A transformer winding fault diagnosis method based on group delay residual network

Abstract

The application relates to the general power transformer winding fault diagnosis technical field, in particular to a transformer winding fault diagnosis method based on a group delay residual network, which comprises the following steps: after injecting a sweep signal into one end of a transformer winding, measuring a response signal output from the other end, and obtaining a phase frequency response according to the measurement result; performing unwinding processing on the phase frequency response to obtain unwound phase frequency response; constructing a discrete Gaussian smoothing filter; performing noise filtering processing on the unwound phase frequency response by using the discrete Gaussian smoothing filter to obtain denoised phase frequency response; performing derivation processing on the diagonal frequency of the denoised phase frequency response to obtain a group delay spectrum; inputting the group delay spectrum into a one-dimensional residual network for feature extraction and classification to obtain a fault diagnosis result of the transformer winding to be measured.

Description

Technical Field

[0001] This application relates to the field of power transformer winding fault diagnosis technology, and in particular to a transformer winding fault diagnosis method based on a group delay residual network. Background Technology

[0002] Transformers are core hub equipment in power systems. During long-term operation, especially under short-circuit electrodynamic impacts or mechanical vibrations, transformers are prone to winding deformation and other faults. If these faults are not detected and addressed promptly, they often evolve into serious insulation breakdowns or even equipment failures over time, posing a significant threat to the safe and stable operation of the power grid. Therefore, fault diagnosis of transformer windings has extremely important engineering application value.

[0003] Currently, the industry widely uses swept-frequency response analysis to test and diagnose transformer winding deformation. Traditional swept-frequency response analysis mainly relies on the macroscopic morphological comparison of amplitude-frequency response curves. However, the inside of a transformer winding is an extremely complex network of distributed inductance and capacitance. Minor changes in local parameters caused by slight mechanical deformation of the winding are often not significant on the amplitude-frequency curve, leading to insufficient sensitivity of traditional methods for diagnosing winding faults. Therefore, a better transformer winding fault diagnosis scheme based on group delay residual networks is needed to overcome these difficulties. Summary of the Invention

[0004] In view of this, this application provides a transformer winding fault diagnosis method based on a group delay residual network, in order to improve the diagnostic sensitivity of transformer winding faults.

[0005] In a first aspect, this application provides a transformer winding fault diagnosis method based on a group delay residual network, the transformer winding fault diagnosis method based on a group delay residual network comprising: After injecting a sweep frequency signal into one end of the transformer winding, the response signal output from the other end is measured, and the phase frequency response is obtained based on the measurement results. The transformer winding includes a normal transformer winding and a transformer winding under test. The phase frequency response is unwound to obtain the unwound phase frequency response. The unwound process is used to eliminate the phase jump generated when calculating the phase frequency response. A discrete Gaussian smoothing filter is constructed, and the discrete Gaussian smoothing filter is used to filter out noise from the unwound phase frequency response to obtain the denoised phase frequency response. The group delay spectrum is obtained by differentiating the denoised phase frequency response with respect to the angular frequency. The group delay spectrum is input into a one-dimensional residual network for feature extraction and classification to obtain the fault diagnosis results of the transformer winding under test.

[0006] Furthermore, the frequency range of the sweep signal is 1-2MHz and consists of 2000 frequency points, and the sweep signal constitutes a sweep signal sequence; the response signal constitutes a response signal sequence; the phase frequency response is the phase angle of the ratio of the sweep signal to the response signal, and the phase frequency response constitutes a phase frequency response sequence.

[0007] Furthermore, the unwinding process of the phase frequency response to obtain the unwound phase frequency response includes: Substitute the phase frequency responses of adjacent frequency points in the phase frequency response sequence into the following formula. ΔH(ω)=H(ω)-H(ω-1) The phase difference is calculated and the phase difference constitutes a phase difference sequence, where H(ω) is the phase frequency response at frequency point ω, H(ω-1) is the phase frequency response at frequency point ω-1, ΔH(ω) is the phase difference at frequency point ω, and ω∈[0,2000]. Substitute the phase difference value into the following formula

[0008] The phase compensation amount is calculated and constitutes a phase compensation amount sequence, where C(ω) is the phase compensation amount at frequency point ω and C(ω-1) is the phase compensation amount at frequency point ω-1. Substitute the phase frequency response and the phase compensation amount into the following formula H un (ω)=H(ω)+C(ω) The phase frequency response after unwinding is calculated, and the phase frequency response after unwinding constitutes a sequence of phase frequency responses after unwinding, wherein H un (ω) is the phase frequency response after unwinding at frequency point ω.

