA-N 0.000 claims description 8
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 claims description 8
- 239000005977 Ethylene Substances 0.000 claims description 8
- 125000002534 ethynyl group Chemical group [H]C#C* 0.000 claims description 8
- 229910052739 hydrogen Inorganic materials 0.000 claims description 8
- 239000001257 hydrogen Substances 0.000 claims description 8
- 238000003064 k means clustering Methods 0.000 claims description 5
- 206010020843 Hyperthermia Diseases 0.000 claims description 4
- 238000010891 electric arc Methods 0.000 claims description 4
- 230000036031 hyperthermia Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 2
- 230000003213 activating effect Effects 0.000 claims 1
- 125000004435 hydrogen atom Chemical class [H]* 0.000 claims 1
- 230000006870 function Effects 0.000 description 65
- 230000001537 neural effect Effects 0.000 description 16
- 238000012544 monitoring process Methods 0.000 description 12
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 description 8
- 150000002431 hydrogen Chemical class 0.000 description 6
- 238000004458 analytical method Methods 0.000 description 5
- UGFAIRIUMAVXCW-UHFFFAOYSA-N Carbon monoxide Chemical compound [O+]#[C-] UGFAIRIUMAVXCW-UHFFFAOYSA-N 0.000 description 4
- 229910002092 carbon dioxide Inorganic materials 0.000 description 4
- 239000001569 carbon dioxide Substances 0.000 description 4
- 229910002091 carbon monoxide Inorganic materials 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 4
- 238000012423 maintenance Methods 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 230000005611 electricity Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000004613 tight binding model Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000002405 diagnostic procedure Methods 0.000 description 1
- 238000012631 diagnostic technique Methods 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000013024 troubleshooting Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/245—Classification techniques relating to the decision surface
- G06F18/2453—Classification techniques relating to the decision surface non-linear, e.g. polynomial classifier
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Mathematical Physics (AREA)
- Nonlinear Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to a kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm, comprising: is trained by the training sample set to Weak Classifier；Weak Classifier is integrated into the strong classifier of more high-class precision after circuit training and weighed value adjusting；Using the test sample as the input of strong classifier, to obtain corresponding fault type.The present invention solves the problems, such as that strong classifier is difficult to obtain, in addition, operation of the present invention is simple, carrying out Classification and Identification to transformer fault mode has preferable practicability by integrating Weak Classifier.
Description
Technical field
The invention belongs to transformer fault on-line monitoring technique field, the gas pair that is generated when refering in particular to according to transformer fault
Transformer carries out fault diagnosis, and in particular to a kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm.
Background technique
With the high speed development of power grid construction, China's power grid from city island network, develops into large regional grid, west electricity
East is sent, north and south supply mutually, and on national network pattern is being formed.Electric system be one by it is numerous send out, give, it is defeated, match, electrical equipment
The big system being formed by connecting, the reliability and operation conditions of these equipment directly determine the stabilization and safety of whole system, also determine
Power supply quality and power supply reliability are determined, as electric system is to high voltage, large capacity, internet development and each electricity consumption portion
The raising that door requires, the requirement to the security reliability index of electric system are also higher and higher.Power transformer is electric system
One of the main reason for important component part, normal reliable operation is guarantee entire power grid operation.Shape in decades
At preventative maintenance system, the reliability for improving operation of power networks has been served very important, but this cannot find in time
Insulation hidden danger inside equipment.
In addition, the expense of preventative maintenance is also high.As power grid develops with national economy to increasingly automated direction to confession
The requirement of electric reliability is higher and higher, there is an urgent need to change to existing maintenance of equipment system, with on-line monitoring and failure
CBM System Based based on diagnostic techniques gradually replaces preventative maintenance system or becomes for the development of tracing and monitoring failure
Gesture has defined.Therefore how reliably diagnosis is carried out to the potential failure of transformer in time to have a very important significance.
At present there are many kinds of Method of Fault Diagnosis in Transformer, such as: BP neural network is the troubleshooting issue of transformer
A kind of relatively good structural system is provided, but that there is convergence speeds is slow, is easily trapped into the shortcomings that local minimum point；Specially
Family system effectively simulated failure diagnosing human expert can complete failure diagnostic process, but that there is also knowledge acquisition is tired
Many technical problems such as difficult, uncertain inference and self study difficulty；Fuzzy control can be with accurate mathematical tool by mould
The concept or natural language sharpening of paste, to can reasonably be quantified to phenomenon of the failure, but its fuzzy membership letter
Number needs expertise or repetition test just to can determine that.
To solve the above-mentioned problems, the invention proposes a kind of transformer fault diagnosis sides based on Adaboost algorithm
Method.
Summary of the invention
Deficiency regarding to the issue above, the present invention propose a kind of transformer fault diagnosis side based on Adaboost algorithm
Method.It is an object of the invention to solve at least one above problem or defect, and provide at least one will be described later it is excellent
Point.
