RESEARCH ON TRANSFORMER CONDITION EVALUATION METHOD BASED ON ASSOCIATION RULE SET PAIR ANALYSIS THEORY

Combining the advantages of set pair analysis and association rules, This paper proposes a transformer condition evaluation based on association rule with set pair analysis theory. In this paper, by analyzing the correlation between the various fault symptoms of transformer, a set of fault types is obtained. At the same time, this paper introduces variable weight formula based on the support degree and confidence degree of association rules, and finally the weight coefficients of fault types and fault symptoms are obtained. By comparing and calculating the support and confidence of association rules, while introducing variable weight formulas, the weight coefficients of fault types and fault symptoms are obtained. it effectively avoid the subjectivity of expert opinions or experiences. Based on the scalability of set pair analysis, a 5-element connection degree is adopted to improve the accuracy of handling uncertain factors in transformer fault diagnosis.


INTRODUCTION
Combining the advantages of set pair analysis theory and the advantages of association rules for association evaluation.This paper focuses on 220kV transformer, a method of equipment condition evaluation based on set pair analysis theory of association rules is proposed on the basis of fully considering all relevant fault factors.
Power equipment data is the basis for evaluating equipment status [1].This method mainly analyzes the correlation between transformer-fault symptoms, it normalizes the naming method about fault types according to the manifestation of fault symptoms, which calculate the confidence degree and support degree between fault symptoms in association rule theory.In the end, it introduces variable weight coefficient to obtain the similarity difference, so it inverse evaluation matrix of equipment status which fulled of related fault types and fault symptoms.Some chinese scholars provides a promising novel method for rapid on-site inspection of power equipment [2].For the newly collected data, Shiqi Zhang proposed a method that the incremental learning of new fault modes is achieved by automatic feature extraction of the ResNet and the node expansion of the BLS.
The effectiveness of the proposed method is verified by data-driven from fault diagnosis [3].
Convolutional neural networks (CNNs) have promoted the development of diagnosis focus attention to insulation defect for gas-insulated switchgear (GIS) attribute to their excellent feature extraction and classification capabilities [4].
In terms of transformer status evaluation, known literature has formed a separate set of pairs for each test indicator which calculated by the connection, but it is not considered that the internal connections between various indicator quantities which determine the type of fault that occurred in the transformer.The Analytic Hierarchy Process (AHP) are used to construct a fuzzy judgment matrix which exist transformer set analysis, it cannot completely eliminate the subjectivity of expert systems [5].
In view of this, this article attempts to organically combine set pair analysis and association rules, and apply them to power transformer fault diagnosis.This method establishes a set of fault types by analyzing the correlation between various fault symptom parameters during transformer operation.By calculating the support and confidence of association rules, while introducing variable weight formulas, the weight coefficients of fault types and fault symptoms are obtained.It effectively avoid the subjectivity of expert opinions or experiences.Based DIAGNOSTYKA, Vol. 24, No. 4 (2023) Shanyuan W, Donglin C, Zhonghong C, Menglong B…: Research on transformer condition evaluation… 2 on the scalability of set pair analysis, a 5-element connection degree is adopted to improve the accuracy of handling uncertain factors in transformer fault diagnosis.
The association rule method is essentially a method based on probability reasoning, which mainly describes the random uncertainty of the research object under certain conditions (such as infinite number of experiments).The evaluation of the operation status of power transformers is not only related to randomly uncertain factors, but also involve various factors such as fuzzy uncertainty caused by incomplete information.Set pair analysis (SPA) theory is a new mathematical theory for dealing with uncertainty.Its basic idea is to include certainty and uncertainty in the same system [6].It study the certainty and uncertainty of objects from three aspects: identity, difference, opposite, in order to comprehensively depict the interrelationships, influences, and transformations between objects.
In view of this, this article attempts to organically combine set pair analysis and association rules, it aim to apply them to power transformer fault diagnosis [7].The basic idea is to classify firstly and manage the typical state variables(fault types and symptoms) that can reflect the operating conditions of transformers in the big data pool for detection and monitoring.
Then, the association rule method is used to explore the correlation between fault types and fault symptoms, and its weight coefficient is determined to minimize the impact of subjective opinions of the expert system on the accuracy of the weight.Furthermore, the relative deterioration degree and membership degree in fuzzy theory are introduced to construct an evaluation matrix [8].
Based on set pair analysis theory, the multiple connection of each fault symptom with fault type are sequentially determined [9].Next, the weight coefficients determined by association rules are combined to obtain the overall operating status of the transformer.Finally, transformer health status assessment and fault diagnosis are carried out [10].As shown in Figure 1, the intelligent dispatching control system collects the status information of the transformer, which obtains Fault type according to the alarm information of the equipment, and then we can find the reason according to the Fault type.Some machine design also studies this type of method [11].
Finally, the countermeasures are taken on this basis.Considering that the calculation of confidence is only associated with the data of the previous period, the calculation results can not be quickly reflected according to the real-time data, this chapter analyzed how to optimize the analysis model based on the association rule set.In some paper, ti also be called the impedance behavior [12].
The principle is to optimize the process performing secondary confidence based on the results of the calculation of constant weight.Its variable weight formula adjusts the coefficient according to the scoring results of each fault type, and the coefficient is adjusted in the range of [0,1], When the coefficient is 0, the speed of changing weight is relatively gentle, and when the coefficient is 1, the speed of changing weight is relatively radical.
Variable weight involves historical data and realtime dynamic data, and its variable weight coefficient changes with the two types of data.

