Rapid detection and diagnosis method and system for short-circuit fault of transformer winding

Abstract

The invention provides a rapid detection and diagnosis method and system for a short-circuit fault of a transformer winding, and relates to the technical field of transformers, and the method comprises the steps: collecting the electrical parameters of the transformer winding, calculating equivalent impedance, building a transfer function model and an equivalent impedance model, and extracting a frequency domain fault characteristic quantity and a short-circuit force characteristic; and constructing a characteristic quantity judgment matrix to obtain a weight coefficient, carrying out multi-source information fusion by utilizing a D-S evidence theory, and realizing fault diagnosis by combining fuzzy comprehensive evaluation. The method can accurately identify the short-circuit fault type of the transformer winding, and improves the reliability and accuracy of fault diagnosis.

Description

TECHNICAL FIELD

[0001] The present application relates to transformer technology, in particular to a rapid detection and diagnosis method and system for transformer winding short circuit fault. BACKGROUND

[0002] As a key equipment in power systems, the safe and stable operation of transformers is crucial for reliable power supply of the power grid. Transformer winding short circuit fault is one of the most common and most dangerous fault types, which may cause transformer damage and even trigger system accidents. Currently, the detection and diagnosis of transformer winding short circuit fault mainly relies on the combination of periodic offline detection and online monitoring. Traditional detection methods include winding resistance measurement, frequency response analysis, short circuit impedance test, etc. These methods collect electrical parameters of the transformer and analyze their change rules to determine whether the winding has short circuit fault.

[0003] The existing transformer winding short circuit fault detection and diagnosis technology has the following shortcomings: first, the traditional single detection method can only obtain local feature information, which is difficult to fully reflect the actual fault state of the transformer winding, and is prone to misjudgment or omission. Second, the existing diagnosis methods lack effective fusion of multi-source information, and fail to fully utilize the complementarity between various detection data, resulting in low reliability and accuracy of the diagnosis results. Third, the current fault diagnosis methods generally lack a scientific feature weight distribution mechanism, which makes it difficult to accurately reflect the importance of different feature quantities in fault diagnosis, affecting the accuracy of the final diagnosis results. SUMMARY

[0004] The embodiment of the present application provides a rapid detection and diagnosis method and system for transformer winding short circuit fault, which can solve the problems in the prior art.

[0005] In a first aspect, the embodiment of the present application provides a rapid detection and diagnosis method for transformer winding short circuit fault, comprising:

[0006] Collecting electrical parameters of the transformer winding, the electrical parameters including winding current, winding voltage and frequency response data; calculating the equivalent impedance of the transformer winding according to the electrical parameters, the equivalent impedance including winding resistance, leakage inductance and distributed capacitance;

[0007] Establishing a transfer function model of the transformer winding based on the equivalent impedance, extracting the zero-pole distribution characteristics of the transfer function to obtain frequency domain fault feature quantities; establishing an equivalent impedance model of the transformer winding, calculating the radial short circuit force and the axial short circuit force of the transformer winding under short circuit state, wherein the radial short circuit force is calculated according to the inner and outer winding currents, and the axial short circuit force is calculated according to the radial magnetic induction intensity; the frequency domain fault feature quantities, the radial short circuit force and the axial short circuit force jointly constitute the fault diagnosis feature quantities;

[0008] construct a characteristic quantity judgment matrix, elements in rows and columns of the characteristic quantity judgment matrix are the fault diagnosis characteristic quantities, calculate a characteristic vector of the characteristic quantity judgment matrix based on an analytic hierarchy process, and normalize the characteristic vector to obtain weight coefficients of each fault diagnosis characteristic quantity;

[0009] establish a multi-source information fusion model based on D-S evidence theory, input the fault diagnosis characteristic quantities and the weight coefficients into the multi-source information fusion model, and obtain a fusion result;

[0010] construct a fuzzy comprehensive evaluation matrix, compare the fusion result with an adjustable confidence threshold, and output a fault diagnosis result according to a comparison result.

[0011] establish a transfer function model of the transformer winding based on the equivalent impedance, extract zero-pole distribution characteristics of the transfer function, and obtain the frequency domain fault characteristic quantity, including:

[0012] calculate an input-output voltage ratio of the transformer winding in the frequency domain according to the equivalent impedance, and determine a transfer function of the transformer winding;

[0013] express the transfer function as a ratio of a numerator polynomial and a denominator polynomial, coefficients of the numerator polynomial are determined by a unit gain in a normal working state, and coefficients of the denominator polynomial are determined by the leakage inductance value, the distributed capacitance value and the winding resistance value;

[0014] solve the numerator polynomial and the denominator polynomial by using a Newton iteration method, obtain complex coordinates of the zero point and the pole point, the complex coordinates include a real part and an imaginary part, calculate amplitudes of the zero point and the pole point, and the amplitudes are calculated by a square root of a square sum of the real part and the imaginary part of the complex coordinates;

[0015] calculate change rates of the zero point and the pole point to a fault state parameter, take the change rates as sensitivity coefficients, sort the zero point and the pole point according to the sensitivity coefficients, and select complex coordinates and amplitudes of the zero point and the pole point with the largest sensitivity coefficients as the frequency domain fault characteristic quantity.

[0016] establish an equivalent impedance model of the transformer winding, and calculate radial short-circuit forces and axial short-circuit forces of the transformer winding in a short-circuit state, including:

[0017] establish an equivalent impedance model of the transformer winding, the equivalent impedance model includes a winding direct-current resistance, an eddy current loss equivalent resistance, a winding leakage inductance and a winding-to-ground and interlayer distributed capacitance, and calculate a leakage magnetic field distribution of the transformer winding by using a magnetic scalar potential method based on the equivalent impedance model;

[0018] Obtaining the effective value of inner and outer winding currents of a transformer winding, winding calculation height and inner and outer winding radius, calculating radial force density according to the leakage magnetic field distribution, and obtaining radial short-circuit force by integrating the radial force density along the winding height direction;

[0019] Obtaining the axial distribution of radial magnetic induction intensity, calculating axial force density, and obtaining axial short-circuit force by double integrating the axial force density along the radial and axial directions.