It is a still further object of the present invention to provide a kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm, lead to
It crosses and the radial basis function neural network classifier is constantly trained using AdaBoost, constantly adjusted according to error, then pass through
Nearest Neighbor with Weighted Voting, which is combined, is promoted to final strong classifier, improves fault diagnosis precision.
It is a still further object of the present invention to provide a kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm,
Using the gas on-site data of transformer online monitoring system real-time monitoring as sample set data, according to these transformer faults spy
The concentration variation for levying gas, predicts transformer fault mode type.
It is a still further object of the present invention to provide a kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm,
AdaBoost algorithm is applied in transformer fault diagnosis analysis, according to the event of the concentration variation prediction transformer of gas on-site
Barrier.
It is a still further object of the present invention to provide a kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm,
Computation complexity and calculation amount are substantially reduced, online quick diagnosis is more suitable for.
In order to realize these purposes and other advantages according to the present invention, the present invention provides one kind to be calculated based on Adaboost
The Diagnosis Method of Transformer Faults of method, comprising the following steps:
Step 1 acquires the concentration value of a variety of transformer fault characteristic gas, will make after a variety of concentration value normalization
For the sample of sample set, while encoding the fault mode type of transformer, will corresponding transformer fault mode type target
Exports coding is as the corresponding class label of attribute value in the sample set, using a part of sample set as training sample set, and it is another
A part of sample set is as test sample collection；
Wherein, each of described training sample set training sample has equal initial weight；
Step 2, using radial basis function neural network as Weak Classifier, all samples that the training sample is concentrated
Originally it is set as the center of the radial basis function neural network；
Step 3 carries out circuit training to the radial basis function neural network by the training sample set, until
To anticipation function；
Step 4, according to the prediction effect of the anticipation function, to assign the anticipation function and the initial weight not
Same weight generates final anticipation function combined sequence using the ballot method of Weight；
Step 5 inputs the test sample collection in the step 1 in the final anticipation function combined sequence,
After the final anticipation function combined sequence judgement identification, by the matching class label, the fault mode type is obtained,
Complete diagnosis.
Adaboost algorithm need to only find several classification accuracies, and more slightly higher than random assortment accuracy rate (i.e. accuracy is bigger
In Weak Classifier 50%), the present invention selects the radial basis function neural network, generates anticipation function sequence by training to it
Column, and the final anticipation function, as final strong classifier are generated using Nearest Neighbor with Weighted Voting mechanism.It is thus possible to by weak typing
Device is organically integrated into the higher final strong classifier of a nicety of grading.Using the method to transformer fault mode class
Type, which carries out Classification and Identification, has preferable practicability.The present invention is by doing comparative experiments discovery, using single radial base nerve net
The classification accuracy of network algorithm is about 81.25%, and uses the Diagnosis Method of Transformer Faults method based on Adaboost algorithm
Classification accuracy be 93.75%, whole accuracy rate improves 12.5%.
Preferably, in the step 1, while the fault type for encoding transformer is specific as follows:
By normal condition, medium temperature overheat, hyperthermia and superheating, six class fault mode of shelf depreciation, spark discharge and arc discharge
It is separately encoded as Arabic numerals 1,2,3,4,5 and 6, as class label corresponding to attribute value in the sample set.
The present invention applies Adaboost algorithm in the practical application of transformer fault mode type identification, improves change
Depressor fault identification precision.
Preferably, the transformer fault characteristic gas is hydrogen, methane, ethane, ethylene and acetylene.
Data source of the invention is in the gas on-site data of transformer online monitoring system real-time monitoring, and not only monitoring becomes
The information such as hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, the carbon dioxide that depressor generates when breaking down, and according to this
The concentration variation of a little gases substantially predicts transformer fault.
Preferably, in said step 1, by initializing each of training sample set training sample
Weight coefficient, so that each described sample has equal initial weight.
Preferably, in the step 2, using radial basis function neural network as Weak Classifier, need to be swashed by Green
Function living is activated.
Preferably, in the step 3, the radial basis function neural network is carried out by the training sample set
Circuit training, until obtaining anticipation function, specific step is as follows:
The final Basis Function Center of the radial basis function neural network is sought based on K- means clustering method；
According to the maximum distance between the training sample and the final Basis Function Center, variance yields is obtained；
By the final Basis Function Center and the variance yields, the hidden layer obtained in the radial base neural net is arrived
Connection weight between output layer；
By the connection weight in conjunction with the structure of the radial base neural net, the reality of the radial base neural net is obtained
Within a predetermined range, then training terminates the positive negative difference of border output, the reality output and target output, obtains anticipation function,
The positive negative difference is not in the preset range, then repeatedly above step；
Wherein, the reality output is the reality output coding of corresponding transformer fault mode type, the target output
For the target exports coding of corresponding transformer fault mode type.