[0009] Furthermore, the step of constructing a discrete Gaussian smoothing filter and using the discrete Gaussian smoothing filter to perform noise filtering on the unwound phase frequency response to obtain a denoised phase frequency response includes: Use the following formula

[0010] Construct a discrete Gaussian kernel function, where G(k) is the discrete Gaussian kernel function, σ is the standard deviation of the discrete Gaussian kernel function, the index k∈[-M,+M], and 2M+1 is the window length of the discrete Gaussian smoothing filter. The discrete Gaussian kernel functions constitute a sequence of discrete Gaussian kernel functions; and use the following formula...

[0011] The Gaussian kernel function corresponding to the discrete Gaussian kernel function is normalized to obtain normalized weights. These normalized weights form a normalized weight sequence, where g(k) is the normalized weight. The following formula is used...

[0012] The unwound phase frequency response is discretely convolved with the normalized weights to obtain the denoised phase frequency response. The denoised phase frequency response constitutes a denoised phase frequency response sequence, where H... f (ω) is the denoised phase-frequency response sequence at frequency point ω, H un (ω-k) is the phase frequency response after unwinding at frequency point ω-k.

[0013] Furthermore, the transformer winding fault diagnosis method based on group delay residual network includes: When calculating the convolution at both ends of the denoised phase frequency response sequence, for out-of-bounds indices where ω-k < 1 or ω-k > 2000, the following formula is used.

[0014] Boundary processing is performed on the unwound phase frequency response sequence to achieve edge data extension of the unwound phase frequency response sequence.

[0015] Furthermore, the step of differentiating the denoised phase frequency response with respect to the angular frequency to obtain the group delay spectrum includes: Use the following formula

[0016] Differentiating the angular frequency yields the group delay spectrum, where τ(ω) is the group delay spectrum at frequency ω, and ωk is the angular frequency at the k-th frequency point, based on the following formula.

[0017] We obtain fk, where fk is the frequency of the kth frequency point.

[0018] Furthermore, the step of inputting the group delay spectrum into a one-dimensional residual network for feature extraction and classification to obtain the fault diagnosis result of the transformer winding under test includes: The group delay spectrum is treated as a one-dimensional sequence signal and standardized to obtain a standardized group delay sequence. The standardized group delay sequence is input into a pre-trained one-dimensional residual network, and the deformation features of the group delay spectrum are extracted through the one-dimensional convolutional layer and residual block in the one-dimensional residual network. The extracted local deformation features are input into a fully connected classification layer to calculate and output the probability distribution of the transformer winding under test belonging to each preset fault state. The preset fault state corresponding to the maximum probability value in the probability distribution is selected as the fault diagnosis result of the transformer winding under test. The preset fault states specifically include: normal, slight deformation, moderate deformation, and severe deformation.

[0019] Secondly, this application provides a transformer winding fault diagnosis device based on a group delay residual network, the transformer winding fault diagnosis device based on a group delay residual network comprising: The signal module is used to inject a sweep frequency signal into one end of the transformer winding, measure the response signal output at the other end, and obtain the phase frequency response based on the measurement result. The transformer winding includes a normal transformer winding and a transformer winding under test. The unwinding module is used to unwind the phase frequency response to obtain the unwound phase frequency response. The unwinding process is used to eliminate the phase jump generated when calculating the phase frequency response. The filtering module is used to construct a discrete Gaussian smoothing filter and use the discrete Gaussian smoothing filter to perform noise filtering on the unwound phase frequency response to obtain a denoised phase frequency response. The group delay spectrum module is used to perform derivative processing of the denoised phase frequency response with respect to the angular frequency to obtain the group delay spectrum. The diagnostic module is used to input the group delay spectrum into a one-dimensional residual network for feature extraction and classification, so as to obtain the fault diagnosis result of the transformer winding under test.

