CN109342862A - Based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults - Google Patents
Based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults Download PDFInfo
- Publication number
- CN109342862A CN109342862A CN201811531680.4A CN201811531680A CN109342862A CN 109342862 A CN109342862 A CN 109342862A CN 201811531680 A CN201811531680 A CN 201811531680A CN 109342862 A CN109342862 A CN 109342862A
- Authority
- CN
- China
- Prior art keywords
- sample
- data
- transformer
- svm model
- sample 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Housings And Mounting Of Transformers (AREA)
Abstract
The invention discloses based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults, comprising the following steps: the sample data for obtaining Dissolved Gas Content in Transformer Oil is standardized；Cluster centre number is set according to transformer state quantity, cluster category division is carried out to the sample data after standardization using Non-surveillance clustering method；Using support vector machine classifier, it is input with the sample data after standardization, SVM model is trained as desired output using clustering category result；Dissolved Gas Content in Transformer Oil data to be diagnosed are standardized reason, input the diagnosis of SVM model realization transformer fault classification.This method is based on data-driven, and diagnosis process depends on excavation of the artificial intelligence technology to Oil Dissolved Gases Concentration data implicit information and the regularity of distribution, is not necessarily to any manual intervention and Heuristics, ensure that the objectivity and reasonability of diagnostic result.
Description
Technical field
The invention belongs to transformer equipment operation troubles diagnostic fields, in particular to a kind of based on Non-surveillance clustering and and SVM
The Diagnosis Method of Transformer Faults of classification.
Background technique
Power transformer is the important equipment of electric system, and operating status directly affects the level of security of system, in time
It was found that latent transformer failure, can prevent from thus causing major accident.Dissolved Gas Content in Transformer Oil is with ratio
Reflect transformer fault type and health status important state parameter, based on this formd IEC recommendation three-ratio method,
The conventional methods such as Rogers method, however in operation there is excessively absolute etc. master of understaffed code, encoded limit in discovery ratio method at the scene
Defect is wanted, causes the accuracy of fault diagnosis lower.
In view of the high reliability of power transformer, most of Oil Dissolved Gases Concentration monitoring data are in normal range (NR)
Interior variation, fault data are extremely rare.The fault type judgement of transformer depends on Heuristics and subjective judgement, has relatively strong
Subjectivity, be easy to produce erroneous judgement and fail to judge.The fast development for having benefited from artificial intelligence technology and its logarithm are according to the prominent of excavation
Output capacity, this paper presents based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults, pass sequentially through non-supervisory
The incidence relation of cluster and svm classifier method quantization Oil Dissolved Gases Concentration and transformer fault classification, and then pass through follow-up
The Oil Dissolved Gases Concentration monitor value of disconnected equipment completes the diagnosis of transformer fault type.
This method is based on data-driven, and fault diagnosis result is dependent on artificial intelligence technology to Oil Dissolved Gases Concentration number
According to the regularity of distribution and implicit information excavation, without carrying out any manual intervention and Heuristics, ensure that diagnostic result and
Objectivity and reasonability.
Summary of the invention
The object of the present invention is to provide it is a kind of based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults,
This method is based on data-driven, ensure that the objectivity and reasonability of diagnostic result.
To achieve the above object, the present invention adopts the following technical solutions:
The present invention provides based on Non-surveillance clustering with and the Diagnosis Method of Transformer Faults of svm classifier include following step
It is rapid:
The sample data for obtaining Dissolved Gas Content in Transformer Oil, is standardized；
According to transformer state quantity be arranged cluster centre number, using Non-surveillance clustering method to standardization after
Sample data carries out cluster category division；
Using support vector machine classifier, it is input with the sample data after standardization, is to cluster category result
Desired output is trained SVM model；
Dissolved Gas Content in Transformer Oil data to be diagnosed are standardized reason, input SVM model realization becomes
The diagnosis of depressor fault category.
Further, the sample data for obtaining Dissolved Gas Content in Transformer Oil, is normalized, specifically
Include:
Hydrogen, methane, ethane, ethylene, acetylene in transformer oil is chosen respectively to carry out every class gas as characteristic gas
Z-score standardization.
Further, the sample data includes normal sample data and exceptional sample data, and normal sample data account for sample
Notebook data sum-rate is lower than 50%.
