Power transformer layered progressive diagnosis method based on intelligent algorithm coupling

By combining the improved gray wolf optimization algorithm and least squares support vector machine, a multi-level diagnostic model was constructed, which solved the problem of insufficient accuracy in power transformer fault diagnosis, and achieved accurate identification of fault type, location and cause, thus improving diagnostic efficiency and accuracy.

CN118965132APending Publication Date: 2024-11-15WEIHAI POWER SUPPLY COMPANY OF STATE GRID SHANDONG ELECTRIC POWER COMPANY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410984756.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Existing intelligent algorithms are not accurate enough in power transformer fault diagnosis, especially since fault classification is greatly affected by the values ​​of key parameters, and traditional methods have problems such as large workload and high model complexity.

Method used

An improved gray wolf optimization algorithm (IGWO) is combined with least squares support vector machine (LSSVM), and dimensionality reduction is performed by kernel principal component analysis (KPCA). A multi-level LSSVM diagnostic model is then constructed to achieve hierarchical and progressive diagnosis of power transformer faults.

Benefits of technology

It improves the accuracy and efficiency of power transformer fault diagnosis, can automatically refine the identification of fault type, location and cause, conforms to the actual maintenance process, and has fast calculation speed and intuitive results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965132A_ABST
    Figure CN118965132A_ABST
Patent Text Reader

Abstract

The invention provides a power transformer layering progressive diagnosis method based on intelligent algorithm coupling, and the method comprises the steps: searching an optimal combination solution of a penalty coefficient and a kernel function parameter of each least square support vector machine classifier through employing an improved grey wolf algorithm based on the data of a dissolved gas in oil test, an electrical test and an oil test; and the faults of each layer are classified. The first layer distinguishes fault types; the second layer distinguishes fault parts; and the third layer distinguishes fault reasons. The method has the following advantages: the improved grey wolf algorithm is coupled with the least square support vector machine to construct a layered internal fault comprehensive diagnosis model, and automatic layered diagnosis of data samples is realized, so that the accuracy of internal fault diagnosis of the power transformer is improved. The method has the advantages of high calculation speed, high accuracy, visual result and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of power transformer fault diagnosis, and in particular relates to a hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling. Background Art

[0002] As the intermediate hub equipment for power conversion and energy transmission, power transformers play a vital role in the reliable operation of power grids. Once a power transformer has an abnormal operating state, it is particularly important to perform fault analysis on the data detected by routine tests such as dissolved gas analysis (DGA), electrical tests, and oil tests. They can reflect important information such as the fault circuit, fault location, and fault cause of the power transformer to a certain extent, thereby facilitating maintenance personnel to make maintenance decisions and improve the pertinence and effectiveness of operation and maintenance.

[0003] In recent years, intelligent algorithms such as BP neural network, support vector machine, DS evidence theory, etc. have been increasingly used in power transformer fault diagnosis. However, power transformer faults are complex, test data are numerous, and the fault classification accuracy is greatly affected by the values ​​of key parameters in the intelligent algorithm. Therefore, it is difficult to achieve the ideal accuracy of power transformer fault diagnosis using only one intelligent algorithm. Summary of the invention

[0004] In view of the defects in the prior art, the present invention provides a hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling, which can effectively solve the above problems.

[0005] The technical solution adopted by the present invention is as follows:

[0006] The present invention provides a hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling, comprising the following steps:

[0007] Step 1001, for the power transformer to be diagnosed and analyzed, collect its DGA data, electrical test data and oil test data; wherein the DGA data is the component type and content of each component of the dissolved gas in the power transformer oil;

[0008] Step 1002, at present, there are many studies on transformer status diagnosis with the help of machine learning, but most of them are limited to overheating or discharge fault type diagnosis. Common machine learning methods include BP neural network, decision tree, support vector machine, etc. All kinds of models have some shortcomings in power transformer fault diagnosis: BP neural network requires a large amount of sample data for support, which increases the difficulty and workload of actually collecting transformer fault samples, and the node structure of BP network is not easy to determine, and a unified and complete theoretical system has not yet been established; for the problem of too many categories, the decision tree needs to be pruned, and the rules generated by the model are too complicated.

[0009] Preferably, the method improves the gray wolf optimization algorithm; the convergence factor in the traditional gray wolf algorithm is calculated using a linear iteration method s=2×(1–t / T), and its convergence speed and convergence accuracy are difficult to meet the optimal performance requirements of the algorithm. One of the improvements of this method is to convert the convergence factor according to the exponential law s=2e -tT Iterate to adapt to the different convergence speed requirements before and after the algorithm iteration; in order to reflect the importance of different optimization results of the wolf group and avoid the algorithm falling into the local optimal solution, the weight coefficient Q of different search results is introduced i , an adaptive displacement update strategy with improved step-size Euclidean distance proportional weight is proposed Finally, the improved grey wolf algorithm (IGWO) was obtained;

[0010] In detail, in step 1002, the improved grey wolf optimization algorithm includes:

[0011] Step 1002-1: Surround the prey:

[0012] D=|P(t)-VP L (t)|

[0013] AD=P L (t)-P(t+1)

[0014] Where D is the distance to the prey; t is the number of updates; V and A are coefficient vectors; P and P L are the position vectors of the gray wolf and the prey respectively. The formulas of V and A are:

[0015] A=2sr1-s

[0016] V=2r2

[0017] Where r1 and r2 are random values ​​in [0,1]; s is the convergence factor s = 2e -tT , e is a natural constant, t is the current update number, and T is the set iteration number.