[0020] Thirdly, embodiments of this application also provide a transformer winding fault diagnosis system based on a group delay residual network. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the methods described above.

[0021] Fourthly, this application provides a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement the above-described method. The computer-readable storage medium may be volatile or non-volatile.

[0022] Fifthly, this application provides an electronic device, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above-described method when executing instructions stored in the memory.

[0023] Sixthly, this application provides a computer program product including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0024] In the embodiments of this specification, the physical characteristic of high sensitivity of group delay to transformer resonant point shift is utilized to extract the group delay features of winding deformation for fault diagnosis, significantly improving diagnostic sensitivity and effectively overcoming the shortcomings of traditional amplitude-frequency response analysis methods in being insensitive to winding deformation. Simultaneously, through the combined processing of phase dewinding and discrete Gaussian smoothing filtering, the inherent phase jump in instrument measurement is eliminated and noise is filtered out. Furthermore, without introducing artificial frequency shifts, the noise amplification effect caused by discrete differentiation is suppressed, achieving noise-resistant winding feature extraction. In addition, by inputting the group delay spectrum into a one-dimensional residual network for feature extraction and classification, subjective qualitative assessment relying on human experience is transformed into quantitative analysis, realizing an objective quantitative and hierarchical diagnostic system encompassing normal to severe deformation, improving the reliability of transformer condition monitoring and its practical engineering application value. Attached Figure Description

[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0026] Figure 1 The diagram above illustrates a flowchart of a transformer winding fault diagnosis method based on a group delay residual network.

[0027] Figure 2 The diagram above illustrates a schematic of the equipment used in a transformer winding fault diagnosis method based on a group delay residual network.

[0028] Figure 3 The example shown is a schematic diagram of a process for processing different transformer windings separately.

[0029] Figure 4The diagram illustrates a phase-frequency response plot under different fault levels.

[0030] Figure 5 The diagram illustrates the phase-frequency response after unwinding under different fault conditions.

[0031] Figure 6 The example shown is a phase frequency response diagram after denoising under different fault levels.

[0032] Figure 7 The example shown is a group delay spectrum under different fault levels.

[0033] Figure 8 The diagram above illustrates a schematic of a one-dimensional residual network.

[0034] Figure 9 The diagram above illustrates a flowchart of a transformer winding fault diagnosis device based on a group delay residual network.

[0035] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0036] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0037] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0038] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0039] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0040] The phase frequency response obtained in transformer winding deformation testing should contain extremely rich information on resonant poles and zeros. However, due to the inherent phase jump phenomenon generated when the instrument calculates the arctangent function and the noise caused by the electromagnetic environment on site, the original phase data is extremely difficult to be directly and effectively used in actual engineering.

[0041] Group delay, as the derivative of the phase frequency response with respect to angular frequency, is extremely sensitive to the shifts in poles and zeros in a transformer's high-frequency resonant network. It can transform minute changes in the distributed parameters of the windings into significant spectral polarity reversals and frequency shifts. However, without effective data preprocessing, directly performing discrete numerical differentiation on the noisy and abrupt raw phase data will produce an extremely severe error amplification effect, causing the true physical characteristics to be completely obscured by noise glitches.

[0042] Furthermore, current winding deformation diagnosis still largely relies on the subjective experience and judgment of testers, lacking a complete system capable of extracting high-fidelity features from complex phase-frequency data for quantitative and automatic classification diagnosis. Therefore, there is an urgent need to develop a quantitative and automatic classification diagnosis method for transformer winding faults that is both noise-resistant and highly sensitive, to overcome the many shortcomings of existing technologies.

[0043] Figure 1 A flowchart illustrating a transformer winding fault diagnosis method based on a group delay residual network according to an embodiment of this disclosure is shown. This method can be applied to a transformer winding fault diagnosis device based on a group delay residual network. The transformer winding fault diagnosis device based on a group delay residual network can be a terminal device, a server, or other processing device. The terminal device can be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, vehicle-mounted device, wearable device, etc.

[0044] In some possible implementations, this transformer winding fault diagnosis method based on group delay residual networks can be implemented by a processor calling computer-readable instructions stored in memory.