[0020] Establishing a multi-source information fusion model based on D-S evidence theory, inputting the fault diagnosis characteristic quantity and the weight coefficient into the multi-source information fusion model to obtain a fusion result, including:

[0021] Defining a transformer winding fault type set, the fault type set including inter-turn short-circuit fault, inter-layer short-circuit fault and winding deformation fault; constructing a basic probability assignment function based on the fault type set, the basic probability assignment function satisfying the condition that the sum of probabilities of all subsets is 1 and the probability of the empty set is 0;

[0022] Normalizing the fault diagnosis characteristic quantity, constructing the normalized fault diagnosis characteristic quantity as a plurality of evidence sources; calculating the evidence distance between the evidence sources, the evidence distance being calculated by the quadratic form of the basic probability assignment vector and the Jacobian matrix of the evidence source; calculating the credibility of each evidence source based on the evidence distance, the credibility being calculated by the normalized weighted average value of the evidence distance;

[0023] Fusing the weight coefficient and the credibility to obtain a comprehensive weight, and modifying the basic probability assignment of the evidence source according to the comprehensive weight; fusing the modified evidence source by using an improved D-S combination rule, the improved D-S combination rule considering the evidence conflict factor and eliminating the influence of evidence height conflict by normalization processing;

[0024] Calculating the trust degree function and the likelihood function of the fusion result, the trust degree function being obtained by probability accumulation of subset inclusion relationship, and the likelihood function being obtained by probability accumulation of subset intersection relationship; constructing the trust degree function and the likelihood function as an uncertainty interval;

[0025] Calculating the maximum difference value of the basic probability assignment of adjacent two iterations, outputting the final fusion result when the maximum difference value is less than a preset convergence threshold, and returning the calculation step of the evidence distance when the maximum difference value is greater than or equal to the preset convergence threshold.

[0026] Fusing the modified evidence source by using an improved D-S combination rule, and calculating the trust degree function and the likelihood function of the fusion result, including:

[0027] According to the credibility, the basic probability assignment data of each evidence source is corrected to obtain corrected basic probability assignment data; the corrected basic probability assignment data is substituted into an improved D-S combination rule for fusion, and a numerator of the improved D-S combination rule is a sum of products of the corrected basic probability assignment data when intersections of the evidence sources are equal to a target subset, and a denominator is 1 minus a conflict coefficient, and the conflict coefficient is a sum of products of the corrected basic probability assignment data when intersections of the evidence sources are empty sets;

[0028] Uncertainty analysis is performed on the fusion result, including: a trust degree function is calculated, and the trust degree function is a sum of basic probability assignments of all subsets of the target subset; a likelihood function is calculated, and the likelihood function is a sum of basic probability assignments of all subsets having intersections with the target subset;

[0029] A comprehensive uncertainty is calculated by subtracting the trust degree function from the likelihood function; an evidence conflict degree is calculated based on the conflict coefficient, and the evidence conflict degree is a ratio of the conflict coefficient to 1 minus the conflict coefficient; a credibility of the fusion result is calculated according to a ratio of the trust degree function to the likelihood function.

[0030] A fuzzy comprehensive evaluation matrix is constructed, and the fusion result is compared with an adjustable confidence threshold value, and a fault diagnosis result is output according to a comparison result, including:

[0031] A fuzzy membership matrix is constructed, a row vector of the fuzzy membership matrix represents a fault type, a column vector represents a fault degree, and a matrix element is calculated by a membership function, and the membership function adopts an adaptive fuzzy rule based on a kernel function;

[0032] A fuzzy comprehensive evaluation matrix is constructed based on the fusion result and the fuzzy membership matrix, and each element of the fuzzy comprehensive evaluation matrix is obtained by weighted operation of the fusion result and a corresponding membership;

[0033] A dynamic confidence threshold value is set, the dynamic confidence threshold value is adaptively adjusted according to a cumulative accuracy rate of fault diagnosis; a decision result of the fuzzy comprehensive evaluation matrix is compared with the dynamic confidence threshold value, when the decision result is greater than the dynamic confidence threshold value, a corresponding fault type is output, and when the decision result is less than or equal to the dynamic confidence threshold value, an uncertain state is determined.

[0034] In a second aspect of the embodiment of the application, a quick detection and diagnosis system for a transformer winding short circuit fault is provided, and the system includes:

[0035] The first unit is configured to collect electrical parameters of the transformer winding, the electrical parameters including winding current, winding voltage and frequency response data; and calculate equivalent impedance of the transformer winding according to the electrical parameters, the equivalent impedance including winding resistance, leakage inductance and distributed capacitance.

[0036] The second unit is configured to establish a transfer function model of the transformer winding based on the equivalent impedance, extract zero-pole distribution characteristics of the transfer function to obtain a frequency domain fault feature quantity; establish an equivalent impedance model of the transformer winding, and calculate radial short-circuit force and axial short-circuit force of the transformer winding in a short-circuit state, wherein the radial short-circuit force is calculated according to inner and outer winding currents, and the axial short-circuit force is calculated according to radial magnetic induction intensity; and combine the frequency domain fault feature quantity, the radial short-circuit force and the axial short-circuit force to form a fault diagnosis feature quantity.

[0037] The third unit is configured to construct a feature quantity judgment matrix, row elements and column elements of the feature quantity judgment matrix being the fault diagnosis feature quantity, calculate a feature vector of the feature quantity judgment matrix based on an analytic hierarchy process, and normalize the feature vector to obtain a weight coefficient of each fault diagnosis feature quantity.

[0038] The fourth unit is configured to establish a multi-source information fusion model based on D-S evidence theory, input the fault diagnosis feature quantity and the weight coefficient into the multi-source information fusion model, and obtain a fusion result.

[0039] The fifth unit is configured to construct a fuzzy comprehensive evaluation matrix, compare the fusion result with an adjustable confidence threshold, and output a fault diagnosis result according to a comparison result.

[0040] The third aspect of the embodiment of the application

[0041] An electronic device is provided, comprising:

[0042] a processor;

[0043] a memory for storing processor-executable instructions;

[0044] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0045] The fourth aspect of the embodiment of the application

[0046] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0047] The beneficial effects of the present application are as follows:

[0048] The present invention collects the electrical parameters of the transformer winding and establishes a transfer function model, and combines it with the equivalent impedance model to calculate the short-circuit force, thereby constructing a complete fault diagnosis feature system, which can comprehensively reflect the characteristic information of the transformer winding short-circuit fault and improve the accuracy and reliability of fault diagnosis.

[0049] The present invention adopts the hierarchical analysis method to calculate the eigenvectors of the feature quantity judgment matrix, and obtains the weight coefficients of each fault diagnosis feature quantity through normalization processing, so that the importance of different feature quantities in the fault diagnosis process is reasonably distributed, avoiding the one-sidedness and misjudgment that may be caused by the diagnosis of a single feature quantity.

