CN104809328A - Transformer fault diagnosis method based on information bottleneck - Google Patents
Transformer fault diagnosis method based on information bottleneck Download PDFInfo
- Publication number
- CN104809328A CN104809328A CN201410527345.2A CN201410527345A CN104809328A CN 104809328 A CN104809328 A CN 104809328A CN 201410527345 A CN201410527345 A CN 201410527345A CN 104809328 A CN104809328 A CN 104809328A
- Authority
- CN
- China
- Prior art keywords
- sample
- data
- transformer
- information bottleneck
- diagnosis method
- 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
Abstract
The invention relates to a transformer fault diagnosis method based on information bottleneck, and belongs to the technical field of transformer fault diagnosis. The method is that the information bottleneck method is adopted to cluster data to obtain a result cluster with the maximum similarity in the cluster; a cluster label of a sample to be detected is determined by voting training samples in the cluster so as to determine the fault class. With the adoption of the method, the classification result is superior to that of the rule-based Duval triangle method, the mode recognizing based BPNN algorithm and the mode recognizing based Bayes method; the diagnosis accuracy can be effectively increased during diagnosing the transformer fault.
Description
Technical field
The present invention relates to a kind of Diagnosis Method of Transformer Faults based on information bottleneck, belong to transformer fault diagnosis technical field.
Background technology
As one of important component part in electric system, the running status of transformer directly determines the stability of whole system.In order to ensure electric power netting safe running, need to carry out early warning to transformer fault.Dissolved gas analysis (Dissolved Gas Analysis, DGA) is one of transformer fault diagnosis main method, can make efficient diagnosis to transformer fault.At present, the DGA diagnostic method of industry mainstream applications is rule-based David's triangulation method, and the method is based on three kinds of oil dissolved gas: methane (CH
4), ethene (C
2h
4) and acetylene (C
2h
2) quantity carry out computing and carry out failure judgement type according to the position of result points in David's triangle.The method has simply, efficiently and intuitively advantage.Except rule-based diagnostic method, mode identification method is extensively incorporated in transformer fault diagnosis problem in recent years, as artificial neural network (ANN), support vector machine (SVM), Bayes's (Bayes) method and semisupervised classification method etc. all achieve certain effect.But the accuracy at transformer fault diagnosis of said method is not too high.
Summary of the invention
The object of this invention is to provide a kind of Diagnosis Method of Transformer Faults based on information bottleneck, with the problem that the accuracy solving existing transformer fault diagnosis is not high.
The present invention is for solving the problems of the technologies described above and providing a kind of Diagnosis Method of Transformer Faults based on information bottleneck, and this method for diagnosing faults comprises the following steps:
1) transformer DGA data are gathered as sample data collection;
2) adopt IB method to carry out cluster to sample data collection, obtain k result bunch;
3) to each sample to be tested, by the class label choosing sample to be tested in a vote of training sample in each bunch, thus the diagnosis to transformer fault is realized.
Described each data sample comprises eight attributes, is respectively the content of H2, CH4, C2H6, C2H4, C2H2, ZTING, CO, CO2 and the data class label of expert's demarcation.
Described transformer DGA data also need to carry out data cleansing and regular, and its process is as follows:
A) sample of the classification number not having expert to demarcate is deleted;
B) repeated sample is deleted;
C) sample having disappearance attribute is deleted;
D) delete property value is the sample of 0.
Described step 1) in transformer DGA data also need to carry out data transformation.
Described data transformation is any one in Max-Min method, logarithmic characteristic converter technique and arctan function method.
Described step 2) detailed process as follows:
A) sample data that sample data is concentrated is carried out training set and test set division, be divided into test sample book and training sample;
B) successively each test sample book is merged into training set;
C) adopt IB clustering and obtain k result bunch.
Described step 3) in the ballot mode that adopts comprise simple majority ballot and Nearest Neighbor with Weighted Voting two kinds, described simple majority ballot refers to that simple majority is elected, and described Nearest Neighbor with Weighted Voting refers to and is weighted ballot with the prior probability of each classification for weighing.
The invention has the beneficial effects as follows: the present invention adopts information bottleneck method to carry out cluster with the maximum result bunch of similarity in obtaining bunch to data, then by bunch in the class label choosing sample to be tested in a vote of training sample, thus determine fault category, classification results of the present invention is better than rule-based David's deltic method, the BPNN algorithm based on pattern-recognition and the Bayes method based on pattern-recognition, in transformer fault diagnosis, effectively can improve the accuracy of diagnosis.
Accompanying drawing explanation
Fig. 1 is the process flow diagram of the Diagnosis Method of Transformer Faults that the present invention is based on information bottleneck.
