CN112881827A - Oil-immersed transformer fault diagnosis method based on improved grey correlation analysis - Google Patents
Oil-immersed transformer fault diagnosis method based on improved grey correlation analysis Download PDFInfo
- Publication number
- CN112881827A CN112881827A CN202110037452.7A CN202110037452A CN112881827A CN 112881827 A CN112881827 A CN 112881827A CN 202110037452 A CN202110037452 A CN 202110037452A CN 112881827 A CN112881827 A CN 112881827A
- Authority
- CN
- China
- Prior art keywords
- feature
- fault
- sample
- characteristic
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Housings And Mounting Of Transformers (AREA)
Abstract
The invention discloses a fault diagnosis method for an oil-immersed transformer, which comprises the following steps: s1, collecting and sorting multi-type fault samples of the oil-immersed transformer, and constructing a feature set FS1 and a fault sample set D1; s2, calculating the feature weight of a feature set FS1 by two feature selection methods, establishing low-dimensional feature subsets FS21 and FS22, and further establishing low-dimensional fault sample sets D21 and D22; s3, solving clustering centers C1 and C2 of fault sample sets D21 and D22 by using a fuzzy clustering algorithm, and taking the obtained clustering centers C1 and C2 as two groups of reference sequences in improved gray correlation analysis; and normalizing the feature weights corresponding to the low-dimensional features FS21 and FS22, calculating corresponding association values based on the Mahalanobis distance and the dynamic identification coefficient, and selecting the fault type corresponding to the maximum association value as a diagnosis result of the sample to be detected after calculating the average value. The oil-immersed transformer fault diagnosis method established by the invention has the advantages of simple structure, high calculation efficiency, good diagnosis performance and the like.
Description
Technical Field
The invention belongs to the technical field of electrical equipment, and particularly relates to an oil-immersed transformer fault diagnosis method based on improved grey correlation analysis.
Background
The oil-immersed transformer is used as expensive and important component equipment in a high-voltage power system, bears the functions of voltage conversion and electric energy distribution, and plays an especially important role in reliable operation of the power system. Because the operation environment of the oil-immersed transformer is severe, various defects and faults inevitably occur, and potential safety hazards are brought to the reliable operation of a power system; when the oil-immersed transformer suffers from a fault, the power production department faces huge economic loss, so that the fault of the transformer can be found and processed as soon as possible, and the establishment of an efficient and reliable fault diagnosis system has important significance.
At present, among the fault diagnosis methods for oil-immersed transformers, a Dissolved Gas Analysis (DGA) method in oil is one of the simplest and most widely applied methods. However, the DGA-based ratio method has the defects of absolute coding boundary, limited coding types, low diagnosis accuracy, inconsistent diagnosis performance and the like; in recent years, experts and scholars at home and abroad adopt technical theories of machine learning, data mining and the like to establish a plurality of intelligent fault diagnosis models based on dissolved gas in oil and obtain positive effects, but the problems of complex diagnosis models, redundant characteristics, low diagnosis precision, low diagnosis efficiency and the like exist at the same time. The problems of characteristic redundancy, difficulty in determining characteristic weight, incapability of reflecting mutual influence among resolution results and the like exist in the conventional oil-immersed transformer fault diagnosis method based on grey correlation analysis. Therefore, a model for diagnosing potential faults of the oil immersed transformer, which is simple in structure, accurate, efficient and convenient, is established, the working state of the power transformer is mastered in time, and the method has important significance for reducing transformer maintenance, shortening power failure time and improving equipment and power grid reliability.
Disclosure of Invention
The invention aims to provide an oil-immersed transformer fault diagnosis method based on improved grey correlation analysis, and solves the problems of complex diagnosis model, redundant characteristics, low diagnosis precision and low diagnosis efficiency of the existing oil-immersed transformer fault diagnosis method.
In order to solve the technical problems, the invention is realized by the following technical scheme:
an oil-immersed transformer fault diagnosis method based on improved grey correlation analysis comprises the following steps:
s1, collecting the oil-immersed transformer, sorting multiple types of fault samples, and constructing a feature set FS1 according to the component values and the combinations of conventional gases dissolved in oil; a fault sample set D1 is established together based on a gas sample dissolved in oil of the oil-immersed transformer and the characteristic set FS 1;
s2, calculating feature weights of a feature set FS1 by adopting two different feature selection methods, sorting the feature weights according to size values, reserving features with weight values meeting the requirements of preset thresholds, establishing low-dimensional feature subsets FS21 and FS22, and further establishing low-dimensional fault sample sets D21 and D22;
s3, solving clustering centers C1 and C2 of fault sample sets D21 and D22 by using a fuzzy clustering algorithm, and respectively using the obtained clustering centers as reference sequences in improved gray correlation analysis; normalization processing is carried out on the feature weights corresponding to the low-dimensional features FS21 and FS22, and then the correlation coefficient and the correlation degree of the sample to be detected are calculated respectively according to the weight values and the reference sequence; and calculating the average value of the two groups of correlation degrees, and outputting the fault type corresponding to the maximum value of the average correlation degrees as the diagnosis result of the sample to be detected.