Further, described that cluster centre number is arranged according to transformer state quantity, using Non-surveillance clustering method pair
Sample data after standardization carries out cluster category division, specifically includes:
Transformer state C is divided into k kind state, sets categorization vector C={ C1,C2,…,Ck}；
K sample data is randomly selected as initial cluster center vector μ={ μ1,…,μk}；It is measured using distance function
Each sample data and all cluster centre vector μiSimilarity will be true apart from nearest center vector according to calculated result
It is set to the classification of the sample, until completing all sample classifications；
The sample average in each class is calculated, and as new cluster centre vector μ *={ μ1*,…,μk* }, again to institute
There is sample to carry out similarity calculation and category division；
Repeat above-mentioned calculating, as μ *={ μ1*,…,μk* it } is no longer changed, shows that cluster is completed and stops counting
It calculates, obtains each sample generic.
It further, is input with the sample data after standardization, to cluster class using support vector machine classifier
Other result is that output is trained SVM model, is specifically included:
It is training sample and test sample that data, which are pressed customized ratio cut partition, and training sample data are inputted SVM model
And it is trained；
The SVM model that training is completed is obtained, test sample is inputted in the SVM model that training is completed and is tested；
Classification output result by test sample in SVM model is compared with its true classification results, when accuracy reaches
When to given lower threshold T, show that the training effect of SVM model reaches accuracy requirement, terminate training, saves training and complete
SVM model；If not reaching accuracy requirement, SVM model is trained again.
Further, described to be standardized Dissolved Gas Content in Transformer Oil data to be diagnosed, input instruction
After the diagnosis algorithm for practicing the SVM model realization transformer fault classification completed, further includes:
It enters data into training completion SVM model to be calculated, obtains SVM and export classification, using the category as the transformation
The condition diagnosing of device is as a result, complete equipment state diagnosis；If diagnostic result is abnormal condition and diagnostic result is correct, by this
Oil Dissolved Gases Concentration data are added in sample database as effective sample, increase the sample size of sample data.
The effect provided in summary of the invention is only the effect of embodiment, rather than invents all whole effects, above-mentioned
A technical solution in technical solution have the following advantages that or the utility model has the advantages that
The present invention is a kind of based on Non-surveillance clustering and and SVM points by conjunction with SVM algorithm, proposing Non-surveillance clustering
The Diagnosis Method of Transformer Faults of class.This method is based on data-driven, and diagnosis process is dependent on artificial intelligence technology to molten in oil
The excavation of gas content data implicit information and the regularity of distribution is solved, any manual intervention and Heuristics is not necessarily to, ensure that diagnosis
As a result objectivity and reasonability.
Detailed description of the invention
Fig. 1 is present invention method flow diagram.
Specific embodiment
In order to clarify the technical characteristics of the invention, below by specific embodiment, and its attached drawing is combined, to this hair
It is bright to be described in detail.Following disclosure provides many different embodiments or example is used to realize different knots of the invention
Structure.In order to simplify disclosure of the invention, hereinafter the component of specific examples and setting are described.In addition, the present invention can be with
Repeat reference numerals and/or letter in different examples.This repetition is that for purposes of simplicity and clarity, itself is not indicated
Relationship between various embodiments and/or setting is discussed.It should be noted that illustrated component is not necessarily to scale in the accompanying drawings
It draws.Present invention omits the descriptions to known assemblies and treatment technology and process to avoid the present invention is unnecessarily limiting.
As shown in Figure 1, based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults, comprising the following steps:
S1, the sample data for obtaining Dissolved Gas Content in Transformer Oil, are standardized；
S2, cluster centre number is arranged according to transformer state quantity, using Non-surveillance clustering method to standardization
Sample data afterwards carries out cluster category division；
S3, using support vector machine classifier, be input with the sample data after standardization, to cluster category result
SVM model is trained for output；
S4, Dissolved Gas Content in Transformer Oil data to be diagnosed are standardized, input the SVM that training is completed
The diagnosis of model realization transformer fault classification.
In step S1, the sample data of Dissolved Gas Content in Transformer Oil is obtained, is standardized, it is specific to wrap
It includes: choosing five kinds of hydrogen, methane, ethane, ethylene, acetylene oil dissolved gas in transformer oil, and be denoted as x=[x1,x2,x3,
x4,x5], as sample data, sample data includes normal sample data and exceptional sample data, and meter sample number is M, normal sample
Notebook data accounts for sample data sum-rate lower than 50%.
