Oil-paper insulation condition assessment method using fuzzy k-nearest neighbor and evidence theory

Abstract

This invention relates to a method for assessing the insulation state of oil-paper insulation using fuzzy K-nearest neighbors and evidence theory. First, a multi-feature parameter database is constructed, and a basic probability allocation method for evidence is proposed based on the database. Then, a combined weighting method is used to integrate the subjective and objective weights of the feature parameters, while simultaneously redistributing the basic probabilities of the evidence using an evidence discount factor to avoid conflicts inherent in D-S evidence theory. Next, fusion reasoning is performed on the various pieces of evidence to obtain a confidence result for the insulation state proposition. Finally, five transformers not included in the database are selected for actual testing, and the conclusions are compared with their actual insulation states to verify the effectiveness of the insulation state assessment method.

Description

Technical Field

[0001] This invention relates to a method for evaluating the insulation status of oil-paper using fuzzy K-nearest neighbors and evidence theory. Background Technology

[0002] Currently, in research on transformer oil-paper insulation condition assessment, some scholars use decision analysis methods such as evidence theory, rough set theory, and improved ideal solutions for insulation condition assessment. Existing research has shown good results by normalizing the indicators and then using membership functions to obtain the membership distribution of transformer insulation condition propositions and synthesizing assessment results. However, the threshold setting for normalization and the selection of membership functions are significantly influenced by subjective factors and cannot handle conflicting data from different indicator assessments. Summary of the Invention

[0003] The purpose of this invention is to provide a method for evaluating the state of oil-paper insulation using fuzzy K-nearest neighbors and evidence theory. This method can accurately assess the insulation state and, to a certain extent, reflect the trend of insulation deterioration, providing a more intuitive reference for transformer maintenance.

[0004] To achieve the above objectives, the technical solution of the present invention is: a method for evaluating the insulation state of oil-paper using fuzzy K-nearest neighbors and evidence theory, comprising the following steps:

[0005] Step 1: Select characteristic quantities for the recovery voltage method based on the extended Debye model;

[0006] Step 2: Perform basic probability assignment based on the evidence theory of FKNN;

[0007] Step 3: Introduce an evidence discount factor to correct the basic probability of the proposition;

[0008] Step 4: Use the DS evidence theory to infer and identify the state level of the transformer oil-paper insulation.

[0009] In one embodiment of the present invention, step 1 is specifically implemented as follows:

[0010] The state information of the insulating medium contained in the recovery voltage polarization spectrum obtained by the recovery voltage tester is shown on the horizontal and vertical axes, respectively, representing the charging time t. c and the maximum recovery voltage U rmax The maximum value of the recovery voltage U rmax Peak voltage U rmp The corresponding charging time t c This is called the recovery voltage principal time constant T. cdom The initial slope peak of the recovery voltage is denoted as S. rmp ;

[0011] The extended Debye model is used to characterize the polarization properties of insulating dielectrics, where R g For insulation resistance, C g For geometric capacitance, R pi For polarization resistance, C pi Polarizing capacitor; insulation resistance R g The smaller the value, the more severe the degradation of the insulating medium; geometric capacitance C g Geometric capacitance C reflects the energy storage capacity of insulating materials. g The larger the value, the more severe the degradation of the insulating medium.

[0012] To assess the condition of transformer oil-paper insulation, the necessary evidence for assessment includes characteristic quantities extracted from the recovery voltage polarization spectrum, including peak voltage U. rmp , recovery voltage principal time constant T cdom , initial slope peak S of the recovery voltage rmp And features extracted based on the extended Debye model, including insulation resistance R. g Geometric capacitance C g .