Preferably, in the step S1, the feature set FS1 is characterized by 9 fault types of oil chromatogram samples, the oil chromatogram samples are composed of content features of 5 feature gases, and the 5 feature gases are methane (CH4), hydrogen (H2), ethane (C2H6), ethylene (C2H4), and acetylene (C2H 2);
wherein, the content characteristic of each characteristic gas consists of three characteristics of absolute content, relative content and mutual ratio;
among them, 9 kinds of fault types include No Fault (NF), Partial Discharge (PD), low energy discharge (LD), high energy discharge (HD), low temperature overheat (LT), medium temperature overheat (MT), high temperature overheat (HT), low energy discharge and overheat (LTD), and high energy discharge and overheat (HTD);
and extracting the contained characteristic values from the characteristic set FS1 so as to construct a fault sample set D1.
Preferably, the step S2 includes the following steps:
s21, aiming at the established fault sample set D1, 80% of fault samples are used for establishing a fault diagnosis model, and 20% of fault samples are used for testing the diagnosis performance of the model; in order to eliminate dimension inconsistency among different samples, the data of the original fault data set is normalized, and the extreme value normalization formula is as follows:
wherein: x is the number ofikIs the kth feature data in the ith sample, and xik maxAnd xik minRespectively representing the maximum value and the minimum value of the kth characteristic;
s22, calculating a feature weight value in the feature set FS1 by using a feature selection method (the Fisher Score method and the Relief method are respectively used in the invention) based on the feature set FS1 normalized in the step S21, wherein the larger the weight value is, the more important the feature is, otherwise, the feature is insensitive to fault classification, and the feature can be removed.
Wherein, calculating the characteristic weight value w by using a Fisher Scoer method1(fi) The calculation method of (2) is as follows:
assume that the dataset contains class c samples, nkIs the total number of class k samples, fj,iIs the jth sample value in the ith feature; u. ofiDenotes the ith characteristiciAverage value of (a), uikRepresents the mean value, x, in the k class under the ith featurei,kjFor the jth sample in the kth class sample in the ith feature fiValue of (i) isScore of feature F (F)i) And weight w1(fi) Can be calculated as follows:
wherein, the characteristic weight value w is calculated by using a Relief F method2(fi) The calculation method of (2) is as follows:
firstly, setting all feature weights as 0; then, k neighbor sample sets of samples belonging to the same class as the random sample R and k samples not belonging to the same class as R are searched from the data set D, and each characteristic weight w is updated2(fi);
Wherein, diff (f)i,Ri,Hj) Denotes samples R and HjIn characteristic fiDifference of above, HjIs the jth nearest neighbor sample of the same class as R. C is the total number of classes, class (R), which is different from sample Ri) Represents the class of the sample R, P (C) represents the probability of class C, diff (f)i,Ri,Mj(C) Are sample R and sample M)j(C) In characteristic fiDifference of above, Mj(C) Representing the jth nearest neighbor sample to sample R in class C. Wherein diff (f)i,Ri,Hj) The calculation can be as follows:
in the above formula, R < f >i]、Hj[fi]Respectively represent samples R and HiIn characteristic fiSample values of the following. Wherein the score is F (F)i) The higher or weight ω (f)i) The larger the size, the stronger the ability of the feature to distinguish different categories; and conversely, the characteristic is insensitive to fault classification and can be eliminated.
S23, step S22The calculated characteristic weight values are sorted according to the weight values, and a characteristic weight threshold value w is presetρ(ii) a When feature fiWeight value of omega1(fi)、ω2(fi) Greater than a threshold value wρIf so, the feature is reserved, otherwise, the feature is removed as an unimportant feature; and respectively establishing low-dimensional feature subsets FS21 and FS2 by using the reserved features, and further establishing corresponding low-dimensional fault sample sets D21 and D22.
Preferably, the feature selection method in step S2 is a Fisher Score feature selection method and a Relief F feature selection method;
the dimensions of the low-dimensional feature subsets FS21, F22 are lower than the feature subset FS 1.
