Transformer winding turn-to-turn short circuit fault diagnosis method

Abstract

The invention discloses a transformer winding turn-to-turn short circuit fault diagnosis method, and relates to the technical field of transformer fault diagnosis. After adaptive noise reduction, fundamental frequency alignment and normalization are carried out on the obtained oscillation wave signals, various time-frequency features of the preprocessed signals are calculated, dynamic feature weighting and nonlinear fusion are carried out on the obtained various features, the features are substituted into a fault coefficient expression, and the transformer winding turn-to-turn short circuit degree is obtained according to a fault coefficient FCI for diagnosis.

Description

Technical Field

[0001] This invention relates to the field of transformer fault diagnosis technology, specifically a method for diagnosing inter-turn short-circuit faults in transformer windings. Background Technology

[0002] Power transformers are core equipment for power grid conversion and transmission. Their operating status directly affects power supply reliability and energy utilization efficiency. Under the goal of green and low-carbon development, their energy-saving and environmentally friendly operation is particularly critical. If a transformer operates abnormally due to a short circuit fault between winding turns, it will not only significantly increase power transmission losses, but may also cause grid load imbalance due to fault shutdown, leading to additional energy waste and carbon emissions.

[0003] However, traditional transformer winding testing methods have significant limitations: First, there is a lack of research on the correlation between differentiated inter-turn short circuit conditions and diagnostic methods, resulting in low fault location accuracy, long testing cycles, frequent repeated tests, and high energy consumption throughout the process; Second, some technologies require long-term offline testing, during which backup equipment needs to be activated to maintain power supply, which increases energy consumption, and some testing methods can damage the cable body, shorten the equipment life, and indirectly increase resource consumption and environmental burden.

[0004] The oscillating wave method, as a novel testing technology, effectively solves the aforementioned problems and also boasts energy-saving and environmental advantages: multiple fault simulation tests can be completed with a single wiring connection, reducing the energy consumption of multiple wiring and repeated tests in traditional testing; its core principle utilizes a closed-loop system formed by a high-voltage DC power supply and capacitors, and achieves periodic charging and discharging through regular control of switches, eliminating the need for continuous high-power supply and significantly reducing testing energy consumption; it also features short testing time, minimal cable damage, and accurate defect location, reducing excessive repair and ineffective testing, and minimizing resource waste. Currently, this technology has gained widespread attention in the field of power insulation testing, providing theoretical and practical support for solving the above problems.

[0005] The frequent occurrence of extreme weather and the widening difference between peak and off-peak loads have led to an increase in transformer failure rates. If faults are not diagnosed in time, the long-term inefficient operation of transformers will result in significant energy loss, and power outages due to faults may also trigger a chain reaction of energy consumption problems such as power supply fluctuations and the startup of backup equipment. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a method for diagnosing inter-turn short-circuit faults in transformer windings, thus solving the technical problems mentioned in the background section.

[0008] (II) Technical Solution

[0009] 1. To achieve the above objectives, the present invention provides the following technical solution: a method for diagnosing inter-turn short-circuit faults in transformer windings, characterized in that it includes:

[0010] S1: Obtain the transformer oscillation wave signal x(t), t=1,2,…,N (N is the number of sampling points);

[0011] S2: Signal preprocessing:

[0012] S3: Feature extraction from the preprocessed signal:

[0013] The features include waveform distortion rate F1, zero-crossing rate of change F2, peak energy ratio F3, fundamental frequency energy ratio F4, total harmonic energy ratio F5, and spectral flatness F6.

[0014] S4: Input feature matrix F∈R S*6 ;

[0015] Calculate contrast intensity ;

[0016] in: The contrast intensity of the j-th feature measures the degree of dispersion of the feature value; Let be the j-th feature value of the i-th sample; Let S be the mean of the feature values ​​of the j-th feature; S is the number of samples;

[0017] Calculate the conflict coefficient ;

[0018] Where: c jk Let be the Pearson correlation coefficient between the j-th feature and the k-th feature, which measures the degree of linear correlation between the two features. The Pearson correlation coefficient has a value range of [-1, 1].

[0019] The correlation coefficient matrix is ​​obtained as follows:

[0020] ;

[0021] Wherein: F ij F is the j-th feature value of the i-th sample; ik Let k be the feature value of the i-th sample; Let be the mean of the j-th feature; Let be the mean of the k-th feature; Let be the contrast intensity of the j-th feature; The contrast intensity of the k-th feature;

[0022] Calculate the conflict coefficient:

[0023]

[0024] Among them: Conflict j The conflict coefficient of the j-th feature measures the degree of difference between this feature and other features (the larger the value, the stronger the conflict); c jk Let be the correlation coefficient between the j-th feature and the k-th feature;

[0025] S8: Input feature vector f into the historical fault database to train the XGBoost model. i ∈R 6 Given the failure probability pi∈[0,1], define the sensitivity coefficient:

[0026]

[0027] Where: βj is the fault sensitivity coefficient of the j-th feature, which measures the sensitivity of the feature to faults; FeatureImportancej is the importance score of the j-th feature output by the XGBoost model; Aj is the feature difference of the j-th feature in fault / normal states (KS distance, the larger the KS distance, the more significant the difference); α is the weight coefficient, with a value range of [0,1], used to balance the contribution of feature importance and feature difference.