[0013] In one embodiment of the present invention, step 2 is specifically implemented as follows:

[0014] First, based on a database, the evidence is quantitatively expressed using the k nearest neighbor samples of the object to be evaluated and their attributes, thus obtaining the basic probabilistic quality of the proposition. Then, based on the database, different evidence p is used in the n samples. i Find the k nearest neighbor samples respectively, and determine the distance between them using the Euclidean distance formula:

[0015] (1)

[0016] Among them, the object to be evaluated is x s With the j-th sample x j The distance is determined by This means that the samples in the database are divided into Q1, Q2, ..., Q6 according to their insulation class. N If there are several categories, then the (indicator) evidence p of the transformer T to be evaluated at this time... i Assigned to proposition The basic probability assignment function is expressed as:

[0017] (2)

[0018] In the formula: Θ represents the complete set of propositions on transformer insulation class (also known as the identification framework). , These represent the evidence for the transformer to be evaluated and the evidence from the l-th neighbor of the transformer to be evaluated in the database, respectively. pi (Qj ) is evidence p i Assigned to proposition Q j The basic probability mass, m pi (Θ) is evidence p. i The basic probability mass assigned to the identification frame, that is, the evidence p i The uncertainty of T l Let u represent the l-th neighbor of the transformer to be evaluated in the database, and d be the fuzzy intensity coefficient, representing the weight of the distance between the transformer to be evaluated and each nearest neighbor; where u jl The formula for calculation is:

[0019] (3)

[0020] Where: n j This indicates that among the k nearest neighbor samples in the database of the transformer to be evaluated, the insulation class belongs to Q. j The number of neighbors, c l T represents l Insulation class; obviously, when the insulation class of the transformer to be evaluated is Q in all k neighbors in the database. j When, the basic probability mass m pi (Q j =1.

[0021] In one embodiment of the present invention, step 3 is specifically implemented as follows:

[0022] An evidence discount factor is introduced to modify the basic probability allocation function of the evidence based on the reliability of each piece of evidence. In order to comprehensively consider the decision-maker's subjective judgment on the importance of the evidence in the synthesis process and the amount of insulation state information objectively contained in the evidence source data, a combined weighting method is used to determine the weight coefficient of each piece of evidence by combining subjective and objective factors.

[0023] First, assume that the insulation condition assessment database contains a samples and b pieces of evidence.

[0024] The objective weights are determined using the entropy method of information entropy assignment; the original data matrix is ​​shown below X, and it is made dimensionless using the range standardization method as shown below:

[0025] (4)

[0026] (5)

[0027] Regarding the evidence p at this point j entropy value e j The weights are calculated using the following formula and by formula (7):

[0028] (6)

[0029] (7)

[0030] In the formula: the weight of the i-th sample evidence under the j-th evidence is q. ij And e j Let represent the entropy value of the j-th piece of evidence.

[0031] The subjective weights were determined using an improved analytic hierarchy process based on expert experience; the judgment matrix was constructed using the 9-scale method as follows:

[0032] (8)

[0033] In the formula: e ij The relative importance of evidence i to evidence j relative to the higher-level evidence.

[0034] Based on the judgment matrix, e is calculated from equations (9) to (11). ij ’ After combination, the optimal fitting matrix E' is obtained. Then, E' is normalized column-wise using equation (12), and the weight allocation g of each level of single sorting is calculated based on equation (13). i ;

[0035] (9)

[0036] (10)

[0037] (11)

[0038] (12)

[0039] (13)

[0040] In the formula: b ij d is the process quantity for optimally fitting the elements in the judgment matrix E. ij Let e ​​represent the elements in the optimal transfer matrix D, and e represent the elements in the optimal transfer matrix D. ’ ij For each element in the optimal fitting matrix E', g i For the weight allocation of single sorting at each level, e ij ’’ The matrix elements are E' normalized, and N is the number of evidence levels calculated. Finally, the weights of each sub-factor layer are multiplied by the corresponding subjective factor layer weights to perform a hierarchical ranking, yielding the subjective weights s of each piece of evidence for the insulation status assessment proposition as determined by the decision-maker. j ;

[0041] Objective and subjective weights are determined separately. To compensate for the shortcomings of each approach, a combined weighting method is used to determine the final weight coefficients for each piece of evidence; evidence p j The weighting coefficients are as follows (14), and let the evidence discount factor be α, W max The maximum value of the weight coefficients among all pieces of evidence is obtained as evidence p. i The formula for calculating the evidence discount factor is as shown in equation (15).