Sample data is divided into training sample and test sample, and it is standardized.Using z-score standard
Change method is respectively standardized 5 kinds of gas datas.With x1For standardization, as shown in formula (1):
X in formula1={ x1(1),x1(2),…,x1(i),…,x1It (M) } is sample vector.After i-th of standardization
Sample.mean(x1) and std (x1) it is respectively x1Average and standard deviation.
In step S2, cluster centre number is arranged according to transformer state quantity, using Non-surveillance clustering method to standard
Change that treated that sample data carries out cluster category division, specifically include:
It is 6 classes by transformer state C points, vector C is respectively adopted1=[0,0,0,0,0,1]T、C2=[0,0,0,0,1,0]T、C3
=[0,0,0,1,0,0]T、C4=[0,0,1,0,0,0]T、C5=[0,1,0,0,0,0]TAnd C6=[1,0,0,0,0,0]TIt indicates.
In this example, distance metric function selects Euclidean distance, and clustering method selects Kmeans clustering algorithm.In fact, away from
It can also be using mahalanobis distance, manhatton distance and included angle cosine etc. from metric function；Clustering method can also be equal using Fuzzy C
It is worth cluster, hierarchical clustering, density clustering etc..
6 sample datas are randomly selected as initial cluster center vector μ={ μ1,…,μ6}；It is measured using distance function
Each sample data and all cluster centre vector μiSimilarity will be true apart from nearest center vector according to calculated result
It is set to the classification of the sample, until completing all sample classifications；
The sample average in each class is calculated, and as new cluster centre vector μ *={ μ1*,…,μ6* }, again to institute
There is sample to carry out similarity calculation and category division；
Repeat above-mentioned calculating, as μ *={ μ1*,…,μ6* it } is no longer changed, shows that cluster is completed and stops counting
It calculates, obtains each sample generic.
In step S3, using support vector machine classifier, with the sample data after standardizationTo input, with cluster
Category result is that desired output is trained SVM model, is specifically included:
By M oil dissolved gas sample data according to 8:2 ratio cut partition be training data xtrainAnd test data
xtest, by training sample xtrainIn 5 class gases be input, the corresponding classification C of each samplejFor output, it is input to initialization
SVM classifier, and SVM is trained.Specifically, construction gaussian radial basis function is as kernel function；Using population
Optimization algorithm (PSO) optimizes the parameter σ and punishment parameter C of gaussian radial basis function, using particle group optimizing theory
It optimizes that detailed process is as follows: 1) initializing；2) initial suitable on search space by calculating each particle in population
Angle value is answered, particle fitness is evaluated；3) according to the speed of iterative formula more new particle and position；If 4) optimizing reach it is maximum into
Change algebra Tmax or evaluation of estimate is less than given accuracy, then terminates searching process, otherwise continue fitness evaluation.Using particle
The target that group's optimum theory optimizes is the classification capacity for enhancing SVM.
Accuracy threshold value T=80% is set, by test sample xtestIn 5 class gases be input in previous step training completion
SVM classifier, obtain the classification results C of test sampletest,j.And with its true tag CjIt is compared, calculates the standard of classification
Exactness ACC, as shown in formula (2).
As ACC >=T, show that SVM training effect meets accuracy requirement, SVM is completed into training and is saved.If ACC < T, table
Bright SVM training effect is unsatisfactory for accuracy requirement, needs to SVM re -training.
It, will Dissolved Gas Content in Transformer Oil data x be diagnosed in step S4dIt is standardized, inputs SVM mould
Type realizes the diagnosis of transformer fault classification, specifically includes:
By input data xd=[xd,1,xd,2,xd,3,xd,4,xd,5] be standardized, with xd,1For, such as formula (3) institute
Show:
It willIt is input in the svm classifier model kept, the classification results C being calculatedjIt is examined for final equipment state
Disconnected result.If diagnostic result is abnormal condition and diagnostic result is correct, using the Oil Dissolved Gases Concentration data as having
Effect sample is added in sample database, increases the sample size of sample data.
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (6)
1. based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults, characterized in that the following steps are included:
The sample data for obtaining Dissolved Gas Content in Transformer Oil, is standardized；
Cluster centre number is set according to transformer state quantity, using Non-surveillance clustering method to the sample after standardization
Data carry out cluster category division；
It is input with the sample data after standardization using support vector machine classifier, to cluster category result as expectation
Output is trained SVM model；
Dissolved Gas Content in Transformer Oil data to be diagnosed are standardized reason, input SVM model realization transformer
The diagnosis of fault category.