Preferably, the final Basis Function Center packet of the radial basis function neural network is sought based on K- means clustering method
Include following steps:
A, n training sample of the training sample concentration is chosen as cluster centre, is gathered according to input sample with described
The input sample is assigned in each cluster set by the Euclidean distance between class center；Calculate each cluster set
The average value of middle training sample obtains new cluster centre, judge the new cluster centre whether change decide whether into
The center of row next round solves；It is to then follow the steps B, it is no, then follow the steps C；
B, the new cluster centre of above-mentioned steps is the final Basis Function Center of radial basis function neural network；
C, the n training samples are chosen again as cluster centre, into the solution of the cluster centre of next round；
Wherein, input sample is the sample that the training sample is concentrated, and the input sample is according between center
The classification that Euclidean distance carries out.
Preferably, in the step 4, assigned according to the prediction effect of the anticipation function sequence its with it is described initial
The weight of weighted, generating final anticipation function combined sequence using the ballot method of Weight, specific step is as follows:
The weight training error for calculating the anticipation function, obtains the prediction effect of the anticipation function；
According to the prediction effect, first weight of anticipation function sequence is assigned, and first weight is updated,
Obtain the second weight；
Second weight is normalized, and sums up to normalized respective weights, is generated by ballot method final pre-
Survey sequence of function combination.
When the prediction effect (training error) of the anticipation function is greater than 0.5, to the second of the final anticipation function
Weight is normalized, and sums up to normalized respective weights, generates final anticipation function sequence group by ballot method
It closes.
Adaboost algorithm need to only find several classification accuracies, and more slightly higher than random assortment accuracy rate (i.e. accuracy is bigger
In Weak Classifier 50%), anticipation function sequence is generated by training, and final prediction letter is generated using Nearest Neighbor with Weighted Voting mechanism
Number, so as to which Weak Classifier to be organically integrated into the higher strong classifier of nicety of grading, i.e., the described final prediction letter
Number, carrying out Classification and Identification to transformer fault mode using the method has preferable practicability, solves strong classifier and is difficult to
The problem of acquisition, while accuracy of identification also can be improved.
Preferably, the preset range is 0.01-0.05.
Preset range is generally set between 0-1, but the preset range of the invention is narrower, is become so that the present invention identifies
The error of depressor fault mode type is small, and accuracy is higher.
Beneficial effects of the present invention:
1, the Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, to radial ba-sis function network
Network classifier is trained, then is combined by Nearest Neighbor with Weighted Voting and is promoted to final strong classifier, and fault diagnosis precision is improved.
2, the Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, in Fault Diagnosis Method of Power Transformer mould
In formula type procedure, the influence of the subjective factor of people is avoided, keeps selection more objective, classification accuracy rate is higher.Experiment shows this
The classification accuracy of invention Fault Diagnosis Method of Power Transformer types of models is 93.75%, than point of single radial base neural net algorithm
Class accuracy rate improves 12.5%.
3, the Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm is calculated by integrating weak study
Method solves the problems, such as that strong learning algorithm is difficult to obtain, while precision of prediction also can be improved.
4, the Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, uses radial base neural net
(RBF) it is used as Weak Classifier, RBF has the excellent spy that accuracy rate is high, structure adaptive is determining, output is unrelated with initial weight
Property, so that the time for training network is far smaller than other network trainings, shortens the time of Fault Diagnosis Method of Power Transformer types of models.
5, the Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, the sample number of the analysis method
According to the transformer fault characteristic gas for deriving from transformer online monitoring system real-time monitoring, not only monitors transformer and break down
When the hydrogen, methane, ethane, ethylene, acetylene, the carbon monoxide, carbon dioxide isoconcentration information that generate, and according to these gases
Concentration variation substantially predicts transformer fault.
6, the Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, can greatly reduce and be calculated as
This, meets the requirement of online Fast Classification diagnosis.
Detailed description of the invention
Fig. 1 is radial base neural net structural representation of the present invention；
Fig. 2 is the structure chart of the Diagnosis Method of Transformer Faults based on Adaboost algorithm；
Fig. 3 is the flow chart of the Diagnosis Method of Transformer Faults based on Adaboost algorithm.
Specific embodiment
The present invention will be further described with reference to the accompanying drawings and detailed description.
Flow chart of the invention is as shown in Fig. 3, comprising the following steps:
Step 1 acquires the concentration value of a variety of transformer fault characteristic gas, will make after a variety of concentration value normalization
For the sample of sample set, while the fault mode type of transformer is encoded, by normal condition, medium temperature overheat, hyperthermia and superheating, part
Electric discharge, spark discharge and six class fault mode of arc discharge are separately encoded as Arabic numerals 1,2,3,4,5 and 6, as described
Class label corresponding to attribute value in sample set, using a part of sample set as training sample set, and another part sample set is made
For test sample collection；
Wherein, the transformer fault characteristic gas is hydrogen, methane, ethane, ethylene and acetylene, or hydrogen,
The combination of methane, ethane, ethylene, acetylene, carbon monoxide and carbon dioxide；
Training sample S={ (x1, y1), (x2, y2) ... ..., (xn, yn) } after being normalized, wherein xi ∈ X,
(concentration of 5 kinds of characteristic gas when xi is transformer fault), yi ∈ Y={ 1,2 ... ..., k }, the failure classes of corresponding transformer
Type.The weight coefficient of each sample is initialized simultaneouslyI=1,2 ... ..., n.I.e. each sample has equal
Initial weight.
