Transformer winding inter-strand short circuit diagnosis method based on frequency response loss characteristics

Abstract

This invention relates to the field of transformer fault condition monitoring technology, and particularly to a method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics. The method includes: applying a sweep frequency excitation signal to a healthy transformer's multi-strand parallel winding in an offline state, and establishing a health baseline curve based on a distributed parameter equivalent model; applying the same signal to a target winding in a test state and acquiring the test frequency response curve; comparing the two sets of curves to extract frequency response attenuation characteristics and curve distortion characteristics characterizing the internal loss state of the winding; using these characteristics to calculate the rate of change of the quality factor and the rate of change of distortion of the resonance peak, respectively, to quantify the electromagnetic energy loss state inside the winding; constructing a multi-threshold loss change criterion, determining the existence of an inter-strand short circuit fault when any rate of change exceeds the threshold; and evaluating the fault location and severity by frequency band based on the attenuation degree and distortion trend of the curves in the low, medium, and high frequency bands.

Description

Technical Field

[0001] This invention relates to the field of transformer fault condition monitoring technology, and in particular to a method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics. Background Technology

[0002] Large power transformers typically employ a multi-strand parallel winding structure to reduce eddy current losses caused by the skin effect and proximity effect. Frequency response analysis (FRA) is currently the primary method for detecting the condition of transformer windings. It involves applying a sweep frequency excitation signal to the offline transformer windings, acquiring voltage and current signals at the response terminals, and then calculating the frequency response impedance curve, which reflects the microstructure of the windings. By comparing the waveform similarity or resonant point drift between the measured curve and a healthy baseline curve, maintenance personnel can assess whether a winding fault has occurred.

[0003] Existing frequency response analysis methods primarily rely on significant changes in capacitance and inductance caused by mechanical deformation to detect winding anomalies, but they are insensitive to the early microscopic circulating currents and energy loss states between multiple parallel conductors within the winding. Since the winding as a whole has not yet undergone significant macroscopic mechanical and geometric deformation in the early stages of an inter-strand short-circuit fault, the shape of the measured frequency response curve and the location of the main resonant point remain almost unchanged. This makes traditional methods unable to perform microscopic quantification of early inter-strand short-circuit hazards and highly prone to underreporting. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics, aiming to improve the problem of the inability to microscopically quantify early-stage inter-strand short circuit hazards and the high likelihood of missed detection.

[0005] This invention provides the following technical solution: a method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics, comprising: S1. When the transformer is offline, apply a sweep frequency excitation signal to the multi-strand parallel winding of the transformer in a healthy state, obtain the frequency response impedance curve in the healthy state, and establish a healthy reference curve based on the distributed parameter equivalent model of the multi-strand parallel winding. S2. In the test state, apply the same frequency sweep excitation signal as in step S1 to the target winding and collect the test frequency response curve of the target winding. S3. Compare the frequency response curve to be tested with the healthy baseline curve, and extract the frequency response attenuation characteristics and curve distortion characteristics that characterize the internal loss change state of the winding. S4. Calculate the rate of change of the quality factor of the resonance peak using the extracted frequency response attenuation characteristics, and calculate the corresponding rate of change of distortion based on the curve distortion characteristics, so as to quantify the electromagnetic energy loss state inside the target winding. S5. Construct a loss change criterion composed of the quality factor change rate and the distortion change rate. When either change rate exceeds the corresponding preset threshold, it is determined that there is an inter-strand short circuit fault inside the target winding. S6. Based on the attenuation degree and curve distortion trend of the frequency response curves under test in the low-frequency, mid-frequency, and high-frequency bands, the location and severity of the inter-strand short-circuit fault are evaluated by frequency band.

[0006] Preferably, in S1, obtaining the frequency response impedance curve under the healthy state includes: A sinusoidal sweep frequency signal is applied to a multi-strand parallel winding of a transformer in a healthy state. The frequency range of the sinusoidal sweep frequency signal covers the low-frequency sensitive segment, the medium-frequency sensitive segment, and the high-frequency sensitive segment of the multi-strand parallel winding. The application method of the sinusoidal sweep frequency signal includes continuous sweep frequency or segmented sweep frequency. The response terminal voltage and response terminal current of the multi-strand parallel winding are measured at various frequency points, and the frequency response impedance curve under healthy conditions is calculated using the response terminal voltage and response terminal current.

[0007] Preferably, in S1, establishing the health baseline curve includes: Construct a distributed network model of the multi-strand parallel winding, wherein the distributed network model includes the resistance parameters and self-inductance parameters corresponding to each strand of conductor, the inter-strand electromagnetic coupling parameters and inter-strand capacitance parameters between any two strands of conductor, and the insulation equivalent impedance parameters used to characterize the inter-strand insulation state. The frequency response impedance curve obtained under the healthy state is substituted into the distributed network model for parameter identification to obtain the winding fingerprint parameter matrix under the healthy state. The frequency response impedance curve under the healthy state is smoothed at the boundary and noise is corrected using the winding fingerprint parameter matrix to obtain the healthy baseline curve.

[0008] Preferably, in S2, applying the frequency sweep excitation signal includes: Generate a sinusoidal sweep signal with a set amplitude, and inject the sinusoidal sweep signal into the input terminal of the target winding; During the application of the sinusoidal sweep frequency signal, the characteristic impedance of the injection channel is adaptively adjusted to match the input impedance of the target winding.

[0009] Preferably, in S2, acquiring the frequency response curve of the target winding includes: The starting voltage signal and the ending current signal of the target winding at each frequency point are collected simultaneously. The first-end voltage signal and the last-end current signal are synchronously sampled and digitally converted to obtain discrete digital signals; The discrete digital signal is subjected to discrete Fourier transform to calculate the impedance magnitude and phase angle at each frequency point, and the frequency response curve to be measured is synthesized.