Preferably, the step S3 includes the following steps:
s31, solving the clustering centers of the fault sample sets D21 and D22 by adopting a fuzzy C-means clustering method, and taking the obtained clustering centers as a reference sequence X in improved grey correlation analysis01And X02;
S32, respectively carrying out normalization processing on the feature weight values corresponding to the low-dimensional feature subsets FS21 and FS22 to obtain a weight vector omega01(k) And ω02(k);
S33, based on the fault sample set D21 obtained in the step S31, the low-dimensional feature subset FS21 and the reference sequence X01Calculating the correlation coefficient of each characteristic point in the sample to be detected and the reference sequence; wherein, for a certain characteristic point k, the correlation coefficient xik-fisher scoreThe calculation formula of (a) is as follows:
in the formula, x0(k) Is a reference sequence X0The value at the kth feature; x is the number ofi(k) The value of the ith sequence to be tested at the kth characteristic is taken as the value of the ith sequence to be tested; the maximum and minimum values of the corresponding characteristic differences of the fault standard sequence x0 and the sequence xi to be detected are represented by delta max and delta min; deltaoi(k) Is the difference between the test sequence and the reference sequence; to be more accurateMore clearly describing the differences between sequences, the present invention uses the Mahalanobis Distance (MD); cov is a covariance matrix among different dimensional variables, rho is a resolution coefficient, and the indirect influence of the attention degree and the characteristics of the delta max on the correlation degree is reflected; in order to improve the diagnosis precision, a dynamic resolution coefficient is adopted, and the calculation method is shown as the following formula:
s34, based on the weight vector omega obtained in step S22 and step S331(fi) And correlation coefficient ξk-fisher scoreCalculating the correlation degree gamma between the sample to be measured and the reference sequencei-fisher scoer(ii) a Wherein, the correlation calculation formula is as follows:
wherein ξkThe correlation coefficient of a certain characteristic point k; w is a01Solving the retained and normalized characteristic weight value based on a Fisher Score method; n is the number of key features; m is the total number of fault types in the reference sequence;
s35, based on the fault sample set D22 obtained in the step S31, the low-dimensional feature subset FS22 and the reference sequence X02Calculating the correlation coefficient of each characteristic point in the sample to be detected and the reference sequence; the calculation steps can be repeated S33 and S34, and then the second group gamma is obtainedi-ReliefF。
S36, based on the magnitude γ of the relevance value obtained in step S34 and step S35 respectivelyi-Fisher ScoerAnd gammai-ReliefFAnd calculating the average relevance value, wherein the calculation formula is as follows:
according toCalculating a value, and outputting the fault type corresponding to the maximum correlation value as a diagnosis result of the sample to be detected.
The invention has the following beneficial effects:
by adopting the dynamic identification coefficient, the diagnosis method of the oil-immersed transformer has the advantages of simple structure, high calculation efficiency, no redundancy characteristic, high diagnosis precision, high diagnosis efficiency and the like, so that the transformer fault diagnosis, maintenance and overhaul work can be carried out more effectively and reliably.
Of course, it is not necessary for any product in which the invention is practiced to achieve all of the above-described advantages at the same time.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
S1, collecting and sorting multi-type fault samples of the oil-immersed transformer, and constructing a feature set FS1 according to the conventional gas component values dissolved in oil and the combination of the conventional gas component values; a fault sample set D1 is established together based on a gas sample dissolved in oil of the oil-immersed transformer and the characteristic set FS 1;
s2, calculating feature weights of a feature set FS1 by adopting two different feature selection methods, sorting the feature weights according to size values, reserving features with weight values meeting the requirements of preset thresholds, establishing low-dimensional feature subsets FS21 and FS22, and further establishing low-dimensional fault sample sets D21 and D22;
s3, solving clustering centers C1 and C2 of fault sample sets D21 and D22 by using a fuzzy clustering algorithm, and respectively using the obtained clustering centers as reference sequences in improved gray correlation analysis; normalization processing is carried out on the feature weights corresponding to the low-dimensional features FS21 and FS22, and then the correlation coefficient and the correlation degree of the sample to be detected are calculated respectively according to the weight values and the reference sequence; and calculating the average value of the two groups of correlation degrees, and outputting the fault type corresponding to the maximum value of the average correlation degrees as the diagnosis result of the sample to be detected.
In step S1, the feature set FS1 is characterized by 9 fault types of oil chromatogram samples, the oil chromatogram samples are composed of content features of 5 feature gases, and the 5 feature gases are methane (CH4), hydrogen (H2), ethane (C2H6), ethylene (C2H4), and acetylene (C2H 2);
wherein, the content characteristic of each characteristic gas consists of three characteristics of absolute content, relative content and mutual ratio;
among them, 9 kinds of fault types include No Fault (NF), Partial Discharge (PD), low energy discharge (LD), high energy discharge (HD), low temperature overheat (LT), medium temperature overheat (MT), high temperature overheat (HT), low energy discharge and overheat (LTD), and high energy discharge and overheat (HTD);
and extracting the contained characteristic values from the characteristic set FS1 so as to construct a fault sample set D1.