The sample data of the analysis method derives from the transformer fault feature of transformer online monitoring system real-time monitoring
Gas not only monitors hydrogen, methane, ethane, ethylene, acetylene, carbon monoxide, the carbon dioxide generated when transformer breaks down
Isoconcentration information, and transformer fault is substantially predicted according to the variation of the concentration of these gases.
Step 2, using radial basis function neural network as Weak Classifier, all samples that the training sample is concentrated
Originally it is set as the center of the radial basis function neural network；
Using radial basis function neural network RBFNN as Weak Classifier, RBF neural is first constructed.Choose hidden node
Number is equal to input sample number, and the activation primitive of hidden node is Green function, specific functional expression are as follows:
In formula, | | xn-ci| | it is European norm, c is the center of Green's function, and σ is the variance of Green's function.Simultaneously by institute
There is input sample to be set as the center of radial basis function, each radial basis function takes unified extension constant.Network structure such as Fig. 1 institute
Show.
Use radial base neural net (RBF) as Weak Classifier, RBF has accuracy rate height, structure adaptive determining, defeated
The good characteristic unrelated with initial weight out makes the time for training network be far smaller than other network trainings, shortens diagnosis
The time of transformer fault types of models.
Step 3 carries out circuit training to the radial basis function neural network by the training sample set, until
It is specific as follows to anticipation function:
The final Basis Function Center of the radial basis function neural network is sought based on K- means clustering method；What selection was given
N training sample is as cluster centre ci (i=1,2 ..., n)；According to the Euclidean distance of xp and center between ci by xp points
It is fitted in each cluster set θ p (p=1,2 ..., P) of input sample；Calculate it is each cluster set θ p in training sample put down
Mean value, i.e., new cluster centre ci, if new cluster centre is no longer changed, resulting ci is RBF neural
Otherwise final Basis Function Center redistributes training sample set, solve into the center of next round.
According to the maximum distance between the training sample and the final Basis Function Center, variance yields, solution side are obtained
Poor σ i, formula are as follows
In formula, maximum distance of the cmax between selected center, i=1,2 ..., n
By the final Basis Function Center and the variance yields, the hidden layer obtained in the radial base neural net is arrived
Connection weight between output layer, the formula for calculating the weight between hidden layer and output layer are as follows:
Wherein, n=1,2 ..., n；I=1,2 ..., n
By the connection weight in conjunction with the structure of the radial base neural net, the reality of the radial base neural net is obtained
Border exports yj, within a predetermined range, then training terminates the positive negative difference that the reality output and target export, and obtains prediction letter
Number, the positive negative difference is not in the preset range, then repeatedly above step；The formula of the reality output is as follows:
Wherein, in j=1,2 ..., n formula,For n-th of input sample, n=1,2 ..., N, N table
Showing total sample number, ci is the center of network hidden layer node, and ω ij is connection weight of the hidden layer to output layer, i=1,2 ...,
N is the nodal point number of hidden layer.
Wherein, the reality output is the reality output coding of corresponding transformer fault mode type, the target output
For the target exports coding of corresponding transformer fault mode type.
Wherein, the preset range given to this invention is 0.01-0.05.
Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, can greatly reduce and be calculated as
This, meets the requirement of online Fast Classification diagnosis.
Step 4, according to the prediction effect of the anticipation function, to assign the anticipation function and the initial weight not
It is specific as follows to generate final anticipation function combined sequence using the ballot method of Weight for same weight:
The weight training error for calculating the anticipation function, obtains the prediction effect of the anticipation function；Calculate the power of ht
Retraining error is exactly mistake point rate, formula is as follows:
Wherein, if yi≠ht(xi), then I=1；Otherwise, I=0.
According to the prediction effect, first weight α of anticipation function sequence is assigned(t), and first weight is carried out
It updates, obtains the second weight；
Wherein, the formula of first weight is as follows:
And the update of the first weight is carried out by above formula, update the formula of coefficient are as follows:
Wherein, ZtFor normalization coefficient, may make
When the training error of anticipation function is greater than 0.5, second weight for finally obtaining anticipation function is returned
One changes, and obtains strong classifier:
Work as ht(x)=k when, corresponding weight is summed up, passes through ballot method and generates final anticipation function combined sequence:
Wherein, arg max g (t), expression be domain a subset, either element can all make function in the subset
G () is maximized.
Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, to radial basis function neural network
Classifier is trained, then is combined by Nearest Neighbor with Weighted Voting and is promoted to final strong classifier, and fault diagnosis precision is improved,
That is solving the problems, such as that strong learning algorithm is difficult to obtain, while precision of prediction also can be improved by integrating weak learning algorithm.