[0042] (14)

[0043] (15)

[0044] In the formula: μ represents the credibility of the evidence source. Based on the evidence discount factor α i Evidence p i For the insulation condition assessment proposition Q j The basic probability allocation is corrected as follows:

[0045] (16)

[0046] Thus, the revised basic probability incorporates the degree of distrust in the evidence into the identification framework, avoiding the inaccuracy of Dempster's rule when synthesizing conflicting evidence.

[0047] In one embodiment of the present invention, step 4 is specifically implemented as follows:

[0048] The identification framework Θ for the transformer insulation class proposition is a set of mutually exclusive and complete propositions, and the set of all possible propositions in Θ is denoted as 2. Θ Define the basic probability assignment function m: 2 Θ →[0, 1], 2 Θ Let be the power set of Θ, and satisfy the following equation:

[0049] (17)

[0050] In the formula: m(B) represents the basic probability mass assigned to proposition B, where the basic probability mass assigned to the identification frame is denoted as m(Θ), which represents the degree of global ignorance;

[0051] Dempster's evidence theory uses Dempster's rules of composition to combine and reason about evidence. The rules of composition are as follows:

[0052] (18)

[0053] In the formula: A represents a proposition in the identification frame Θ, m pi (B i ) is evidence p iAssigned to proposition B i The basic probability mass is the final result of the combined probability mass of proposition A obtained by synthesizing x pieces of evidence.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] This invention utilizes fuzzy K-nearest neighbor theory and evidence theory to evaluate the state of transformer oil-paper insulation. This evaluation method has the following advantages:

[0056] 1) The evidence basic probability allocation method based on the fuzzy K-nearest neighbor algorithm contains more information due to the introduction of the concept of class membership degree and distance weight, and can more objectively and accurately realize the basic probability allocation of each proposition.

[0057] 2) Taking into account the subjective and objective weights of the evidence, an evidence discount factor is introduced based on the comprehensive weight to allocate the degree of unreliability of the evidence to the global unknown, thereby solving the problem of unsatisfactory evidence synthesis results caused by conflicting evidence.

[0058] 3) Based on the DS evidence theory, multiple insulation condition assessment feature values ​​are comprehensively considered and synthesized for inference. The results are presented in the form of confidence distribution, which can accurately assess the insulation condition and reflect the insulation degradation trend to a certain extent, providing a more intuitive reference for transformer maintenance. Attached Figure Description

[0059] Figure 1 This is a flowchart of the method of the present invention.

[0060] Figure 2 To recover the voltage meter test curve.

[0061] Figure 3 To extend the equivalent circuit of the Debye model.

[0062] Figure 4 Hierarchical model for assessing the condition of transformer oil-paper insulation Detailed Implementation

[0063] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0064] This invention discloses a method for evaluating the insulation state of oil-paper using fuzzy K-nearest neighbors and evidence theory, comprising the following steps:

[0065] Step 1: Selecting characteristic quantities for the recovery voltage method based on the extended Debye model.

[0066] The recovery voltage polarization spectrum obtained by the recovery voltage tester contains rich information about the state of the insulating medium, such as... Figure 2 The horizontal and vertical axes of the recovered voltage polarization spectrum represent the charging time t, respectively. c and the maximum recovery voltage Urmax The maximum value of the recovery voltage U rmax The maximum value U rmp The corresponding charging time t c This is called the principal time constant T. cdom The initial slope peak of the recovery voltage is denoted as S. rmp .