The following table is the feature content of the feature set FS 1:
TABLE 1 DFA-based FS1 feature set
Characteristic name | Feature content | Characteristic name | Feature content |
f1 | H2% | f13 | CH4/C2H2 |
f2 | CH4% | f14 | CH4/TH |
f3 | C2H6% | f15 | C2H6/C2H4 |
f4 | C2H4% | f16 | C2H6/C2H2 |
f5 | C2H2% | f17 | C2H6/TH |
f6 | H2/CH4 | f18 | C2H4/C2H2 |
f7 | H2/C2H6 | f19 | C2H4/TH |
f8 | H2/C2H4 | f20 | C2H2/TH |
f9 | H2/C2H2 | f21 | C2H2/TD |
f10 | H2/TH | f22 | C2H4/TD |
f11 | CH4/C2H6 | f23 | CH4/TD |
f12 | CH4/C2H4 | f24 | TS |
Wherein, H2% is H2/(H2+ CH4+ C2H6+ C2H4+ C2H2), and the rest is similar; TH ═ CH4+ C2H6+ C2H4+ C2H 2; TD ═ C2H4+ C2H2+ CH 4; TS ═ H2+ CH4+ C2H6+ C2H4+ C2H 2; all by mass.
In step S2, the method includes the following steps:
s21, aiming at the established fault sample set D1, 80% of fault samples are used for establishing a fault diagnosis model, and 20% of fault samples are used for testing the diagnosis performance of the model; in order to eliminate dimension inconsistency among different samples, the data of the original fault data set is normalized, and the extreme value normalization formula is as follows:
wherein: x is the number ofikIs the kth feature data in the ith sample, and xik maxAnd xik minRespectively representing the maximum value and the minimum value of the kth characteristic;
s22, calculating a feature weight value in the feature set FS1 by using a feature selection method (the Fisher Score method and the Relief method are respectively used in the invention) based on the feature set FS1 normalized in the step S21, wherein the larger the weight value is, the more important the feature is, otherwise, the feature is insensitive to fault classification, and the feature can be removed.
Wherein, calculating the characteristic weight value w by using a Fisher Scoer method1(fi) The calculation method of (2) is as follows:
assume that the dataset contains class c samples, nkIs the total number of class k samples, fj,iIs the jth sample value in the ith feature; u. ofiDenotes the ith characteristiciAverage value of (a), uikRepresents the mean value, x, in the k class under the ith featurei,kjFor the jth sample in the kth class sample in the ith feature fiThe value of the i-th feature is the score F (F)i) And weight w1(fi) Can be calculated as follows:
wherein, the characteristic weight value w is calculated by using a Relief F method2(fi) The calculation method of (2) is as follows:
firstly, setting all feature weights as 0; then, k neighbor sample sets of samples belonging to the same class as the random sample R and k samples not belonging to the same class as R are searched from the data set D, and each characteristic weight w is updated2(fi);
Wherein, diff (f)i,Ri,Hj) Denotes samples R and HjIn characteristic fiDifference of above, HjIs the jth nearest neighbor sample of the same class as R. C is the total number of classes, class (R), which is different from sample Ri) Represents the class of the sample R, P (C) represents the probability of class C, diff (f)i,Ri,Mj(C) Are sample R and sample M)j(C) In characteristic fiDifference of above, Mj(C) Representing the jth nearest neighbor sample to sample R in class C. Wherein diff (f)i,Ri,Hj) The calculation can be as follows:
in the above formula, R < f >i]、Hj[fi]Respectively represent samples R and HiIn characteristic fiSample values of the following. Wherein the score is F (F)i) The higher or weight ω (f)i) The larger the size, the stronger the ability of the feature to distinguish different categories; and conversely, the characteristic is insensitive to fault classification and can be eliminated.
S23, based on the characteristic weight value calculated in step S22Sorting according to the weight value, and presetting a characteristic weight threshold value wρ(ii) a When feature fiWeight value of omega1(fi)、ω2(fi) Greater than a threshold value wρIf so, the feature is reserved, otherwise, the feature is removed as an unimportant feature; and establishing low-dimensional feature subsets FS21 and FS2 for the reserved features, and further establishing low-dimensional fault sample sets D21 and D22.
In step S3, the method includes the following steps:
s31, solving the clustering centers of the fault sample sets D21 and D22 by adopting a fuzzy C-means clustering method, and taking the obtained clustering centers as a reference sequence X in improved grey correlation analysis01And X02;
S32, respectively carrying out normalization processing on the feature weight values corresponding to the low-dimensional feature subsets FS21 and FS22 to obtain weight vectors omega 01(k) and omega 02(k), wherein the calculation formula is as follows;
wherein p and q respectively represent the weight preset value omega which is calculated and ordered by utilizing a Fisher Socre method and a Relief F methodρThe number of features of (2).