[0028] Calculate the combined weights w j :

[0029]

[0030] Where: w j The initial combined weight of the j-th feature is obtained by multiplying the contrast strength, conflict coefficient, and fault sensitivity coefficient. The contrast intensity of the j-th feature; Conflict j The conflict coefficient of the j-th feature; β j Let j be the sensitivity coefficient for the j-th characteristic fault;

[0031] Normalize the combined weights w j ;

[0032] ;

[0033] in: is the normalized weight for the j-th feature, with a value range of [0,1].

[0034] S9: Select the top 3 pairs of related features with the largest mutual information values ​​{(j1,k1),(j2,k2),(j3,k3)} using mutual information;

[0035] ;

[0036] Where: I(j,k) is the mutual information value between the j-th feature and the k-th feature, which measures the statistical correlation between the two features (the larger the value, the stronger the correlation). Let be the joint probability density of the j-th eigenvalue and the k-th eigenvalue; The marginal probability density of the j-th eigenvalue; The marginal probability density of the k-th eigenvalue;

[0037] S10: Generate FCI metrics:

[0038] ;

[0039] ;

[0040] ;

[0041] ;

[0042] Among them, FCI is a comprehensive fault index used to comprehensively evaluate the fault status of the signal; α is the linear term weight coefficient, used to adjust the contribution of the linear term in the calculation of FCI, which can be adjusted according to the actual situation to adapt to different fault detection scenarios; LinearTerm is a linear term, which is obtained by linear combination of each feature and its weight, and integrates the contribution of a single feature to the fault; γ is the cross term weight coefficient, used to adjust the contribution of the cross term in the calculation of FCI, which can also be adjusted according to actual needs.

[0043] CrossTerm refers to the cross term, which captures the impact of the interaction between features on the fault by combining and calculating selected feature pairs. The weight of the j-th feature reflects the importance of this feature in fault detection. The larger the weight, the more significant the impact of this feature on the fault. The value of the j-th feature represents the specific numerical value of the j-th feature extracted from the oscillating wave signal; The weight of the j-th feature in the m-th feature pair reflects the importance of that feature in the interaction; The weight of the k-th feature in the m-th feature pair also reflects the importance of this feature in the interaction; Let j be the value of the j-th feature in the m-th feature pair; Let be the value of the k-th feature in the m-th feature pair.

[0044] Preferably, if the FCI is less than 0.3, the transformer under test is considered to be normal.

[0045] If 0.3 ≤ FCI < 0.5, the transformer under test is considered to have a minor inter-turn short circuit fault.

[0046] If 0.5 ≤ FCI < 0.7, the transformer under test is considered to have a medium-sized inter-turn short circuit fault.

[0047] If 0.7 < FCI, the transformer under test is considered to have a serious inter-turn short circuit fault.

[0048] Preferably, the S2 feature extraction includes the following steps:

[0049] S21: Remove interference noise;

[0050] S22: Corrects signal phase shift, making different test data comparable;

[0051] S23: Amplitude normalization processing.

[0052] Preferably, step S21, which removes interference noise, employs an improved wavelet threshold function:

[0053] Where: ψj,k is the db8 wavelet basis function, dj,k is the wavelet coefficient, and the threshold function η() is defined as follows:

[0054]

[0055] Adaptive threshold σj is the median absolute deviation (MAD) estimate of the wavelet coefficients at the j-th level:

[0056]

[0057] S22 performs baseband synchronization alignment: extracting instantaneous phase:

[0058]

[0059] in: For Hilbert transform operators; This is the data after noise reduction; arg() is the argument function, used to solve for the phase angle of the analytic signal;

[0060] Calculate the phase offset:

[0061]

[0062] in: It represents the overall phase shift of the signal, reflecting the phase difference between the signal and the standard fundamental frequency signal; f0 is the instantaneous phase of the signal; f0 is the fundamental frequency of the signal; t is the time index, which takes values ​​from 1 to N;

[0063] Aligned signal:

[0064]

[0065] in: This is the phase-aligned signal; f0 is the calculated phase offset; f0 is the fundamental frequency of the signal.

[0066] The amplitude normalization of S23:

[0067]

[0068] in: The signal after normalization This is the signal after amplitude normalization, with a value range of [0,1]. This is the phase-aligned signal; This represents the minimum value of the signal after phase alignment. This represents the maximum value of the signal after phase alignment.

[0069] Preferably, the ; It is a standard sine wave;

[0070] The Where Z0 is the basic zero-crossing rate, and the frequency at which the statistical signal crosses the zero axis is: Z d Zero-crossing rate of differential signals measures the zero-crossing characteristic of the rate of change of a signal.

[0071] The F3 is the peak energy ratio, which measures the proportion of energy in the peak region of the signal to the total energy.

[0072] The ;

[0073] Where: F4 is the fundamental frequency energy percentage, which measures the proportion of energy near the fundamental frequency to the total frequency domain energy; X(f) is the amplitude spectrum of the normalized signal; f0 is the fundamental frequency; f is the neighborhood bandwidth of the fundamental frequency (used to define the calculation range of the fundamental frequency energy); s f is the sampling frequency of the signal; s / 2 represents the upper limit frequency for frequency domain analysis;

[0074] The F5 is the total energy ratio of the 3rd to 7th harmonics, which measures the proportion of the sum of the 3rd to 7th harmonic energies to the total frequency domain energy.