Embodiment
Below in conjunction with accompanying drawing, the specific embodiment of the present invention is further described.
Information bottleneck method (Information Bottleneck, IB) is a kind of based on information-theoretical mode identification method, is widely used in the fields such as clustering documents, graphical analysis, DNA process and video image retrieval.For improving the accuracy of transformer fault diagnosis, information bottleneck method is applied in transformer DGA diagnosis by the present invention, and propose a kind of Diagnosis Method of Transformer Faults based on information bottleneck method, as shown in Figure 1, detailed process is as follows.
1. adopt information bottleneck method to carry out cluster to sample data collection (comprising training sample and sample to be tested), obtain k result bunch.
2. to each sample to be tested, by the class label choosing sample to be tested in a vote of training sample in each bunch.
First get a test sample book, and be merged in training set; Adopt IB method cluster and obtain k result bunch; If x belongs to certain bunch of t, then use all training samples in this bunch to vote, ballot mode adopts simple majority ballot and Nearest Neighbor with Weighted Voting two kinds of modes.Simple majority ballot is simple majority way to elect, is weighted ballot during Nearest Neighbor with Weighted Voting with the prior probability of each classification for weighing.Such as hypothetical target class c=(1,2,3), prior probability Pc=(1/4,1/2,1/4).If comprise 10 training samples and 1 sample to be tested in bunch A, and the class label of 10 training samples is [1,3,1,1,2,1,2,2,1].Then the simple majority of these 10 training samples is voted as result is class label 1, namely generic 1 is demarcated as to sample to be tested, and Nearest Neighbor with Weighted Voting result is class label 2, namely generic number is decided to be to sample to be tested table, therefrom can find out, simple majority and Nearest Neighbor with Weighted Voting can produce different class labels, and the degree of reliability of Nearest Neighbor with Weighted Voting depends on priori.
When employing is diagnosed DGA data based on mode identification method, different algorithms has different data prediction requirements.Such as, neural network algorithm requires that data do normalized, to avoid the attribute left and right arithmetic result with higher value territory; Bayesian algorithm requires to carry out discretize to carry out probability calculation to continuous data.And for IB algorithm, only require that data are non-negative data, so that can computing information entropy and association relationship judge the cost that sample is assigned again with this.Therefore, the present invention in an experiment common-sense have employed some data conversion methods and carries out pre-service to data, according to DGA data characteristics, the present invention can select in Max-Min method, logarithmic characteristic converter technique and arctan function method three kinds of data conversion methods any one.
Experimental data of the present invention is taken from actual production data, altogether collected 609 samples, each sample comprises the data class label that eight attributes (H2, CH4, C2H6, C2H4, C2H2, ZTNC, CO, CO2) and expert demarcate altogether, data cleansing and regular step as follows:
1) sample of the classification number not having expert to demarcate is deleted;
2) repeated sample is deleted;
3) sample having disappearance attribute is deleted;
4) delete property value is the sample of 0.
Through above-mentioned steps process, the valid data number of samples finally obtained is 350, and the fault type distribution of data sample is in table 1.
Table 1
10 groups of experimental data collection are obtained after carrying out training set and test set division for these 350 data samples.Division methods adopts ten folding cross-validation methods, is divided into ten parts at random by data set, in turn will wherein 9 parts as training data, 1 part as test data.Adopt the mean value of ten ten folding cross validation accuracy as the estimation to arithmetic accuracy of the present invention.
Simple majority and the comparison of Nearest Neighbor with Weighted Voting result, DGAsIB algorithm has three parameters, and in experiment, classification k is set as DGA fault type number, i.e. k=7, cluster cycle index l=10, balance parameters.Classification accuracy rate after table 2 gives raw data under two kinds of ballot modes, max-min conversion, after log-transformation and after arc tangent conversion.
Table 2
Therefrom can observe:
1) vote in the contrast of modes for two kinds, in 10 experiments, simple majority is dominant for 5 times, and Nearest Neighbor with Weighted Voting is dominant for 5 times, and the average that is folded flat for ten times ten of simple majority is 58.37%, and the average that is folded flat for ten times ten of Nearest Neighbor with Weighted Voting is 57.83%.This illustrates that simple majority ballot method is slightly better than weighted voting algorithm, its main cause is due to fault data Limited Number obtainable from actual production, the prior distribution probability (see table 1) of each fault type is caused to fail complete reaction actual distribution, therefore the weighted value of weighted voting algorithm is not ideal enough, along with the accumulation gradually of data, the result of Nearest Neighbor with Weighted Voting can improve.
2) in the contrast of various data transformation form, only once the result of log-transformation is optimum, this illustrates for IB method, only need ensure the nonnegativity (DGA data itself have had the characteristic of non-negative) of raw data, and not need normalization or the conversion process of other algorithm for pattern recognition requirement.
It should be noted last that: above embodiment is the non-limiting technical scheme of the present invention in order to explanation only, although with reference to above-described embodiment to invention has been detailed description, those of ordinary skill in the art is to be understood that; Still can modify to the present invention or equivalent replacement, and not depart from any modification or partial replacement of the spirit and scope of the present invention, it all should be encompassed in the middle of right of the present invention.
Claims (7)
1. based on a Diagnosis Method of Transformer Faults for information bottleneck, it is characterized in that, this method for diagnosing faults comprises the following steps:
1) transformer DGA data are gathered as sample data collection;
2) adopt IB method to carry out cluster to sample data collection, obtain k result bunch;
3) to each sample to be tested, by the class label choosing sample to be tested in a vote of training sample in each bunch, thus the diagnosis to transformer fault is realized.
2. the Diagnosis Method of Transformer Faults based on information bottleneck according to claim 1, it is characterized in that, described each data sample comprises eight attributes, is respectively the content of H2, CH4, C2H6, C2H4, C2H2, ZTING, CO, CO2 and the data class label of expert's demarcation.
3. the Diagnosis Method of Transformer Faults based on information bottleneck according to claim 2, is characterized in that, described transformer DGA data also need to carry out data cleansing and regular, and its process is as follows:
A) sample of the classification number not having expert to demarcate is deleted;
B) repeated sample is deleted;
C) sample having disappearance attribute is deleted;
D) delete property value is the sample of 0.
4. the Diagnosis Method of Transformer Faults based on information bottleneck according to claim 3, is characterized in that, described step 1) in transformer DGA data also need to carry out data transformation.
5. the Diagnosis Method of Transformer Faults based on information bottleneck according to claim 4, is characterized in that, described data transformation is any one in Max-Min method, logarithmic characteristic converter technique and arctan function method.
6. the Diagnosis Method of Transformer Faults based on information bottleneck according to claim 4, is characterized in that, described step 2) detailed process as follows:
A) sample data that sample data is concentrated is carried out training set and test set division, be divided into test sample book and training sample;
B) successively each test sample book is merged into training set;
C) adopt IB clustering and obtain k result bunch.
7. the Diagnosis Method of Transformer Faults based on information bottleneck according to claim 6, it is characterized in that, described step 3) in adopt ballot mode comprise simple majority ballot and Nearest Neighbor with Weighted Voting two kinds, described simple majority ballot refers to that simple majority is elected, and described Nearest Neighbor with Weighted Voting refers to and is weighted ballot with the prior probability of each classification for weighing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410527345.2A CN104809328A (en) | 2014-10-09 | 2014-10-09 | Transformer fault diagnosis method based on information bottleneck |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410527345.2A CN104809328A (en) | 2014-10-09 | 2014-10-09 | Transformer fault diagnosis method based on information bottleneck |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104809328A true CN104809328A (en) | 2015-07-29 |
Family
ID=53694144
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410527345.2A Pending CN104809328A (en) | 2014-10-09 | 2014-10-09 | Transformer fault diagnosis method based on information bottleneck |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104809328A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106596900A (en) * | 2016-12-13 | 2017-04-26 | 贵州电网有限责任公司电力科学研究院 | Transformer fault diagnosis method based on improved semi-supervised classification of graph |
| CN107831711A (en) * | 2017-10-27 | 2018-03-23 | 山东大学 | Bull-dozer power assembly fault diagnosis system construction method and device based on cluster |
| CN109740013A (en) * | 2018-12-29 | 2019-05-10 | 深圳英飞拓科技股份有限公司 | Image processing method and image search method |
| 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 (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW584732B (en) * | 2002-02-20 | 2004-04-21 | Mang-Hui Wang | CMAC_based fault diagnosis of power transformers |
| CN102944769A (en) * | 2012-10-18 | 2013-02-27 | 上海市电力公司 | Fault diagnosis method of power transformer based on extreme learning machine |
| CN103198175A (en) * | 2013-03-04 | 2013-07-10 | 辽宁省电力有限公司鞍山供电公司 | Transformer fault diagnosis method based on fuzzy cluster |
| CN103398769A (en) * | 2013-08-05 | 2013-11-20 | 国家电网公司 | Transformer on-line fault detecting method based on sampling integrated SVM (support vector machine) under wavelet GGD (general Gaussian distribution) feather and unbalanced K-mean value |
-
2014
- 2014-10-09 CN CN201410527345.2A patent/CN104809328A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW584732B (en) * | 2002-02-20 | 2004-04-21 | Mang-Hui Wang | CMAC_based fault diagnosis of power transformers |
| CN102944769A (en) * | 2012-10-18 | 2013-02-27 | 上海市电力公司 | Fault diagnosis method of power transformer based on extreme learning machine |
| CN103198175A (en) * | 2013-03-04 | 2013-07-10 | 辽宁省电力有限公司鞍山供电公司 | Transformer fault diagnosis method based on fuzzy cluster |
| CN103398769A (en) * | 2013-08-05 | 2013-11-20 | 国家电网公司 | Transformer on-line fault detecting method based on sampling integrated SVM (support vector machine) under wavelet GGD (general Gaussian distribution) feather and unbalanced K-mean value |
Non-Patent Citations (4)
| Title |
|---|
| 《2011 EIGHTH INTERNATIONAL CONFERENCE ON FUZZY SYSTEMS AND KNOWLEDGE DISCOVERY》 * |
| 《INTERNATIONAL JOURNAL OF PATTERN RECOGNITION AND ARTIFCIAL INTELLIGENCE》 * |
| 姬波: "信息瓶颈方法的特征权重研究", 《中国博士学位论文全文数据库》 * |
| 张燕: "基于改进的BP神经网络的变压器故障诊断研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106596900A (en) * | 2016-12-13 | 2017-04-26 | 贵州电网有限责任公司电力科学研究院 | Transformer fault diagnosis method based on improved semi-supervised classification of graph |
| CN107831711A (en) * | 2017-10-27 | 2018-03-23 | 山东大学 | Bull-dozer power assembly fault diagnosis system construction method and device based on cluster |
| CN107831711B (en) * | 2017-10-27 | 2019-10-15 | 山东大学 | Bull-dozer power assembly fault diagnosis system construction method and device based on cluster |
| CN109740013A (en) * | 2018-12-29 | 2019-05-10 | 深圳英飞拓科技股份有限公司 | Image processing method and image search method |
| 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 |
|---|---|---|
| CN108985380B (en) | Point switch fault identification method based on cluster integration | |
| US10325150B2 (en) | System and method for electric load identification and classification employing support vector machine | |
| CN105760888B (en) | A kind of neighborhood rough set integrated learning approach based on hierarchical cluster attribute | |
| Shen et al. | Time series anomaly detection with multiresolution ensemble decoding | |
| CN104951764B (en) | Hot-short Activity recognition method based on secondary spectral clustering and HMM-RF mixed models | |
| CN104809328A (en) | Transformer fault diagnosis method based on information bottleneck | |
| CN101738998B (en) | System and method for monitoring industrial process based on local discriminatory analysis | |
| CN108333468B (en) | The recognition methods of bad data and device under a kind of active power distribution network | |
| CN107656152B (en) | One kind being based on GA-SVM-BP Diagnosis Method of Transformer Faults | |
| CN109934245B (en) | Point switch fault identification method based on clustering | |
| CN108664538A (en) | A kind of automatic identification method and system of the doubtful familial defect of power transmission and transforming equipment | |
| Zhan et al. | Fabric defect classification using prototypical network of few-shot learning algorithm | |
| CN110765587A (en) | Complex petrochemical process fault diagnosis method based on dynamic regularization judgment local retention projection | |
| CN104832418A (en) | Hydraulic pump fault diagnosis method based on local mean conversion and Softmax | |
| CN103493075A (en) | Outlier detecting apparatus, outlier detecting method, and vehicle trouble diagnosis system | |
| CN103337248A (en) | Airport noise event recognition method based on time series kernel clustering | |
| CN113255236A (en) | Non-invasive load self-adaptive identification method based on twin network | |
| Li et al. | Distance measures in building informatics: An in-depth assessment through typical tasks in building energy management | |
| CN116796271A (en) | Resident energy abnormality identification method | |
| CN102945238A (en) | Fuzzy ISODATA (interactive self-organizing data) based feature selection method | |
| Tehrani et al. | Frequency-based multi task learning with attention mechanism for fault detection in power systems | |
| CN106224067B (en) | A kind of method for diagnosing faults of the automobile three-way catalysts based on data-driven | |
| Zhao et al. | A hierarchical framework for drone identification based on radio frequency machine learning | |
| CN116432027A (en) | Transformer running state detection method and system based on vibration signals | |
| CN116433049A (en) | Power consumption abnormality detection method based on fuzzy rough entropy |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| EXSB | Decision made by sipo to initiate 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: 20150729 |