[0010] Preferably, in S3, the extraction of frequency response attenuation features and curve distortion features includes: By comparing the corresponding resonant peaks of the health baseline curve and the frequency response curve under test, the peak value difference is calculated as the peak value reduction, and the bandwidth difference at the set decibel drop in amplitude is calculated as the half-power bandwidth increase, so as to form the frequency response attenuation characteristics. The high-frequency sensitive segments of the health baseline curve and the frequency response curve under test are extracted, and the overall slope difference between the two is calculated as the change in the slope of the amplitude decrease in the high-frequency segment. The statistical correlation coefficient between the two in the corresponding frequency band is calculated as the curve correlation coefficient to form the curve distortion characteristics.

[0011] Preferably, in S4, the calculation of the rate of change of the quality factor of the resonance peak includes: Based on the center frequencies of the corresponding resonant peaks of the health reference curve and the increase in half-power bandwidth, the decrease in the quality factor of the frequency response curve under test relative to the health reference curve is calculated. The rate of change of the quality factor is obtained by calculating the ratio of the decrease in the quality factor to the original quality factor of the health baseline curve.

[0012] Preferably, in S4, the calculation of the corresponding distortion rate includes: The high-frequency band attenuation rate in the distortion rate is obtained by calculating the ratio of the change in the slope of the amplitude drop in the high-frequency band to the original slope of the health baseline curve in the corresponding high-frequency sensitive segment. The correlation distortion rate in the distortion rate is obtained by calculating the difference between the curve correlation coefficient of the frequency response curve under test and the healthy baseline curve. The high-frequency band attenuation change rate and the correlation distortion rate are weighted and summed to obtain the distortion change rate.

[0013] Preferably, in S5, the criteria for determining loss variation include: Set a first preset threshold corresponding to the rate of change of the quality factor and a second preset threshold corresponding to the rate of change of distortion, respectively. The rate of change of the quality factor is compared with a first preset threshold, and the rate of change of distortion is compared with a second preset threshold to construct a loss change criterion. When the rate of change of the quality factor is greater than the first preset threshold, or the rate of change of the distortion is greater than the second preset threshold, a judgment signal that satisfies the loss change criterion is output.

[0014] Preferably, in S6, the step of determining the attenuation degree and curve distortion trend of the frequency response curve under test in the low-frequency, mid-frequency, and high-frequency bands includes: In the low-frequency range, the overall inductance change state of the multi-strand parallel winding is evaluated based on the change in loop inductance characteristics of the frequency response curve under test relative to the healthy reference curve. In the mid-frequency band, the inter-strand coupling change state of the multi-strand parallel winding is evaluated based on the offset of the resonant peak position of the frequency response curve under test relative to the healthy reference curve. In the high-frequency band, the inter-strand additional loss and insulation state of the multi-strand parallel winding are evaluated based on the rate of change of the quality factor and the rate of change of the distortion. By combining the overall inductance change state, inter-strand coupling change state, inter-strand additional loss change state, and insulation state change trend, the fault location of the inter-strand short circuit fault is determined, and the severity of the inter-strand short circuit fault is determined based on the characteristic rate of change amplitude in each corresponding frequency band.

[0015] The present invention has the following beneficial effects: 1. In this invention, by extracting the peak value reduction and half-power bandwidth increase at the resonance peak and calculating the rate of change of the quality factor, the microscopic quantitative analysis of the microscopic circulating current and energy loss state inside the winding is realized, which solves the problem that traditional frequency response diagnosis mainly relies on macroscopic mechanical deformation, which easily leads to missed detection in the early stage of inter-strand short circuit faults.

[0016] 2. In this invention, by adaptively adjusting the characteristic impedance during the signal injection process to dynamically match the input impedance of the target winding, the excitation energy is fed into the transmission boundary without reflection and with full and stable input. This solves the problem of power reflection and waveform distortion of the test signal caused by the drastic fluctuation of the wideband complex impedance of the multi-strand parallel winding.

[0017] 3. In this invention, by constructing a high- and low-frequency multi-index "OR logic" loss criterion and combining it with the electromagnetic dominance mechanism of different frequency bands of the winding to perform multi-dimensional matrix mapping calculation, the overall anti-interference capability of the diagnostic system is improved, as well as the quantitative classification assessment of the spatial location and severity of the fault is achieved. This solves the problem that single-dimensional waveform indicators are easily affected by the interference of field lead capacitors, causing false alarms and failing to accurately guide the core repair. Attached Figure Description

[0018] Figure 1 This is a flowchart of the transformer winding inter-strand short circuit diagnosis method based on frequency response loss characteristics proposed in this invention; Figure 2 This is a flowchart illustrating the establishment of a healthy baseline curve for the transformer winding inter-strand short-circuit diagnosis method based on frequency response loss characteristics proposed in this invention. Figure 3This is a flowchart illustrating the frequency-segmented assessment of fault location and severity in the transformer winding inter-strand short-circuit diagnosis method based on frequency response loss characteristics proposed in this invention. Detailed Implementation

[0019] The technical solutions in 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.

[0020] In embodiments of the present invention, the present invention provides a method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics, such as... Figures 1-3 As shown, it includes the following steps: S1. When the transformer is offline, apply a sweep frequency excitation signal to the multi-strand parallel winding of the transformer in a healthy state, obtain the frequency response impedance curve in a healthy state, and establish a healthy baseline curve based on the distributed parameter equivalent model of the multi-strand parallel winding. Furthermore, in S1, obtaining the frequency response impedance curve under healthy conditions includes: A sinusoidal sweep frequency signal is applied to the multi-strand parallel winding of a transformer in a healthy state. The frequency range of the sinusoidal sweep frequency signal covers the low-frequency sensitive segment, the medium-frequency sensitive segment, and the high-frequency sensitive segment of the multi-strand parallel winding. The application method of the sinusoidal sweep frequency signal includes continuous sweep frequency or segmented sweep frequency. The response terminal voltage and response terminal current of the multi-strand parallel winding are measured at various frequency points, and the frequency response impedance curve under healthy conditions is calculated using the response terminal voltage and response terminal current.