CONDITION EVALUATION SYSTEM OF TRANSFORMER
Combining the advantages in set pair analysis theory for uncertainty evaluation and the advantages in association rules for association evaluation, a method of equipment condition evaluation based on set pair analysis theory of association rules is proposed on the basis of fully considering all relevant fault factors.Due to the variety of equipment, the fault symptoms and types of each equipment are different, so this paper focuses on 220kV transformer, it could extend this method to the state assessment of other equipment.
This method mainly analyzes the correlation between transformer fault symptoms, which standardize the naming method of fault types according to the manifestation of fault symptoms.In the end, it could calculate the confidence degree and support degree of fault symptoms in association rule theory.It introduces variable weight coefficient to obtain the evaluation matrix of equipment status from fault types and fault symptoms, which avoid the subjectivity and fuzziness of experts in setting weights.In view of the accuracy of the evaluation formula, the multivariate relation degree theory of set pair analysis theory is introduced to improve the accuracy of uncertainty evaluation under equipment's condition above fault diagnosis.
Set pair analysis theory is an analysis method research based on system engineering and mathematical theory proposed by Zhao Keqin, a famous mathematician in China.Its method is characterized by clear logic, high calculation Shanyuan W, Donglin C, Zhonghong C, Menglong B…: Research on transformer condition evaluation… accuracy, and intuitive evaluation method.It is often used to deal with uncertainty problems caused by fuzzy objectives and incomplete information.Its theory has great advantages in more complex evaluation systems.Now, with the improvement of the set pair analysis theory itself, it is often used in the research of power equipment risk assessment, basin water resources assessment and other important scientific research fields.
Set pair analysis (SPA) theory is often used to deal with system engineering problems of uncertainty evaluation.Its main idea combined certainty evaluation and uncertainty evaluation in the same analysis framework, it study the certainty and uncertainty of the target for engineering issues from three measurement directions about identity, difference and opposition.In the end, it further establish the relationship between measurement directions.At present, many experts and scholars have applied set pair analysis theory to power grid planning.Under the research background of an engineering problem G, a set pair D is composed of two related sets, which is defined as D=(A, B).The engineering characteristics of the unit whose set belongs are discussed and subdivided into T characteristics.Among the T characteristics, set A and set B both contain X characteristics, and set A and set B are independent of each other in Z characteristics and only one set owns them, while the rest belong to the characteristics that set A and set B do not have.In conclusion, in the context of engineering problem G, the correlation function is: In formula(1):u(A, B) is the degree of connection between two related sets: A and B, and the value range of u(A, B) is [-1, 1].In this section, the average value takes method of the different coefficient of the set pair analysis theory, which is combined in the establishment of the equipment-risk assessment model.Because the proportional value is more accurate for the assessment results in the hierarchical process of the connection degree, that is, i is divided into a certain range according to a certain proportion.In this paper, i is divided in an equal proportion way according to the equipment maintenance strategy.We correspond a, b and c respectively into the degree of identity, difference and opposition of engineering problem G, where a=X/T, b=Z/T, c=(T-X-Z)/T;I is the difference coefficient, and generally i is between [-1, 1]; When j is the coefficient of opposites, j is usually taken as -1.
The linkage degree expression in formula (1) divides the precision into three levels, which is called ternary linkage degree.In the application of many engineering problems, the ternary relation degree has sufficient accuracy, but for other systems (fault diagnosis of power equipment) that require high-precision evaluation values [13] , such accuracy still cannot meet the requirements.Therefore, according to the extensibility of set pair analysis, we have made scalability for b in the linkage degree function (1).According to the actual situation, b is set up according to specific levels, and relevant scientific research experts name the expanded connection degree function as the multivariate connection degree function, as shown in formula (2): In this section, formula(2) is named as the Lelement connection degree function, R=[a, b1, . . ., bL-2, c] is named as the same-difference inverse evaluation matrix, and E=[1, i1, . . ., iL-2, j] is named as the same-difference inverse coefficient matrix.