[0050] The present invention establishes a multi-source information fusion model based on DS evidence theory, effectively fuses the information of multiple fault diagnosis feature quantities, and compares them with an adjustable confidence threshold through a fuzzy comprehensive evaluation matrix. It not only makes full use of various fault feature information, but also provides a flexible judgment standard, thereby realizing rapid and accurate diagnosis of transformer winding short-circuit faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 Schematic diagram of the flow of a method for rapid detection and diagnosis of transformer winding short-circuit faults according to an embodiment of the present invention;

[0052] Figure 2 Schematic diagram for comparing fusion performance under different conflict levels;

[0053] Figure 3 Schematic diagram comparing BPA before and after evidence source correction. DETAILED DESCRIPTION

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0055] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0056] Figure 1 FIG. 1 is a flow chart of a method for rapid detection and diagnosis of a transformer winding short-circuit fault according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0057] Collecting electrical parameters of the transformer winding, the electrical parameters including winding current, winding voltage, and frequency response data; calculating the equivalent impedance of the transformer winding based on the electrical parameters, the equivalent impedance including winding resistance, leakage inductance, and distributed capacitance;

[0058] A transfer function model of the transformer winding is established based on the equivalent impedance, and the zero-pole distribution characteristics of the transfer function are extracted to obtain a frequency domain fault feature quantity; an equivalent impedance model of the transformer winding is established, and the radial short-circuit force and axial short-circuit force of the transformer winding in a short-circuit state are calculated, wherein the radial short-circuit force is calculated based on the inner and outer winding currents, and the axial short-circuit force is calculated based on the radial magnetic induction intensity; the frequency domain fault feature quantity, the radial short-circuit force, and the axial short-circuit force are collectively used to form a fault diagnosis feature quantity;

[0059] Constructing a feature value judgment matrix, wherein the row elements and column elements of the feature value judgment matrix are all the fault diagnosis feature values, calculating the eigenvectors of the feature value judgment matrix based on the hierarchical analysis method, and normalizing the eigenvectors to obtain weight coefficients of each fault diagnosis feature value;

[0060] Establishing a multi-source information fusion model based on DS evidence theory, inputting the fault diagnosis feature quantity and the weight coefficient into the multi-source information fusion model to obtain a fusion result;

[0061] A fuzzy comprehensive evaluation matrix is ​​constructed, the fusion result is compared with an adjustable confidence threshold, and a fault diagnosis result is output according to the comparison result.

[0062] In an optional embodiment, a transfer function model of the transformer winding is established based on the equivalent impedance, and the zero-pole distribution characteristics of the transfer function are extracted to obtain the frequency domain fault feature quantity, which includes:

[0063] Calculating the input-output voltage ratio of the transformer winding in the frequency domain based on the equivalent impedance to determine the transfer function of the transformer winding;

[0064] The transfer function is expressed as a ratio of a numerator polynomial to a denominator polynomial, wherein the coefficient of the numerator polynomial is determined by the unity gain under normal working conditions, and the coefficient of the denominator polynomial is determined by the leakage inductance value, the distributed capacitance value, and the winding resistance value;

[0065] Solving the numerator polynomial and the denominator polynomial using a Newton iteration method to obtain complex coordinates of zeros and poles, the complex coordinates including real and imaginary parts; calculating amplitudes of the zeros and poles, the amplitudes being calculated as the square root of the sum of the squares of the real and imaginary parts of the complex coordinates;

[0066] Calculate the rate of change of the zero point and the pole to the fault state parameter, and use the rate of change as the sensitivity coefficient; sort the zero point and the pole according to the size of the sensitivity coefficient, and select the complex coordinates and amplitudes of the zero point and the pole with the largest sensitivity coefficient as the frequency domain fault feature.

[0067] This paper proposes a transformer winding fault diagnosis method based on equivalent impedance. This method first establishes a transfer function model for the transformer winding, then extracts the zero-pole distribution characteristics of the transfer function, and ultimately obtains frequency-domain fault signatures. The specific implementation steps are as follows:

[0068] The transfer function of the transformer winding is determined by calculating the frequency-domain input-output voltage ratio based on the equivalent impedance of the transformer winding. The equivalent impedance, which can be obtained through measurement or simulation, includes parameters such as leakage inductance, distributed capacitance, and winding resistance. The transfer function expression can be derived by expressing the ratio of input voltage to output voltage as a function of frequency.

[0069] The transfer function is expressed as the ratio of the numerator polynomial to the denominator polynomial. The coefficients of the numerator polynomial are determined by the unity gain under normal operating conditions and are usually set to 1. The coefficients of the denominator polynomial are determined by the leakage inductance, distributed capacitance, and winding resistance.

[0070] The Newton iteration method is used to solve the numerator and denominator polynomials to obtain the complex coordinates of the zeros and poles. The Newton iteration method is a commonly used numerical solution method that approximates the roots of a polynomial through multiple iterations. For each zero and pole, its complex coordinates are obtained, including both real and imaginary parts. For example, a zero can be represented as -2+3j.

[0071] Calculate the magnitude of the zeros and poles. The magnitude is calculated by taking the square root of the sum of the squares of the real and imaginary parts of the complex coordinates. Taking the zero point above as an example, its magnitude is sqrt((-2)^2+3^2)≈

[0072] 3.61. The amplitude reflects the distance of the zero and pole from the origin on the complex plane and has a significant impact on the system characteristics.

[0073] Calculate the rate of change of the zero and pole to the fault condition parameters and use this rate as the sensitivity coefficient. Specifically, vary the leakage inductance, distributed capacitance, and winding resistance, observe the changes in the positions of the zero and pole, and calculate the rate of change. For example, if the leakage inductance increases by 10%, the real part of a pole changes from -2 to -2.2, and the sensitivity coefficient of this pole to the leakage inductance is 0.1.

[0074] The zeros and poles are sorted according to the size of the sensitivity coefficient, and the complex coordinates and amplitudes of the zeros and poles with the largest sensitivity coefficients are selected as the frequency domain fault characteristics. These characteristics are most sensitive to faults and can effectively reflect the fault status of the transformer winding.

[0075] In practical applications, by comparing frequency-domain fault signatures under normal and fault conditions, it is possible to determine whether a transformer winding fault has occurred, as well as the type and severity of the fault. For example, if the real part of a pole changes from -2 to -1.5, the imaginary part changes from 3 to 2.8, and the amplitude changes from 3.61 to 3.17, this may indicate a change in winding resistance, necessitating further inspection and maintenance.

[0076] This method achieves accurate diagnosis of transformer winding faults by establishing a transfer function model of the transformer winding and extracting the zero-pole distribution characteristics. Compared with traditional methods, this method can more comprehensively reflect the electrical characteristics of the transformer winding and improve the accuracy and reliability of fault diagnosis.

[0077] The Newton iteration method is used to solve the zeros and poles of the transfer function and calculate their complex coordinates and amplitudes, providing rich information for fault feature extraction. This method not only considers the location of the zeros and poles, but also their amplitudes, which can more comprehensively describe the system characteristics and help improve the sensitivity of fault diagnosis.

[0078] By calculating the rate of change of zero-pole points to fault state parameters and introducing the concept of sensitivity coefficient, the optimal selection of fault characteristic quantities is achieved. This method can effectively identify the characteristic quantity that is most sensitive to the fault, improve the efficiency and accuracy of fault diagnosis, and also provide a basis for judging the fault type.