2. the method as described in claim 1, characterized in that the sample number for obtaining Dissolved Gas Content in Transformer Oil
According to being standardized, specifically include:
It chooses hydrogen, methane, ethane, ethylene, acetylene in transformer oil and z- is carried out to every class gas respectively as characteristic gas
Score standardization.
3. method according to claim 2, characterized in that the sample data includes normal sample data and exceptional sample number
According to normal sample data account for sample data sum-rate lower than 50%.
4. method as claimed in claim 3, characterized in that it is described that cluster centre number is arranged according to transformer state quantity,
Cluster category division is carried out to the sample data after standardization using Non-surveillance clustering method, is specifically included:
Transformer state C is divided into k kind state, sets categorization vector C={ C1,C2,…,Ck}；
K sample data is randomly selected as initial cluster center vector μ={ μ1,…,μk}；It is each using distance function measurement
Sample data and all cluster centre vector μiSimilarity will be determined as apart from nearest center vector according to calculated result
The classification of the sample, until completing all sample classifications；
The sample average in each class is calculated, and as new cluster centre vector μ*={ μ1 *,…,μk *, again to all samples
This progress similarity calculation and category division；
Repeat above-mentioned calculating, works as μ*={ μ1 *,…,μk *Be no longer changed, show that cluster is completed and stops calculating, obtains
Each sample generic.
5. method as claimed in claim 4, characterized in that support vector machine classifier is used, with the sample after standardization
Notebook data is input, is trained, is specifically included to SVM model with clustering category result as output:
By customized ratio cut partition it is training sample and test sample by data, training sample data is inputted into SVM model and right
It is trained；
The SVM model that training is completed is obtained, test sample is inputted in the SVM model that training is completed and is tested；
Classification output result by test sample in SVM model is compared with its true classification results, when accuracy reach to
When fixed lower threshold T, show that the training effect of SVM model reaches accuracy requirement, terminate training, saves what training was completed
SVM model；If not reaching accuracy requirement, SVM model is trained again.
6. method as claimed in claim 5, characterized in that it is described will Dissolved Gas Content in Transformer Oil data be diagnosed into
Row standardization inputs after the diagnosis algorithm for the SVM model realization transformer fault classification that training is completed, further includes:
It enters data into training completion SVM model to be calculated, obtains SVM and export classification, using the category as the transformer
Condition diagnosing is as a result, complete equipment state diagnosis；If diagnostic result is abnormal condition and diagnostic result is correct, will be in the oil
Dissolved gas content data is added in sample database as effective sample, increases the sample size of sample data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811531680.4A CN109342862A (en) | 2018-12-14 | 2018-12-14 | Based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811531680.4A CN109342862A (en) | 2018-12-14 | 2018-12-14 | Based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109342862A true CN109342862A (en) | 2019-02-15 |
Family
ID=65303864
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811531680.4A Pending CN109342862A (en) | 2018-12-14 | 2018-12-14 | Based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109342862A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109934281A (en) * | 2019-03-08 | 2019-06-25 | 电子科技大学 | A kind of unsupervised training method of two sorter networks |
| CN112733878A (en) * | 2020-12-08 | 2021-04-30 | 国网辽宁省电力有限公司锦州供电公司 | Transformer fault diagnosis method based on kmeans-SVM algorithm |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3342989A (en) * | 1960-12-27 | 1967-09-19 | Westinghouse Air Brake Co | Track fullness system |
| CN103823158A (en) * | 2014-02-19 | 2014-05-28 | 福州大学 | Resonant earthed system fault line selection method with invariant moments |
| CN105868770A (en) * | 2016-03-23 | 2016-08-17 | 国网山东省电力公司电力科学研究院 | High-voltage circuit breaker fault diagnosis method based on unsupervised learning model |