Association rule technology is a kind of data mining theory, which was first used to study the habit correlation of customer's purchase behavior at this early stage.This method can be used to study the correlation between different items in the same event.The data representing various indicators provide a favorable data basis for transformer condition assessment.These data bases which have polymorphism characteristics are often heterogeneous.Then, how to mine the standard data that can effectively and comprehensively express the equipment operation status from the above massive information assessment system is the prerequisite for the research of equipment risk assessment.
The status of transformer can be classified as normal, attention, minor, abnormal, serious, etc.
Its status is related to the fault type of equipment.The damage degree of fault type to equipment is calculated from the support and confidence of historical data.This chapter divides the fault type into nine states, and the fault type can be calculated from 24 fault symptoms.In the judgment of fault type, the corresponding fault symptoms are taken according to the support degree and it is evaluated according to the rules of the same difference opposite evaluation matrix.
According to the related concepts of association rules, this paper defines association rules as follows: Definition 1:If an association rule set is D, it is named as a transaction database, and the database D contains N-subset transactions, then the association rule set D can be defined as: D={d1, d2, . . ., dN}, and d is a subset of the association rule set.
Definition 2: In subset d, each subset consists of multiple events.At this time, subset d can be defined as:d={ψ1, ψ2, …, ψM}, there are M transaction items in the subset.Therefore, there are a total of L transaction items in association rule set D, The set of transaction items contained in association rule set D is:d={ψ1, ψ2, …, ψL}。 Definition 3: We define the set containing k transaction items as dk subset, and name it k itemset, and then define the frequency of k-itemset in transaction database as f(dk).The proportion of f(dk) in transaction of database which are from N subsets is called the support degree of k itemset, as shown in Formula (3): In this paper, we define the minimum support as supportmin, and we use this as a screening basis to obtain a k-item set that is greater than the minimum support [14] .
Definition 4: Association rules are represented by A and B, which are transaction subsets.A∈d={ ψ1, ψ2, …, ψL}, B∈d={ψ1, ψ2, …, ψL}, A∩B is an empty set, and defines A as the premise.If B is the subsequent conclusion, so it can represent A→B.
Definition 5: A∪B is defined as a transaction set containing transaction subsets A and B. On the basis of definition 4, the probability when association rule set D contains transaction subset A, which define it as confidence(A→B), as shown in formula ( 4): In association rules, support and confidence represent the validity and certainty of the related transaction set respectively.When the support of kitem set in definition 3 is higher, it represents the closer correlation between the front and back items.The higher the confidence level between transaction A-subsets and B-subsets in definition 5, the more stable the relationship between transaction subsets.
There are many types of faults in transformers, it is difficult for a certain classification method to classify types.This article is mainly based on the "Guide for condition evaluation of Oil -immersed power transformers (reactors)", it is based on actual operating experience referring to the fault classification set in previous experience, the common fault types of transformers are finally divided into 9 types, as shown in Table 1 and Table 2.
It can be roughly divided into winding fault, core fault, current circuit overheating, etc.According to the insulation characteristics of equipment, faults can be divided into insulation damp, arc discharge, insulation aging, insulation oil deterioration, etc; According to the circuit fault, the fault can be divided into partial discharge and oil flow discharge.The above fault types are shown in Table 1 In this chapter, the complex logic relationship is established according to the state information in the power system, and then the transformer fault is evaluated.However, some state information cannot be simply analyzed quantitatively, which brings certain challenges to the evaluation of transformers in this chapter.Although the amount of monitoring data collected is large, the value density is low.Therefore, it is necessary to select the most representative fault symptoms of transformers as specific indicators to evaluate the health of transformers.
The state information of power transformers is numerous.If all state information is considered, the fault diagnosis system will be extremely complex, and some state information is relatively vague and not suitable for quantitative description, which is not conducive for a evaluation of transformers to be comprehensive and accurate.To verify the correctness and effectiveness of the method proposed in this article, a 220kV oil-filled transformer(SFPS9-180000/220) was analyzed as a case in a substation where in Liaoning Province.
By reviewing the relevant information of this transformer [15], it was found that both the initial values of the 24 state variables in Table 2