[0045] like Figure 1 As shown, the transformer winding fault diagnosis method based on group delay residual network may include: In step S11, after injecting a sweep frequency signal into one end of the transformer winding, the response signal output from the other end is measured, and the phase frequency response is obtained based on the measurement results.

[0046] The transformer windings include normal transformer windings and transformer windings under test.

[0047] Figure 2 A schematic diagram of the equipment used in the transformer winding fault diagnosis method based on group delay residual networks is shown. Figure 2 As shown, the equipment involved in the transformer winding fault diagnosis method based on group delay residual networks includes at least: a transformer winding, which can be set to normal, slight, moderate, and severe deformation; a frequency response analyzer for obtaining phase frequency response data, which includes test clamps for injecting sweep signals and measuring response signals; and a processor for collecting and processing the acquired phase frequency response data. At least three lines are included between the transformer winding and the frequency response analyzer: a sweep signal injection line, a sweep signal measurement line, and a response signal measurement line.

[0048] Figure 3 This is a flowchart illustrating the separate processing of the windings of a normal transformer and the windings of the transformer under test. (For example...) Figure 3 As shown, a sweep frequency signal can be injected into one end of a normal transformer winding. The frequency range of this sweep frequency signal can be 1kHz-2MHz, and it can form a sweep frequency signal sequence containing 2000 frequency points. The response signal is then measured at the other end of the normal transformer winding, and this response signal can form a response signal sequence.

[0049] Furthermore, based on the measurement results, the phase frequency response corresponding to the normal transformer winding can be obtained. This phase frequency response can be the phase angle of the ratio of the sweep signal to the response signal (corresponding to the sweep signal). When the sweep signal is labeled U1 and the response signal is labeled U2, the phase angle can be labeled arg(U2 / U1). This phase frequency response constitutes a phase frequency response sequence. Obviously, the number of frequency points in the response signal sequence and the phase frequency response sequence is also 2000.

[0050] Similarly, a sweep frequency signal can be injected into one end of the transformer winding under test, and the response signal can be measured at the other end of the winding. Based on the measurement results, the phase frequency response of the transformer winding under test can be obtained. The specific process for determining the phase frequency response of the transformer winding under test can be found in the description of the process for determining the phase frequency response of a normal transformer winding, and will not be repeated here.

[0051] Figure 4 The phase-frequency response diagrams under different fault degrees are shown. For example... Figure 4As shown, the four lines represent the phase frequency response of the transformer winding under normal, mild, moderate, and severe deformation conditions.

[0052] In step S12, the phase frequency response is unwound to obtain the unwound phase frequency response.

[0053] The unwinding process is used to eliminate phase jumps generated during the calculation of the phase frequency response. For example... Figure 3 As shown, the phase frequency response of both the normal transformer winding and the transformer winding under test can be de-wound to eliminate phase jumps and obtain their respective de-wound phase frequency responses.

[0054] Specifically, a sequence of phase compensation values ​​required to eliminate phase jumps can be defined, with its frequency points being the same as the phase frequency response. The first value of this phase compensation value sequence can be 0; then, for frequency points ω=2,3,…,2000, the phase difference of the phase frequency response at adjacent frequency points can be calculated sequentially according to Formula 1: ΔH(ω)=H(ω)-H(ω-1) (Formula 1) Where H(ω) is the phase frequency response at frequency point ω, H(ω-1) is the phase frequency response at frequency point ω-1, ΔH(ω) is the phase difference at frequency point ω, and frequency point ω∈[0,2000].

[0055] According to Formula 2, the phase compensation amount at frequency points ω=2,3,…,2000 can be calculated based on the phase difference: (Formula 2) Where C(ω) is the phase compensation amount at frequency point ω, C(ω-1) is the phase compensation amount at frequency point ω-1, ΔH(ω) is the phase difference at frequency point ω, and frequency point ω∈[0,2000].

[0056] Furthermore, according to Formula 3, the phase frequency response can be added to the phase compensation amount to obtain the unwound phase frequency. The unwound phase frequency response constitutes the unwound phase frequency response sequence. H un (ω)=H(ω)+C(ω)(Formula 3) Among them, H un (ω) is the phase frequency response of frequency point ω after unwinding, C(ω) is the phase compensation amount of frequency point ω, H(ω) is the phase frequency response of frequency point ω, and frequency point ω∈[0,2000].