S33, based on the fault sample set D21 obtained in the step S31, the low-dimensional feature subset FS21 and the reference sequence X01Calculating the correlation coefficient of each characteristic point in the sample to be detected and the reference sequence; wherein, for a certain characteristic point k, the correlation coefficient xik-fisher scoreThe calculation formula of (a) is as follows:
in the formula, x0(k) Is a reference sequence X0The value at the kth feature; x is the number ofi(k) The value of the ith sequence to be tested at the kth characteristic is taken as the value of the ith sequence to be tested; the maximum and minimum values of the corresponding characteristic differences of the fault standard sequence x0 and the sequence xi to be detected are represented by delta max and delta min; deltaoi(k) Is the difference between the test sequence and the reference sequence; in order to describe the difference between sequences more accurately and obviously, the Mahalanobis Distance (MD) is adopted in the invention; cov is a covariance matrix among different dimensional variables, rho is a resolution coefficient, and the indirect influence of the attention degree and the characteristics of the delta max on the correlation degree is reflected; in order to improve the diagnosis precision, a dynamic resolution coefficient is adopted, and the calculation method is shown as the following formula:
s34, based on the weight vector omega obtained in step S22 and step S3301(fi) And correlation coefficient ξk-fisher scoreCalculating the correlation degree gamma between the sample to be measured and the reference sequencei-fisher scoer(ii) a Wherein, the correlation calculation formula is as follows:
wherein ξkThe correlation coefficient of a certain characteristic point k; w is akThe characteristic weight value is normalized; n is the number of key features; m is the fault type in the reference sequence;
s35, based on the fault sample set D22 obtained in the step S31, the low-dimensional feature subset FS22 and the reference sequence X02Calculating the correlation coefficient of each characteristic point in the sample to be detected and the reference sequence; the calculation steps can be repeated S33 and S34, and then the second group gamma is obtainedi-ReliefF。
S36, based on the magnitude γ of the relevance value obtained in step S34 and step S35 respectivelyi-Fisher ScoerAnd gammai-ReliefFAnd calculating the average relevance value, wherein the calculation formula is as follows:
according toCalculating a value, and outputting the fault type corresponding to the maximum correlation value as a diagnosis result of the sample to be detected.
According to the oil-immersed transformer fault diagnosis method based on the improved grey correlation analysis, provided by the invention, the diagnosed fault characteristic information is more comprehensive, the selected characteristic subset has better identification capability and lower input characteristic dimension, and the problem of difficulty in determining the characteristic weight is solved; and by adopting the dynamic identification coefficient, the whole set of diagnosis method has the advantages of simple structure, high calculation efficiency, good diagnosis performance and the like, so that the transformer fault diagnosis, maintenance and overhaul work can be carried out more effectively and reliably.
In the description herein, references to the description of "one embodiment," "an example," "a specific example" or the like are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The preferred embodiments of the invention disclosed above are intended to be illustrative only. The preferred embodiments are not intended to be exhaustive or to limit the invention to the precise embodiments disclosed. Obviously, many modifications and variations are possible in light of the above teaching. The embodiments were chosen and described in order to best explain the principles of the invention and the practical application, to thereby enable others skilled in the art to best utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims (5)
1. An oil-immersed transformer fault diagnosis method based on improved grey correlation analysis is characterized by comprising the following steps:
s1, collecting and sorting multi-type fault samples of the oil-immersed transformer, and constructing a feature set FS1 according to the conventional gas component values dissolved in oil and the combination of the conventional gas component values; a fault sample set D1 is established together based on a gas sample dissolved in oil of the oil-immersed transformer and the characteristic set FS 1;
s2, calculating feature weights of a feature set FS1 by adopting two different feature selection methods, sorting the feature weights according to size values, reserving features with weight values meeting the requirements of preset thresholds, establishing low-dimensional feature subsets FS21 and FS22, and further establishing low-dimensional fault sample sets D21 and D22;
s3, solving clustering centers C1 and C2 of fault sample sets D21 and D22 by using a fuzzy clustering algorithm, and respectively using the obtained clustering centers as reference sequences in improved gray correlation analysis; normalization processing is carried out on the feature weights corresponding to the low-dimensional features FS21 and FS22, and then the correlation coefficient and the correlation degree of the sample to be detected are calculated respectively according to the weight values and the reference sequence; and calculating the average value of the two groups of correlation degrees, and outputting the fault type corresponding to the maximum value of the average correlation degrees as the diagnosis result of the sample to be detected.