[0075] The ;

[0076] The Where: F6 is the spectral flatness, which measures the flatness of the spectrum; For the m-th frequency point f m The corresponding spectral amplitude; M is the number of frequency points in the frequency domain analysis; f mLet m be the frequency value at the m-th frequency point.

[0077] Preferably, the .

[0078] Preferably, the .

[0079] Preferably, the historical fault data of the XGBoost model comes from the fault simulation device.

[0080] Preferably, the fault simulation device includes a transformer winding, an infrared heating lamp, a second L-shaped bracket, a second guide rail, a power supply mechanism, a spring shock absorber, a first guide rail, a first L-shaped bracket, a rotating clamping mechanism, a first fixed crossbar, a second fixed crossbar, a power unit for the rotating clamping mechanism, a first multi-degree-of-freedom clamp, a second multi-degree-of-freedom clamp, a third multi-degree-of-freedom clamp, a fourth multi-degree-of-freedom clamp, a fifth multi-degree-of-freedom clamp, a sixth multi-degree-of-freedom clamp, a seventh multi-degree-of-freedom clamp, an eighth multi-degree-of-freedom clamp, an insulation layer breaking cutter, a rotatable turntable, a main crossbar of the support mechanism, a second support crossbar, a second longitudinal support column, an oblique support column, a first support crossbar, a first longitudinal support column, a transformer winding positioning and holding frame, a second moving roller, a second guide rail drive belt, a first guide rail drive belt, a first moving roller, a tail end lead wire, a signal acquisition device, a first section lead wire, and an oscillating wave power supply system;

[0081] The support structure is used to support and fix the transformer windings and the transverse guide rail;

[0082] The transverse guide rail includes two guide rails connected end-to-end, which are fixed to the support mechanism. These are a first guide rail and a second guide rail. The first guide rail is connected to the rotary clamping mechanism, and the second guide rail is connected to the infrared heating mechanism. Both the rotary clamping mechanism and the infrared heating mechanism can move on the transverse guide rail.

[0083] The rotary clamping mechanism includes an L-shaped bracket, a rotatable turntable, and a multi-degree-of-freedom clamp. The rotatable turntable is fixed on the L-shaped bracket and drives the multi-degree-of-freedom clamp to rotate and move.

[0084] The infrared heating mechanism includes an L-shaped bracket and an infrared heating lamp, and the infrared heating lamp can be fixed on the L-shaped bracket.

[0085] The signal acquisition device is connected to one end of the transformer winding and is used to obtain transformer oscillation waveform data under fault conditions.

[0086] The support structure includes a power supply device, which is fixed below the first guide rail;

[0087] The transverse guide rail has a slot along its length, and rollers are provided on both sides below the L-shaped bracket. The rotating clamping mechanism and the infrared heating mechanism are configured to move along the extension direction of the slot in the length direction of the transverse guide rail.

[0088] The rotary clamping mechanism is connected to the L-shaped bracket and includes multiple multi-degree-of-freedom clamps. The multiple multi-degree-of-freedom clamps are located around the rotatable turntable and are evenly distributed at 45-degree intervals. The rotary clamping mechanism is provided with a rotary drive mechanism, and the output end of the rotary drive mechanism is connected to the rotatable turntable.

[0089] Both the transverse guide rail and the rotary clamping mechanism are signal-connected to the control system.

[0090] The high-voltage oscillating wave power supply system includes a high-frequency high-voltage switch and a high-frequency high-voltage DC power supply; the high-frequency high-voltage switch, the high-frequency high-voltage DC power supply and the signal acquisition device are connected in series at both ends of the transformer winding; the high-frequency high-voltage switch is used to control the high-frequency high-voltage DC power supply to periodically charge and discharge the transformer winding.

[0091] Preferably, the historical fault data is obtained through the following steps:

[0092] 1) Fix the transformer windings by placing them in the transformer winding fixing slots on the supporting structure;

[0093] 2) Insert the first guide rail into the transformer winding and fix it horizontally with the transformer winding axis. At the same time, adjust the rotary clamping mechanism on the first guide rail to be on the same horizontal axis as the transformer winding. Fix the cutter head on the first multi-degree-of-freedom fixture on the rotary clamping mechanism. Install and fix the resistors and capacitors of different specifications on the second to eighth multi-degree-of-freedom fixtures on the rotary clamping mechanism and move them to the appropriate position through the first guide rail.

[0094] 3) Adjust the infrared heating mechanism located on the second guide rail to be on the same horizontal axis as the transformer winding, and move it to a suitable position via the second guide rail;

[0095] 4) The first guide rail drives the rotary clamping mechanism to move and extend into the transformer winding to the position where fault simulation is required. The rotatable turntable is controlled to make the cutter head on the first multi-degree-of-freedom fixture destroy the insulation layer at the fault simulation position. The rotatable turntable is controlled to make the resistor-capacitor components on other multi-degree-of-freedom fixtures rotate to the insulation layer destruction point to simulate the inter-turn short circuit fault of the transformer winding.