[0021] Furthermore, in S1, establishing a health baseline curve includes: A distributed network model of multiple parallel windings is constructed. The distributed network model includes the resistance and self-inductance parameters of each conductor, the electromagnetic coupling parameters and capacitance parameters between any two conductors, and the insulation equivalent impedance parameters used to characterize the insulation state between the conductors. The frequency response impedance curve obtained under healthy conditions is substituted into the distributed network model for parameter identification to obtain the winding fingerprint parameter matrix under healthy conditions. By using the winding fingerprint parameter matrix, the frequency response impedance curve under healthy conditions is smoothed at the boundary and noise is corrected to obtain the healthy baseline curve.

[0022] Specifically, to achieve high-precision parameter identification and noise correction of measured data using a distributed network model, this scheme introduces a least-squares iterative algorithm in the parameter identification step. This algorithm, through a combination of mathematical analysis and physical mechanism models, achieves accurate inversion of the microscopic distribution parameters inside the winding.

[0023] In practical implementation, the frequency response impedance curve obtained under the measured healthy state is first expressed as a discrete function of frequency. ,in Indicates the first The frequency values ​​of each sweep point are then calculated. Simultaneously, the forward circuit is solved using the constructed distributed network model to obtain the theoretical impedance function, which is jointly determined by the model's internal resistance parameters, self-inductance parameters, inter-strand electromagnetic coupling parameters, inter-strand capacitance parameters, and insulation equivalent impedance parameters. ,in Let be the distribution parameter matrix vector to be identified. To achieve the best fit between the model output and the actual measurements, the objective function of the residual sum of squares is constructed as shown below. : ; In the formula, This represents the total number of frequency points within the sensitive segment of the applied sinusoidal sweep signal. Calculation of the magnitude difference of complex impedance.

[0024] Based on this, the algorithm minimizes the objective function using the Gauss-Newton iterative method. During each iteration, the update logic of the parameter matrix vector is executed as follows: ; In the formula, The order represents the current iteration step. Indicates the first The estimated values ​​of the distribution parameters during each iteration. This indicates the updated estimate for the next step. It is a length of The residual vector, whose internal elements are the differences between the measured impedance and the theoretical impedance of the current model at each frequency point. ; This represents the Jacobian matrix with respect to the parameter vector, where each element is the partial derivative of the theoretical impedance function with respect to each distributed parameter. The structure is used to characterize the sensitivity of different distributed parameters to impedance changes at various frequency points; The symbol represents the matrix transpose operation, and the -1 symbol represents the matrix inversion operation.

[0025] When the difference between the objective functions of two consecutive iterations is less than the set convergence threshold, the iteration stops, and the output parameter matrix vector is used as the winding fingerprint parameter matrix in the final healthy state.

[0026] When applying this parameter matrix for boundary smoothing and noise correction, the finally identified fingerprint parameter matrix is ​​substituted back into the forward theoretical impedance function for solution, resulting in a theoretical fingerprint curve that fully conforms to the physical constraints of the distributed network. Because the measured curves contain high-frequency environmental glitches, the system uses the following dynamic weighted smoothing algorithm to complete the health baseline curve. Synthesis: ; In the formula, The weighting coefficients are dynamically adjusted based on the frequency sweep band. In the high-frequency sensitive band where noise is concentrated on-site, the system is lowered. The value of the fingerprint curve is increased to increase the constraint weight of the theoretical fingerprint curve, thereby filtering out abrupt noise caused by zero-point drift or high-frequency radiation in the measurement path. Finally, a healthy reference curve that combines the measured frequency response characteristics and circuit mechanism constraints is calculated.

[0027] S2. In the test state, apply the same frequency sweep excitation signal as in step S1 to the target winding and collect the test frequency response curve of the target winding. Furthermore, in S2, applying the sweep frequency excitation signal includes: Generate a sinusoidal sweep signal with a set amplitude and inject the sinusoidal sweep signal into the input terminal of the target winding; During the application of the sinusoidal sweep frequency signal, the characteristic impedance of the injection channel is adaptively adjusted to match the input impedance of the target winding.

[0028] Furthermore, in S2, the frequency response curve of the target winding to be measured is collected, including: Synchronously acquire the start-end voltage signal and end-end current signal of the target winding at various frequency points; The voltage signal at the beginning and the current signal at the end are simultaneously sampled and digitally converted to obtain discrete digital signals. The discrete digital signal is subjected to discrete Fourier transform, and the impedance magnitude and phase angle at each frequency point are calculated to synthesize the frequency response curve to be measured.

[0029] Specifically, the transformer winding inter-strand short circuit diagnosis method enters the stage of actual signal injection and response acquisition of the target winding under test. The operation process of this stage begins with the generation and injection of a standard excitation signal, followed by ensuring distortion-free signal transmission through dynamic impedance harmonization, and finally synthesizing the test frequency response curve for subsequent fault diagnosis through synchronous dual-channel acquisition and frequency domain transformation.

[0030] In practice, the process of applying a sweep frequency excitation signal to the target winding is triggered by a signal source generation mechanism. The signal generation unit within the test system generates a sinusoidal sweep frequency signal with a set amplitude. This signal amplitude remains constant throughout the sweep cycle to ensure uniform energy input. The generated sinusoidal sweep frequency signal is directly injected into the input terminal of the transformer target winding under test via a transmission channel. Because the transformer winding exhibits drastically fluctuating complex impedance characteristics at different frequencies, an impedance matching adjustment mechanism is introduced throughout the dynamic process of applying the sinusoidal sweep frequency signal to prevent power reflection loss and waveform distortion at the transmission boundary. The system adaptively adjusts the characteristic impedance of the injection channel, ensuring that its physical impedance characteristics track and match the current input impedance of the target winding in real time. This adaptive adjustment process operates based on the principle of minimizing the reflection coefficient. It uses a hardware adjustment network to change the equivalent impedance distribution of the injection channel, thereby eliminating waveform distortion caused by transmission line effects and ensuring that the set excitation energy can be completely and stably fed into the target winding.

[0031] As the sinusoidal sweep signal propagates through the target winding, the microscopic electromagnetic losses and structural changes within the winding directly modulate the response signal. To fully extract these electromagnetic responses characterizing inter-strand short circuits, the process of acquiring the frequency response curve under test is simultaneously initiated. The first-end acquisition unit at the input of the target winding and the last-end acquisition unit at the output of the target winding work together to synchronously acquire the first-end voltage signal and the last-end current signal of the target winding at each frequency point through which the sinusoidal sweep signal flows. To ensure absolute accuracy of the phase information, the acquisition system employs a dual-channel synchronous trigger clock to perform perfectly synchronous sampling and digitization of the acquired first-end voltage signal and last-end current signal. Through analog-to-digital conversion, the continuous time-domain analog signal is converted into a discrete digital signal containing discrete time series points.

[0032] After acquiring the discrete digital signal, the system calls the Discrete Fourier Transform algorithm to project the time-domain data into the frequency domain to calculate the precise impedance information at each frequency point. The time-domain sampled discrete voltage sequence... and current discrete sequence When performing series transformations, the corresponding transformation formulas are as follows: ; ; In the above formula, This represents the total number of discrete sampling points within one frequency sweep period. The cardinality index of the discrete time-domain sample points, with values ​​ranging from 0 to 1. Integers; This represents the corresponding discrete spectral line number in the frequency domain, and its value range is also from 0 to... Integers; This represents the imaginary unit in mathematical calculations. Using this algorithm, time-domain sequences are transformed into frequency-domain complex vectors. and .

[0033] After obtaining the frequency domain complex vector, the system uses the solved voltage frequency domain complex vector. With current frequency domain complex vector The comprehensive complex impedance of the target winding is calculated by complex division at each frequency point. Based on the rules of complex number operations, the impedance magnitude and phase angle at each frequency point are calculated, and the specific calculation formulas are shown below: ; ; In the above formula system, the physical meaning and mathematical definition of each symbolic parameter are strictly limited as follows: Indicates the target winding in the th... Comprehensive complex impedance at discrete spectral points in the frequency domain; Indicates the first The impedance modulus at discrete spectral points in the frequency domain, with the unit being ohms; Indicates the first The impedance phase angle at discrete spectral points in the frequency domain is expressed in radians. Represents a complex voltage vector The modulus; Represents the complex vector of current The modulus; Represents a complex voltage vector The real part value; Represents a complex voltage vector The value of the imaginary part; Represents the complex vector of current The real part value; Represents the complex vector of current The value of the imaginary part; Represents a complex voltage vector The phase angle; Represents the complex vector of current The phase angle; This represents the arctangent function in mathematical operations.

[0034] By traversing all the set sweep frequencies, the system sequentially combines the calculated impedance magnitude and phase angle at each frequency point according to frequency order, ultimately synthesizing the frequency response curve to be measured. This curve is then transmitted to the main control processing module for multi-dimensional loss feature extraction compared with a pre-stored health baseline curve.

[0035] S3. Compare the frequency response curve under test with the healthy reference curve to extract the frequency response attenuation characteristics and curve distortion characteristics that characterize the internal loss change state of the winding. Furthermore, in S3, the extraction of frequency response attenuation features and curve distortion features includes: By comparing the corresponding resonant peaks of the health baseline curve and the frequency response curve under test, the peak value difference is calculated as the peak value reduction, and the bandwidth difference at the set decibel drop in amplitude is calculated as the half-power bandwidth increase, so as to form the frequency response attenuation characteristics. The high-frequency sensitive segments of the health baseline curve and the frequency response curve under test are extracted, and the overall slope difference between the two is calculated as the change in the slope of the amplitude decrease in the high-frequency segment. The statistical correlation coefficient between the two in the corresponding frequency band is calculated as the curve correlation coefficient to form the curve distortion characteristics.

[0036] Specifically, the diagnostic method for inter-strand short circuits in transformer windings has entered the data processing and core feature extraction stage. This stage primarily focuses on comparing the frequency response curve under test with a healthy baseline curve. Through quantitative calculations, it delves into the geometric deformations hidden within the curves, thereby extracting frequency response attenuation characteristics and curve distortion characteristics that characterize the internal loss changes of the windings. This provides valuable data support for subsequent quantitative assessment and fault diagnosis.

[0037] In practical implementation, the extraction of frequency response attenuation characteristics begins with the comparison of the resonant peaks of the two sets of curves. The system automatically identifies and locates the corresponding resonant peaks in the healthy reference curve and the frequency response curve under test, and locks the coordinate data of the peak vertices of each resonant peak. By calculating the amplitude difference between the highest points of the corresponding resonant peaks, the peak value reduction is obtained, which characterizes the phenomenon of increased equivalent loop resistance and intensified electromagnetic energy dissipation caused by inter-strand short circuits. Next, the system finds the bandwidth at which the amplitude of each resonant peak drops from its maximum value by a set decibel. In specific processing, this set decibel is usually determined based on the definition of the half-power point. By calculating the difference between the bandwidth of the frequency response curve under test at the set decibel drop in amplitude and the bandwidth of the healthy reference curve at the corresponding position, the half-power bandwidth increase is obtained. The change in quality factor directly depends on the sharpness of the resonant peak, while the half-power bandwidth increase quantitatively reflects the increase in loop damping from the perspective of frequency domain broadening. The peak value reduction and the half-power bandwidth increase work together to form a complete frequency response attenuation characteristic.

[0038] Subsequently, the system initiates the curve distortion feature extraction process in parallel, extending the technical perspective to the high-frequency region, which is more susceptible to the influence of inter-strand micro-distributed capacitance losses. The system first extracts high-frequency sensitive segment data from the healthy baseline curve and the frequency response curve under test, and uses a univariate linear regression algorithm to fit the high-frequency waveforms of both sets of curves. The univariate linear regression algorithm calculates the fitting slope value, representing the overall downward trend of the high-frequency segment of the two curves, by minimizing the sum of squared residuals from the waveform data points to the fitted line. By calculating the difference between the high-frequency fitting slope of the frequency response curve under test and the high-frequency fitting slope of the healthy baseline curve, the change in the high-frequency amplitude decrease slope is obtained, which characterizes the degree of deterioration of the high-frequency additional iron loss and inter-strand insulation dielectric loss of the winding. To further capture abrupt changes and distortions in local waveforms, the system simultaneously calculates the statistical correlation coefficient of the two curves within the corresponding high-frequency sensitive segments. This statistical correlation coefficient is calculated using the Pearson correlation coefficient algorithm, and its specific calculation formula is shown below: ; In the above formula system, the physical meaning and mathematical definition of each symbolic parameter are strictly limited as follows: It represents the statistical correlation coefficient between the health baseline curve and the frequency response curve under test in the corresponding high-frequency sensitive segment, and its value ranges from negative one to positive one. This indicates the total number of discrete frequency sweep points contained within the extracted high-frequency sensitive segment; This represents the cardinality index of discrete data points within the high-frequency sensitive range, with values ​​ranging from one to... Integers; The health baseline curve represents the first... Impedance magnitude at each high-frequency data point; This indicates that the frequency response curve to be measured is in the first... Impedance magnitude at each high-frequency data point; It represents the arithmetic mean of all impedance magnitudes within the selected high-frequency sensitive segment of the health baseline curve; It represents the arithmetic mean of all impedance moduli within the selected high-frequency sensitive segment of the frequency response curve under test.

[0039] The statistical correlation coefficient calculated using the Pearson correlation coefficient algorithm is used as the curve correlation coefficient, which can highly sensitively reflect the overall waveform similarity break caused by the drift, disappearance, or addition of local resonant points due to inter-strand short circuits. Finally, the change in the slope of the high-frequency amplitude decrease is combined with the curve correlation coefficient to jointly constitute the curve distortion characteristics. By completely separating and extracting the frequency response attenuation characteristics and curve distortion characteristics from the measured curve, the system successfully transforms complex physical phenomena into deterministic quantitative feature vectors, providing an impeccable data foundation for the next stage of constructing multi-condition fault determination.

[0040] S4. Calculate the rate of change of the quality factor of the resonance peak using the extracted frequency response attenuation characteristics, and calculate the corresponding rate of change of distortion based on the curve distortion characteristics, so as to quantify the electromagnetic energy loss state inside the target winding. Furthermore, in S4, the calculation of the rate of change of the quality factor of the resonance peak includes: Based on the center frequencies of the corresponding resonant peaks of the health baseline curve and the increase in half-power bandwidth, the decrease in the quality factor of the frequency response curve under test relative to the health baseline curve is calculated. The rate of change of the quality factor is obtained by calculating the ratio of the decrease in quality factor to the original quality factor of the health baseline curve.

[0041] Furthermore, in S4, the calculation of the corresponding distortion rate includes: The high-frequency attenuation rate in the distortion rate is obtained by calculating the ratio of the change in the slope of the high-frequency amplitude decrease to the original slope of the health baseline curve in the corresponding high-frequency sensitive segment. The correlation distortion rate in the distortion rate is obtained by calculating the difference between the curve correlation coefficient of the frequency response curve under test and the healthy baseline curve. The distortion rate is obtained by weighted summation of the high-frequency attenuation change rate and the correlation distortion rate.

[0042] Specifically, the transformer winding inter-strand short circuit diagnosis method has entered the stage of quantitative calculation of core parameters. The operation process of this stage mainly focuses on transforming the frequency response attenuation characteristics and curve distortion characteristics extracted from the previous stage into two high-dimensional control indicators that can directly characterize the degree of winding fault through specific mathematical mapping and composite solution algorithms. These indicators are the rate of change of quality factor and the rate of change of distortion, thereby achieving accurate quantification of the electromagnetic energy loss state inside the target winding.

[0043] In practical implementation, the process of calculating the rate of change of the quality factor of the resonant peak begins by establishing the baseline parameters under healthy conditions. The system reads the center frequencies of the corresponding resonant peaks of the healthy baseline curve and obtains the half-power bandwidth increase calculated by the previous stage. To quantitatively characterize the energy dissipation caused by the short-circuit loop, the system first calculates the decrease in the quality factor of the frequency response curve under test relative to the healthy baseline curve. Since the quality factor is inversely proportional to the resonant bandwidth, when an inter-strand short-circuit fault occurs, the increased loop damping leads to bandwidth broadening, and the quality factor decreases accordingly. After calculating the decrease in the quality factor of each resonant peak, the system uses the ratio of this decrease to the original quality factor of the healthy baseline curve to calculate the rate of change of the quality factor, which reflects the relative attenuation of the quality factor. This indicator, based on the microscopic physical mechanism of energy dissipation, accurately quantifies the degree of electromagnetic energy loss of the winding near the resonant point.

[0044] Simultaneously, the system initiates the parallel calculation process for the distortion rate of change to quantitatively assess the overall waveform deformation caused by changes in inter-strand distributed capacitance and increased dielectric loss. This calculation process consists of three sub-steps: First, the system retrieves the change in the slope of the high-frequency amplitude drop extracted from the previous stage and calculates its ratio with the original slope of the healthy reference curve in the corresponding high-frequency sensitive segment, thereby solving for the first component of the distortion rate of change, namely the high-frequency attenuation rate of change; Second, the system uses the curve correlation coefficient of the frequency response curve under test relative to the healthy reference curve calculated from the previous stage and calculates its difference with the perfectly linearly correlated reference value, thereby solving for the second component of the distortion rate of change, namely the correlation distortion rate. The high-frequency attenuation rate of change primarily characterizes the drift of the high-frequency trend, while the correlation distortion rate primarily characterizes the abrupt change of the local waveform; the two describe curve distortion from different dimensions.

[0045] To organically integrate the waveform distortion at these two levels, the system introduces a weighted synthesis mechanism in the final step. This mechanism weights and sums the high-frequency attenuation change rate and the correlation distortion rate to synthesize the final distortion change rate. The specific calculation formula for this weighted synthesis process is shown below: ; In the above formula system, the physical meaning and mathematical definition of each symbolic parameter are strictly limited as follows: This represents the distortion rate obtained from the final synthesis, used to comprehensively and quantitatively characterize the degree of deformation of the high-frequency curve; This represents the rate of change of attenuation in the high-frequency band, which is calculated from the ratio of the change in the slope of the amplitude decrease in the high-frequency band to the original slope. This represents the correlation distortion rate, which is derived from the calculation of the difference between the curve correlation coefficient and the benchmark value. This represents the adaptive weighting coefficient assigned to the attenuation rate of the high-frequency band, and its value ranges from zero to one. This represents the adaptive weighting coefficient assigned to the correlation distortion rate, and its value also ranges from zero to one. and The arithmetic sum of these values ​​is always equal to one. Before using the above formula for weighted summation, the system must first normalize the high-frequency attenuation change rate and the correlation distortion rate. Since the high-frequency attenuation change rate originates from the slope ratio, and the correlation distortion rate originates from the correlation coefficient difference, their original numerical ranges and dimensions are completely different. To eliminate the influence of dimensional differences on the synthesis results, the system uses a linear mapping algorithm to scale both values ​​to [the appropriate scale]. Within a unified numerical range, it is transformed into a standardized dimensionless value with unified dimensions.

[0046] In practical applications, the weighting coefficients are adjusted in real time based on the intensity of electromagnetic interference noise, thereby ensuring that the synthesized distortion rate has extremely high anti-interference stability. Ultimately, the calculated quality factor rate of change and the distortion rate of change together constitute a two-dimensional feature vector, perfectly quantifying the electromagnetic energy loss state inside the target winding, providing direct criterion input for the next stage of multi-threshold fault determination.

[0047] S5. Construct a loss change criterion composed of the quality factor change rate and the distortion change rate. When either change rate exceeds the corresponding preset threshold, it is determined that there is an inter-strand short circuit fault inside the target winding. Furthermore, in S5, the criteria for determining loss changes include: Set a first preset threshold corresponding to the rate of change of quality factor and a second preset threshold corresponding to the rate of change of distortion, respectively. The rate of change of quality factor is compared with the first preset threshold, and the rate of change of distortion is compared with the second preset threshold to construct a criterion for loss change. When the rate of change of quality factor is greater than the first preset threshold, or the rate of change of distortion is greater than the second preset threshold, a judgment signal that satisfies the loss change criterion is output.

[0048] Specifically, the transformer winding inter-strand short circuit diagnosis method has entered the core fault determination and decision-making stage. The operation process of this stage mainly focuses on using the quality factor change rate and distortion change rate calculated by the previous stage, and constructing a two-dimensional multi-threshold loss change criterion to achieve the final logical decision on whether an inter-strand short circuit fault has occurred inside the transformer under test.

[0049] In practical implementation, the process of constructing loss change criteria begins with system initialization and threshold configuration. The system sets a first preset threshold for the rate of change of the quality factor and a second preset threshold for the rate of change of distortion in the expert strategy library. These two preset thresholds serve as safety boundary lines for determining whether the electromagnetic state inside the winding has undergone fundamental deterioration. Their values ​​are typically generated based on the data distribution patterns of historical healthy transformers under large-sample frequency sweep tests. Subsequently, the system compares the actual rate of change of the quality factor calculated in the previous stage with the first preset threshold, and simultaneously compares the actual rate of change of distortion with the second preset threshold. Through these two sets of parallel data comparison actions, the system establishes a loss change criterion with multi-threshold collaborative verification at the logical level.

[0050] To transform the comparison results into physical actions with controllable characteristics, the execution mechanism of the criterion employs an "OR logic" threshold-triggered control algorithm. The core of this algorithm is to perform a logical OR operation on two independent multi-threshold comparison results. When the rate of change of the quality factor exceeds a first preset threshold, or the rate of change of distortion exceeds a second preset threshold, the control logic is triggered, outputting a judgment signal that satisfies the loss change criterion. The specific control decision formula is shown below: ; In the above formula system, the physical meaning and mathematical definition of each symbolic parameter are strictly limited as follows: This represents the final output short-circuit fault judgment signal. When its value is one, it means that the loss change criterion is met and there is an inter-strand short-circuit fault. When its value is zero, it means that the criterion is not met and the winding is in a normal state. This represents the rate of change of the quality factor obtained from the previous stage calculation, used to measure the degree of energy dissipation inside the winding; This represents the first preset threshold corresponding to the rate of change of the quality factor; This represents the distortion rate obtained by weighted synthesis after normalization of the preceding stages, used to measure the overall degree of deformation of the high-frequency curve; This represents the second preset threshold corresponding to the distortion rate. The logical OR operator is used in mathematical calculations and computer programming. This represents the standard step activation function, whose internal operational logic is defined as follows: when the independent variable is greater than zero, the function output value is always equal to one; when the independent variable is less than or equal to zero, the function output value is always equal to zero.

[0051] In practical applications, this control decision signal is transmitted in real time to the alarm trigger module of the transformer status monitoring terminal. If either of the two judgment conditions is met—either a short circuit between strands causing excessive loop losses leading to a sharp drop in the quality factor, or distortion of the distributed capacitance causing severe deformation of the high-frequency waveform—the system can immediately output a judgment signal that meets the criteria through an activation function. This dual-indicator parallel underlying technology effectively prevents missed detections that may occur under complex electromagnetic interference conditions in the field, ensuring the comprehensiveness and high reliability of the early-stage short circuit hazard judgment conclusions.

[0052] S6. Based on the attenuation degree and curve distortion trend of the frequency response curves under test in the low-frequency, mid-frequency, and high-frequency bands, the location and severity of the inter-strand short-circuit fault are evaluated by frequency band.

[0053] Furthermore, in S6, based on the attenuation degree and curve distortion trend of the frequency response curve under test in the low-frequency, mid-frequency, and high-frequency bands, the following are included: In the low-frequency range, the overall inductance change of the multi-strand parallel winding is evaluated based on the change in loop inductance characteristics of the frequency response curve under test relative to the healthy reference curve. In the mid-frequency band, the inter-strand coupling change state of the multi-strand parallel winding is evaluated based on the offset of the resonant peak position of the frequency response curve under test relative to the healthy reference curve. In the high-frequency band, the inter-strand additional loss and insulation status of multi-strand parallel windings are evaluated based on the rate of change of quality factor and the rate of change of distortion. By comprehensively considering the overall inductance change state, inter-strand coupling change state, inter-strand additional loss change state, and insulation state change trend, the fault location of the inter-strand short circuit fault is determined, and the severity of the inter-strand short circuit fault is determined based on the characteristic rate of change amplitude in each corresponding frequency band.

[0054] Specifically, the diagnostic method for inter-strand short circuits in transformer windings has entered the stage of refined assessment and location based on frequency bands. The operational process in this stage mainly focuses on using a frequency-division diagnostic mechanism triggered by the preceding judgment signal to perform multi-dimensional calculations on the different physical manifestations of the frequency response curves under test in the low-frequency, mid-frequency, and high-frequency bands, ultimately achieving a joint quantitative assessment of the location and severity of inter-strand short circuit faults.

[0055] In practice, the frequency band evaluation process is carried out in parallel or sequentially, proceeding from low to high frequency. In the low-frequency band, since the physical characteristics of the transformer windings are mainly dominated by the leakage magnetic field and the main magnetic flux, the waveform shape strongly depends on the macroscopic geometry of the windings. The system evaluates the overall inductance change of the multi-strand parallel windings based on the change in loop inductance characteristics of the measured frequency response curve relative to the healthy reference curve. The change in loop inductance characteristics is obtained by extracting the slope difference of the impedance modulus in the low-frequency non-resonant region, reflecting the degree of overall leakage inductance collapse caused by the reduction of parallel branches and the enhancement of reverse magnetic flux due to inter-strand short circuits.

[0056] In the mid-frequency range, the mutual inductance between parallel conductors and the distributed capacitance to ground and between strands begin to resonate together. The system evaluates the inter-strand coupling change state of the multi-strand parallel winding based on the resonant peak position offset of the measured frequency response curve relative to the healthy reference curve. The resonant peak position offset is obtained by locking the center frequency of a specific order resonant peak and calculating its frequency difference, which reflects the destruction of the inter-strand electromagnetic coupling coefficient and the reconstruction state of the local equivalent capacitance grid caused by short-circuit circulating current in the local winding.

[0057] In the high-frequency range, the distributed capacitance of the windings dominates, and the skin effect and proximity effect are sharply enhanced. Based on the rate of change of quality factor and the rate of change of distortion calculated by the previous stage, the system evaluates the changes in the inter-strand additional losses and the trend of insulation status of the multi-strand parallel windings. This step quantitatively captures the microscopic electromagnetic deformation caused by the degradation of the insulating medium and the sharp increase in eddy current losses near the short-circuit point by monitoring the damped dissipation of the high-frequency resonant peak and the high-frequency waveform distortion.

[0058] After completing the independent evaluation of the three frequency bands mentioned above, the system initiates a multi-dimensional matrix mapping algorithm. This algorithm integrates the overall inductance change state, the inter-strand coupling change state, the inter-strand additional loss change state, and the insulation state change trend to construct a position-sensitive feature vector. To determine the location of an inter-segment short-circuit fault. Location-sensitive feature vector. The relationship with the axial longitudinal position of the winding is achieved through the following mapping matrix formula: ; In the above formula system, the physical meaning and mathematical definition of each symbolic parameter are strictly limited as follows: This represents a position-sensitive feature vector, which contains regional confidence probability indices representing different circumferential orientations and axial heights of the winding. This represents a preset frequency band-space coupling mapping matrix, where each matrix element represents the sensitivity contribution coefficient of the physical state of different frequency bands to faults at specific locations in space. The input frequency band characteristic state vector is a four-dimensional column vector whose ordered elements are composed of the change in loop inductance characteristics in the low-frequency band, the offset of the resonant peak position in the mid-frequency band, and the combination of the rate of change of quality factor and the rate of change of distortion in the high-frequency band.

[0059] While determining the fault location through the matrix multiplication method described above, the decision module uses the Euclidean distance weighted calculation formula to determine the severity index of the inter-strand short-circuit fault based on the characteristic rate of change amplitude within each corresponding frequency band. : ; In the above severity assessment formula system, the physical meaning and mathematical definition of each symbolic parameter are strictly limited as follows: The index represents the severity of the inter-strand short circuit fault in the final output. The higher the value, the higher the degree of inter-strand short circuit burnout or the higher the proportion of damaged parallel strands. This represents the rate of change of the loop inductance characteristic calculated in the low-frequency range; The low-frequency severity weighting coefficient represents the rate of change of the inductance characteristics of the loop. This represents the scaled value of the resonant peak position offset calculated in the mid-frequency band after normalization to the maximum bandwidth. This represents the mid-frequency severity weighting coefficient that assigns the resonant peak position offset. This represents the rate of change of the quality factor obtained from the previous calculation. This represents the distortion rate obtained from the previous calculation. This represents the high-frequency severity weighting coefficient that assigns high-frequency loss distortion characteristics.

[0060] In practical applications, the calculated position-sensitive feature vector Severity index The data is input into the transformer operation and maintenance management system. Through the visualization of this data, on-site operation and maintenance personnel can not only accurately identify which part of the internal parallel winding has an inter-strand contact fault, but also take targeted operation and maintenance strategies according to the severity of the fault, such as online load restriction or direct shutdown and core hoisting for maintenance. This enables proactive prevention and precise maintenance of latent electromagnetic hazards inside the transformer.