S7
The content of H2

S8
Core grounding current S9 Insulation resistance of iron core S10 The content of C2H6

S11
The content of C2H4 The content of C2H2

S20
Gas content in oil

S21
The content of CH4

S22 Oil static current(neutral point) S23
The content of furfural

Degree of polymerization
As shown in Table 2, according to the electrical test of the transformer, the following indicators can be quantified:insulation resistance absorption ratio, polarization index, volume resistivity and other 11 characteristic indicators.it quantifies the following indexes according to oil chromatographic analysis: 7 characteristic indexes such as H2 content, C2H6 content and C2H4 content.According to the insulating oil test, the following indexes are quantified:dielectric loss of insulating oil, water Shanyuan W, Donglin C, Zhonghong C, Menglong B…: Research on transformer condition evaluation… content in oil, breakdown voltage of oil, etc.Finally, the structural characteristics are quantified according to the physical properties of the materials, and the quantitative index is the degree of polymerization of the paperboard.

Fig. 2. Correlation between Symptom set and Fault type
This article extracts 9 types of power transformer faults [17], and it also select 24 types of transformer variables of monitoring state as symptom values.The symptom quantities contained in the fault types have a cross relationship, as shown in Figure 2.
The evaluation values of each fault state of power transformers vary in dimension level.If the data is not processed in a standardized form, it will lead to big data devouring small data.Therefore, this paper introduces the relative degree to quantify the data in a standardized way, so that the data of weak dimensions also show the same importance.This section combined with the characteristics of power data acquisition is based on the relative deterioration degree, The normalized treatment method of fault symptoms is introduced, as shown in Formula (5): As shown in Table 3, this section introduces the corresponding connection function according to the relative deterioration of the transformer [18].In this paper, traditional power equipment is generally summarized as qualified or unqualified.This paper subdivides the equipment into five states according to the degree of loss: normal, attention, slight, abnormal and serious.The fault symptom is defined as Sm, n, which represent that a fault symptom-n is from fault type m.According to formula (3.5), the relative deterioration degree of each fault symptom is substituted into the corresponding formula to obtain the corresponding correlation degree values of each fault symptom under five status levels as r1(Sm, n), r2(Sm, n), r3(Sm, n), r4(Sm, n), r5(Sm, n).