[0018] Step 1002-2: Catching prey:

[0019] When the gray wolf is hunting, the θ wolf updates its position through the position information of the other three wolves. The mathematical model of this behavior is as follows:

[0020]

[0021] Among them, P α is the position of wolf α; P β is the position of β wolf; P δ is the position of δ wolf; D α is the distance from the wolf to the prey; D β is the distance from β wolf to prey; D δ is the distance from δ wolf to the prey; A1 is the random vector of α wolf; V1 is the adaptive vector of α wolf; A2 is the random vector of β wolf; V2 is the adaptive vector of β wolf; A3 is the random vector of δ wolf; V3 is the adaptive vector of δ wolf; P1 is the position vector of guiding wolf No. 1; P2 is the position vector of guiding wolf No. 2; P3 is the position vector of guiding wolf No. 3. θ wolf will advance with the guidance of the three guiding wolves.

[0022] In view of the problem that the average value of P1, P2, and P3 in the traditional gray wolf algorithm cannot reflect the importance of α, β, and δ, this paper adopts the proportional weight of the improved step length Euclidean distance. The expression of the distance weight between the current gray wolf individual and α, β, and δ is as follows:

[0023]

[0024] In specific operations, the denominator may be zero. To avoid this, a very small constant μ is added to the above formula, which is generally 10. –9 ; Q1 is the weight of θ wolf to α wolf; Q2 is the weight of θ wolf to β wolf; Q3 is the weight of θ wolf to δ wolf.

[0025] The final gray wolf position update equation is as follows:

[0026]

[0027] In detail, in step 1002, the least squares support vector machine (LSSVM) can be generally described as:

[0028] In the sample set Q = {x i ,y i}(i=1,2,3,…,n), set the input x i and the target output y i ; n is the total number of samples. The vector of the sample set Q is nonlinearly mapped to the high-dimensional feature space F, and the optimal decision function is constructed in F:

[0029]

[0030] Among them, a i is the Lagrangian operator, b is the bias factor, and their values ​​are obtained by the least squares method, K(x,x i ) is the kernel function.

[0031] Step 1003, combining the improved Grey Wolf Optimization Algorithm with the Least Squares Support Vector Machine (LSSVM), and seeking the optimal penalty coefficient C and kernel function parameter g in each LSSVM classifier through IGWO to ensure the accuracy of diagnosis and classification; adding an if loop statement in the programming of the diagnosis algorithm to perform corresponding progressive hierarchical diagnosis on the data samples of the faulty transformer;

[0032] Step 1004, the state quantities obtained from various tests of transformer fault diagnosis affect each other. If all state quantities are used as the basis for fault diagnosis, on the one hand, it will increase the workload of test personnel, and on the other hand, it will have side effects on power transformer state assessment and fault location due to redundant information. Based on the collected common fault information of power transformers, this method uses kernel principal component analysis (KPCA) to reduce the dimension. This method uses a Gaussian kernel function, whose hyperparameter γ represents the distribution of data points in high-dimensional space. The larger the value of γ, the more concentrated the distribution of data points in high-dimensional space. The common value range is 10 -3 to 10 3 , this method takes the value of 2. The linearly independent characteristic state quantities with cumulative contribution rates higher than 94% are screened out to enhance the independence between state quantities.

[0033] Step 1005, this method adopts a multi-level LSSVM diagnosis method, by increasing the number of LSSVM layers, reducing the number of classifications in each layer, gradually refining the diagnosis results, and increasing the discrimination. The physical meaning and diagnosis process of progressive fault diagnosis are consistent with the actual maintenance work. Seven improved IGWO-LSSVM models are used for fault diagnosis classification, and all samples are put into the model at the same time. In the algorithm, through the if loop statement, the samples are automatically substituted into the classifier corresponding to the next layer for diagnosis. At the same time, the diagnostic accuracy and diagnosis time of each classifier can be obtained, and it is verified whether the predicted result of each sample is consistent with the actual result, so as to realize three-layer automatic progressive diagnosis. The DGA data, electrical test data, and oil test data of the power transformer to be diagnosed and analyzed are used as the input values ​​of the IGWO-LSSVM1 classifier, and the first-level fault circuit diagnosis is performed on the power transformer to be diagnosed and analyzed; wherein, the output result of the first-level diagnostic analysis is one of the following four types: normal, circuit fault, magnetic circuit fault, oil circuit fault; judge whether the following conditions are met;