Step 5 inputs the test sample collection in the step 1 in the final anticipation function combined sequence,
After the final anticipation function combined sequence judgement identification, by the matching class label, the fault mode type is obtained,
Complete diagnosis.
Diagnosis Method of Transformer Faults provided by the invention based on Adaboost algorithm, to radial basis function neural network
Classifier is trained, then is combined by Nearest Neighbor with Weighted Voting and is promoted to final strong classifier, and fault diagnosis precision is improved.
On the other hand, the present invention can greatly reduce calculating cost, meet the requirement of online Fast Classification diagnosis.
One embodiment provided by the invention, choose hydrogen (H2), methane (CH4), ethane (C2H6), ethylene (C2H4),
The component of acetylene (C2H2) this five kinds of characteristic feature gases first carries out dissolved gas constituent content as primitive attribute data
Normalized makes it all in the range of [- 1,1], as the attribute value in sample set.
Then transformer fault type is encoded, normal condition, medium temperature overheat, hyperthermia and superheating, shelf depreciation, fire
It is 1,2,3,4,5,6 that flower, which discharges, arc discharge is separately encoded, as class label corresponding to attribute value in sample set.
Trained and test sample is used as using 305 groups of Gases Dissolved in Transformer Oil components for having determined that fault type.?
Training sample is concentrated, and each fault type has 35 groups of sample datas, remaining 95 groups of test sample as model.Utilize 210 groups
Data propose that the Diagnosis Method of Transformer Faults analysis model of Adaboost algorithm is trained to this patent, using 95 groups of numbers
According to being tested, wherein the part test sample collection is 16 groups, as shown in table 1.
16 groups of data of test sample collection described in 1 part of table
In order to verify effectiveness of the invention and accuracy, and compared with the classification of single radial base neural net algorithm
Performance superiority and inferiority has carried out one group of comparative experiments, the accuracy of two kinds of classification methods such as table 2.
The comparison result of table 2 present invention and single radial base neural net algorithm classification accuracy
By table 2, it can be concluded that, experiment discovery is about using the classification accuracy of single radial base neural net algorithm
81.25%, and use the Diagnosis Method of Transformer Faults analysis model classification accuracy of Adaboost algorithm for 93.75%, this
For invention the method compared with single radial base neural net algorithm, classification accuracy improves 12.5%.
The present invention uses the Adaboost algorithm of the transformer fault figure penalties function based on Adaboost algorithm.
Adaboost is that a kind of realization is simple, using also very simple algorithm, and will not over-fitting.In addition, being based on figure penalties letter
Several Adaboost algorithms algorithm substantially reduces the required precision to Weak Classifier, and algorithm is simple and clear, can direct solution it is more
Class classification problem, greatly reduces computation complexity and calculation amount, is a kind of algorithm for being suitably applied transformer fault diagnosis.
The present invention there are also the data of other embodiments, is just not listed one by one herein.
Although the embodiments of the present invention have been disclosed as above, but its is not only in the description and the implementation listed
With it can be fully applied to various fields suitable for the present invention, for those skilled in the art, can be easily
Realize other modification, therefore without departing from the general concept defined in the claims and the equivalent scope, the present invention is simultaneously unlimited
In specific details and legend shown and described herein.
Claims (8)
1. a kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm, which comprises the following steps:
Step 1 acquires the concentration value of a variety of transformer fault characteristic gas, will be used as sample after a variety of concentration value normalization
The sample of this collection, while the fault mode type of transformer is encoded, the target of the fault mode type of corresponding transformer is exported
Coding is as the corresponding class label of attribute value in the sample set, using a part of sample set as training sample set, and another portion
Divide sample set as test sample collection；
Wherein, each of described training sample set training sample has equal initial weight；
Step 2, using radial basis function neural network as Weak Classifier, all samples that the training sample is concentrated are set
For the center of the radial basis function neural network；
Step 3 carries out circuit training to the radial basis function neural network by the training sample set, until obtaining pre-
Surveying function, specific step is as follows:
The final Basis Function Center of the radial basis function neural network is sought based on K- means clustering method；
According to the maximum distance between the training sample and the final Basis Function Center, variance yields is obtained；
By the final Basis Function Center and the variance yields, the hidden layer obtained in the radial basis function neural network is arrived
Connection weight between output layer；
By the connection weight in conjunction with the structure of the radial basis function neural network, the radial basis function neural network is obtained
Reality output, the reality output and target output positive negative difference within a predetermined range, then training terminate, obtain prediction letter
Number, the positive negative difference is not in the preset range, then repeatedly above step；
Wherein, the reality output is the reality output coding of corresponding transformer fault mode type, and the target output is pair
Answer the target exports coding of transformer fault mode type；
Step 4 is different from the initial weight to assign the anticipation function according to the prediction effect of the anticipation function
Weight generates final anticipation function combined sequence using the ballot method of Weight；
Step 5 inputs the test sample collection in the step 1 in the final anticipation function combined sequence, through institute
After stating final anticipation function combined sequence judgement identification, by matching the class label, obtains the fault mode type, complete
Diagnosis.