[0057] Figure 5 The phase-frequency response diagrams after unwinding are shown under different fault conditions. For example... Figure 5As shown, the four lines represent the phase frequency response of the transformer winding after unwinding under normal, light, moderate, and heavy deformation conditions.

[0058] In step S13, a discrete Gaussian smoothing filter is constructed, and the discrete Gaussian smoothing filter is used to perform noise filtering on the unwound phase frequency response to obtain the denoised phase frequency response.

[0059] Set the window length of the discrete Gaussian smoothing filter to 2M+1, and set the standard deviation of the Gaussian function to σ. Construct the discrete Gaussian kernel function G(k), where the index k = -M, -M+1, ..., M. The discrete Gaussian kernel function can be calculated using Equation 4: (Formula 4) Where G(k) is the discrete Gaussian kernel function, σ is the standard deviation of the discrete Gaussian kernel function, the index k∈[-M,+M], and 2M+1 is the window length of the discrete Gaussian smoothing filter.

[0060] To ensure that the overall energy and phase reference of the signal do not shift before and after filtering, the Gaussian kernel function can be normalized using Equation 5 to obtain normalized weights. These normalized weights form a normalized weight sequence. (Formula 5) Where g(k) is the normalized weight, G(k) is the discrete Gaussian kernel function, the indices j and k ∈ [-M, +M], and 2M+1 is the window length of the discrete Gaussian smoothing filter.

[0061] Furthermore, Formula 6 can be used to perform a discrete convolution operation between the unwound phase frequency response and the normalized weights to obtain the denoised phase frequency response. The denoised phase frequency responses constitute a denoised phase frequency response sequence. (Formula 6) Among them, H f (ω) is the denoised phase frequency response, g(k) is the normalization weight, and H un (ω-k) is the unwound phase frequency response at frequency point ω-k, where frequency point ω∈[0,2000], index k∈[-M,+M], and 2M+1 is the window length of the discrete Gaussian smoothing filter.

[0062] Figure 6 The denoised phase-frequency response diagrams are shown for different fault levels. Figure 6 As shown, the four lines represent the denoised phase frequency response of the transformer winding under normal, light, moderate, and heavy deformation conditions.

[0063] When calculating the convolution at both ends of the denoised phase frequency response sequence, for out-of-bounds indices where ω-k < 1 or ω-k > 2000, boundary processing can be performed on the unwound phase frequency response sequence using edge data extension, according to Formula 7. (Formula 7) Among them, H un (ω) is the unwound phase frequency response at frequency point ω, where frequency point ω∈[0,2000], index k∈[-M,+M], and 2M+1 is the window length of the discrete Gaussian smoothing filter.

[0064] Steps S12-S13, through the combined processing of phase dewinding and discrete Gaussian smoothing filtering, eliminate the inherent phase jump in instrument measurement and filter out noise. Without introducing artificial frequency shift, the noise amplification effect caused by discrete differentiation is suppressed, and the noise-resistant winding characteristics (i.e., the phase frequency response after denoising) are extracted.

[0065] In step S14, the denoised phase frequency response is differentiated with respect to the angular frequency to obtain the group delay spectrum.

[0066] Among them, the group delay spectrum is a graph describing the delay characteristics of the swept frequency signal.

[0067] like Figure 3 As shown, the denoising phase frequency response of the normal transformer winding and the transformer under test winding can be differentiated with respect to the angular frequency to obtain their respective group delay spectra.

[0068] Since the denoised phase frequency response sequence is a discrete sequence, the group delay spectrum can be obtained by discrete numerical differentiation using the central difference method. First, Equation 8 can be used to convert the frequency of a frequency point into its angular frequency: (Formula 8) Where, ω k f is the angular frequency of the k-th index. k Let k be the frequency of the k-th index, where k ∈ [-M, +M], and 2M+1 is the window length of the discrete Gaussian smoothing filter.