| CN106443259A (en) * | 2016-09-29 | 2017-02-22 | 国网山东省电力公司电力科学研究院 | Transformer fault diagnosis new method based on Euclidean clustering and SPO-SVM |
| CN107545307A (en) * | 2017-07-28 | 2018-01-05 | 上海交通大学 | Predicting model for dissolved gas in transformer oil method and system based on depth belief network |
-
2018
- 2018-12-14 CN CN201811531680.4A patent/CN109342862A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3342989A (en) * | 1960-12-27 | 1967-09-19 | Westinghouse Air Brake Co | Track fullness system |
| CN103823158A (en) * | 2014-02-19 | 2014-05-28 | 福州大学 | Resonant earthed system fault line selection method with invariant moments |
| CN105868770A (en) * | 2016-03-23 | 2016-08-17 | 国网山东省电力公司电力科学研究院 | High-voltage circuit breaker fault diagnosis method based on unsupervised learning model |
| CN106443259A (en) * | 2016-09-29 | 2017-02-22 | 国网山东省电力公司电力科学研究院 | Transformer fault diagnosis new method based on Euclidean clustering and SPO-SVM |
| CN107545307A (en) * | 2017-07-28 | 2018-01-05 | 上海交通大学 | Predicting model for dissolved gas in transformer oil method and system based on depth belief network |
Non-Patent Citations (2)
| Title |
|---|
| 荣智海等: "基于油中溶解气体Canopy 高维模型的变压器异常状态快速识别方法", 《中国电机工程学报》 * |
| 郑元兵: "3.1 聚类算法与支持向量机结合的变压器故障诊断组合模型", 《变压器故障诊断与预测集成学习方法及维修决策模型研究》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109934281A (en) * | 2019-03-08 | 2019-06-25 | 电子科技大学 | A kind of unsupervised training method of two sorter networks |
| CN112733878A (en) * | 2020-12-08 | 2021-04-30 | 国网辽宁省电力有限公司锦州供电公司 | Transformer fault diagnosis method based on kmeans-SVM algorithm |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109931678B (en) | Air conditioner fault diagnosis method based on deep learning LSTM | |
| CN113496262B (en) | Data-driven active power distribution network abnormal state sensing method and system | |
| Dai et al. | Dissolved gas analysis of insulating oil for power transformer fault diagnosis with deep belief network | |
| Lu et al. | An MDL approach to the climate segmentation problem | |
| CN111523603B (en) | Ship power equipment fault identification method | |
| CN106843195B (en) | The Fault Classification differentiated based on adaptive set at semi-supervised Fei Sheer | |
| CN108268905A (en) | A kind of Diagnosis Method of Transformer Faults and system based on support vector machines | |
| CN111898690B (en) | Power transformer fault classification method and system | |
| CN108536107A (en) | Colony intelligence optimizing fault diagnosis system based on hybrid optimized parameter | |
| CN111291783A (en) | Intelligent fault diagnosis method, system, terminal and storage medium for gas pressure regulating equipment | |
| CN109298633A (en) | Chemical production process fault monitoring method based on adaptive piecemeal Non-negative Matrix Factorization | |
| CN111340248A (en) | Transformer fault diagnosis method and system based on intelligent integration algorithm | |
| CN111680875A (en) | Unmanned aerial vehicle state risk fuzzy comprehensive evaluation method based on probability baseline model | |
| Laayati et al. | Smart Energy Management System: Oil Immersed Power Transformer Failure Prediction and Classification Techniques Based on DGA Data | |
| CN107239789A (en) | A kind of industrial Fault Classification of the unbalanced data based on k means | |
| CN109342862A (en) | Based on Non-surveillance clustering with and svm classifier Diagnosis Method of Transformer Faults | |
| CN113225346A (en) | Network operation and maintenance situation assessment method based on machine learning | |
| CN116562114A (en) | Power transformer fault diagnosis method based on graph convolution neural network | |
| CN111695288A (en) | Transformer fault diagnosis method based on Apriori-BP algorithm | |
| CN110443481B (en) | Power distribution automation terminal state evaluation system and method based on hybrid K-nearest neighbor algorithm | |
| CN115392782A (en) | Method and system for monitoring and diagnosing health state of process system of nuclear power plant | |
| CN114384423A (en) | Fuel cell health state identification method based on typical operation parameters | |
| CN117056678B (en) | Machine pump equipment operation fault diagnosis method and device based on small sample | |
| CN105759787B (en) | A kind of method for diagnosing faults having supervision linear dynamic system model based on switching | |
| CN116756939A (en) | Transformer fault diagnosis method based on ANOVA-IWAA-SVM hybrid model |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190215 |