2. the Diagnosis Method of Transformer Faults according to claim 1 based on Adaboost algorithm, which is characterized in that described
In step 1, while the fault mode type for encoding transformer is specific as follows:
Normal condition, medium temperature overheat, hyperthermia and superheating, shelf depreciation, spark discharge and six class fault mode of arc discharge are distinguished
Arabic numerals 1,2,3,4,5 and 6 are encoded to, as class label corresponding to attribute value in the sample set.
3. the Diagnosis Method of Transformer Faults according to claim 1 based on Adaboost algorithm, which is characterized in that described
A variety of transformer fault characteristic gas are hydrogen, methane, ethane, ethylene and acetylene.
4. the Diagnosis Method of Transformer Faults according to claim 1 based on Adaboost algorithm, which is characterized in that in institute
It states in step 1, by initializing the weight coefficient of each of training sample set training sample, so that each institute
Training sample is stated with equal initial weight.
5. the Diagnosis Method of Transformer Faults as claimed in any of claims 2 to 4 based on Adaboost algorithm,
Be characterized in that, in the step 2, using radial basis function neural network as Weak Classifier, need to by Green's activation primitive into
Line activating.
6. the Diagnosis Method of Transformer Faults according to claim 1 based on Adaboost algorithm, which is characterized in that be based on
K- means clustering method seek the final Basis Function Center of the radial basis function neural network the following steps are included:
A, n training sample of the training sample concentration is chosen as cluster centre, according in input sample and the cluster
The input sample is assigned in each cluster set by the Euclidean distance between the heart；It calculates and is instructed in each cluster set
The average value for practicing sample, obtains new cluster centre, judges whether the new cluster centre changes to decide whether to carry out down
The center of one wheel solves；It is to then follow the steps B, it is no, then follow the steps C；
B, the new cluster centre of above-mentioned steps is the final Basis Function Center of the radial basis function neural network；
C, the n training samples are chosen again as cluster centre, into the solution of the cluster centre of next round.
7. the Diagnosis Method of Transformer Faults according to claim 1 based on Adaboost algorithm, which is characterized in that described
In step 4, its weight different from the initial weight is assigned according to the prediction effect of the anticipation function, using cum rights
The ballot method of weight generates final anticipation function combined sequence, and specific step is as follows:
The weight training error for calculating the anticipation function, obtains the prediction effect of the anticipation function；
According to the prediction effect, first weight of anticipation function is assigned, and first weight is updated, obtains
Two weights；
Second weight is normalized, and normalized respective weights are summed up, is generated by ballot method final
Anticipation function combined sequence.
8. the Diagnosis Method of Transformer Faults according to claim 1 based on Adaboost algorithm, which is characterized in that institute
Stating preset range is 0.01-0.05.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610227946.0A CN105930861B (en) | 2016-04-13 | 2016-04-13 | A kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610227946.0A CN105930861B (en) | 2016-04-13 | 2016-04-13 | A kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105930861A CN105930861A (en) | 2016-09-07 |
| CN105930861B true CN105930861B (en) | 2019-08-06 |
Family
ID=56837998
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610227946.0A Active CN105930861B (en) | 2016-04-13 | 2016-04-13 | A kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105930861B (en) |
Families Citing this family (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106530082A (en) * | 2016-10-25 | 2017-03-22 | 清华大学 | Stock predication method and stock predication system based on multi-machine learning |
| CN106501693A (en) * | 2016-12-08 | 2017-03-15 | 贵州电网有限责任公司电力科学研究院 | A kind of Diagnosis Method of Transformer Faults based on fuzzy Boltzmann machine |
| CN107180140B (en) * | 2017-06-08 | 2019-12-10 | 中南大学 | Shafting fault identification method based on dual-tree complex wavelet and AdaBoost |
| CN107194465A (en) * | 2017-06-16 | 2017-09-22 | 华北电力大学（保定） | A kind of method that utilization virtual sample trains Neural Network Diagnosis transformer fault |
| CN108021945A (en) * | 2017-12-07 | 2018-05-11 | 广东电网有限责任公司电力科学研究院 | A kind of transformer state evaluation model method for building up and device |
| CN108717149A (en) * | 2018-05-25 | 2018-10-30 | 西安工程大学 | Diagnosis Method of Transformer Faults based on M-RVM fusion dynamic weightings AdaBoost |