[0096] 5) The second guide rail drives the infrared heating mechanism to move to a suitable position to locally heat the simulated short-circuit fault area between transformer winding turns in order to better match the actual operating conditions.

[0097] 6) Repeat steps 2) to 5) to complete the simulation of transformer winding inter-turn short circuit faults with different fault levels.

[0098] (III) Beneficial Effects

[0099] This invention provides a method for diagnosing inter-turn short-circuit faults in transformer windings. It has the following beneficial effects:

[0100] (1) The method for diagnosing short-circuit faults between turns of the transformer winding performs adaptive wavelet denoising, fundamental frequency alignment and normalization preprocessing on the collected oscillating wave signal; then extracts multi-dimensional features in the time domain (waveform distortion rate, zero-crossing rate change, etc.) and frequency domain (fundamental energy ratio, harmonic energy ratio, etc.); through dynamic feature weighting algorithm (considering contrast strength, conflict and XGBoost model sensitivity) and feature nonlinear fusion, the final output is a comprehensive fault index FCI in the range of 0-1, realizing four levels of fault degree quantification (normal <0.3, slight 0.3-0.5, moderate 0.5-0.7, severe ≥0.7).

[0101] (2) The transformer winding inter-turn short circuit fault diagnosis method uses a fault simulation device to provide training data for the XGBoost model. The simulation device can adjust the fault position of the transformer winding to be simulated by the guide rail so that the fault simulation can be performed after reaching the predetermined position. The rotating clamping mechanism can realize the rotation of the multi-degree-of-freedom clamp on the rotatable turntable. During the rotation, the contact between the transformer winding and the resistors and capacitors of different specifications held by the multi-degree-of-freedom clamp can be changed, thereby simulating different degrees of inter-turn short circuit faults. The infrared heating mechanism is used to locally heat the simulated position of the transformer winding inter-turn short circuit fault to better match the temperature conditions under actual working conditions. The transverse guide rail can allow the rotating clamping mechanism and the infrared heating mechanism to move arbitrarily in the length direction of the guide rail, realizing a more reliable simulation of the transformer winding inter-turn short circuit fault. Attached Figure Description

[0102] Figure 1 This is a schematic diagram of the transformer winding inter-turn short-circuit fault simulation device in an embodiment of the present invention;

[0103] Figure 2 This is a front view of the transformer winding inter-turn short-circuit fault simulation device in an embodiment of the present invention;

[0104] Figure 3 This is a top view of the transformer winding inter-turn short-circuit fault simulation device in an embodiment of the present invention;

[0105] Figure 4 This is a schematic diagram of the rotatable turntable, i.e., the multi-degree-of-freedom fixture structure in an embodiment of the present invention;

[0106] Figure 5This is a schematic diagram of inter-turn short-circuit fault detection in a transformer winding according to an embodiment of the present invention.

[0107] Figure 6 This is a flowchart of the transformer winding inter-turn short circuit fault diagnosis method of the present invention. Detailed Implementation

[0108] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0109] Transformer winding inter-turn short-circuit oscillation wave testing and fault diagnosis, including;

[0110] 1) The connection drawn from the beginning of transformer winding 1 is called the beginning connection 36, and the connection drawn from the end of transformer winding 1 is called the end connection 34.

[0111] 2) Connect the lead-out terminal 36 of transformer winding 1 to the high-frequency high-voltage DC power supply and the high-frequency high-voltage switch. The high-frequency high-voltage switch will perform periodic operation to periodically charge and discharge transformer winding 1.

[0112] 3) Repeat the above process to obtain transformer oscillation waveform data under different inter-turn short circuit fault conditions through signal acquisition device 35.

[0113] 4) Preprocessing of oscillating wave signal: Input the original oscillating wave signal x(t), t=1,2,…,N (N is the number of sampling points);

[0114] First, adaptive wavelet denoising is performed: an improved wavelet threshold function is used.

[0115]

[0116] Where: ψj,k is the db8 wavelet basis function, dj,k is the wavelet coefficient, and the threshold function η() is defined as follows:

[0117]

[0118] Adaptive threshold σj is the median absolute deviation (MAD) estimate of the wavelet coefficients at the j-th level:

[0119]

[0120] Then perform fundamental frequency synchronization alignment: extract instantaneous phase:

[0121]

[0122] Where: H is the Hilbert transform operator; The data is after noise reduction; arg() is the argument function, used to solve for the phase angle of the analytic signal.

[0123] Calculate the phase offset:

[0124]

[0125] in: It represents the overall phase shift of the signal, reflecting the phase difference between the signal and the standard fundamental frequency signal; f0 is the instantaneous phase of the signal; f0 is the fundamental frequency of the signal; t is the time index, which takes values ​​from 1 to N.

[0126] Aligned signal:

[0127]

[0128] in: This is the phase-aligned signal; f0 is the calculated phase offset; f0 is the fundamental frequency of the signal.

[0129] Amplitude normalization:

[0130]

[0131] in: This is the signal after amplitude normalization, with a value range of [0,1]. This is the phase-aligned signal; This represents the minimum value of the signal after phase alignment. This represents the maximum value of the signal after phase alignment.