[0079] In an optional embodiment, establishing an equivalent impedance model of a transformer winding and calculating a radial short-circuit force and an axial short-circuit force of the transformer winding in a short-circuit state include:

[0080] Establishing an equivalent impedance model of the transformer winding, the equivalent impedance model includes the winding DC resistance, eddy current loss equivalent resistance, winding leakage inductance, and winding-to-ground and interlayer distributed capacitance; calculating the leakage magnetic field distribution of the transformer winding using the magnetic scalar potential method based on the equivalent impedance model;

[0081] Obtaining the effective values ​​of the inner and outer winding currents, the calculated winding height, and the inner and outer winding radii of the transformer winding, calculating the radial force density based on the leakage magnetic field distribution, and obtaining the radial short-circuit force by integrating the radial force density along the winding height direction;

[0082] The axial distribution of the radial magnetic induction intensity is obtained, the axial force density is calculated, and the axial short-circuit force is obtained by performing double integration of the axial force density along the radial and axial directions.

[0083] The present invention provides a method for establishing an equivalent impedance model of a transformer winding and calculating the short-circuit force. First, an equivalent impedance model of the transformer winding is established. The model includes the winding DC resistance, eddy current loss equivalent resistance, winding leakage inductance, and the winding-to-ground and interlayer distributed capacitance. These parameters can be obtained through actual measurement or theoretical calculation. For example, the winding DC resistance can be measured using a DC resistance tester, the eddy current loss equivalent resistance can be obtained through an AC impedance test, the winding leakage inductance can be calculated through a short-circuit test, and the distributed capacitance can be measured using a capacitance tester.

[0084] Based on the established equivalent impedance model, the magnetic scalar potential method is used to calculate the leakage magnetic field distribution of the transformer winding. This method is an effective numerical calculation method that can accurately describe the magnetic field distribution in the transformer winding. In this step, the winding is first divided into several small units, and the magnetic scalar potential is calculated for each unit. By solving the Laplace equation for the magnetic scalar potential, the magnetic field distribution of the entire winding space can be obtained.

[0085] Obtain the RMS current values ​​for the inner and outer windings, the calculated winding height, and the inner and outer winding radii of the transformer windings. These parameters can be obtained from transformer design drawings or actual measurements. For example, for a typical 110kV / 35kV transformer, the inner winding RMS current might be 500A, the outer winding RMS current 1500A, the calculated winding height 2m, the inner winding radius 0.5m, and the outer winding radius 0.6m.

[0086] Based on the calculated leakage magnetic field distribution, the radial force density is calculated. The radial force density is generated by the interaction between the current in the winding and the axial magnetic field. By integrating the radial force density along the winding height, the radial short-circuit force can be obtained. This step can be performed using numerical integration methods such as the trapezoidal method or the Simpson method.

[0087] Obtain the axial distribution of the radial magnetic flux density. This can be obtained by processing the previously calculated leakage magnetic field distribution. Next, calculate the axial force density. The axial force density is generated by the interaction between the current in the winding and the radial magnetic field. By double integrating the axial force density along the radial and axial directions, the axial short-circuit force can be obtained. This step can also be performed using numerical integration methods.

[0088] In practical applications, short-circuit force calculations can be considered under different operating conditions. For example, different short-circuit current levels can be simulated, such as 5 times or 10 times the rated current, and the corresponding short-circuit force can be calculated. Furthermore, the effect of temperature on winding parameters can be considered, and short-circuit force calculations can be performed at different temperatures to assess the impact of temperature changes on the transformer's short-circuit withstand capability.

[0089] Through the above method, the mechanical stress of transformer windings under short-circuit conditions can be comprehensively evaluated, providing important reference for the design and operation of transformers.

[0090] The method improves the accuracy of transformer winding leakage magnetic field distribution calculation by establishing a detailed equivalent impedance model, thereby more accurately evaluating the short-circuit force. This is of great significance for the design optimization and safe operation of transformers.

[0091] Using the magnetic scalar potential method to calculate the leakage magnetic field distribution not only has high computational efficiency, but also can handle complex winding structures, making it suitable for various types of transformers. The versatility of this method makes it widely applicable to short-circuit force calculation for transformers of different specifications and structures.

[0092] The method takes into account both radial and axial short-circuit forces, providing comprehensive short-circuit force evaluation. This helps identify weak links in transformer windings, guides structural optimization design, and improves the short-circuit bearing capacity and operation reliability of transformers.

[0093] In an optional implementation, a multi-source information fusion model based on D-S evidence theory is established, the fault diagnosis feature quantity and the weight coefficient are input into the multi-source information fusion model, and a fusion result is obtained, which includes:

[0094] A set of transformer winding fault types is defined, including inter-turn short circuit fault, inter-layer short circuit fault, and winding deformation fault; a basic probability assignment function is constructed based on the set of fault types, which satisfies the condition that the sum of all subset probabilities is 1 and the empty set probability is 0;

[0095] The fault diagnosis feature quantity is normalized, and the normalized fault diagnosis feature quantity is constructed as multiple evidence sources; the evidence distance between the evidence sources is calculated, which is calculated by the quadratic form of the basic probability assignment vector and the Jacobian matrix of the evidence source; the credibility of each evidence source is calculated based on the evidence distance, which is calculated by the normalized weighted average value of the evidence distance;

[0096] The weight coefficient and the credibility are fused to obtain a comprehensive weight, and the basic probability assignment of the evidence source is modified according to the comprehensive weight; the modified evidence source is fused by using an improved D-S combination rule, which considers the evidence conflict factor,

[0097] The influence of high evidence conflict is eliminated by normalization processing;

[0098] Calculating a confidence function and a likelihood function of the fusion result, wherein the confidence function is obtained by accumulating the probabilities of subset inclusion relations, and the likelihood function is obtained by accumulating the probabilities of subset intersection relations; constructing the confidence function and the likelihood function into uncertainty intervals;

[0099] Calculate the maximum difference between the basic probability distributions of two adjacent iterations, and output the final fusion result when the maximum difference is less than a preset convergence threshold. When the maximum difference is greater than or equal to the preset convergence threshold, return to the step of calculating the evidence distance.

[0100] This paper proposes a multi-source information fusion model based on DS evidence theory for transformer winding fault diagnosis. The model fuses multiple fault diagnosis features and weight coefficients to obtain the final fusion result. The specific implementation steps are as follows:

[0101] Define a set of transformer winding fault types. This set includes three types: turn-to-turn short circuit faults, layer-to-layer short circuit faults, and winding deformation faults. Based on this set of fault types, construct a basic probability distribution function. This function must satisfy two conditions: the sum of all subset probabilities is 1, and the probability of the empty set is 0.

[0102] Normalize the fault diagnosis features. The purpose of normalization is to unify features of different dimensions onto the same scale for ease of subsequent processing. The normalized fault diagnosis features are then constructed into multiple evidence sources. Each evidence source represents a fault diagnosis feature.