The expression of Similarity Difference Inverse
Evaluation Matrix of Sm, n of fault symptom is Formula (6): Combining set pair analysis theory and IDR evaluation matrix, the fault diagnosis process of transformer is constructed as shown in Figure 3, and the work steps are as follows: Normal or not?

Overall operation condition of transformer Fault type
Output results Combining set pair analysis theory and IDR evaluation matrix, the fault diagnosis process of transformer is constructed as shown in Figure 3.The work steps are as follows: 1.According to the association rule theory, the historical fault data of transformers in actual operation are classified into fault types and fault symptom sets.We establish nine typical fault types were shown in Table 1 and 24 typical fault symptoms were shown in Table 2. 2. The measurement information of each part of the power transformer is collected according to step1.

Confidence Support Weight
Finally we obtain the relative deterioration degree and score value of the fault type according to confidence formula(4) and relative deterioration formula(5), 3. It collect the calculation information of the fault symptoms under the operation state of the transformer.Then, it determine the number of element from the contact degree function according to the transformer fault state evaluation information in Table 3.
4. The correlation coefficient of each fault type of the transformer and the overall operation condition of the equipment were determined according to the theory of contact degree, the formula of the same different opposite evaluation matrix of fault symptoms (6), and the same different opposite evaluation matrix of fault types (7).The process is shown in Figure 4. 5. Refer to Table 3, it determine whether the equipment operates normally according to the output results.Otherwise, it determine the status information and fault type of the equipment according to the specific coefficient and the diagnosis results.

EXAMPLE ANALYSIS BASED ON EVALUATION MODEL
This section collects the historical data of faults, and it counts the times exceeding the standard of historical data and fault cases [19].The times of exceeding the standard refer to the times of single symptom set in any state of the equipment.As for the support factor, the basic screening with 70%-support are described above(the symptoms related to the fault are included in the fault cases).In Appendix, the positions before the separator "/" represent the times of exceeding the standard and the positions after the separator "/" represent the degree of support respectively.According to the fault symptom collection data in Table A, we take the constant weight calculation process of fault case F1 as an example, and then extending to other fault types, it first calculate the fault case F1 according to the support formula (3) of the times of exceeding the standard.The process steps are as follows: 1.This paper counts the number of winding faults as 143, and it counts the number of cases where the fault symptom exceeds the standard when the winding fault occurs.According to According to the support formula (3), this paper calculates the fault symptom support in the first step: According to the above setting of 70% of the support threshold, the groups S7, S8, S15 and S16 with the highest correlation of the fault cases will be selected.3. The paper Supplement the previous step, we calculate and screen according to the support degree of formula (3) to obtain the four groups that calculated in this step.Then, we obtain the confidence degree of fault symptoms according to formula(4).The calculation process is as follows: C1, 1=confidence(S1, 7→F1) = P(F1| S1, 7) = f(F1∪S1,  4, the fault types are divided into: winding fault, core fault, current circuit overheating, insulation moisture, arc discharge, insulation aging, insulation oil degradation, partial discharge, oil flow discharge.The physical morphology and chemical reaction corresponding to the fault type are composed of 24 fault symptoms, which are in a cross form.The same fault symptoms may correspond to multiple fault types.The paper takes the arc discharge fault type as an example, the fault symptoms include H2-content, winding DC resistance difference, C2H2 content, and partial discharge, and these are based on the calculation of the analysis theory of the association rule set.The constant weight coefficients are 0. 2102, 0. 2579, 0. 2794 and 0. 2525 respectively.
In this paper, the collected information of a 220kV transformer in a 220kVsubstation in a city is taken as an example for checking calculation.According to the standard processing of fault symptoms whose test data is shown in Table 5 in this paper.The table contains the standard value, initial value and measured value of 24 groups of fault symptoms, and the standard value is often defined as the attention value in the project, which means that when the measured value is close to the standard value, the fault symptom may show abnormalities.
In addition, Table 5 shows the abnormal status of the real-time test data of power transformer on several fault symptoms which including H2 content, core grounding current, initial value difference of winding short-circuit impedance, and winding insulation dielectric loss.The standard value of H2 content is 150uL/L, but the measured value is 142uL/L; The standard value of iron core grounding current is 0. 1A, but the actual measured value is 0. 29A that has seriously exceeded the standard.The standard value of the initial value difference of the winding short-circuit impedance is 3%, but the actual measured value is 4. 5% that at position of attention.The standard value of winding insulation dielectric loss is 0. 8%, but the measured value is 1. 0%.
According to the relative deterioration formula (5), the reference standard value, test value and initial value in Table 5 are substituted to obtain the relative deterioration of the fault symptoms of the 220kV power transformer.The results are shown in Table 6, the relative deterioration of each fault feature is collected.According to the relative deterioration of the measured and standardized fault symptoms at this time, the evaluation logic of the first fault type about "winding fault", actually, it is analyzed and elaborated based on the calculation of the IDR evaluation matrix.Firstly, the IDR evaluation matrix of the equipment fault type is calculated.According to the calculation method of this fault type, the fault type relation degree equation of other fault types is obtained, and then it is also based on the variable weight coefficient.The identity difference inverse evaluation matrix about the operation state of power equipment is carried out in the end.Refer to Table 4 for constant weight coefficient of fault symptom, W1=[0.1306, 0. 2173, 0. 4432, 0. 2089], and according to the connection degree formula (7), the connection degree equation of winding fault is obtained as follows:   The expression form of the connection degree equation for this fault type is based on five reference quantities, and each value is assigned according to the equipartition method in the set pair analysis theory.In this chapter, i1, i2, i3, and j are assigned [-0.5, 0, 0. 5, 1] respectively, and their calculation results will be described in detail in the subsequent steps.
According to the above method, this paper firstly constructs the relative deterioration degree of the DIAGNOSTYKA, Vol. 24, No. 4 (2023) Shanyuan W, Donglin C, Zhonghong C, Menglong B…: Research on transformer condition evaluation… 8 fault characteristics corresponding to each fault type, and then we make the relative deterioration degree matrix.Finally, it is constructed according to the constant weight coefficient of each fault type, and it calculate the similarity difference inverse evaluation correlation equation of the remaining eight fault types: Based on the relative deterioration after standardized treatment in Table 5 and Table 6, according to the scoring formula (7) of the fault type, the constant weight coefficient converted in Table 4 is substituted to obtain the calculation information of the power transformer fault type.With the help of the scoring value of each fault type, the variable weight coefficient of each fault type is calculated.The results are shown in Table 7. Refer to the variable weight coefficient of each fault type in Table 7, and it calculate the overall health status of the transformer according to Formula (7).The calculation process of the contact function is as follows: During the operation of the power grid, the occurrence of any fault in the equipment is random and uncertain.Therefore, it is common for multiple equipment to be in an abnormal operating state, so them require maintenance/repair.However, the amount of resources allocated by enterprises is limited [18].According to the research ideas in this article, the fault state can be quantified as a score between 0 and 1.The lower the score, the higher the probability of the fault occurrence and the higher the urgency of maintenance.Based on the different degradation stages of fault stages, this article divides the maintenance work of all faults into three priority levels, as shown in Figure 5 [20].

Fig. 5. Prioritization of Power Equipment Maintenance
In this paper, the values of [i1, i2, i3, j] are respectively assigned as 0. 5, 0, -0. 5 and -1 by the equal division method, then E=[1, 0. 5, 0, -0. 5, -1] T , so the value of u is 0. 0030.At this time, the transformer is in the priority arrangement state referring to Table 3.When the transformer is under maintenance, the personnel maintenance shall use the same difference-opposite evaluation contacting degree equation.These data are described above to calculate the contact degree equation of 9 fault types in the process of the fault location.The process is as follows: As shown in Figure 6, the method calculates the overall rating value of the transformer based on its variable weight coefficient and comprehensive rating value [21].By comparing with these failure type in the equipment condition maintenance, the overall rating value calculated by this method could determine the scope and severity of transformer faults, then we provide corresponding maintenance opinions.
Therefore, through the comparison of the fault type correlation value, the winding fault is -0.8109, which is the lowest in the whole group.Therefore, it is speculated that the possibility of winding fault is greater than other fault types, so the personnel maintenance can focus on the winding fault.Shanyuan W, Donglin C, Zhonghong C, Menglong B…: Research on transformer condition evaluation… Fig. 6.Correlation between Failure type and Transformer