[0132] 5) Temporal feature extraction:

[0133] Waveform distortion rate (F1):

[0134]

[0135] Where: F1 is the waveform distortion rate, which measures the degree of difference between the actual signal and the standard sine wave; the larger the value, the more severe the distortion. The signal is normalized. The standard sine wave is represented by N, which is the number of sampling points.

[0136] Change in zero-crossing rate (F2):

[0137] Basic zero-crossing rate:

[0138]

[0139] Where: Z0 is the basic zero-crossing rate, and the frequency at which the statistical signal crosses the zero axis; The signal is normalized; N is the number of sampling points.

[0140] Differential signal zero-crossing rate:

[0141]

[0142] Where: Zd is the zero-crossing rate of the differential signal, which measures the zero-crossing characteristic of the signal's rate of change; N is the number of sampling points.

[0143] Change in zero crossing rate

[0144]

[0145] Where: Z0 is the basic zero-crossing rate; Zd is the differential signal zero-crossing rate; N is the number of sampling points.

[0146] Peak energy ratio (F3):

[0147]

[0148] Where: F3 is the peak energy ratio, which measures the proportion of energy in the peak region of the signal to the total energy; This is the normalized signal.

[0149] 6) Frequency domain feature extraction:

[0150] Let the spectrum

[0151] in: The amplitude spectrum of the normalized signal reflects the magnitude of the signal at different frequency components; This is the normalized signal.

[0152] Fundamental wave energy percentage (F4)

[0153]

[0154] Where: F4 is the fundamental frequency energy percentage, which measures the proportion of energy near the fundamental frequency to the total frequency domain energy; X(f) is the amplitude spectrum of the normalized signal; f0 is the fundamental frequency; fs is the neighborhood bandwidth of the fundamental frequency (used to define the calculation range of the fundamental energy); fs is the sampling frequency of the signal; fs / 2 is the upper limit frequency for frequency domain analysis.

[0155] Total energy ratio of 3rd to 7th harmonics (F5)

[0156]

[0157] Wherein: F5 is the total energy ratio of the 3rd to 7th harmonics, which measures the proportion of the sum of the 3rd to 7th harmonic energies to the total frequency domain energy.

[0158] Spectral flatness (F6):

[0159]

[0160] Where: F6 is the spectral flatness, which measures the flatness of the spectrum (close to 1 is a flat spectrum, close to 0 is a peak spectrum). fm represents the spectral amplitude corresponding to the m-th frequency point; M is the number of frequency points in the frequency domain analysis; and fm is the frequency value of the m-th frequency point.

[0161] 7) Dynamic feature weighting:

[0162] Input feature matrix (S is the sample size)

[0163] Calculate the contrast intensity:

[0164]

[0165] in: The contrast intensity of the j-th feature measures the degree of dispersion of the feature value; Let be the j-th feature value of the i-th sample; Let S be the mean of the feature values ​​of the j-th feature; S is the number of samples.

[0166] Calculate the conflict coefficient

[0167] Where: cjk is the Pearson correlation coefficient between the j-th feature and the k-th feature, which measures the degree of linear correlation between the two features (value range [-1,1]).

[0168] Correlation coefficient matrix:

[0169]

[0170] Where: Fij is the j-th feature value of the i-th sample; Fik is the k-th feature value of the i-th sample; Let be the mean of the j-th feature; Let be the mean of the k-th feature; Let be the contrast intensity of the j-th feature; denoted as the contrast intensity of the k-th feature.

[0171] Conflict factor:

[0172]

[0173] Where: Conflictj is the conflict coefficient of the j-th feature, which measures the degree of difference between this feature and other features (the larger the value, the stronger the conflict); cjk is the correlation coefficient between the j-th feature and the k-th feature.

[0174] Training an XGBoost model based on a historical fault database:

[0175] Input: Feature vector

[0176] Output: Fault probability pi∈[0,1] Define sensitivity coefficient:

[0177]

[0178] Where: βj is the fault sensitivity coefficient of the j-th feature, which measures the sensitivity of the feature to faults (the larger the value, the higher the sensitivity); FeatureImportancej is the importance score of the j-th feature output by the XGBoost model; Aj is the feature difference of the j-th feature in fault / normal states (KS distance, the larger the KS distance, the more significant the difference); α is the weight coefficient (value range [0,1]), used to balance the contribution of feature importance and feature difference.

[0179] Combined weights

[0180]

[0181] Where: the initial combination weight of the j-th feature wj is obtained by multiplying the contrast strength, conflict coefficient and fault sensitivity coefficient; βj is the contrast intensity of the j-th feature; Conflictj is the conflict coefficient of the j-th feature; βj is the fault sensitivity coefficient of the j-th feature.

[0182] Normalized weights:

[0183]

[0184] Where: wj is the initial combination weight of the j-th feature; is the normalized weight of the j-th feature, with a value range of [0,1].

[0185] 8) Nonlinear fusion (generating FCI metrics):

[0186] Selecting associated feature pairs using mutual information:

[0187]

[0188] Select the top 3 feature pairs: {(j1,k1),(j2,k2),(j3,k3)} (the top 3 pairs with the largest mutual information values).

[0189] Where: I(j,k) is the mutual information value between the j-th feature and the k-th feature, which measures the statistical correlation between the two features (the larger the value, the stronger the correlation). Let be the joint probability density of the j-th eigenvalue and the k-th eigenvalue; The marginal probability density of the j-th eigenvalue; Let be the marginal probability density of the k-th eigenvalue.

[0190] 9) Calculation of comprehensive indicators

[0191] in:

[0192]

[0193]

[0194]

[0195]

[0196] Among them, FCI is a comprehensive fault index used to comprehensively evaluate the fault status of the signal; α is the linear term weight coefficient, used to adjust the contribution of the linear term in the calculation of FCI, which can be adjusted according to the actual situation to adapt to different fault detection scenarios; LinearTerm is a linear term, which is obtained by linear combination of each feature and its weight, and integrates the contribution of a single feature to the fault; γ is the cross term weight coefficient, used to adjust the contribution of the cross term in the calculation of FCI, which can also be adjusted according to actual needs.

[0197] CrossTerm refers to the cross term, which captures the impact of the interaction between features on the fault by combining and calculating selected feature pairs. The weight of the j-th feature reflects the importance of this feature in fault detection. The larger the weight, the more significant the impact of this feature on the fault. The value of the j-th feature represents the specific numerical value of the j-th feature extracted from the oscillating wave signal; The weight of the j-th feature in the m-th feature pair reflects the importance of that feature in the interaction; The weight of the k-th feature in the m-th feature pair also reflects the importance of this feature in the interaction; Let j be the value of the j-th feature in the m-th feature pair; Let be the value of the k-th feature in the m-th feature pair.

[0198] Fault severity mapping:

[0199] .

[0200] The historical fault data for the XGBoost model was obtained through the following methods.

[0201] Simulate different degrees of inter-turn short-circuit faults in transformer windings, including:

[0202] 1) The transformer winding 1 is placed in the transformer winding positioning and retaining bracket 29 on the supporting structure to complete the fixation;

[0203] 2) Insert the first guide rail 7 into the transformer winding 1 and fix it horizontally with the axis of the transformer winding 1. At the same time, adjust the rotary clamping mechanism 9 on the first guide rail 7 to be on the same horizontal axis as the transformer winding 1. Fix the insulation layer breaking cutter head 21 on the multi-degree-of-freedom clamps 13~20 on the rotary clamping mechanism 9. Install and fix the resistors and capacitors of different specifications on the first to eighth multi-degree-of-freedom clamps 13~20 on the rotary clamping mechanism 9 and move them to the appropriate position through the first guide rail 7.

[0204] 3) Adjust the infrared heating mechanism 2 located on the second guide rail 4 to be on the same horizontal axis as the transformer winding 1, and move it to a suitable position via the second guide rail 4;

[0205] 4) The first guide rail 7 drives the rotary clamping mechanism to move 9 and extend into the transformer winding 1 to the position where fault simulation is required. The rotatable turntable 22 is controlled to make the insulation layer breaking cutter 21 on the first multi-degree-of-freedom fixture 13 break the insulation layer at the fault simulation position. The rotatable turntable 22 is controlled to make the resistor and capacitor components on other multi-degree-of-freedom fixtures rotate to the insulation layer breaking point to simulate the inter-turn short circuit fault of the transformer winding 1.

[0206] 5) The second guide rail 4 drives the infrared heating mechanism 9 to move to a suitable position to locally heat the simulated short-circuit fault between the transformer winding turns so as to better match the actual working conditions.

[0207] 6) Repeat steps 2) to 5) to complete the simulation of transformer winding inter-turn short circuit faults with different fault levels.

[0208] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0209] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for diagnosing inter-turn short-circuit faults in transformer windings, characterized in that, include: S1: Obtain the transformer oscillation wave signal x(t), t=1,2,…,N (N is the number of sampling points); S2: Signal preprocessing: S3: Feature extraction from the preprocessed signal: The features include waveform distortion rate F1, zero-crossing rate of change F2, peak energy ratio F3, fundamental energy ratio F4, total harmonic energy ratio F5, and spectral flatness F6. S4: Input feature matrix F∈R S*6 ; Calculate contrast intensity ; in: The contrast intensity of the j-th feature measures the degree of dispersion of the feature value; Let be the j-th feature value of the i-th sample; Let S be the mean of the feature values ​​of the j-th feature; S is the number of samples. Calculate the conflict coefficient ; Where: c jk Let be the Pearson correlation coefficient between the j-th feature and the k-th feature, which measures the degree of linear correlation between the two features. The Pearson correlation coefficient has a value range of [-1, 1]. The correlation coefficient matrix is ​​obtained as follows: ； Wherein: F ij F is the j-th feature value of the i-th sample; ik Let k be the feature value of the i-th sample; Let be the mean of the j-th feature; Let be the mean of the k-th feature; Let be the contrast intensity of the j-th feature; The contrast intensity of the k-th feature; Calculate the conflict coefficient: ； Among them: Conflict j The conflict coefficient of the j-th feature measures the degree of difference between this feature and other features (the larger the value, the stronger the conflict); c jk Let be the correlation coefficient between the j-th feature and the k-th feature; S8: Input feature vector f into the historical fault database to train the XGBoost model. i ∈R 6 Given the failure probability pi∈[0,1], define the sensitivity coefficient: ； Where: βj is the fault sensitivity coefficient of the j-th feature, which measures the sensitivity of the feature to faults; FeatureImportancej is the importance score of the j-th feature output by the XGBoost model; Aj is the feature difference of the j-th feature in fault / normal states (KS distance, the larger the KS distance, the more significant the difference); α is the weight coefficient, with a value range of [0,1], used to balance the contribution of feature importance and feature difference. Calculate the combined weights w j : ; Where: w j The initial combined weight of the j-th feature is obtained by multiplying the contrast strength, conflict coefficient, and fault sensitivity coefficient. The contrast intensity of the j-th feature; Conflict j The conflict coefficient of the j-th feature; β j Let j be the sensitivity coefficient for the j-th characteristic fault; Normalize the combined weights w j ; ； in: is the normalized weight for the j-th feature, with a value range of [0,1]. S9: Select the top 3 pairs of related features with the largest mutual information values ​​{(j1,k1),(j2,k2),(j3,k3)} using mutual information; ； Where: I(j,k) is the mutual information value between the j-th feature and the k-th feature, which measures the statistical correlation between the two features (the larger the value, the stronger the correlation). Let be the joint probability density of the j-th eigenvalue and the k-th eigenvalue; The marginal probability density of the j-th eigenvalue; The marginal probability density of the k-th eigenvalue; S10: Generate FCI metrics: ； ； ； ； Among them, FCI is a comprehensive fault index used to comprehensively evaluate the fault status of the signal; α is the linear term weight coefficient, used to adjust the contribution of the linear term in the calculation of FCI, which can be adjusted according to the actual situation to adapt to different fault detection scenarios; LinearTerm is a linear term, which is obtained by linear combination of each feature and its weight, and integrates the contribution of a single feature to the fault; γ is the cross term weight coefficient, used to adjust the contribution of the cross term in the calculation of FCI, which can also be adjusted according to actual needs. CrossTerm refers to the cross term, which captures the impact of the interaction between features on the fault by combining and calculating selected feature pairs. The weight of the j-th feature reflects the importance of this feature in fault detection. The larger the weight, the more significant the impact of this feature on the fault. The value of the j-th feature represents the specific numerical value of the j-th feature extracted from the oscillating wave signal; The weight of the j-th feature in the m-th feature pair reflects the importance of that feature in the interaction; The weight of the k-th feature in the m-th feature pair also reflects the importance of this feature in the interaction; Let j be the value of the j-th feature in the m-th feature pair; Let be the value of the k-th feature in the m-th feature pair.

2. The method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 1, characterized in that: If FCI < 0.3, the transformer under test is considered normal. If 0.3 ≤ FCI < 0.5, the transformer under test is considered to have a minor inter-turn short circuit fault. If 0.5 ≤ FCI < 0.7, the transformer under test is considered to have a medium-sized inter-turn short circuit fault. If 0.7 < FCI, the transformer under test is considered to have a serious inter-turn short circuit fault.

3. The method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 1, characterized in that: The S2 feature extraction includes the following steps: S21: Remove interference noise; S22: Corrects signal phase shift, making different test data comparable; S23: Amplitude normalization processing.

4. The method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 3, characterized in that: The S21 step removes interference noise using an improved wavelet threshold function: ; Where: ψj,k is the db8 wavelet basis function, dj,k is the wavelet coefficient, and the threshold function η() is defined as follows: ； Adaptive threshold σj is the median absolute deviation (MAD) estimate of the wavelet coefficients at the j-th level: ； S22 performs baseband synchronization alignment: extracting instantaneous phase: ； in: For Hilbert transform operators; This is the data after noise reduction; arg() is the argument function, used to solve for the phase angle of the analytic signal; Calculate the phase offset: ； in: It represents the overall phase shift of the signal, reflecting the phase difference between the signal and the standard fundamental frequency signal; f0 is the instantaneous phase of the signal; f0 is the fundamental frequency of the signal; t is the time index, which takes values ​​from 1 to N; Aligned signal: ； in: This is the phase-aligned signal; f0 is the calculated phase offset; f0 is the fundamental frequency of the signal. The amplitude normalization of S23: ； in: The signal after normalization This is the signal after amplitude normalization, with a value range of [0,1]. This is the phase-aligned signal; This represents the minimum value of the signal after phase alignment. This represents the maximum value of the signal after phase alignment.

5. The method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 4, characterized in that: The ; It is a standard sine wave; The Where Z0 is the basic zero-crossing rate, and the frequency at which the statistical signal crosses the zero axis is: Z d Zero-crossing rate of differential signals measures the zero-crossing characteristic of the rate of change of a signal. The F3 is the peak energy ratio, which measures the proportion of energy in the peak region of the signal to the total energy. The ; Where: F4 is the fundamental frequency energy percentage, which measures the proportion of energy near the fundamental frequency to the total frequency domain energy; X(f) is the amplitude spectrum of the normalized signal; f0 is the fundamental frequency; f is the neighborhood bandwidth of the fundamental frequency (used to define the calculation range of the fundamental frequency energy); s f is the sampling frequency of the signal; s / 2 represents the upper limit frequency for frequency domain analysis; The F5 is the total energy ratio of the 3rd to 7th harmonics, which measures the proportion of the sum of the 3rd to 7th harmonic energies to the total frequency domain energy. The ; The Where: F6 is the spectral flatness, which measures the flatness of the spectrum; For the m-th frequency point f m The corresponding spectral amplitude; M is the number of frequency points in the frequency domain analysis; f m Let m be the frequency value at the m-th frequency point.

6. The method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 5, characterized in that: The .

7. The method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 5, characterized in that: The .

8. The method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 1, characterized in that: The historical fault data of the XGBoost model comes from the fault simulation device.

9. A method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 8, characterized in that: The fault simulation device includes a transformer winding, an infrared heating lamp, a second L-shaped bracket, a second guide rail, a power supply mechanism, a spring shock absorber, a first guide rail, a first L-shaped bracket, a rotating clamping mechanism, a first fixed crossbar, a second fixed crossbar, a power unit for the rotating clamping mechanism, a first multi-degree-of-freedom clamp, a second multi-degree-of-freedom clamp, a third multi-degree-of-freedom clamp, a fourth multi-degree-of-freedom clamp, a fifth multi-degree-of-freedom clamp, a sixth multi-degree-of-freedom clamp, a seventh multi-degree-of-freedom clamp, an eighth multi-degree-of-freedom clamp, an insulation layer breaking cutter, a rotatable turntable, a main crossbar of the support mechanism, a second support crossbar, a second longitudinal support column, an oblique support column, a first support crossbar, a first longitudinal support column, a transformer winding positioning and holding frame, a second moving roller, a second guide rail drive belt, a first guide rail drive belt, a first moving roller, a tail end lead wire, a signal acquisition device, a first section lead wire, and an oscillating wave power supply system. The support structure is used to support and fix the transformer windings and the transverse guide rail; The transverse guide rail includes two guide rails connected end-to-end, which are fixed to the support mechanism. These are a first guide rail and a second guide rail. The first guide rail is connected to the rotary clamping mechanism, and the second guide rail is connected to the infrared heating mechanism. Both the rotary clamping mechanism and the infrared heating mechanism can move on the transverse guide rail. The rotary clamping mechanism includes an L-shaped bracket, a rotatable turntable, and a multi-degree-of-freedom clamp. The rotatable turntable is fixed on the L-shaped bracket and drives the multi-degree-of-freedom clamp to rotate and move. The infrared heating mechanism includes an L-shaped bracket and an infrared heating lamp, and the infrared heating lamp can be fixed on the L-shaped bracket. The signal acquisition device is connected to one end of the transformer winding and is used to obtain transformer oscillation waveform data under fault conditions. The support structure includes a power supply device, which is fixed below the first guide rail; The transverse guide rail has a slot along its length, and rollers are provided on both sides below the L-shaped bracket. The rotating clamping mechanism and the infrared heating mechanism are configured to move along the extension direction of the slot in the length direction of the transverse guide rail. The rotary clamping mechanism is connected to the L-shaped bracket and includes multiple multi-degree-of-freedom clamps. The multiple multi-degree-of-freedom clamps are located around the rotatable turntable and are evenly distributed at 45-degree intervals. The rotary clamping mechanism is provided with a rotary drive mechanism, and the output end of the rotary drive mechanism is connected to the rotatable turntable. Both the transverse guide rail and the rotary clamping mechanism are signal-connected to the control system. The high-voltage oscillating wave power supply system includes a high-frequency high-voltage switch and a high-frequency high-voltage DC power supply; the high-frequency high-voltage switch, the high-frequency high-voltage DC power supply and the signal acquisition device are connected in series at both ends of the transformer winding; the high-frequency high-voltage switch is used to control the high-frequency high-voltage DC power supply to periodically charge and discharge the transformer winding.

10. A method for diagnosing inter-turn short-circuit faults in transformer windings according to claim 9, characterized in that: The historical fault data is obtained through the following steps: 1) Fix the transformer windings by placing them in the transformer winding fixing slots on the supporting structure; 2) Insert the first guide rail into the transformer winding and fix it horizontally with the transformer winding axis. At the same time, adjust the rotary clamping mechanism on the first guide rail to be on the same horizontal axis as the transformer winding. Fix the cutter head on the first multi-degree-of-freedom fixture on the rotary clamping mechanism. Install and fix the resistors and capacitors of different specifications on the second to eighth multi-degree-of-freedom fixtures on the rotary clamping mechanism and move them to the appropriate position through the first guide rail. 3) Adjust the infrared heating mechanism located on the second guide rail to be on the same horizontal axis as the transformer winding, and move it to a suitable position via the second guide rail; 4) The first guide rail drives the rotary clamping mechanism to move and extend into the transformer winding to the position where fault simulation is required. The rotatable turntable is controlled to make the cutter head on the first multi-degree-of-freedom fixture destroy the insulation layer at the fault simulation position. The rotatable turntable is controlled to make the resistor-capacitor components on other multi-degree-of-freedom fixtures rotate to the insulation layer destruction point to simulate the inter-turn short circuit fault of the transformer winding. 5) The second guide rail drives the infrared heating mechanism to move to a suitable position to locally heat the simulated short-circuit fault between the turns of the transformer winding to better match the actual operating conditions. 6) Repeat steps 2) to 5) to complete the simulation of transformer winding inter-turn short circuit faults with different fault levels.

Images