| CN109063734B (en) * | 2018-06-29 | 2022-02-25 | 广东工业大学 | Oil-immersed transformer fault state evaluation method combining multi-level local density clustering |
| CN109188162A (en) * | 2018-07-17 | 2019-01-11 | 广东工业大学 | It is a kind of based on the Transformer condition evaluation that can open up radial base neural net |
| CN109325519A (en) * | 2018-08-20 | 2019-02-12 | 中国铁道科学研究院集团有限公司电子计算技术研究所 | Fault recognition method and device |
| CN109726767A (en) * | 2019-01-13 | 2019-05-07 | 胡燕祝 | A kind of perceptron network data classification method based on AdaBoost algorithm |
| CN109858564B (en) * | 2019-02-21 | 2023-05-05 | 上海电力学院 | Improved Adaboost-SVM model generation method suitable for wind power converter fault diagnosis |
| CN110618340A (en) * | 2019-03-11 | 2019-12-27 | 广东工业大学 | Transformer state evaluation method |
| CN110286161A (en) * | 2019-03-28 | 2019-09-27 | 清华大学 | Main transformer method for diagnosing faults based on adaptive enhancing study |
| CN110163531A (en) * | 2019-06-02 | 2019-08-23 | 南京邮电大学盐城大数据研究院有限公司 | Network transformer abnormality method for early warning based on K- cluster |
| CN110263837A (en) * | 2019-06-13 | 2019-09-20 | 河海大学 | A kind of circuit breaker failure diagnostic method based on multilayer DBN model |
| CN110289097A (en) * | 2019-07-02 | 2019-09-27 | 重庆大学 | A kind of Pattern Recognition Diagnosis system stacking model based on Xgboost neural network |
| CN110516339B (en) * | 2019-08-21 | 2022-03-22 | 西北工业大学 | Adaboost algorithm-based method for evaluating reliability of sealing structure in multiple failure modes |
| CN111723518A (en) * | 2020-05-29 | 2020-09-29 | 国网四川省电力公司电力科学研究院 | Transformer fault diagnosis device and method based on condition inference tree and AdaBoost |
| CN111767675A (en) * | 2020-06-24 | 2020-10-13 | 国家电网有限公司大数据中心 | Transformer vibration fault monitoring method and device, electronic equipment and storage medium |
| CN111860658A (en) * | 2020-07-24 | 2020-10-30 | 华北电力大学（保定） | Transformer fault diagnosis method based on cost sensitivity and integrated learning |
| CN112580715B (en) * | 2020-12-16 | 2024-05-07 | 珠海格力电器股份有限公司 | Household equipment fault detection method, device, equipment and medium |
| CN112710956B (en) * | 2020-12-17 | 2023-08-04 | 四川虹微技术有限公司 | Expert system-based battery management system fault detection system and method |
| CN113391172A (en) * | 2021-05-31 | 2021-09-14 | 国网山东省电力公司电力科学研究院 | Partial discharge diagnosis method and system based on time sequence integration and used for multi-source ultrasonic detection |
| CN113705405B (en) * | 2021-08-19 | 2023-04-18 | 电子科技大学 | Nuclear pipeline fault diagnosis method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5640103A (en) * | 1994-06-30 | 1997-06-17 | Siemens Corporate Research, Inc. | Radial basis function neural network autoassociator and method for induction motor monitoring |
| CN103207950A (en) * | 2013-04-16 | 2013-07-17 | 郑州航空工业管理学院 | Intelligent transformer fault diagnostic method based on RBF (radial basis function) neural network |
| CN103235973A (en) * | 2013-04-16 | 2013-08-07 | 郑州航空工业管理学院 | Transformer fault diagnosis method based on radial basis function neural network |
| CN104299035A (en) * | 2014-09-29 | 2015-01-21 | 国家电网公司 | Method for diagnosing fault of transformer on basis of clustering algorithm and neural network |
| CN104616033A (en) * | 2015-02-13 | 2015-05-13 | 重庆大学 | Fault diagnosis method for rolling bearing based on deep learning and SVM (Support Vector Machine) |
-
2016
- 2016-04-13 CN CN201610227946.0A patent/CN105930861B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5640103A (en) * | 1994-06-30 | 1997-06-17 | Siemens Corporate Research, Inc. | Radial basis function neural network autoassociator and method for induction motor monitoring |
| CN103207950A (en) * | 2013-04-16 | 2013-07-17 | 郑州航空工业管理学院 | Intelligent transformer fault diagnostic method based on RBF (radial basis function) neural network |
| CN103235973A (en) * | 2013-04-16 | 2013-08-07 | 郑州航空工业管理学院 | Transformer fault diagnosis method based on radial basis function neural network |
| CN104299035A (en) * | 2014-09-29 | 2015-01-21 | 国家电网公司 | Method for diagnosing fault of transformer on basis of clustering algorithm and neural network |
| CN104616033A (en) * | 2015-02-13 | 2015-05-13 | 重庆大学 | Fault diagnosis method for rolling bearing based on deep learning and SVM (Support Vector Machine) |
Non-Patent Citations (3)
| Title |
|---|
| Application of support vector machines for fault diagnosis in power transmission system;B. Ravikumar 等;《IET Generation, Transmission & Distribution》;20080121;第2卷(第1期);第119-130页 |
| 基于径向基概率神经网络的变压器故障诊断;高宏岩;《煤矿机械》;20071031;第198-200页 |
| 径向基函数神经网络在电力变压器故障诊断中的应用;麻闽政;《广东电力》;20120131;第25卷(第1期);第80-83页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105930861A (en) | 2016-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105930861B (en) | A kind of Diagnosis Method of Transformer Faults based on Adaboost algorithm | |