[0034] Case 5.1, if the result diagnosed by the IGWO-LSSVM1 classifier is normal, the first-layer diagnosis result is determined to be normal, and the process ends;

[0035] Case 5.2: If the result diagnosed by the IGWO-LSSVM1 classifier is a circuit fault, it is determined that the first-layer diagnosis result is a circuit fault, and then step 1006 is executed;

[0036] Case 5.3: If the result diagnosed by the IGWO-LSSVM1 classifier is a magnetic circuit fault, the first-layer diagnosis result is determined to be a magnetic circuit fault, and then step 1009 is executed;

[0037] Case 5.4, if the result diagnosed by the IGWO-LSSVM1 classifier is an oil circuit fault, it is determined that the first layer diagnosis result is an oil circuit fault, and then step 1012 is executed;

[0038] Step 1006, using the IGWO-LSSVM2 classifier to automatically perform a second-level fault diagnosis on the power transformer diagnosed as a circuit fault in the first level; wherein the diagnosed fault location is one of the following two types: winding, tap changer; and determining whether the following conditions are met;

[0039] Case 6.1: If the diagnosis result of the IGWO-LSSVM2 classifier is that the winding part is faulty, the diagnosis result of the second layer is determined to be the winding, and then step 1007 is executed;

[0040] Case 6.2: If the result diagnosed by the IGWO-LSSVM2 classifier is that the tap changer is faulty, the second layer diagnosis result is determined to be the tap changer, and then step 1008 is executed;

[0041] Step 1007, using the IGWO-LSSVM5 classifier to automatically perform the third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is the winding, determine that the fault cause is one of the following two: short circuit between strands, layers, turns or deformation of the winding, and end the process;

[0042] Step 1008, for the power transformer with a fault at the tap switch, the method mainly studies the fault cause as poor contact of the tap switch, takes this as the cause diagnosis result of the fault at the third-layer tap switch, and ends the process;

[0043] Step 1009, using the IGWO-LSSVM3 classifier to automatically perform a second-level fault diagnosis on the power transformer diagnosed as a magnetic circuit fault in the first level; wherein the diagnosed fault location is one of the following two types: iron core, metal accessories; and determining whether the following conditions are met;

[0044] Case 9.1, if the result diagnosed by the IGWO-LSSVM3 classifier is that the core part is faulty, then the diagnosis result of the second layer is determined to be the core, and then step 1010 is executed;

[0045] Case 9.2: If the result diagnosed by the IGWO-LSSVM3 classifier is that the metal accessory part is faulty, then the diagnosis result of the third layer is determined to be a metal accessory, and then step 1011 is executed;

[0046] Step 1010, for the power transformer with a fault in the core, the method mainly studies the fault cause of the core multi-point grounding, takes this as the diagnosis result of the cause of the third-layer core fault, and ends the process;

[0047] Step 1011, using the IGWO-LSSVM6 classifier to automatically perform the third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is a metal accessory, determine that the fault cause is one of the following three types: suspended discharge, magnetic leakage or magnetic shielding heating, and bare metal heating, and end the process;

[0048] Step 1012, using the IGWO-LSSVM4 classifier to automatically perform a second-level fault diagnosis on the power transformer diagnosed as an oil circuit fault in the first level; wherein the fault location diagnosed is one of the following three types: enclosure, oil channel, insulating oil paper; and judging whether the following conditions are met;

[0049] Case 12.1, if the result diagnosed by the IGWO-LSSVM4 classifier is that the enclosure is faulty, then the second layer diagnosis result is determined to be the enclosure, and then step 1013 is executed;

[0050] Case 12.2, if the result diagnosed by the IGWO-LSSVM4 classifier is that the oil channel is faulty, then the second layer diagnosis result is determined to be the oil channel, and then step 1014 is executed;

[0051] Case 12.3, if the result diagnosed by the IGWO-LSSVM4 classifier is that the insulation oil paper part is faulty, then the diagnosis result of the second layer is determined to be insulation oil paper, and then step 1015 is executed;

[0052] Step 1013, for the power transformer with fault in the enclosure, the method mainly studies the fault cause as enclosure discharge, takes this as the diagnosis result of the cause of the third-layer enclosure fault, and ends the process;

[0053] Step 1014, for the power transformer with oil channel fault, the method mainly studies the fault cause of the oil channel obstruction, takes this as the diagnosis result of the third-level oil channel fault cause, and ends the process;

[0054] Step 1015, using IGWO-LSSVM7 classifier to perform third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is insulating oil-paper, determine the cause of the fault is one of the following two: oil-paper aging, oil-paper dampness, and end the process.

[0055] Preferably, the input values ​​in step 1004 are all the following characteristic state quantities, so as to perform hierarchical diagnosis of internal faults of power transformers: C2H4 / C2H6, CH4 / H2, three-phase unbalance rate of DC resistance, core grounding current, C2H2 / C2H4, partial discharge, CO2 / CO, absorption ratio, water content in oil, and oil dielectric loss.