2. The oil-filled transformer fault diagnosis method based on improved gray correlation analysis according to claim 1, wherein in step S1, the feature set FS1 is characterized by 9 fault types of oil chromatogram samples, the oil chromatogram samples are composed of content features of 5 feature gases, and the 5 feature gases are methane (CH4), hydrogen (H2), ethane (C2H6), ethylene (C2H4), and acetylene (C2H 2);
wherein, the content characteristic of each characteristic gas consists of three characteristics of absolute content, relative content and mutual ratio;
among them, 9 kinds of fault types include No Fault (NF), Partial Discharge (PD), low energy discharge (LD), high energy discharge (HD), low temperature overheat (LT), medium temperature overheat (MT), high temperature overheat (HT), low energy discharge and overheat (LTD), and high energy discharge and overheat (HTD);
and extracting the contained characteristic values from the characteristic set FS1 so as to construct a fault sample set D1.
3. The oil-filled transformer fault diagnosis method based on improved grey correlation analysis according to claim 2, wherein the step S2 includes the following steps:
s21, aiming at the established fault sample set D1, 80% of fault samples are used for establishing a fault diagnosis model, and 20% of fault samples are used for testing the diagnosis performance of the model; in order to eliminate dimension inconsistency among different samples, the data of the original fault data set is normalized, and the extreme value normalization formula is as follows:
wherein: x is the number ofikIs the kth feature data in the ith sample, and xik maxAnd xik minRespectively representing the maximum value and the minimum value of the kth characteristic;
s22, calculating a feature weight value in the feature set FS1 by using a feature selection method (the Fisher Score method and the Relief F method are respectively used in the invention) based on the feature set FS1 normalized in the step S21, wherein the larger the weight value is, the more important the feature is, otherwise, the feature is insensitive to fault classification;
wherein, calculating the characteristic weight value w by using a Fisher Scoer method1(fi) The calculation method of (2) is as follows:
assume that the dataset contains class c samples, nkIs the total number of class k samples, fj,iIs the jth sample value in the ith feature; u. ofiDenotes the ith characteristiciAverage value of (a), uikRepresents the mean value, x, in the k class under the ith featurei,kjFor the jth sample in the kth class sample in the ith feature fiThe value of the i-th feature is the score F (F)i) And weight w1(fi) Can be calculated as follows:
wherein, the characteristic weight value w is calculated by using a Relief F method2(fi) The calculation method of (2) is as follows:
firstly, setting all feature weights as 0; then, k neighbor sample sets of samples belonging to the same class as the random sample R and k samples not belonging to the same class as R are searched from the data set D, and each characteristic weight w is updated2(fi);
Wherein, diff (f)i,Ri,Hj) Denotes samples R and HjIn characteristic fiDifference of above, HjIs the jth nearest neighbor sample of the same class as R. C is the total number of classes, class (R), which is different from sample Ri) Represents the class of the sample R, P (C) represents the probability of class C, diff (fi, R)i,Mj(C) Are sample R and sample M)j(C) In characteristic fiDifference of above, Mj(C) Representing the jth nearest neighbor sample to sample R in class C. Wherein diff (f)i,Ri,Hj) The calculation can be as follows:
in the above formula, R < f >i]、Hj[fi]Respectively represent samples R and HiIn characteristic fiSample values of the following. Wherein the score is F (F)i) The higher or weight ω (f)i) The larger the size, the stronger the ability of the feature to distinguish different categories; otherwise, the characteristic is not sensitive to fault classification;
s23, sorting according to the weight value based on the characteristic weight value calculated in the step S22, and presetting a characteristic weight threshold wρ(ii) a When feature fiWeight value of omega1(fi)、ω2(fi) Greater than a threshold value wρIf so, the feature is reserved, otherwise, the feature is removed as an unimportant feature; and respectively establishing low-dimensional feature subsets FS21 and FS2 by using the reserved features, and further establishing corresponding low-dimensional fault sample sets D21 and D22.
4. The oil-filled transformer fault diagnosis method based on the improved grey correlation analysis according to claim 3, wherein the feature selection method in the step S2 is a Fisher Score feature selection method and a Relief feature selection method;
the dimensions of the low-dimensional feature subsets FS21, F22 are lower than the feature subset FS 1.