CONCLUSION
The current detection and monitoring methods for power equipment in power grids are complex, and the analysis of a large amount of data obtained from operational detection is crucial for the state assessment and fault diagnosis in the power grid.The following conclusions can be drawn from this study: 1.The association rules for monitoring data analysis of transformer fault types and fault symptoms is practical and effective.By calculating support and confidence, the coupling relationship and weight between fault types and fault symptoms could be obtained, which avoid the subjectivity of expert opinions or experience.2. This article attempts to organically combine set pair analysis and association rules, and apply them to power transformer fault diagnosis, which has certain advantages in dealing with uncertain problems.The concept of fuzzy theoretical membership degree was introduced in the process of constructing a multiple evaluation matrix, which objectively reflects the corresponding relationship between state evaluation indicators and operational state.3. Comparative analysis of examples show that the fusion and application of multiple data analysis achieve good state evaluation and get well diagnosis results, and it also perform well in handling multi fault problems.Moreover, the process of the algorithm in this paper is easy to program with computers, which has good operability and scalability.Finally, the working state of the transformer is classified, and the associated fault symptoms are determined.In the quantification process of fault symptom set, the relative deterioration degree and membership degree are introduced, the evaluation matrix is constructed, and then the relationship matrix between fault symptom set and fault type is determined according to the factors in factor space.According to the division of the health level, the health status of the transformer is diagnosed.Finally, the accuracy and effectiveness of the method in this chapter are verified by an example.

Fig. 1 .
Fig. 1.Condition monitoring system of transformer sum up, the same-different and opposite evaluation matrix of fault type can be expressed by formula (

Fig. 3 .
Fig. 3. Transformer fault diagnosis method based on spatial factor weight theory

Fig. 4 .
Fig. 4. The basic framework of the application of similarity, difference and counter evaluation Health Source of funding: This work is supported by Science Foundation of State Grid Fujian Electric Power Company(Grant, No. 52133121001F)． Author contributions: research concept and design, S.W., D.C., Z.C., M.B., H.O. X.L.; Collection and/or assembly of data, S.W., H.O., W.H., C.C.; Data analysis and interpretation, S.W., D.C., Z.C., M.B., X.L.; Writing the article, S.W.; Critical revision of the article, Y.S.; Final approval of the article, S.W., D.C., Z.C., M.B., H.O., X.L., W.H., C.C., Y.S.; Declaration of competing interest: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.Shanyuan W, Donglin C, Zhonghong C, Menglong B…: Research on transformer condition evaluation…

Table 1 .
: Types of faults in power transformers

Table 2 :Table 2 .
[16]t.Therefore, these 24 state variables can be selected as fault symptoms for this diagnosis.The relevant data of fault symptoms are shown in Table 5.The Liaoning Electric Power Research Institute provided 1238 sets of historical test data from transformers with similar operating environments.These known fault types from tested transformers are specific.The results after sorting are shown in Table A. According to the Inspection Specifications for On-line Monitoring Device of Substation Equipment, the device of Transformer Core Grounding Current and other relevant regulations in the Test Procedures for Condition are based on Maintenance of Power Transmission and Transformation Equipment issued by the State Grid in January 2014[16], the fault characteristics closely related to the fault state item set and with a quantitative process are selected, as shown in Fault characteristics of power transformers AbsorptanceS5Polarization index

Table 5 .
Test data of fault symptom of power transformer

Table 7 .
Calculation information of fault type of power transformer

Table B .
Classification function coefficients of fault types