[0103] The evidence distance reflects the degree of difference between different evidence sources. It is calculated using the quadratic form of the basic probability distribution vector and the Jacobian matrix of the evidence source. Specifically, the Jacobian matrix is ​​first constructed, whose elements represent the similarity between different fault types. Then, the basic probability distribution vector is multiplied by the Jacobian matrix to obtain the quadratic form, which is the evidence distance.

[0104] Based on the calculated evidence distances, the credibility of each evidence source is calculated. Credibility indicates the importance of that evidence source in the overall fusion process. This is calculated by taking the normalized weighted average of the evidence distances. Specifically, the evidence distances are first normalized, and then the weighted average is calculated to obtain the credibility of each evidence source.

[0105] The weight coefficients are combined with the credibility to obtain a comprehensive weight. This step aims to comprehensively consider the pre-set weight coefficients and the credibility calculated based on the evidence distance to obtain a more reasonable weight distribution. The fusion method can use weighted averaging or other appropriate fusion algorithms.

[0106] Based on the obtained comprehensive weights, the basic probability distribution of the evidence sources is revised. The purpose of the revision is to adjust the influence of each evidence source in the fusion process, so that the evidence sources with high credibility have greater weight.

[0107] The modified evidence sources are fused using an improved DS combination rule. The traditional DS combination rule can produce unreasonable results when dealing with highly conflicting evidence. Therefore, this method adopts an improved DS combination rule. This rule takes into account conflicting evidence and eliminates the impact of highly conflicting evidence through normalization. In its implementation, the conflict degree between the evidence sources is first calculated, then the combination rule is modified based on the conflict degree, and finally normalization is performed.

[0108] Calculate the confidence and likelihood functions of the fusion results. The confidence function represents the degree of support for a particular fault type and is obtained by summing the probabilities of subset inclusion relationships. The likelihood function represents the likelihood of a particular fault type and is obtained by summing the probabilities of subset intersection relationships. Construct the confidence and likelihood functions as uncertainty intervals to represent the uncertainty of the fusion results.

[0109] Calculate the maximum difference between the basic probability distributions of two consecutive iterations. Compare this difference with the preset convergence threshold. If the maximum difference is less than the preset convergence threshold, output the final fusion result. If the maximum difference is greater than or equal to the preset convergence threshold, return to the step of calculating the evidence distance and proceed to the next round of iteration.

[0110] To better understand the implementation process of this method, a specific data case is given below:

[0111] Assume that there are three types of faults: turn-to-turn short circuit fault (A), layer-to-layer short circuit fault (B), and winding deformation fault (C). There are three fault diagnosis characteristics: partial discharge, winding resistance, and short circuit impedance.

[0112] First, the three feature quantities are normalized, and the normalized values ​​are: [0.8, 0.6, 0.7], [0.5, 0.7, 0.6] and [0.7, 0.5, 0.8] respectively.

[0113] Then, construct the basic probability distribution function. Assume that the initial basic probability distribution is:

[0114] m1(A)=0.4, m1(B)=0.3, m1(C)=0.2, m1(A,B,C)=0.1;

[0115] m2(A)=0.3, m2(B)=0.4, m2(C)=0.2, m2(A,B,C)=0.1;

[0116] m3(A)=0.3, m3(B)=0.2, m3(C)=0.4, m3(A,B,C)=0.1;

[0117] Calculate the evidence distance. Assume that the obtained evidence distance matrix is:

[0118] [0,0.15,0.2];

[0119] [0.15,0,0.25];

[0120] [0.2,0.25,0];

[0121] Calculate the credibility based on the evidence distance. Assume that the credibility is: [0.4, 0.3, 0.3];

[0122] Assume that the pre-set weight coefficients are: [0.35, 0.35, 0.3]. Combining the weight coefficients with the credibility, we get the comprehensive weight: [0.375, 0.325, 0.3];

[0123] The basic probability distribution is revised according to the comprehensive weight to obtain the revised basic probability distribution.

[0124] The improved DS combination rule is used for fusion. Assume that the fusion result is:

[0125] m(A)=0.45, m(B)=0.35, m(C)=0.15, m(A,B,C)=0.05;

[0126] Calculate the confidence function and likelihood function and construct the uncertainty interval:

[0127] Bel(A)=0.45, Pl(A)=0.5;

[0128] Bel(B)=0.35, Pl(B)=0.4;

[0129] Bel(C)=0.15, Pl(C)=0.2;

[0130] Finally, the maximum difference between the basic probability distributions of two adjacent iterations is calculated and compared with the preset convergence threshold. If the convergence condition is met, the final result is output; otherwise, the iteration continues.

[0131] By fusing multiple fault diagnosis features, the information contained in each feature is fully utilized, improving the accuracy and reliability of fault diagnosis. By considering the correlation and conflict between different features, the potential misjudgment caused by a single feature is effectively reduced. The improved DS combination rule effectively resolves the paradox that can arise in traditional DS theory when dealing with highly conflicting evidence. By introducing the concepts of evidence distance and credibility, the evidence sources are appropriately weighted, improving the credibility of the fusion results. The fusion results are optimized through multiple iterations until the preset convergence conditions are met. This approach enables continuous adjustment and optimization of the fusion process, ultimately resulting in more stable and reliable diagnostic results. At the same time, by constructing uncertainty intervals, decision makers are provided with more comprehensive fault diagnosis information, helping them make more reasonable maintenance decisions.

[0132] In an optional embodiment, the modified evidence sources are fused using an improved DS combination rule, and the confidence function and likelihood function of the fusion result are calculated, including:

[0133] The basic probability distribution data of each evidence source is modified according to the credibility to obtain modified basic probability distribution data; the modified basic probability distribution data is substituted into the improved DS combination rule for fusion, wherein the numerator of the improved DS combination rule is: the sum of the products of the modified basic probability distribution data when the intersection of each evidence source is equal to the target subset, and the denominator is: 1 minus the conflict coefficient, wherein the conflict coefficient is the sum of the products of the modified basic probability distribution data when the intersection of each evidence source is an empty set;

[0134] Perform uncertainty analysis on the fusion result, including: calculating a confidence function, which is the sum of the basic probability distribution of all subsets of the target subset; calculating a likelihood function, which is the sum of the basic probability distribution of all subsets that have an intersection with the target subset;

[0135] The difference between the trust function and the likelihood function is calculated to obtain a comprehensive uncertainty; the evidence conflict degree is calculated based on the conflict coefficient, where the evidence conflict degree is the ratio of the conflict coefficient to 1 minus the conflict coefficient; and the credibility of the fusion result is calculated according to the ratio of the trust function to the likelihood function.

[0136] Correct the basic probability distribution data for each source of evidence. Adjust the original basic probability distribution for each source of evidence based on a predetermined credibility indicator. For example, assume there are three sources of evidence: E1, E2, and E3, with credibility of 0.9, 0.8, and 0.7, respectively. For the identification framework Θ = {A, B, C}, the original basic probability distribution for E1 is m1(A) = 0.5, m1(B) = 0.3, and m1(C) = 0.2. The corrected basic probability distribution is then: m1'(A) = 0.5*0.9 = 0.45, m1'(B) = 0.3*0.9 = 0.27, and m1'(C) = 0.2*0.9 = 0.18. Correct E2 and E3 in the same way.

[0137] The modified evidence sources are fused using the improved DS combination rule. The numerator of the improved combination rule is the sum of the products of the modified basic probability distributions when the intersection of each evidence source is equal to the target subset. For example, for the target subset A, the numerator is m1'(A)*m2'(A)*m3'(A)+

[0138] m1'(A)*m2'(A)*m3'(Θ)+m1'(A)*m2'(Θ)*m3'(A)+

[0139] m1'(Θ)*m2'(A)*m3'(A). The denominator is 1 minus the conflict coefficient k, where k is the sum of the products of the corrected basic probability distributions when the intersection of each evidence source is an empty set. This approach can effectively reduce unreasonable fusion results caused by evidence conflicts.

[0140] Perform uncertainty analysis on the fusion results. First, calculate the confidence function Bel(A), which is the sum of the basic probability distributions of all subsets of the target subset A. For A = {a, b}, Bel(A) = m(a) + m(b) + m({a, b}). Next, calculate the likelihood function Pl(A), which is the sum of the basic probability distributions of all subsets that intersect with the target subset A. For A = {a, b}, Pl(A) = m(a) + m(b) + m({a, b}) + m({a, c}) + m({b, c}) + m({a, b, c}).

[0141] Next, we calculate the difference between the confidence function and the likelihood function to obtain the combined uncertainty. For example, for subset A, the combined uncertainty is Pl(A) - Bel(A). The larger this value, the more uncertain our judgment of A is.

[0142] Furthermore, the evidence conflict degree is calculated based on the conflict coefficient k. The evidence conflict degree is defined as k / (1-k), where k is the conflict coefficient calculated previously. This metric reflects the degree of contradiction between multiple sources of evidence, with larger values ​​indicating more severe conflict.

[0143] Finally, the fusion result credibility is calculated according to the ratio of the credibility function and the likelihood function. The credibility is defined as Bel(A) / Pl(A). The closer the value is to 1, the more credible the judgment of A is.

[0144] Through the above steps, more reliable and accurate multi-source information fusion results can be obtained, thereby providing strong support for subsequent decision-making.

[0145] Figure 2 In order to compare the fusion performance under different conflict degrees, the basic probability assignment (BPA) of the three evidence sources is modified based on the credibility. The BPA distribution before and after the modification under the Monte Carlo simulation is shown in the figure. It can be seen that the modification reduces the original BPA value. Figure 3 The BPA comparison diagram before and after the modification of the evidence source is shown in Figure 2. Figure 3 The fusion accuracy of the traditional D-S and the improved D-S under different conflict degrees is compared. The horizontal axis is the evidence conflict degree, and the vertical axis is the fusion accuracy. It can be seen that as the conflict degree increases, the advantage of the improved D-S method is more obvious.

[0146] In an optional embodiment, a fuzzy comprehensive evaluation matrix is constructed, the fusion result is compared with an adjustable confidence threshold, and a fault diagnosis result is output according to the comparison result, including:

[0147] A fuzzy membership matrix is constructed, the row vector of the fuzzy membership matrix represents a fault type, the column vector represents a fault degree, and the matrix element is calculated by a membership function. The membership function adopts an adaptive fuzzy rule based on a kernel function;

[0148] A fuzzy comprehensive evaluation matrix is constructed based on the fusion result and the fuzzy membership matrix. Each element of the fuzzy comprehensive evaluation matrix is obtained by weighted operation of the fusion result and the corresponding membership;

[0149] A dynamic confidence threshold is set, the dynamic confidence threshold is adaptively adjusted according to the cumulative accuracy rate of fault diagnosis, the decision result of the fuzzy comprehensive evaluation matrix is compared with the dynamic confidence threshold, when the decision result is greater than the dynamic confidence threshold, the corresponding fault type is output, and when the decision result is less than or equal to the dynamic confidence threshold, an uncertain state is determined.

[0150] The present application provides a fault diagnosis method based on fuzzy comprehensive evaluation. The method first constructs a fuzzy membership matrix, then constructs a fuzzy comprehensive evaluation matrix based on the fusion result and the fuzzy membership matrix, and finally compares the dynamic confidence threshold to obtain the fault diagnosis result.

[0151] The row vectors of the fuzzy membership matrix represent different fault types, and the column vectors represent the severity of the faults. Each element in the matrix is calculated by a membership function. This method uses adaptive fuzzy rules based on kernel functions to construct the membership function. Specifically, a Gaussian kernel function can be chosen as the basis, and the parameters of the kernel function are automatically adjusted based on historical data, so that the membership function can adaptively reflect the characteristics of different fault types and degrees.

[0152] For example, assuming there are 3 fault types (A, B, C) and 4 fault severity levels (mild, moderate, severe, extreme), a 3x4 fuzzy membership matrix can be constructed. Each element in the matrix represents the membership of a certain fault type to a certain degree level. An example matrix calculated by an adaptive kernel function may be as follows:

[0153] A: [0.8, 0.5, 0.2, 0.1];

[0154] B: [0.6, 0.7, 0.4, 0.2];

[0155] C: [0.3, 0.6, 0.8, 0.5];

[0156] Based on the fusion result and the fuzzy membership matrix, a fuzzy comprehensive evaluation matrix is constructed. The fusion result is the preliminary judgment obtained by multiple fault detection methods, which can be represented as a vector. Weighted operation of this vector and the fuzzy membership matrix can obtain the fuzzy comprehensive evaluation matrix. The weighted method can use weighted average or weighted sum, etc.

[0157] For example, assuming the fusion result is [0.5, 0.3, 0.2], representing the preliminary judgment probability of the three fault types. Weighted average operation with the above fuzzy membership matrix can obtain the fuzzy comprehensive evaluation matrix:

[0158] [0.65, 0.56, 0.32, 0.20];

[0159] This matrix represents the final evaluation result of the system on the four fault severity levels after considering the preliminary judgment and membership.

[0160] The dynamic confidence threshold is adaptively adjusted with the cumulative accuracy rate of fault diagnosis. The sliding window method can be used to calculate the accuracy rate of the last N diagnoses, and the threshold is dynamically adjusted according to the trend of the accuracy rate. For example, when the accuracy rate continues to rise, the threshold can be appropriately increased to reduce false positives; when the accuracy rate decreases, the threshold is lowered to increase the sensitivity of diagnosis.

[0161] Compare the decision results of the fuzzy comprehensive evaluation matrix with the dynamic confidence threshold. The maximum value in the matrix is ​​selected as the decision result. If this value is greater than the dynamic confidence threshold, the corresponding fault type and severity are output. If it is less than or equal to the threshold, the state is determined to be uncertain, requiring further observation or other diagnostic methods.