[0061] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics, characterized in that, include: S1. When the transformer is offline, apply a sweep frequency excitation signal to the multi-strand parallel winding of the transformer in a healthy state, obtain the frequency response impedance curve in the healthy state, and establish a healthy reference curve based on the distributed parameter equivalent model of the multi-strand parallel winding. S2. In the test state, apply the same frequency sweep excitation signal as in step S1 to the target winding and collect the test frequency response curve of the target winding. S3. Compare the frequency response curve to be tested with the healthy baseline curve, and extract the frequency response attenuation characteristics and curve distortion characteristics that characterize the internal loss change state of the winding. S4. Calculate the rate of change of the quality factor of the resonance peak using the extracted frequency response attenuation characteristics, and calculate the corresponding rate of change of distortion based on the curve distortion characteristics, so as to quantify the electromagnetic energy loss state inside the target winding. S5. Construct a loss change criterion composed of the quality factor change rate and the distortion change rate. When either change rate exceeds the corresponding preset threshold, it is determined that there is an inter-strand short circuit fault inside the target winding. S6. Based on the attenuation degree and curve distortion trend of the frequency response curves under test in the low-frequency, mid-frequency, and high-frequency bands, the location and severity of the inter-strand short-circuit fault are evaluated by frequency band.

2. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S1, obtaining the frequency response impedance curve under the healthy state includes: A sinusoidal sweep frequency signal is applied to a multi-strand parallel winding of a transformer in a healthy state. The frequency range of the sinusoidal sweep frequency signal covers the low-frequency sensitive segment, the medium-frequency sensitive segment, and the high-frequency sensitive segment of the multi-strand parallel winding. The application method of the sinusoidal sweep frequency signal includes continuous sweep frequency or segmented sweep frequency. The response terminal voltage and response terminal current of the multi-strand parallel winding are measured at various frequency points, and the frequency response impedance curve under healthy conditions is calculated using the response terminal voltage and response terminal current.

3. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S1, establishing the health baseline curve includes: Construct a distributed network model of the multi-strand parallel winding, wherein the distributed network model includes the resistance parameters and self-inductance parameters corresponding to each strand of conductor, the inter-strand electromagnetic coupling parameters and inter-strand capacitance parameters between any two strands of conductor, and the insulation equivalent impedance parameters used to characterize the inter-strand insulation state. The frequency response impedance curve obtained under the healthy state is substituted into the distributed network model for parameter identification to obtain the winding fingerprint parameter matrix under the healthy state. The frequency response impedance curve under the healthy state is smoothed at the boundary and noise is corrected using the winding fingerprint parameter matrix to obtain the healthy baseline curve.

4. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S2, the applied frequency sweep excitation signal includes: Generate a sinusoidal sweep signal with a set amplitude, and inject the sinusoidal sweep signal into the input terminal of the target winding; During the application of the sinusoidal sweep frequency signal, the characteristic impedance of the injection channel is adaptively adjusted to match the input impedance of the target winding.

5. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S2, the frequency response curve of the target winding to be measured is acquired, including: The starting voltage signal and the ending current signal of the target winding at each frequency point are collected simultaneously. The first-end voltage signal and the last-end current signal are synchronously sampled and digitally converted to obtain discrete digital signals; The discrete digital signal is subjected to discrete Fourier transform to calculate the impedance magnitude and phase angle at each frequency point, and the frequency response curve to be measured is synthesized.

6. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S3, the extraction of frequency response attenuation features and curve distortion features includes: By comparing the corresponding resonant peaks of the health baseline curve and the frequency response curve under test, the peak value difference is calculated as the peak value reduction, and the bandwidth difference at the set decibel drop in amplitude is calculated as the half-power bandwidth increase, so as to form the frequency response attenuation characteristics. The high-frequency sensitive segments of the health baseline curve and the frequency response curve under test are extracted, and the overall slope difference between the two is calculated as the change in the slope of the amplitude decrease in the high-frequency segment. The statistical correlation coefficient between the two in the corresponding frequency band is calculated as the curve correlation coefficient to form the curve distortion characteristics.

7. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S4, the calculation of the rate of change of the quality factor of the resonance peak includes: Based on the center frequencies of the corresponding resonant peaks of the health reference curve and the increase in half-power bandwidth, the decrease in the quality factor of the frequency response curve under test relative to the health reference curve is calculated. The rate of change of the quality factor is obtained by calculating the ratio of the decrease in the quality factor to the original quality factor of the health baseline curve.

8. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S4, the calculation of the corresponding distortion rate includes: The high-frequency band attenuation rate in the distortion rate is obtained by calculating the ratio of the change in the slope of the amplitude drop in the high-frequency band to the original slope of the health baseline curve in the corresponding high-frequency sensitive segment. The correlation distortion rate in the distortion rate is obtained by calculating the difference between the curve correlation coefficient of the frequency response curve under test and the healthy baseline curve. The high-frequency band attenuation change rate and the correlation distortion rate are weighted and summed to obtain the distortion change rate.

9. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S5, the criteria for determining loss variation include: Set a first preset threshold corresponding to the rate of change of the quality factor and a second preset threshold corresponding to the rate of change of distortion, respectively. The rate of change of the quality factor is compared with a first preset threshold, and the rate of change of distortion is compared with a second preset threshold to construct a loss change criterion. When the rate of change of the quality factor is greater than the first preset threshold, or the rate of change of the distortion is greater than the second preset threshold, a judgment signal that satisfies the loss change criterion is output.

10. The method for diagnosing inter-strand short circuits in transformer windings based on frequency response loss characteristics according to claim 1, characterized in that, In S6, the step of determining the attenuation degree and curve distortion trend of the frequency response curve under test in the low-frequency, mid-frequency, and high-frequency bands includes: In the low-frequency range, the overall inductance change state of the multi-strand parallel winding is evaluated based on the change in loop inductance characteristics of the frequency response curve under test relative to the healthy reference curve. In the mid-frequency band, the inter-strand coupling change state of the multi-strand parallel winding is evaluated based on the offset of the resonant peak position of the frequency response curve under test relative to the healthy reference curve. In the high-frequency band, the inter-strand additional loss and insulation state of the multi-strand parallel winding are evaluated based on the rate of change of the quality factor and the rate of change of the distortion. By combining the overall inductance change state, inter-strand coupling change state, inter-strand additional loss change state, and insulation state change trend, the fault location of the inter-strand short circuit fault is determined, and the severity of the inter-strand short circuit fault is determined based on the characteristic rate of change amplitude in each corresponding frequency band.

Images