[0069] Furthermore, we can use Equation 9 to differentiate with respect to the angular frequency and calculate the group delay spectrum: (Formula 9) Where τ(ω) is the group delay spectrum at frequency ω, where ω∈[0,2000], index k∈[-M,+M], 2M+1 is the window length of the discrete Gaussian smoothing filter, and ω k+1 Let ω be the angular frequency of the (k+1)th index. k-1H is the angular frequency of the (k-1)th index. f (ω+1) is the denoised phase frequency response at frequency point ω+1, H f (ω-1) is the denoised phase frequency response at frequency point ω-1.

[0070] Figure 7 The group delay spectrum is shown under different fault levels. For example... Figure 7 As shown, the four lines represent the group delay spectrum of the transformer winding under normal, light, moderate, and heavy deformation conditions.

[0071] Furthermore, the step of inputting the group delay spectrum into a one-dimensional residual network for feature extraction and classification to obtain the fault diagnosis result of the transformer winding under test includes: The group delay spectrum is treated as a one-dimensional sequence signal and standardized to obtain a standardized group delay sequence. The standardized group delay sequence is input into a pre-trained one-dimensional residual network, and the deformation features of the group delay spectrum are extracted through the one-dimensional convolutional layer and residual block in the one-dimensional residual network. The extracted local deformation features are input into a fully connected classification layer to calculate and output the probability distribution of the transformer winding under test belonging to each preset fault state. The preset fault state corresponding to the maximum probability value in the probability distribution is selected as the fault diagnosis result of the transformer winding under test.

[0072] The preset fault states specifically include: normal, slight deformation, moderate deformation, and severe deformation.

[0073] In step S15, the group delay spectrum is input into a one-dimensional residual network for feature extraction and classification to obtain the fault diagnosis result of the transformer winding under test.

[0074] Figure 8 A schematic diagram of a one-dimensional residual network is shown. Figure 8 As shown, the input to the one-dimensional residual network is the group delay spectrum, and the output is the fault diagnosis result. This output can be multiple, meaning the fault diagnosis result can be graded using the one-dimensional residual network.

[0075] In the embodiments of this specification, the physical characteristic of high sensitivity of group delay to transformer resonant point offset is utilized to extract the group delay features of winding deformation and realize fault diagnosis based on a one-dimensional residual network. This significantly improves the sensitivity of diagnosis and effectively overcomes the shortcomings of traditional amplitude-frequency response analysis methods in being insensitive to winding deformation. Simultaneously, through the combined processing of phase dewinding and discrete Gaussian smoothing filtering, the inherent phase jump in instrument measurement is eliminated and noise is filtered out. Furthermore, without introducing artificial frequency shifts, the noise amplification effect caused by discrete differentiation is suppressed, achieving noise-resistant winding feature extraction. In addition, by inputting the group delay spectrum into a one-dimensional residual network for feature extraction and classification, subjective qualitative assessments relying on human experience are transformed into quantitative analysis indicators. This realizes an objective quantitative and hierarchical diagnostic system encompassing normal to severe deformation, improving the reliability of transformer condition monitoring and its practical engineering application value.

[0076] The present invention also provides a transformer winding fault diagnosis device based on a group delay residual network. Figure 9 This diagram illustrates a block diagram of a transformer winding fault diagnosis device based on a group delay residual network according to an embodiment of this specification. This transformer winding fault diagnosis device based on a group delay residual network can be a terminal device, a server, or other processing equipment. The terminal device can be a user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, vehicle-mounted device, wearable device, etc.

[0077] In some possible implementations, the transformer winding fault diagnosis device based on the group delay residual network can be implemented by a processor calling computer-readable instructions stored in memory.

[0078] like Figure 9 As shown, the transformer winding fault diagnosis device 90 based on group delay residual network may include: Signal module 91 is used to inject a sweep frequency signal into one end of the transformer winding, measure the response signal output at the other end, and obtain the phase frequency response based on the measurement result. The transformer winding includes a normal transformer winding and a transformer winding under test. The unwinding module 92 is used to unwind the phase frequency response to obtain the unwound phase frequency response. The unwinding process is used to eliminate the phase jump generated when calculating the phase frequency response. Filtering module 93 is used to construct a discrete Gaussian smoothing filter and use the discrete Gaussian smoothing filter to perform noise filtering on the unwound phase frequency response to obtain a denoised phase frequency response. Coefficient module 94 is used to perform derivative processing of the denoised phase frequency response with respect to the angular frequency to obtain the group delay spectrum; The diagnostic module 95 is used to input the group delay spectrum into a one-dimensional residual network for feature extraction and classification, so as to obtain the fault diagnosis result of the transformer winding under test.

[0079] Thirdly, embodiments of this application also provide a transformer winding fault diagnosis system based on a group delay residual network. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the methods described above.

[0080] Fourthly, embodiments of this application also provide a computer-readable storage medium storing computer program instructions thereon, which, when executed by a processor, implement the above-described method. The computer-readable storage medium may be volatile or non-volatile.

[0081] Fifthly, embodiments of this application also provide an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing instructions stored in the memory.

[0082] Sixthly, this application provides a computer program product including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the above-described method.

[0083] This invention is now complete.

[0084] In summary, the embodiments described in this specification utilize the physical characteristic of high sensitivity of group delay to transformer resonant point offset to extract group delay features of winding deformation and perform fault diagnosis based on a one-dimensional residual network. This significantly improves the sensitivity of the diagnosis and effectively overcomes the shortcomings of traditional amplitude-frequency response analysis methods in being insensitive to winding deformation. Simultaneously, through the combined processing of phase dewinding and discrete Gaussian smoothing filtering, the inherent phase jump in instrument measurement is eliminated and noise is filtered out. Furthermore, without introducing artificial frequency shifts, the noise amplification effect caused by discrete differentiation is suppressed, achieving noise-resistant winding feature extraction. Moreover, by inputting the group delay spectrum into a one-dimensional residual network for feature extraction and classification, subjective qualitative assessments relying on human experience are transformed into quantitative analysis indicators. This realizes an objective quantitative and hierarchical diagnostic system encompassing normal to severe deformation, improving the reliability of transformer condition monitoring and its practical engineering application value.

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for diagnosing transformer winding faults based on a group delay residual network, characterized in that, The transformer winding fault diagnosis method based on group delay residual network includes: After injecting a sweep frequency signal into one end of the transformer winding, the response signal output from the other end is measured, and the phase frequency response is obtained based on the measurement results. The transformer winding includes a normal transformer winding and a transformer winding under test. The phase frequency response is unwound to obtain the unwound phase frequency response. The unwound process is used to eliminate the phase jump generated when calculating the phase frequency response. A discrete Gaussian smoothing filter is constructed, and the discrete Gaussian smoothing filter is used to filter out noise from the unwound phase frequency response to obtain the denoised phase frequency response. The group delay spectrum is obtained by differentiating the denoised phase frequency response with respect to the angular frequency. The group delay spectrum is input into a one-dimensional residual network for feature extraction and classification to obtain the fault diagnosis results of the transformer winding under test.

2. The transformer winding fault diagnosis method based on group delay residual network according to claim 1, characterized in that, The frequency range of the sweep signal is 1-2MHz and consists of 2000 frequency points, forming a sweep signal sequence; the response signal forms a response signal sequence; the phase frequency response is the phase angle of the ratio of the sweep signal to the response signal, forming a phase frequency response sequence.

3. The transformer winding fault diagnosis method based on group delay residual network according to claim 2, characterized in that, The unwinding process of the phase frequency response to obtain the unwound phase frequency response includes: Substitute the phase frequency responses of adjacent frequency points in the phase frequency response sequence into the following formula. ΔH(ω)=H(ω)-H(ω-1) The phase difference value is calculated and constitutes a phase difference value sequence, where H(ω) is the phase frequency response at frequency point ω, H(ω-1) is the phase frequency response at frequency point ω-1, ΔH(ω) is the phase difference value at frequency point ω, and ω∈[0,2000]. Substitute the phase difference value into the following formula The phase compensation amount is calculated and constitutes a phase compensation amount sequence, where C(ω) is the phase compensation amount at frequency point ω and C(ω-1) is the phase compensation amount at frequency point ω-1. Substitute the phase frequency response and the phase compensation amount into the following formula H un (ω)=H(ω)+C(ω) The phase frequency response after unwinding is calculated, and the phase frequency response after unwinding constitutes a sequence of phase frequency responses after unwinding, wherein H un (ω) is the phase frequency response after unwinding at frequency point ω.

4. The transformer winding fault diagnosis method based on group delay residual network according to claim 3, characterized in that, The process of constructing a discrete Gaussian smoothing filter and using the discrete Gaussian smoothing filter to filter noise from the unwound phase frequency response to obtain a denoised phase frequency response includes: Use the following formula Construct a discrete Gaussian kernel function, where G(k) is the discrete Gaussian kernel function, σ is the standard deviation of the discrete Gaussian kernel function, the index k∈[-M,+M], and 2M+1 is the window length of the discrete Gaussian smoothing filter. The discrete Gaussian kernel functions constitute a sequence of discrete Gaussian kernel functions; and use the following formula... The Gaussian kernel function corresponding to the discrete Gaussian kernel function is normalized to obtain normalized weights. These normalized weights form a normalized weight sequence, where g(k) is the normalized weight. The following formula is used... The unwound phase frequency response is discretely convolved with the normalized weights to obtain the denoised phase frequency response. The denoised phase frequency response constitutes a denoised phase frequency response sequence, where H... f (ω) is the denoised phase-frequency response sequence at frequency point ω, H un (ω-k) is the phase frequency response after unwinding at frequency point ω-k.

5. The transformer winding fault diagnosis method based on group delay residual network according to claim 4, characterized in that, The transformer winding fault diagnosis method based on group delay residual network includes: When calculating the convolution at both ends of the denoised phase frequency response sequence, for out-of-bounds indices where ω-k < 1 or ω-k > 2000, the following formula is used. Boundary processing is performed on the unwound phase frequency response sequence to achieve edge data extension of the unwound phase frequency response sequence.

6. The transformer winding fault diagnosis method based on group delay residual network according to claim 3, characterized in that, The step of differentiating the denoised phase frequency response with respect to the angular frequency to obtain the group delay spectrum includes: Use the following formula Differentiating the angular frequency yields the group delay spectrum, where τ(ω) is the group delay spectrum at frequency ω. k The angular frequency of the k-th index is based on the following formula We obtain, where f k The frequency of the k-th index.

7. The transformer winding fault diagnosis method based on group delay residual network according to claim 3, characterized in that, The step of inputting the group delay spectrum into a one-dimensional residual network for feature extraction and classification to obtain the fault diagnosis result of the transformer winding under test includes: The group delay spectrum is treated as a one-dimensional sequence signal and standardized to obtain a standardized group delay sequence. The standardized group delay sequence is input into a pre-trained one-dimensional residual network, and the deformation features of the group delay spectrum are extracted through the one-dimensional convolutional layer and residual block in the one-dimensional residual network. The extracted local deformation features are input into a fully connected classification layer to calculate and output the probability distribution of the transformer winding under test belonging to each preset fault state. The preset fault state corresponding to the maximum probability value in the probability distribution is selected as the fault diagnosis result of the transformer winding under test. The preset fault states specifically include: normal, slight deformation, moderate deformation, and severe deformation.

8. A transformer winding fault diagnosis device based on a group delay residual network, characterized in that, The transformer winding fault diagnosis device based on group delay residual network includes: The signal module is used to inject a sweep frequency signal into one end of the transformer winding, measure the response signal output at the other end, and obtain the phase frequency response based on the measurement result. The transformer winding includes a normal transformer winding and a transformer winding under test. The unwinding module is used to unwind the phase frequency response to obtain the unwound phase frequency response. The unwinding process is used to eliminate the phase jump generated when calculating the phase frequency response. The filtering module is used to construct a discrete Gaussian smoothing filter and use the discrete Gaussian smoothing filter to perform noise filtering on the unwound phase frequency response to obtain a denoised phase frequency response. The group delay spectrum module is used to perform derivative processing of the denoised phase frequency response with respect to the angular frequency to obtain the group delay spectrum. The diagnostic module is used to input the group delay spectrum into a one-dimensional residual network for feature extraction and classification, so as to obtain the fault diagnosis result of the transformer winding under test.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images