[0056] Compared with the prior art, the power transformer fault hierarchical progressive diagnosis method based on intelligent algorithm coupling provided by the present invention has the following advantages:

[0057] The present invention combines the improved grey wolf algorithm with the least squares support vector machine, adds an if loop statement in the algorithm code, realizes the automatic progressive hierarchical diagnosis of the internal fault of the power transformer, refines the power transformer fault diagnosis result, conforms to the actual diagnosis process, and thus improves the recognition accuracy of the internal fault circuit, fault location, and fault cause of the power transformer. The present invention also has the advantages of fast calculation speed and intuitive results. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0059] Figure 1 It is a schematic diagram of the hierarchical structure of the gray wolf group of the present invention.

[0060] Figure 2 This is a diagnostic flow chart of the improved grey wolf algorithm combined with the least squares support vector machine of the present invention.

[0061] Figure 3 This is a schematic diagram of the architecture of the intelligent algorithm-coupled hierarchical progressive diagnosis method for power transformer faults provided by the present invention.

[0062] Figure 4 This is a diagram of the diagnostic results of the first-layer output fault circuit of the present invention.

[0063] Figure 5 This is a diagram showing the diagnosis results of the circuit fault location in the second layer of the present invention.

[0064] Figure 6This is a diagram showing the diagnostic results of the magnetic circuit fault location in the second layer of the present invention.

[0065] Figure 7 This is a diagram showing the diagnosis results of the oil circuit fault location in the second layer of the present invention.

[0066] Figure 8 This is a diagram showing the diagnosis results of the causes of winding faults in the third layer of the present invention.

[0067] Fig. 9 This is a diagram showing the diagnosis results of the causes of metal accessory failures in the third layer of the present invention.

[0068] Fig.10 This is a diagram showing the diagnosis result of the cause of the oil-paper insulation failure in the third layer of the present invention. DETAILED DESCRIPTION

[0069] In order to make the technical problems, technical solutions and beneficial effects solved by the present invention more clearly understood, the following is a summary of the technical problems, technical solutions and beneficial effects solved by the present invention in combination with the attached drawings. Figures 1 to 10 It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0070] The least square support vector machine (LSSVM) uses a quadratic loss function to replace the non-equality constraints of the optimization problem with equality constraints. It has a faster computing speed and has obvious advantages in solving nonlinear and high-dimensional classification problems. The disadvantage is that two important parameters: the kernel function parameter g and the penalty coefficient C are positional, and their values ​​have a great influence on the performance of the model and need to be carefully determined. The optimization process of the improved gray wolf algorithm (IGWO) simulates the hunting process of wolves. The algorithm has few control parameters and is easy to implement. It can effectively solve the problem of seeking the optimal penalty coefficient C and kernel function parameter g in the LSSVM classifier. Different state quantities will be used to characterize different fault types, fault locations and fault causes of power transformers. In view of the above problems, the present invention couples the improved gray wolf algorithm with the least square support vector machine, and constructs a hierarchical diagnosis model of internal faults of power transformers by analyzing multiple state quantities of power transformers, thereby realizing accurate diagnosis of fault types, fault locations and fault causes of power transformers.

[0071] Specifically, in order to perform more detailed and accurate fault diagnosis, more comprehensive state information is needed as support. Different state quantities of power transformers have different sensitivity to different faults. The present invention proposes a hierarchical diagnosis model for internal faults of power transformers, which not only realizes the classification of fault types, but also realizes the fault location and determination of fault causes, and gradually analyzes the fault conditions from coarse to fine to ensure the accuracy of fault diagnosis. The present invention divides the diagnosis into three layers by analyzing a large number of power transformer fault cases and referring to previous successful fault classification experience.

[0072] The diagnostic process provided by the present invention is as follows Figure 3 As shown, the following steps are included:

[0073] Step 1001, for the power transformer to be diagnosed, obtain its DGA data, electrical test data, and oil test data, wherein the DGA data is the component type and content of each component of the dissolved gas in the power transformer oil;

[0074] Step 1002, using the kernel principal component analysis method to perform dimensionality reduction processing on all the collected state quantities. This method uses the Gaussian kernel function, and its hyperparameter γ represents the distribution of data points in the high-dimensional space. The larger the value of γ, the more concentrated the distribution of data points in the high-dimensional space. The common value range is 10 -3 to 10 3 , this method takes the value of 2. The linearly independent characteristic state quantities with cumulative contribution rates higher than 94% are screened out to enhance the independence between state quantities.