| Dai et al. | Dissolved gas analysis of insulating oil for power transformer fault diagnosis with deep belief network | |
| Feng et al. | Analysis and prediction of students’ academic performance based on educational data mining | |
| CN105930901B (en) | A kind of Diagnosis Method of Transformer Faults based on RBPNN | |
| Thai-Nghe et al. | Factorization Models for Forecasting Student Performance. | |
| CN106646158B (en) | Based on multi-category support vector machines transformer fault diagnosis method for improving | |
| CN107256393A (en) | The feature extraction and state recognition of one-dimensional physiological signal based on deep learning | |
| CN109242149A (en) | A kind of student performance early warning method and system excavated based on educational data | |
| CN105574589B (en) | Transformer oil chromatographic method for diagnosing faults based on niche genetic algorithm | |
| CN107132310B (en) | Transformer equipment health status method of discrimination based on gauss hybrid models | |
| CN110766137A (en) | Power electronic circuit fault diagnosis method based on longicorn whisker optimized deep confidence network algorithm | |
| CN110542819B (en) | Transformer fault type diagnosis method based on semi-supervised DBNC | |
| CN104155574A (en) | Power distribution network fault classification method based on adaptive neuro-fuzzy inference system | |
| CN106650932B (en) | A kind of the intelligent trouble classification method and device of data center's monitoring system | |
| CN108229581A (en) | Based on the Diagnosis Method of Transformer Faults for improving more classification AdaBoost | |
| CN107545307A (en) | Predicting model for dissolved gas in transformer oil method and system based on depth belief network | |
| Xu et al. | Fusing complete monotonic decision trees | |
| Ni et al. | Deepqr: Neural-based quality ratings for learnersourced multiple-choice questions | |
| CN109711707A (en) | A kind of Ship Power Equipment synthetical condition assessment method | |
| Li et al. | Fault identification in power network based on deep reinforcement learning | |
| Zhang et al. | Multi-task learning boosted predictions of the remaining useful life of aero-engines under scenarios of working-condition shift | |
| Li et al. | Snowball: Iterative model evolution and confident sample discovery for semi-supervised learning on very small labeled datasets | |
| Lizhe et al. | BP network model optimized by adaptive genetic algorithms and the application on quality evaluation for class teaching | |
| Deng et al. | RISSAD: Rule-based Interactive Semi-Supervised Anomaly Detection. | |
| Cai et al. | Capturing deep dynamic information for mapping users across social networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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 | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A transformer fault diagnosis method based on AdaBoost algorithm Effective date of registration: 20201119 Granted publication date: 20190806 Pledgee: Xi'an Science and Technology Financial Service Center Co.,Ltd. Pledgor: XI'AN SI-TOP ELECTRIC Co.,Ltd. Registration number: Y2020980008277 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20211202 Granted publication date: 20190806 Pledgee: Xi'an Science and Technology Financial Service Center Co.,Ltd. Pledgor: XI'AN SI-TOP ELECTRIC Co.,Ltd. Registration number: Y2020980008277 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A transformer fault diagnosis method based on AdaBoost algorithm Effective date of registration: 20211203 Granted publication date: 20190806 Pledgee: Xi'an Science and Technology Financial Service Center Co.,Ltd. Pledgor: XI'AN SI-TOP ELECTRIC Co.,Ltd. Registration number: Y2021980014051 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20230109 Granted publication date: 20190806 Pledgee: Xi'an Science and Technology Financial Service Center Co.,Ltd. Pledgor: XI'AN SI-TOP ELECTRIC Co.,Ltd. Registration number: Y2021980014051 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A Transformer Fault Diagnosis Method Based on Adaboost Algorithm Effective date of registration: 20230309 Granted publication date: 20190806 Pledgee: Industrial Bank Limited by Share Ltd. Xi'an branch Pledgor: XI'AN SI-TOP ELECTRIC Co.,Ltd. Registration number: Y2023610000156 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Granted publication date: 20190806 Pledgee: Industrial Bank Limited by Share Ltd. Xi'an branch Pledgor: XI'AN SI-TOP ELECTRIC Co.,Ltd. Registration number: Y2023610000156 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A Transformer Fault Diagnosis Method Based on Adaboost Algorithm Granted publication date: 20190806 Pledgee: Industrial Bank Limited by Share Ltd. Xi'an branch Pledgor: XI'AN SI-TOP ELECTRIC Co.,Ltd. Registration number: Y2024610000077 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right |