[0067] Equivalent models are of great significance for studying the dielectric response of insulating media, and the extended Debye model is widely accepted. The extended Debye model is used to characterize the polarization properties of insulating media, such as... Figure 3 As shown, where R g For insulation resistance, C g For geometric capacitance, R pi For polarization resistor, C pi This is a polarized capacitor. Insulation resistance R g The smaller the value, the more severe the degradation of the insulating medium; geometric capacitance C g The geometric capacitance C reflects the energy storage capacity of the insulating material. g The larger the value, the more severe the degradation of the insulating medium.

[0068] To assess the condition of transformer oil-paper insulation, the required evaluation index, i.e., evidence, includes characteristic quantities extracted from the recovery voltage polarization spectrum (polarization spectrum peak voltage U). rmp , recovery voltage principal time constant T cdom , initial slope peak S of the recovery voltage rmp ) and the feature quantity extracted based on the extended Debye model (insulation resistance R) g Geometric capacitance C g ).

[0069] Step 2: Perform basic probability assignment based on the evidence theory of FKNN.

[0070] Using the FKNN evidence theory for classification can effectively avoid the impact of subjective factors on the evaluation of transformer oil-paper insulation status.

[0071] The method proposed in this invention first establishes a database and quantitatively expresses the evidence based on the k nearest neighbor samples of the object to be evaluated and their sample attributes, thereby obtaining the basic probabilistic quality of the proposition. Based on the database, in n samples, according to different evidence p... i Find the k nearest neighbor samples and determine their distance using the Euclidean distance formula:

[0072] (1)

[0073] Among them, the object to be evaluated is x s With the j-th sample x j The distance is determined by This means that the samples in the database are divided into Q1, Q2, ..., Q6 according to their insulation class. N If there are several categories, then the evidence p of the transformer T to be evaluated at this time... i Assigned to proposition The basic probability assignment function is expressed as:

[0074] (2)

[0075] In the formula: Θ represents the complete set of propositions on transformer insulation class (also known as the identification framework). , These represent the evidence for the transformer to be evaluated and the evidence from the l-th neighbor of the transformer to be evaluated in the database, respectively. pi (Q j ) is evidence p i Assigned to proposition Q j The basic probability mass, m pi (Θ) is evidence p. i The basic probability mass assigned to the identification frame, that is, the evidence p i The uncertainty of T l Let u represent the l-th neighbor of the transformer to be evaluated in the database, and d be the fuzzy intensity coefficient, representing the weight of the distance between the transformer to be evaluated and each nearest neighbor; where u jl The formula for calculation is:

[0076] (3)

[0077] Where: n j This indicates that among the k nearest neighbor samples in the database of the transformer to be evaluated, the insulation class belongs to Q. j The number of neighbors, c l T represents l Insulation class; obviously, when the insulation class of the transformer to be evaluated is Q in all k neighbors in the database. j When, the basic probability mass m pi (Q j =1.

[0078] Step 3: Introduce an evidence discount factor to correct the basic probability of the proposition.

[0079] To avoid discrepancies between the synthesis results of Dempster's combination rule and the actual situation due to conflicting evidence, an evidence discount factor is introduced to modify the basic probability allocation function of the evidence based on the reliability of each piece of evidence. To comprehensively consider the decision-maker's subjective judgment on the importance of the evidence in the synthesis process and the amount of insulation state information objectively contained in the source data of the evidence, a combination weighting method is used to determine the weight coefficient of each piece of evidence by combining subjective and objective factors.

[0080] First, assume that the insulation condition assessment database contains a samples and b pieces of evidence.

[0081] The objective weights are determined using the entropy method of information entropy assignment; the original data matrix is ​​shown below X, and it is made dimensionless using the range standardization method as shown below:

[0082] (4)

[0083] (5)

[0084] Regarding the evidence p at this point j entropy value e j The weights are calculated using the following formula and by formula (7):

[0085] (6)

[0086] (7)

[0087] In the formula: the weight of the i-th sample evidence under the j-th evidence is q. ij And e j Let represent the entropy value of the j-th piece of evidence.

[0088] The subjective weights were determined using an improved analytic hierarchy process based on expert experience; the judgment matrix was constructed using the 9-scale method as follows:

[0089] (8)

[0090] In the formula: e ij The relative importance of evidence i to evidence j relative to the higher-level evidence.

[0091] Based on the judgment matrix, e is calculated from equations (9) to (11). ij ’ After combination, the optimal fitting matrix E' is obtained. Then, E' is normalized column-wise using equation (12), and the weight allocation g of each level of single sorting is calculated based on equation (13). i ;

[0092] (9)

[0093] (10)

[0094] (11)

[0095] (12)

[0096] (13)

[0097] In the formula: b ij d is the process quantity for optimally fitting the elements in the judgment matrix E. ij Let e ​​represent the elements in the optimal transfer matrix D, and e represent the elements in the optimal transfer matrix D. ’ ij For each element in the optimal fitting matrix E', g i For the weight allocation of single sorting at each level, e ij ’’ The matrix elements are E' normalized, and N is the number of evidence levels calculated. Finally, the weights of each sub-factor layer are multiplied by the corresponding subjective factor layer weights to perform a hierarchical ranking, yielding the subjective weights s of each piece of evidence for the insulation status assessment proposition as determined by the decision-maker. j ;

[0098] Objective and subjective weights are determined separately. To compensate for the shortcomings of each approach, a combined weighting method is used to determine the final weight coefficients for each piece of evidence; evidence p j The weighting coefficients are as follows (14), and let the evidence discount factor be α, W max The maximum value of the weight coefficients among all pieces of evidence is obtained as evidence p. i The formula for calculating the evidence discount factor is as shown in equation (15).

[0099] (14)

[0100] (15)

[0101] In the formula: μ represents the credibility of the evidence source. Based on the evidence discount factor α i Evidence p i For the insulation condition assessment proposition Q j The basic probability allocation is corrected as follows:

[0102] (16)

[0103] Thus, the revised basic probability incorporates the degree of distrust in the evidence into the identification framework, avoiding the inaccuracy of Dempster's rule when synthesizing conflicting evidence.

[0104] Step 4: Use the DS evidence theory to infer and identify the transformer oil-paper insulation condition level.

[0105] The identification framework Θ for the transformer insulation class proposition is a set of mutually exclusive and complete propositions, and the set of all possible propositions in Θ is denoted as 2. Θ Define the basic probability assignment function m: 2 Θ →[0, 1], 2 Θ Let be the power set of Θ, and satisfy the following equation:

[0106] (17)

[0107] In the formula: m(B) represents the basic probability mass assigned to proposition B, where the basic probability mass assigned to the identification frame is denoted as m(Θ), which represents the degree of global ignorance;

[0108] Dempster's evidence theory uses Dempster's rules of composition to combine and reason about evidence. The rules of composition are as follows:

[0109] (18)

[0110] In the formula: A represents a proposition in the identification frame Θ, m pi (B i ) is evidence p i Assigned to proposition B i The basic probability mass is the final result of the combined probability mass of proposition A obtained by synthesizing x pieces of evidence.

[0111] Application examples of the method of this invention.

[0112] To verify the feasibility and accuracy of the proposed method integrating fuzzy K-nearest neighbor (KNN) and DS evidence theory for evaluating the oil-paper insulation status of transformers, five feature values ​​extracted in this patent were used as insulation status evaluation indicators. Simultaneously, in accordance with the provisions for preventive testing of power transformers in the "Regulations for Preventive Testing of Power Equipment," the oil-paper insulation status of transformers was classified into three categories—good insulation (G), moderate insulation (M), and severe aging (B)—based on tests such as furfural content detection in the oil. To avoid the influence of sample distribution imbalance when applying fuzzy KNN, data from 30 transformers (10 for each insulation status category) were selected from the recovery voltage test data of nearly 60 transformers to establish a transformer insulation status evaluation database. Partial sample data from the database are shown in Table 1.

[0113] Table 1. Characteristic values ​​and insulation status categories of 30 transformers

[0114]

[0115] Transformer T1, which was not included in the data, was selected as the test object, and its specific data is shown in Table 2 below.

[0116] Table 2 Evidence values ​​of aging characteristics of transformer T1

[0117]

[0118] First, the identification set is constructed. Let the insulation state assessment index, i.e., the evidence set P = {p i |i=1,2,…,5}, where the elements of the evidence to be evaluated are Urmp T cdom S rmp R g C g , Identification frame Θ={Q i │i=1,2,3}, where the insulation state proposition elements are good insulation, moderate insulation and severe aging, respectively.

[0119] Next, probability assignment is performed on the insulation state proposition. Based on equation (1), the k nearest neighbor samples in the database are obtained for each evaluation evidence of the transformer T1 to be evaluated. The value of k should not be too large or too small. Empirically, k≈N is taken. 0.5 N is the number of samples in the database. In this example, the value of k is 5. The basic probability of each insulation state proposition is allocated according to formula (2), and the results are shown in Table 3.

[0120] Table 3 Basic Probability Allocation of Insulation State of Transformer T1

[0121]

[0122] Analysis shows that since the only common proposition supported by the above evidence is Q1, the confidence level of proposition Q1 will be 100% when the evidence is synthesized directly using Dempster's rule. Therefore, the probability allocation results are adjusted based on the basic probability. The objective weights of each of the five pieces of evaluation evidence are calculated according to equations (4) to (7):

[0123] w=[0.0707, 0.1442, 0.2346, 0.4482, 0.1023]

[0124] Construct a hierarchical model for transformer insulation condition assessment, such as... Figure 4 As shown.

[0125] based on Figure 4 A judgment matrix is ​​constructed based on expert experience. This includes the judgment matrix E0 of the main factor layer regarding the target layer, and the judgment matrix U of the sub-factor layer. rmp T cdom S rmp Judgment matrix E1 of characteristic evidence of the principal factor layer recovery voltage method, and sub-factor layer R g C g The judgment matrix E2 for the evidence of the extended Debye model parameters in the main factor layer is as follows:

[0126] , ,

[0127] Based on equations (9) to (13), the weight distribution matrices after single sorting at each level are obtained as follows:

[0128] , ,

[0129] Multiply the weights of each sub-factor layer by the weights of their corresponding principal factor layers and perform a hierarchical ranking to obtain the subjective weight values:

[0130] s=[0.1516, 0.5255, 0.0729, 0.1875, 0.0625]

[0131] Based on equation (14), the subjective and objective weights are combined, and the final weight coefficients of each piece of evidence are:

[0132] W=[0.0552, 0.3906, 0.0881, 0.4331, 0.0330]

[0133] The evidence discount factor obtained from equation (15) is:

[0134] α=[0.1147, 0.8117, 0.1831, 0.9000, 0.0686]

[0135] Substituting the evidence discount factor α into formula (16) corrects the basic probability allocation of the evidence for each insulating proposition. The corrected results are shown in Table 4.

[0136] Table 4 Correction of Basic Reliability Allocation for Transformer T1

[0137]

[0138] According to the combination rule of Equation (18), the evidences p1 to p5 are synthesized, and the confidence distribution of each insulation proposition is shown in Table 5. Among them, the confidence level of the insulation state proposition Q1 is 91.43%. According to the maximum membership principle, the insulation state of transformer T1 is good.

[0139] Similarly, following the steps above, four additional transformers T2 to T5 were tested in sequence. The combined results of all transformers are shown in Table 5 below. The basic information and actual results of transformer T1 are also given in Table 6.

[0140] Table 5 Synthesis Results of Transformer T1

[0141]

[0142] Table 6 Basic Information and Actual Results for T1~T5

[0143]

[0144] Analysis of Tables 5 and 6 shows that transformers T1 and T2 are assessed as having good insulation condition with a confidence level exceeding 90% and an uncertainty of less than 5%, consistent with their short service life and low furfural content. Transformers T3 and T4 are assessed as having moderate insulation condition with a confidence level exceeding 50% and an uncertainty of less than 5%. Transformer T5 is assessed as having severe insulation aging with a confidence level of 70% and an uncertainty of less than 5%, all generally consistent with its insulation condition. Therefore, the method proposed in this patent can reflect the deterioration trend of the assessed transformers to a certain extent. It is foreseeable that when the insulation condition assessment database is expanded to a certain extent, the method of this invention can achieve quantitative assessment of the insulation condition of transformers.

[0145] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for evaluating the insulation state of oil-paper using fuzzy K-nearest neighbors and evidence theory, characterized in that, Includes the following steps: Step 1: Selecting characteristic quantities for the recovery voltage method based on the extended Debye model; the specific implementation is as follows: The state information of the insulating medium contained in the recovery voltage polarization spectrum obtained by the recovery voltage tester is shown on the horizontal and vertical axes, respectively, representing the charging time t. c and the maximum recovery voltage U rmax The maximum value of the recovery voltage U rmax Peak voltage U rmp The corresponding charging time t c This is called the recovery voltage principal time constant T. cdom The initial slope peak of the recovery voltage is denoted as S. rmp ; The extended Debye model is used to characterize the polarization properties of insulating dielectrics, where R g For insulation resistance, C g For geometric capacitance, R pi For polarization resistance, C pi Polarizing capacitor; insulation resistance R g The smaller the value, the more severe the degradation of the insulating medium; geometric capacitance C g Geometric capacitance C reflects the energy storage capacity of insulating materials. g The larger the value, the more severe the degradation of the insulating medium. To assess the condition of transformer oil-paper insulation, the necessary evidence for assessment includes characteristic quantities extracted based on the recovery voltage polarization spectrum, including peak voltage U. rmp , recovery voltage principal time constant T cdom , Initial slope peak value S of the recovery voltage rmp And features extracted based on the extended Debye model, including insulation resistance R. g Geometric capacitance C g ; Step 2: Perform basic probability assignment based on the evidence theory of FKNN; the specific implementation is as follows: First, based on a database, quantitative expression of evidence is achieved using the k nearest neighbor samples of the object to be evaluated and their sample attributes, thus obtaining the basic probabilistic quality of the proposition. Then, based on the database, different indicators (i.e., different pieces of evidence) are used to evaluate the n samples. i Find the k nearest neighbor samples respectively, and determine the distance between them using the Euclidean distance formula: (1) Among them, the object to be evaluated is x s With the j-th sample x j The distance is determined by This means that the samples in the database are divided into Q1, Q2, ..., Q6 according to their insulation class. N If there are several categories, then the evidence p of the transformer T to be evaluated at this time... i Assigned to proposition The basic probability assignment function is expressed as: (2) In the formula: Θ represents the complete set of propositions on transformer insulation class, i.e., the identification framework. , These represent the evidence for the transformer to be evaluated and the evidence for the transformer to be evaluated in the l-th neighbor of the database, respectively; m pi (Q j ) is evidence p i Assigned to proposition Q j The basic probability mass, m pi (Θ) is evidence p i The basic probability mass assigned to the identification frame, that is, the evidence p i The uncertainty of T l Let u represent the l-th neighbor of the transformer to be evaluated in the database, and d be the fuzzy intensity coefficient, representing the weight of the distance between the transformer to be evaluated and each nearest neighbor; where u jl The formula for calculation is: (3) Where: n j This indicates that among the k nearest neighbor samples in the database of the transformer to be evaluated, the insulation class belongs to Q. j The number of neighbors, c l T represents l Insulation class; obviously, when the insulation class of the transformer to be evaluated is Q in all k neighbors in the database. j At that time, the basic probability mass m pi (Q j )=1; Step 3: Introduce an evidence discount factor to correct the basic probability of the proposition; Step 4: Use the DS evidence theory to infer and identify the state level of the transformer oil-paper insulation.

2. The method for evaluating the insulation state of oil-paper using fuzzy K-nearest neighbors and evidence theory according to claim 1, characterized in that, Step 3 is implemented as follows: An evidence discount factor is introduced to modify the basic probability allocation function of the evidence based on the reliability of each piece of evidence. In order to comprehensively consider the decision-maker's subjective judgment on the importance of the evidence in the synthesis process and the amount of insulation state information objectively contained in the evidence source data, a combined weighting method is used to determine the weight coefficient of each piece of evidence by combining subjective and objective factors. First, assume that the insulation condition assessment database contains a samples and b pieces of evidence; The objective weights are determined using the entropy method of information entropy assignment; the original data matrix is ​​shown below X, and it is made dimensionless using the range standardization method as shown below: (4) (5) Regarding the evidence p at this point j entropy value e j The weights are calculated using the following formula and by formula (7): (6) (7) In the formula: the weight of the i-th sample evidence under the j-th evidence is q. ij And e j Let represent the entropy value of the j-th piece of evidence; The subjective weights were determined using an improved analytic hierarchy process based on expert experience; the judgment matrix was constructed using the 9-scale method as follows: (8) In the formula: e ij The relative importance of evidence i to evidence j relative to the higher-level evidence. Based on the judgment matrix, e is calculated from equations (9) to (11). ij ’ After combination, the optimal fitting matrix E' is obtained. Then, E' is normalized column-wise using equation (12), and the weight allocation g of each level of single sorting is calculated based on equation (13). i ; (9) (10) (11) (12) (13) In the formula: b ij d is the process quantity for optimally fitting the elements in the judgment matrix E. ij Let e ​​represent the elements in the optimal transfer matrix D, and e represent the elements in the optimal transfer matrix D. ’ ij For each element in the optimal fitting matrix E', g i For the weight allocation of single sorting at each level, e ij ’’ The matrix elements are E' normalized, and N is the number of evidence levels calculated. Finally, the weights of each sub-factor layer are multiplied by the corresponding subjective factor layer weights to perform a hierarchical ranking, yielding the subjective weights s of each piece of evidence for the insulation status assessment proposition as determined by the decision-maker. j ; Objective and subjective weights are determined separately. To compensate for the shortcomings of each approach, a combined weighting method is used to determine the final weight coefficients for each piece of evidence; evidence p j The weighting coefficients are as follows (14), and let the evidence discount factor be α, W max The maximum value of the weight coefficients among all pieces of evidence is obtained as evidence p. i The formula for calculating the evidence discount factor is as shown in equation (15). (14) (15) In the formula: μ represents the credibility of the evidence source; based on the evidence discount factor α i Evidence p i For the insulation condition assessment proposition Q j The basic probability allocation is corrected as follows: (16) Thus, the revised basic probability incorporates the degree of distrust in the evidence into the identification framework, avoiding the inaccuracy of Dempster's rule when synthesizing conflicting evidence.

3. The method for evaluating the insulation state of oil-paper using fuzzy K-nearest neighbors and evidence theory according to claim 1, characterized in that, Step 4 is implemented as follows: The identification framework Θ for the transformer insulation class proposition is a set of mutually exclusive and complete propositions, and the set of all possible propositions in Θ is denoted as 2. Θ Define the basic probability assignment function m: 2 Θ →[0, 1], 2 Θ Let be the power set of Θ, and satisfy the following equation: (17) In the formula: m(B) represents the basic probability mass assigned to proposition B, where the basic probability mass assigned to the identification frame is denoted as m(Θ), which represents the degree of global ignorance; Dempster's evidence theory uses Dempster's rules of composition to combine and reason about evidence. The rules of composition are as follows: (18) In the formula: A represents a proposition in the identification frame Θ, m pi (B i ) is evidence p i Assigned to proposition B i The basic probability mass is the final result of the combined probability mass of proposition A obtained by synthesizing x pieces of evidence.

Images