5. The oil-filled transformer fault diagnosis method based on improved grey correlation analysis according to claim 4, wherein the step S3 includes the following steps:
s31, solving the clustering centers of the fault sample sets D21 and D22 by adopting a fuzzy C-means clustering method, and taking the obtained clustering centers as a reference sequence X in improved grey correlation analysis01And X02;
S32, respectively carrying out normalization processing on the feature weight values corresponding to the low-dimensional feature subsets FS21 and FS22 to obtain a weight vector omega01(k) And ω02(k);
S33, based on the fault sample set D21 obtained in the step S31, the low-dimensional feature subset FS21 and the reference sequence X01Calculating the correlation coefficient of each characteristic point in the sample to be detected and the reference sequence; whereinFor a certain feature point k, the correlation coefficient ξk-fisher scoreThe calculation formula of (a) is as follows:
in the formula, x0(k) Is a reference sequence X0The value at the kth feature; x is the number ofi(k) The value of the ith sequence to be tested at the kth characteristic is taken as the value of the ith sequence to be tested; the maximum and minimum values of the corresponding characteristic differences of the fault standard sequence x0 and the sequence xi to be detected are represented by delta max and delta min; deltaoi(k) Is the difference between the test sequence and the reference sequence; in order to describe the difference between sequences more accurately and obviously, the Mahalanobis Distance (MD) is adopted in the invention; cov is a covariance matrix among different dimensional variables, rho is a resolution coefficient, and the indirect influence of the attention degree and the characteristics of the delta max on the correlation degree is reflected; in order to improve the diagnosis precision, a dynamic resolution coefficient is adopted, and the calculation method is shown as the following formula:
s34, based on the weight vector omega obtained in step S22 and step S3301(fi) And correlation coefficient ξk-fisher scoreCalculating the correlation degree gamma between the sample to be measured and the reference sequencei-fisherscoer(ii) a Wherein, the correlation calculation formula is as follows:
wherein ξkThe correlation coefficient of a certain characteristic point k; w is a01Solving and normalizing the processed feature weights based on the Fisher Score methodA weight value; n is the number of key features; m is the total number of fault types in the reference sequence;
s35, based on the fault sample set D22 obtained in the step S31, the low-dimensional feature subset FS22 and the reference sequence X02Calculating the correlation coefficient of each characteristic point in the sample to be detected and the reference sequence; the calculation steps can be repeated S33 and S34, and then the second group gamma is obtainedi-ReliefF;
S36, based on the magnitude γ of the relevance value obtained in step S34 and step S35 respectivelyi-Fisher ScoerAnd gammai-ReliefFAnd calculating the average relevance value, wherein the calculation formula is as follows:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110037452.7A CN112881827B (en) | 2021-01-12 | 2021-01-12 | Oil-immersed transformer fault diagnosis method based on improved grey correlation analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110037452.7A CN112881827B (en) | 2021-01-12 | 2021-01-12 | Oil-immersed transformer fault diagnosis method based on improved grey correlation analysis |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112881827A true CN112881827A (en) | 2021-06-01 |
CN112881827B CN112881827B (en) | 2022-11-25 |
Family
ID=76044489
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110037452.7A Active CN112881827B (en) | 2021-01-12 | 2021-01-12 | Oil-immersed transformer fault diagnosis method based on improved grey correlation analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112881827B (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103064019A (en) * | 2013-01-04 | 2013-04-24 | 河海大学常州校区 | High-voltage circuit interrupter fault diagnosis method based on grey correlation fuzzy clustering |
CN104156568A (en) * | 2014-07-22 | 2014-11-19 | 国家电网公司 | Transformer fault diagnosis method on basis of weighted gray correlation and fuzzy clustering |
CN107169514A (en) * | 2017-05-05 | 2017-09-15 | 清华大学 | The method for building up of diagnosing fault of power transformer model |
CN108319789A (en) * | 2018-02-07 | 2018-07-24 | 海南电网有限责任公司电力科学研究院 | In conjunction with transformer methods of risk assessment, device, equipment and the medium of abnormal failure |
CN109164343A (en) * | 2018-08-30 | 2019-01-08 | 西华大学 | Quantify the Diagnosis Method of Transformer Faults with weighting KNN based on characteristic information |
CN109492857A (en) * | 2018-09-18 | 2019-03-19 | 中国电力科学研究院有限公司 | A kind of distribution network failure risk class prediction technique and device |
US20190155869A1 (en) * | 2017-11-20 | 2019-05-23 | Hongfujin Precision Electronics (Tianjin) Co.,Ltd. | Device and method for managing risk in area |
CN110595778A (en) * | 2019-06-10 | 2019-12-20 | 内蒙古工业大学 | Wind turbine generator bearing fault diagnosis method based on MMF and IGRA |