[0075] Step 1003, based on the various test data obtained above, the first level fault diagnosis is performed through IGWO-LSSVM1 to classify the diagnosed power transformer fault types into the following four groups: normal, circuit fault, magnetic circuit fault, and oil circuit fault, and determine whether the following conditions are met;

[0076] Case 3.1, if the result diagnosed by the IGWO-LSSVM1 classifier is normal, the first-layer diagnosis result is determined to be normal, and the process ends;

[0077] Case 3.2: If the result diagnosed by the IGWO-LSSVM1 classifier is a circuit fault, it is determined that the first-layer diagnosis result is a circuit fault, and then step 1004 is executed;

[0078] Case 3.3: if the result diagnosed by the IGWO-LSSVM1 classifier is a magnetic circuit fault, the first-layer diagnosis result is determined to be a magnetic circuit fault, and then step 1007 is executed;

[0079] Case 3.4, if the result diagnosed by the IGWO-LSSVM1 classifier is an oil circuit fault, it is determined that the first layer diagnosis result is an oil circuit fault, and then step 1010 is executed;

[0080] Step 1004, using the IGWO-LSSVM2 classifier to perform a second-level fault diagnosis on the power transformer diagnosed as a circuit fault in the first level; wherein the diagnosed fault location is one of the following two types: winding, tap changer; and determining whether the following conditions are met;

[0081] Case 4.1, if the result diagnosed by the IGWO-LSSVM2 classifier is that a fault occurs in the winding part, it is determined that the diagnosis result of the second layer is the winding, and then step 1005 is executed;

[0082] Case 4.2: If the diagnosis result of the IGWO-LSSVM2 classifier is that the tap changer is faulty, the diagnosis result of the second layer is determined to be the tap changer, and then step 1006 is executed;

[0083] Step 1005, using the IGWO-LSSVM5 classifier to perform a third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is the winding, determining that the fault cause is one of the following two: short circuit between strands, layers, turns or winding deformation, and ending the process;

[0084] Step 1006, for the power transformer with a fault at the tap switch, the method mainly studies the fault cause as poor contact of the tap switch, takes this as the cause diagnosis result of the fault at the third-layer tap switch, and ends the process;

[0085] Step 1007, using the IGWO-LSSVM3 classifier to perform a second-level fault diagnosis on the power transformer diagnosed as a magnetic circuit fault in the first level; wherein the diagnosed fault location is one of the following two types: iron core, metal accessories; and determining whether the following conditions are met;

[0086] Case 7.1, if the result diagnosed by the IGWO-LSSVM3 classifier is that the core part is faulty, then the diagnosis result of the second layer is determined to be the core, and then step 1008 is executed;

[0087] Case 7.2: If the result diagnosed by the IGWO-LSSVM3 classifier is that a metal accessory part is faulty, the diagnosis result of the third layer is determined to be a metal accessory, and then step 1009 is executed;

[0088] Step 1008, for the power transformer with a fault in the core, the method mainly studies the fault cause of the core multi-point grounding, takes this as the diagnosis result of the cause of the third-layer core fault, and ends the process;

[0089] Step 1009, using the IGWO-LSSVM6 classifier to perform a third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is a metal accessory, determining that the fault cause is one of the following three types: suspended discharge, magnetic leakage or magnetic shielding heating, and bare metal heating, and ending the process;

[0090] Step 1010, using IGWO-LSSVM4 classifier to perform second-level fault diagnosis on the power transformer diagnosed as oil circuit fault in the first level; wherein the diagnosed fault location is one of the following three types: enclosure, oil channel, insulating oil paper; determine whether the following conditions are met;

[0091] Case 10.1, if the result diagnosed by the IGWO-LSSVM4 classifier is that the enclosure is faulty, then the second layer diagnosis result is determined to be the enclosure, and then step 1011 is executed;

[0092] Case 10.2, if the result diagnosed by the IGWO-LSSVM4 classifier is that the oil channel is faulty, then the second layer diagnosis result is determined to be the oil channel, and then step 1012 is executed;

[0093] Case 10.3, if the diagnosis result of the IGWO-LSSVM4 classifier is that the insulation oil paper part is faulty, then the diagnosis result of the second layer is determined to be insulation oil paper, and then step 1013 is executed;

[0094] Step 1011, for the power transformer with fault in the enclosure, the method mainly studies the fault cause as enclosure discharge, takes this as the diagnosis result of the cause of the third-layer enclosure fault, and ends the process;

[0095] Step 1012, for the power transformer with a fault in the oil channel, the method mainly studies the fault cause of the oil channel being blocked, takes this as the diagnosis result of the third-level oil channel fault cause, and ends the process;

[0096] Step 1013, using IGWO-LSSVM7 classifier to perform third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is insulating oil-paper, determine the cause of the fault is one of the following two: oil-paper aging, oil-paper dampness, and end the process.

[0097] In the present invention, the state quantities used for hierarchical diagnosis of internal faults of power transformers include C2H4 / C2H6, CH4 / H2, three-phase unbalance rate of DC resistance, core grounding current, C2H2 / C2H4, partial discharge, CO2 / CO, absorption ratio, water content in oil, and oil dielectric loss.

[0098] Among them, the first level of diagnosis is divided into 4 fault circuits of the power transformer: normal, circuit fault, magnetic circuit fault, and oil circuit fault; the second level of diagnosis is divided into 7 fault locations: windings, tap changers, iron cores, metal accessories, screens, oil channels, and insulating oil paper; the third level of diagnosis is divided into 11 fault causes: X1 strand, layer, and turn-to-turn short circuits, X2 winding deformation, X3 poor contact of tap changers, X4 multi-point grounding of iron cores, X5 suspended discharge, X6 leakage flux or magnetic shielding heating, X7 bare metal heating, X8 screen discharge, X9 oil channel obstruction, X10 oil paper aging, and X11 oil paper moisture.

[0099] The steps of using each IGWO-LSSVM classifier to perform hierarchical diagnosis of internal faults of power transformers are as follows:

[0100] 1) The DGA, electrical test and oil test data of the power transformer to be diagnosed are used as model input.

[0101] 2) Use the improved grey wolf algorithm to find the optimal penalty coefficient C and kernel function parameter g for each least squares support vector machine.

[0102] 3) Substitute the optimal penalty coefficients and kernel function parameters sought into each least squares support vector machine model for classification.

[0103] 4) Output the classification results of each classifier.

[0104] Verification example:

[0105] The 300 groups of power transformer samples collected were diagnosed using the hierarchical diagnosis method provided by the present invention, and the sample data after dimensionality reduction was divided into training sets and test sets at a ratio of 7:3, including 210 training sets and 90 test sets. In the first layer of fault types, there are 19 normal test sets, 28 circuit fault test sets, 26 magnetic circuit fault test sets, and 17 oil circuit fault test sets. The diagnostic effect is shown in Table 1. In the second layer of fault locations, there are 21 winding test sets, 7 tap changer test sets, 7 core test sets, 19 metal accessory test sets, 7 screen test sets, 4 oil channel test sets, and 6 insulating oil paper test sets. The diagnostic effect is shown in Table 2. Among the third-level fault causes, there are 12 test sets for X1 strand, layer, and turn-to-turn short circuits, 9 test sets for X2 winding deformation, 7 test sets for X3 tap changer poor contact, 7 test sets for X4 core multi-point grounding, 2 test sets for X5 suspended discharge, 11 test sets for X6 leakage flux or magnetic shielding heating, 6 test sets for X7 bare metal heating, 7 test sets for X8 enclosure discharge, 4 test sets for X9 oil channel obstruction, 2 test sets for X10 oil paper aging, and 4 test sets for X11 oil paper moisture. The diagnostic results are shown in Table 3.

[0106] Table 1: Fault type diagnosis effect

[0107]

[0108] Table 2: Fault location diagnosis results

[0109]

[0110] Table 3: Fault cause diagnosis results

[0111]

[0112]

[0113] In order to verify the superiority of this method, the IGWO-LSSVM single-layer output model, the GWO-LSSVM multi-layer output model, the ACO-LSSVM single-layer output model, and the ACO-BP model are compared and analyzed. The accuracy of each model is shown in Table 4.

[0114] Table 4: Comparison of diagnostic accuracy of multiple models

[0115]

[0116] By analyzing Table 4, it can be concluded that this method can effectively diagnose the fault type, fault location and fault cause of the power transformer through hierarchical progressive diagnosis, which is in line with the actual power transformer fault diagnosis process.

[0117] This method was applied to an actual fault case for analysis: After a transformer with model SZ11-63000 / 110 in a substation failed, the operation and maintenance personnel took oil samples for chromatographic analysis, and the acetylene content reached 33uL / L. The oil chromatographic test data is shown in Table 5. During the electrical test, it was found that the DC resistance of the high-voltage winding phase A 1-4 and 14-17 gears exceeded the deviation from the B and C phases. The core grounding current measurement, partial discharge test, transformer oil dielectric loss and oil water content test were all normal.

[0118] Table 5: DGA data of a 110kV power transformer in a substation

[0119]

[0120] Substituting these data into the trained IGWO-LSSVM for multi-level diagnosis, the diagnosis result is circuit fault → winding → winding deformation. The actual core hanging result is that the A-phase high-voltage winding voltage regulating coil is deformed and damaged.

[0121] In view of the accuracy of least squares support vector machine classification and the rapidity of the improved gray wolf algorithm in optimization, the present invention proposes a hierarchical diagnosis method for internal faults of power transformers based on intelligent algorithm coupling, comprehensively considering the dissolved gas test in oil, electrical test and oil test data, and establishing a hierarchical diagnosis model for internal faults of power transformers based on the improved gray wolf algorithm and least squares support vector machine coupling on the basis of analyzing a large number of power transformer failure cases. The model uses a layer structure to gradually subdivide the internal faults of power transformers, preliminarily judge the fault type, fault location and fault cause, and help to provide effective maintenance suggestions for the site. The model has the advantages of fast calculation speed, high accuracy and intuitive results. A large number of examples have also proved the effectiveness of the method provided by the present invention.

[0122] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling, characterized in that: The steps include: Step 1001, for the power transformer to be diagnosed and analyzed, collect its DGA data, electrical test data and oil test data; wherein the DGA data is the component type and content of each component of the dissolved gas in the power transformer oil; Step 1002, improve the Grey Wolf Optimization Algorithm; wherein, the convergence factor is improved from linear convergence to exponential convergence to meet the requirements of different convergence speeds in the early and late stages of algorithm iteration; in order to improve the optimization performance, an adaptive displacement update strategy of improving the step length Euclidean distance proportional weight is proposed; and finally, IGWO is obtained; Step 1003, combining IGWO with LSSVM, and seeking the optimal penalty coefficient C and kernel function parameter g in each LSSVM classifier through IGWO to ensure the accuracy of diagnosis and classification; adding an if loop statement in the programming of the diagnosis algorithm to perform corresponding progressive hierarchical diagnosis on the data samples of the faulty transformer; Step 1004, using the kernel principal component analysis method to perform dimensionality reduction processing, screen out linearly independent characteristic state quantities with cumulative contribution rates higher than 94%, and enhance the independence between state quantities; Step 1005, using the DGA data, electrical test data, and oil test data of the power transformer to be diagnosed and analyzed as input values ​​of the IGWO-LSSVM1 classifier, and performing the first-level fault diagnosis on the power transformer to be diagnosed and analyzed; Step 1006, using the IGWO-LSSVM2 classifier to automatically perform a second-level fault diagnosis on the power transformer diagnosed as a circuit fault at the first level; Step 1007, using the IGWO-LSSVM5 classifier to automatically perform the third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is the winding, determine that the fault cause is one of the following two: short circuit between strands, layers, turns or deformation of the winding, and end the process; Step 1008, for the power transformer with a fault at the tap switch, the method mainly studies the fault cause as poor contact of the tap switch, takes this as the cause diagnosis result of the fault at the third-layer tap switch, and ends the process; Step 1009, using the IGWO-LSSVM3 classifier to automatically perform a second-level fault diagnosis on the power transformer diagnosed as a magnetic circuit fault in the first level; Step 1010, for the power transformer with a fault in the core, the method mainly studies the fault cause of the core multi-point grounding, takes this as the diagnosis result of the cause of the third-layer core fault, and ends the process; Step 1011, using the IGWO-LSSVM6 classifier to automatically perform the third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is a metal accessory, determine that the fault cause is one of the following three types: suspended discharge, magnetic leakage or magnetic shielding heating, and bare metal heating, and end the process; Step 1012, using the IGWO-LSSVM4 classifier to automatically perform a second-level fault diagnosis on the power transformer diagnosed as an oil circuit fault in the first level; Step 1013, for the power transformer with fault in the enclosure, the method mainly studies the fault cause as enclosure discharge, takes this as the diagnosis result of the cause of the third-layer enclosure fault, and ends the process; Step 1014, for the power transformer with oil channel fault, the method mainly studies the fault cause of the oil channel obstruction, takes this as the diagnosis result of the third-level oil channel fault cause, and ends the process; Step 1015, using IGWO-LSSVM7 classifier to perform third-level fault diagnosis on the power transformer whose fault location in the second-level diagnosis is insulating oil-paper, determine the cause of the fault is one of the following two: oil-paper aging, oil-paper dampness, and end the process.

2. The hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling according to claim 1 is characterized in that: In step 1002, the improved grey wolf optimization algorithm includes: Step 1002-1: Surround the prey: D=|P(t)-VP L (t)| AD=P L (t)-P(t+1) Where D is the distance to the prey; t is the number of updates; V and A are coefficient vectors; P and P L are the position vectors of the gray wolf and the prey respectively; the formulas of V and A are: A=2sr1-s V=2r2 Where r1 and r2 are random values ​​in [0,1]; s is the convergence factor s = 2e -tT , e is a natural constant, t is the current update number, and T is the set iteration number; Step 1002-2: Catching prey: When the gray wolf is hunting, the θ wolf updates its position through the position information of the other three wolves. The mathematical model of this behavior is as follows: Among them, P α is the position of wolf α; P β is the position of β wolf; P δ is the position of δ wolf; D α is the distance from the wolf to the prey; D β is the distance from wolf β to prey; D δ is the distance from δ wolf to prey; A1 is the random vector of α wolf; V1 is the adaptive vector of α wolf; A2 is the random vector of β wolf; V2 is the adaptive vector of β wolf; A3 is the random vector of δ wolf; V3 is the adaptive vector of δ wolf; P1 is the position vector of No. 1 guiding wolf; P2 is the position vector of No. 2 guiding wolf; P3 is the position vector of No. 3 guiding wolf; θ wolf will advance with the guidance of the three guiding wolves; In view of the problem that the average value of P1, P2, and P3 in the traditional gray wolf algorithm cannot reflect the importance of α, β, and δ, this paper adopts the proportional weight of the improved step length Euclidean distance. The expression of the distance weight between the current gray wolf individual and α, β, and δ is as follows: In specific operations, the denominator may be zero. To avoid this, a very small constant μ is added to the above formula, which is generally 10. –9 ; Q1 is the weight of θ wolf to α wolf; Q2 is the weight of θ wolf to β wolf; Q3 is the weight of θ wolf to δ wolf; The final gray wolf position update equation is as follows: P(t+1) is the final position of the gray wolf.

3. The hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling according to claim 1 is characterized in that: In step 1005, the output result of the first-level diagnostic analysis is one of the following four: normal, circuit fault, magnetic circuit fault, oil circuit fault; determine whether the following conditions are met; Case 5.1, if the result diagnosed by the IGWO-LSSVM1 classifier is normal, the first-layer diagnosis result is determined to be normal, and the process ends; Case 5.2: If the result diagnosed by the IGWO-LSSVM1 classifier is a circuit fault, it is determined that the first-layer diagnosis result is a circuit fault, and then step 1006 is executed; Case 5.3: If the result diagnosed by the IGWO-LSSVM1 classifier is a magnetic circuit fault, the first-layer diagnosis result is determined to be a magnetic circuit fault, and then step 1009 is executed; Case 5.4: if the diagnosis result obtained by the IGWO-LSSVM1 classifier is an oil circuit fault, it is determined that the first-layer diagnosis result is an oil circuit fault, and then step 1012 is executed.

4. The hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling according to claim 1 is characterized in that: In step 1006, the fault location diagnosed is one of the following two: winding, tap changer; determine whether the following conditions are met; Case 6.1: If the diagnosis result of the IGWO-LSSVM2 classifier is that the winding part is faulty, the diagnosis result of the second layer is determined to be the winding, and then step 1007 is executed; In case 6.2, if the diagnosis result of the IGWO-LSSVM2 classifier is that a fault occurs at the tap changer, the diagnosis result of the second layer is determined to be the tap changer, and then step 1008 is executed.

5. The hierarchical progressive diagnosis method for power transformers based on intelligent algorithm coupling according to claim 1 is characterized in that: In step 1009, the fault location diagnosed is one of the following two types: iron core, metal accessories; determine whether the following conditions are met; Case 9.1, if the result diagnosed by the IGWO-LSSVM3 classifier is that the core part is faulty, then the diagnosis result of the second layer is determined to be the core, and then step 1010 is executed; In case 9.2, if the diagnosis result of the IGWO-LSSVM3 classifier is that a metal accessory part has failed, it is determined that the diagnosis result of the third layer is a metal accessory, and then step 1011 is executed.

6. The power transformer hierarchical progressive diagnosis method based on intelligent algorithm coupling according to claim 1 is characterized in that: In step 1012, the fault location diagnosed is one of the following three types: enclosure, oil channel, insulating oil paper; determine whether the following conditions are met; Case 12.1, if the result diagnosed by the IGWO-LSSVM4 classifier is that the enclosure is faulty, then the second layer diagnosis result is determined to be the enclosure, and then step 1013 is executed; Case 12.2, if the result diagnosed by the IGWO-LSSVM4 classifier is that the oil channel is faulty, then the second layer diagnosis result is determined to be the oil channel, and then step 1014 is executed; In case 12.3, if the diagnosis result of the IGWO-LSSVM4 classifier is that the insulation oil paper part is faulty, it is determined that the diagnosis result of the second layer is insulation oil paper, and then step 1015 is executed.

7. The power transformer hierarchical progressive diagnosis method based on intelligent algorithm coupling according to claim 1 is characterized in that: The selection of the kernel function in step 1004 includes: This method uses a Gaussian kernel function, whose hyperparameter γ represents the distribution of data points in high-dimensional space. The larger the value of γ, the more concentrated the distribution of data points in high-dimensional space. The common value range is 10 -3 to 10 3 , this method takes the value as 2.

8. The power transformer hierarchical progressive diagnosis method based on intelligent algorithm coupling according to claim 1 is characterized by: The optimal combination solution of penalty coefficient C and kernel function parameter g in each least squares support vector machine classifier is sought by improving the grey wolf algorithm; this method adopts a multi-level LSSVM diagnosis method, by increasing the number of LSSVM layers, reducing the number of classifications in each layer, gradually refining the diagnosis results, and increasing the discrimination degree. The physical meaning and diagnosis process of progressive fault diagnosis are consistent with the actual maintenance work; all samples are put into the model at the same time, and the samples are automatically substituted into the corresponding classifier of the next layer for diagnosis through the if loop statement in the algorithm. At the same time, the diagnostic accuracy and diagnosis time of each classifier can be obtained, and it is verified whether the predicted result of each sample is consistent with the actual result, so as to realize three-layer automatic progressive diagnosis.

9. The power transformer hierarchical progressive diagnosis method based on intelligent algorithm coupling according to claim 1 is characterized in that: The input values ​​in step 1004 are all the following characteristic state quantities, so as to perform hierarchical diagnosis of internal faults of power transformers: C2H4 / C2H6, CH4 / H2, three-phase unbalance rate of DC resistance, core grounding current, C2H2 / C2H4, partial discharge, CO2 / CO, absorption ratio, water content in oil, and oil dielectric loss.