[0162] For example, suppose the current dynamic confidence threshold is 0.6, and the maximum value of the fuzzy comprehensive evaluation matrix is ​​0.65, corresponding to a "minor" level. Since 0.65 > 0.6, the diagnostic result can be output as "minor fault." If the maximum value is 0.56, it is less than the threshold and should be considered an uncertain state.

[0163] Through the above steps, this method realizes intelligent fault diagnosis based on fuzzy comprehensive evaluation. This method takes into account multiple factors, can effectively deal with the uncertainty and fuzziness in fault diagnosis, and improves the accuracy and reliability of diagnosis.

[0164] By constructing a fuzzy membership matrix and a fuzzy comprehensive evaluation matrix, the uncertainty and ambiguity in the fault diagnosis process are effectively addressed. Adopting adaptive fuzzy rules based on kernel functions enables the membership function to automatically adjust based on historical data, thereby more accurately reflecting the characteristics of different fault types and severity. This adaptive capability significantly improves the accuracy and robustness of fault diagnosis.

[0165] A dynamic confidence threshold mechanism has been introduced, which adaptively adjusts based on the cumulative diagnostic accuracy. This dynamic adjustment mechanism enables the system to maintain high accuracy while flexibly responding to fault diagnosis needs under varying operating conditions. When system performance is stable, the threshold can be raised to reduce false positives; when system performance fluctuates, the threshold can be lowered to increase diagnostic sensitivity. This adaptive capability significantly enhances the practicality and reliability of the fault diagnosis system.

[0166] By comprehensively considering the fusion results of multiple fault detection methods and combining them with a fuzzy membership matrix, a comprehensive fuzzy comprehensive evaluation matrix was constructed. This multi-source information fusion method fully leverages the advantages of various fault detection technologies, enabling a more comprehensive capture of fault characteristics, thereby improving the accuracy and reliability of fault diagnosis. Furthermore, by introducing the judgment of uncertain states, forced judgments when confidence levels are insufficient are avoided, further enhancing the reliability of the diagnostic results.

[0167] The rapid detection and diagnosis system for transformer winding short-circuit faults according to the embodiment of the present invention includes:

[0168] The first unit is configured to collect electrical parameters of the transformer winding, the electrical parameters including winding current, winding voltage and frequency response data; and calculate equivalent impedance of the transformer winding according to the electrical parameters, the equivalent impedance including winding resistance, leakage inductance and distributed capacitance.

[0169] The second unit is configured to establish a transfer function model of the transformer winding based on the equivalent impedance, extract zero-pole distribution characteristics of the transfer function to obtain a frequency domain fault feature quantity; establish an equivalent impedance model of the transformer winding, and calculate radial short-circuit force and axial short-circuit force of the transformer winding in a short-circuit state, wherein the radial short-circuit force is calculated according to inner and outer winding currents, and the axial short-circuit force is calculated according to radial magnetic induction intensity; and combine the frequency domain fault feature quantity, the radial short-circuit force and the axial short-circuit force to form a fault diagnosis feature quantity.

[0170] The third unit is configured to construct a feature quantity judgment matrix, row elements and column elements of the feature quantity judgment matrix being the fault diagnosis feature quantity, calculate a feature vector of the feature quantity judgment matrix based on an analytic hierarchy process, and normalize the feature vector to obtain a weight coefficient of each fault diagnosis feature quantity.

[0171] The fourth unit is configured to establish a multi-source information fusion model based on D-S evidence theory, input the fault diagnosis feature quantity and the weight coefficient into the multi-source information fusion model, and obtain a fusion result.

[0172] The fifth unit is configured to construct a fuzzy comprehensive evaluation matrix, compare the fusion result with an adjustable confidence threshold, and output a fault diagnosis result according to a comparison result.

[0173] A third aspect of the embodiment of the application,

[0174] An electronic device is provided, comprising:

[0175] a processor;

[0176] a memory for storing processor-executable instructions;

[0177] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0178] A fourth aspect of the embodiment of the application,

[0179] A computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0180] The present application can be a method, an apparatus, a system, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for performing various aspects of the present application.

[0181] Finally, it should be noted that the above-described embodiments are merely intended to illustrate the technical solutions of the present application, and are not intended to limit the present application; even though the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some or all of the technical features thereof; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A rapid detection and diagnosis method for transformer winding short-circuit faults, characterized in that: include: Collecting electrical parameters of the transformer winding, wherein the electrical parameters include winding current, winding voltage and frequency response data; Calculating the equivalent impedance of the transformer winding according to the electrical parameters, wherein the equivalent impedance includes winding resistance, leakage inductance and distributed capacitance; A transfer function model of the transformer winding is established based on the equivalent impedance, and the zero-pole distribution characteristics of the transfer function are extracted to obtain a frequency domain fault feature quantity; an equivalent impedance model of the transformer winding is established, and the radial short-circuit force and axial short-circuit force of the transformer winding in a short-circuit state are calculated, wherein the radial short-circuit force is calculated based on the inner and outer winding currents, and the axial short-circuit force is calculated based on the radial magnetic induction intensity; the frequency domain fault feature quantity, the radial short-circuit force, and the axial short-circuit force are collectively used to form a fault diagnosis feature quantity; Constructing a feature value judgment matrix, wherein the row elements and column elements of the feature value judgment matrix are all the fault diagnosis feature values, calculating the eigenvectors of the feature value judgment matrix based on the hierarchical analysis method, and normalizing the eigenvectors to obtain weight coefficients of each fault diagnosis feature value; Establishing a multi-source information fusion model based on DS evidence theory, inputting the fault diagnosis feature quantity and the weight coefficient into the multi-source information fusion model to obtain a fusion result; A fuzzy comprehensive evaluation matrix is ​​constructed, the fusion result is compared with an adjustable confidence threshold, and a fault diagnosis result is output according to the comparison result.

2. The method according to claim 1, characterized in that A transfer function model of the transformer winding is established based on the equivalent impedance, and the zero-pole distribution characteristics of the transfer function are extracted to obtain the frequency domain fault feature quantity including: Calculating the input-output voltage ratio of the transformer winding in the frequency domain based on the equivalent impedance to determine the transfer function of the transformer winding; The transfer function is expressed as a ratio of a numerator polynomial to a denominator polynomial, wherein the coefficient of the numerator polynomial is determined by the unity gain under normal working conditions, and the coefficient of the denominator polynomial is determined by the leakage inductance value, the distributed capacitance value, and the winding resistance value; Solving the numerator polynomial and the denominator polynomial using a Newton iteration method to obtain complex coordinates of zeros and poles, the complex coordinates including real and imaginary parts; calculating amplitudes of the zeros and poles, the amplitudes being calculated as the square root of the sum of the squares of the real and imaginary parts of the complex coordinates; Calculate the rate of change of the zero point and the pole to the fault state parameter, and use the rate of change as the sensitivity coefficient; sort the zero point and the pole according to the size of the sensitivity coefficient, and select the complex coordinates and amplitudes of the zero point and the pole with the largest sensitivity coefficient as the frequency domain fault feature.

3. The method according to claim 1, characterized in that Establishing the equivalent impedance model of the transformer winding and calculating the radial short-circuit force and axial short-circuit force of the transformer winding under the short-circuit state include: Establishing an equivalent impedance model of the transformer winding, the equivalent impedance model includes the winding DC resistance, eddy current loss equivalent resistance, winding leakage inductance, and winding-to-ground and interlayer distributed capacitance; calculating the leakage magnetic field distribution of the transformer winding using the magnetic scalar potential method based on the equivalent impedance model; Obtaining the effective values ​​of the inner and outer winding currents, the calculated winding height, and the inner and outer winding radii of the transformer winding, calculating the radial force density based on the leakage magnetic field distribution, and obtaining the radial short-circuit force by integrating the radial force density along the winding height direction; The axial distribution of the radial magnetic induction intensity is obtained, the axial force density is calculated, and the axial short-circuit force is obtained by performing double integration of the axial force density along the radial and axial directions.

4. The method according to claim 1, wherein A multi-source information fusion model based on DS evidence theory is established, and the fault diagnosis feature quantity and the weight coefficient are input into the multi-source information fusion model to obtain the fusion results including: Defining a set of transformer winding fault types, the set of fault types including turn-to-turn short circuit faults, layer-to-layer short circuit faults, and winding deformation faults; constructing a basic probability distribution function based on the set of fault types, the basic probability distribution function satisfying the conditions that the sum of all subset probabilities is 1 and the probability of the empty set is 0; Normalizing the fault diagnosis feature quantities and constructing the normalized fault diagnosis feature quantities into multiple evidence sources; calculating the evidence distance between the evidence sources, where the evidence distance is calculated by a quadratic form of a basic probability distribution vector and a Jacobian matrix of the evidence source; and calculating the credibility of each evidence source based on the evidence distance, where the credibility is calculated by a normalized weighted average of the evidence distances. The weight coefficient is combined with the credibility to obtain a comprehensive weight, and the basic probability distribution of the evidence source is modified according to the comprehensive weight; the modified evidence source is combined using an improved DS combination rule, the improved DS combination rule taking into account evidence conflict factors and eliminating the influence of highly conflicting evidence through normalization processing; Calculating a confidence function and a likelihood function of the fusion result, wherein the confidence function is obtained by accumulating the probabilities of subset inclusion relations, and the likelihood function is obtained by accumulating the probabilities of subset intersection relations; constructing the confidence function and the likelihood function into uncertainty intervals; Calculate the maximum difference between the basic probability distributions of two adjacent iterations, and output the final fusion result when the maximum difference is less than a preset convergence threshold. When the maximum difference is greater than or equal to the preset convergence threshold, return to the step of calculating the evidence distance.

5. The method according to claim 4, characterized in that The modified evidence sources are fused using the improved DS combination rule. The confidence function and likelihood function of the fusion result are calculated as follows: The basic probability distribution data of each evidence source is modified according to the credibility to obtain modified basic probability distribution data; the modified basic probability distribution data is substituted into the improved DS combination rule for fusion, wherein the numerator of the improved DS combination rule is: the sum of the products of the modified basic probability distribution data when the intersection of each evidence source is equal to the target subset, and the denominator is: 1 minus the conflict coefficient, wherein the conflict coefficient is the sum of the products of the modified basic probability distribution data when the intersection of each evidence source is an empty set; Perform uncertainty analysis on the fusion result, including: calculating a confidence function, which is the sum of the basic probability distribution of all subsets of the target subset; calculating a likelihood function, which is the sum of the basic probability distribution of all subsets that have an intersection with the target subset; The difference between the trust function and the likelihood function is calculated to obtain a comprehensive uncertainty; the evidence conflict degree is calculated based on the conflict coefficient, where the evidence conflict degree is the ratio of the conflict coefficient to 1 minus the conflict coefficient; and the credibility of the fusion result is calculated according to the ratio of the trust function to the likelihood function.

6. The method according to claim 1, characterized in that Construct a fuzzy comprehensive evaluation matrix, compare the fusion result with the adjustable confidence threshold, and output the fault diagnosis results according to the comparison results, including: Constructing a fuzzy membership matrix, wherein the row vectors of the fuzzy membership matrix represent the fault type, the column vectors represent the fault degree, and the matrix elements are calculated by a membership function, wherein the membership function adopts an adaptive fuzzy rule based on a kernel function; Constructing a fuzzy comprehensive evaluation matrix based on the fusion result and the fuzzy membership matrix, wherein each element of the fuzzy comprehensive evaluation matrix is ​​obtained by weighted operation of the fusion result and the corresponding membership; A dynamic confidence threshold is set, and the dynamic confidence threshold is adaptively adjusted according to the cumulative accuracy of fault diagnosis; the decision result of the fuzzy comprehensive evaluation matrix is ​​compared with the dynamic confidence threshold, and when the decision result is greater than the dynamic confidence threshold, the corresponding fault type is output; when the decision result is less than or equal to the dynamic confidence threshold, it is determined to be an uncertain state.

7. A rapid detection and diagnosis system for transformer winding short-circuit faults, for implementing the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to collect electrical parameters of the transformer winding, wherein the electrical parameters include winding current, winding voltage and frequency response data; Calculating the equivalent impedance of the transformer winding according to the electrical parameters, wherein the equivalent impedance includes winding resistance, leakage inductance and distributed capacitance; The second unit is used to establish a transfer function model of the transformer winding based on the equivalent impedance, extract the zero-pole distribution characteristics of the transfer function, and obtain a frequency domain fault feature; establish an equivalent impedance model of the transformer winding, calculate the radial short-circuit force and axial short-circuit force of the transformer winding in a short-circuit state, wherein the radial short-circuit force is calculated based on the inner and outer winding currents, and the axial short-circuit force is calculated based on the radial magnetic induction intensity; the frequency domain fault feature, the radial short-circuit force, and the axial short-circuit force are collectively used to form a fault diagnosis feature; The third unit is used to construct a feature value judgment matrix, wherein the row elements and column elements of the feature value judgment matrix are all the fault diagnosis feature values, calculate the eigenvectors of the feature value judgment matrix based on the hierarchical analysis method, and normalize the eigenvectors to obtain the weight coefficients of each fault diagnosis feature value; The fourth unit is used to establish a multi-source information fusion model based on DS evidence theory, input the fault diagnosis feature quantity and the weight coefficient into the multi-source information fusion model, and obtain a fusion result; The fifth unit is used to construct a fuzzy comprehensive evaluation matrix, compare the fusion result with an adjustable confidence threshold, and output a fault diagnosis result based on the comparison result.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images