CN111369168A (en) * | 2020-03-18 | 2020-07-03 | 武汉大学 | Associated feature selection method suitable for multiple regulation and control operation scenes of power grid |
CN111382943A (en) * | 2020-03-09 | 2020-07-07 | 西北工业大学 | Fault diagnosis and evaluation method based on weighted grey correlation analysis |
-
2021
- 2021-01-12 CN CN202110037452.7A patent/CN112881827B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103064019A (en) * | 2013-01-04 | 2013-04-24 | 河海大学常州校区 | High-voltage circuit interrupter fault diagnosis method based on grey correlation fuzzy clustering |
CN104156568A (en) * | 2014-07-22 | 2014-11-19 | 国家电网公司 | Transformer fault diagnosis method on basis of weighted gray correlation and fuzzy clustering |
CN107169514A (en) * | 2017-05-05 | 2017-09-15 | 清华大学 | The method for building up of diagnosing fault of power transformer model |
US20190155869A1 (en) * | 2017-11-20 | 2019-05-23 | Hongfujin Precision Electronics (Tianjin) Co.,Ltd. | Device and method for managing risk in area |
CN108319789A (en) * | 2018-02-07 | 2018-07-24 | 海南电网有限责任公司电力科学研究院 | In conjunction with transformer methods of risk assessment, device, equipment and the medium of abnormal failure |
CN109164343A (en) * | 2018-08-30 | 2019-01-08 | 西华大学 | Quantify the Diagnosis Method of Transformer Faults with weighting KNN based on characteristic information |
CN109492857A (en) * | 2018-09-18 | 2019-03-19 | 中国电力科学研究院有限公司 | A kind of distribution network failure risk class prediction technique and device |
CN110595778A (en) * | 2019-06-10 | 2019-12-20 | 内蒙古工业大学 | Wind turbine generator bearing fault diagnosis method based on MMF and IGRA |
CN111382943A (en) * | 2020-03-09 | 2020-07-07 | 西北工业大学 | Fault diagnosis and evaluation method based on weighted grey correlation analysis |
CN111369168A (en) * | 2020-03-18 | 2020-07-03 | 武汉大学 | Associated feature selection method suitable for multiple regulation and control operation scenes of power grid |
Non-Patent Citations (4)
Title |
---|
JIA-YUAN HU 等: "Prediction Method of Breakdown Voltage of Transformer Oil Based on Fuzzy Clustering Theory and BP Neural Network", 《2010 INTERNATIONAL CONFERENCE ON ELECTRICAL AND CONTROL ENGINEERING》 * |
唐勇波: "数据驱动的大型电力变压器故障诊断和预测研究", 《中国优秀博硕士学位论文全文数据库(博士)工程科技Ⅱ辑》 * |
李邦彦 等: "基于Relief F 和灰色关联分析的高压断路器机械故障诊断", 《高压电器》 * |
葛乐 等: "基于改进熵权法和灰色关联分析的变压器故障诊断", 《电测与仪表》 * |
Also Published As
Publication number | Publication date |
---|---|
CN112881827B (en) | 2022-11-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111768082A (en) | Power equipment state evaluation method based on big data analysis | |
Li et al. | Fault diagnosis of power transformers with membership degree | |
CN110609200B (en) | Power distribution network earth fault protection method based on fuzzy metric fusion criterion | |
CN107132310B (en) | Transformer equipment health status method of discrimination based on gauss hybrid models | |
CN104535865A (en) | Comprehensive diagnosing method for operation troubles of power transformer based on multiple parameters | |
CN110297841B (en) | Transformer fault diagnosis and rapid indexing method and system | |
CN105631771A (en) | Variable weight function-based multi-parameter cable state evaluation method | |
CN110542819A (en) | transformer fault type diagnosis method based on semi-supervised DBNC | |
CN111401749A (en) | Dynamic safety assessment method based on random forest and extreme learning regression | |
CN110879373B (en) | Oil-immersed transformer fault diagnosis method with neural network and decision fusion | |
CN113342784A (en) | Database design method for risk assessment of main transformer equipment of power grid | |
CN112308425A (en) | Method for constructing distribution transformer health evaluation index system | |
McGrail et al. | Data mining techniques to assess the condition of high voltage electrical plant | |
CN110689068A (en) | Transformer fault type diagnosis method based on semi-supervised SVM | |
CN113205125A (en) | XGboost-based extra-high voltage converter valve operation state evaluation method | |
CN116595459A (en) | Pollution flashover early warning method and system based on electric field signals | |
CN109033498A (en) | Winding deformation recognition methods based on transmission function feature principal component and neural network | |
CN110059773B (en) | Transformer fault composite diagnosis method | |
CN114372693B (en) | Transformer fault diagnosis method based on cloud model and improved DS evidence theory | |
CN111652478A (en) | Electric power system voltage stability evaluation misclassification constraint method based on umbrella algorithm | |
CN114065495A (en) | Transformer state evaluation method based on reverse cloud combination weight and fuzzy closeness | |
CN112881827B (en) | Oil-immersed transformer fault diagnosis method based on improved grey correlation analysis | |
CN109784777B (en) | Power grid equipment state evaluation method based on time sequence information fragment cloud similarity measurement | |
CN111585277A (en) | Power system dynamic security assessment method based on hybrid integration model | |
CN113961708B (en) | Power equipment fault tracing method based on multi-level graph convolutional network |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |