Partial discharge pulse electric field time domain waveform recovery method and system

Abstract

The application discloses a partial discharge pulse electric field time domain waveform recovery method and system, comprising the following steps: respectively decomposing a wide frequency electric field signal and a UHF sensor output voltage signal into a plurality of first sub-band signals and a plurality of second sub-band signals which are not aliasing with each other; calculating the amplitude flatness and the group delay flatness of each sub-band signal based on the first sub-band signals and the second sub-band signals; judging whether the UHF sensor is a linear system based on the amplitude flatness and the group delay flatness, and obtaining a judgment result; and recovering the time domain waveform according to a preset recovery strategy based on the judgment result. The application is aimed at two cases of whether the transfer function of the UHF sensor is linear, the minimum phase method is used to recover the phase in the linear and distortionless case, and the deep convolution network (DCN) is used to directly recover the time domain electric field waveform in the nonlinear distortion case, so that the problem that the subsequent signal processing is inaccurate due to the incomplete phase information in the traditional method is solved.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment testing technology, and more specifically, to a method and system for recovering the time-domain waveform of a partial discharge pulse electric field. Background Technology

[0002] Partial discharge measurement of the insulation of high-voltage electrical equipment is a common method for assessing the insulation condition of such equipment. Partial discharge measurement techniques include traditional pulse current methods, as well as a series of new live-line detection techniques such as ultra-high frequency methods, high frequency methods, and ultrasonic methods. Among them, the ultra-high frequency method for partial discharge detection has become increasingly widely used due to its appropriate quantitative and localization capabilities and portability.

[0003] The time-domain waveform of the pulsed electric field excited by ultra-high frequency (UHF) partial discharge contains rich information about the characteristics of partial discharge. UHF sensors convert the pulsed electromagnetic field signal into a pulsed voltage signal and output it to the main unit of the detector. However, the time-domain waveform of the pulsed electromagnetic field signal excited by partial discharge is difficult to obtain directly and requires calculation using time-frequency transformation techniques. The conversion from the frequency spectrum to the time-domain waveform can be achieved through inverse fast Fourier transform (IFFT), but this requires simultaneous knowledge of the amplitude and phase information of the spectrum. Equipment manufacturers or metrology units generally only provide the amplitude information of the sensor correction coefficients, lacking phase information. Therefore, phase information is lost during data correction, making it impossible to directly obtain the time-domain waveform through inverse Fourier transform. This also makes it difficult to calculate the time-domain waveform of the pulsed electric field to determine the sensitivity threshold. Therefore, it is necessary to study methods for recovering the time-domain waveform from amplitude spectrum data. A key technical challenge is that the lack of a phase-frequency curve affects the recovery of the pulsed time-domain waveform.

[0004] HAYESM et al. first proposed using the minimum phase method to reconstruct phase information. They assumed that a signal transmission system satisfies the minimum phase condition, and therefore the amplitude and phase of its transfer function satisfy a Hilbert transform relationship, allowing the other to be solved from one. TESCHEF M assumed the system was a minimum (or maximum) phase system and reconstructed its phase information using the Hilbert transform. However, these conventional minimum phase estimation methods for time-domain response introduce additional numerical errors during implementation due to the repeated use of the Fast Fourier Transform (FFT) to convert signals between the time and frequency domains. Careful selection of the FFT parameters is crucial to minimize these errors. Some improved methods use the Prony method, the best approximation element method, and filter modeling to parameterize the frequency-domain transfer function, obtaining a discrete transfer function model of the system, and then estimating the system's impulse response from this model. However, parameter modeling methods like the Prony method also have limitations, such as poor fitting accuracy, unstable poles, and the algorithm's convergence being highly dependent on the initial value selection. All of these methods are ill-suited for recovering the time-domain waveform of ultra-high frequency partial discharge pulse electric fields. Summary of the Invention

[0005] This invention proposes a method and system for recovering the time-domain waveform of a partial discharge pulse electric field, in order to solve the problem of how to recover the time-domain waveform of a partial discharge pulse electric field.

[0006] To address the aforementioned problems, according to one aspect of the present invention, a method for recovering the time-domain waveform of a partial discharge pulse electric field is provided, the method comprising:

[0007] The broadband electric field signal and the ultra-high frequency sensor output voltage signal are respectively decomposed into multiple first sub-band signals and multiple second sub-band signals with non-overlapping frequency bands;

[0008] Calculate the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal, respectively.

[0009] Based on the amplitude flatness and group delay flatness, determine whether the UHF sensor is a linear system and obtain the determination result;

[0010] Based on the judgment result, the time-domain waveform is restored according to the preset recovery strategy.

[0011] Preferably, the method further includes:

[0012] Before calculating the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal respectively, noise judgment is performed on the first sub-band signal and the second sub-band signal respectively, and the first sub-band signal and the second sub-band signal determined to be noise are removed.

[0013] Preferably, the recovery of the time-domain waveform based on the judgment result according to a preset recovery strategy includes:

[0014] If the judgment result indicates that it belongs to a linear system, then the phase is recovered by using the minimum phase method, and then the time domain waveform of the pulse electric field is recovered.

[0015] If the judgment result indicates that it does not belong to a linear system, then the pulse electric field time domain waveform is recovered based on a deep convolutional network.

[0016] Preferably, the recovery of the pulse electric field time-domain waveform after recovering the phase using the minimum phase method includes:

[0017] The output voltage signal of the ultra-high frequency sensor is subjected to a Fast Fourier Transform (FFT) to obtain the complex spectrum sequence V(k);

[0018] Based on the frequency information of V(k), the correction coefficient ln|H(k)| is interpolated to obtain the number of points N of the correction coefficient; where H(k) is the discretized system transfer function, which is obtained by discretizing the Fourier transform of the output signal by the Fourier transform of the input signal.

[0019] Based on the scaling and extension of N on ln|H(k)|, we obtain G(k);

[0020] Perform an inverse Fourier transform on G(k) and then window it to obtain H(k) that satisfies the minimum phase condition.

[0021] Multiply V(k) and H(k) in complex spectral form and perform Fast Fourier Transform (FFT) to obtain the corresponding pulse electric field time-domain waveform e(t).

[0022] Preferably, the recovery of the pulse electric field time-domain waveform based on a deep convolutional network includes:

[0023] Collect the pulse electric field waveform e(t) at the front end of the UHF sensor and the output voltage waveform u(t) at the back end;

[0024] The time-frequency transformation technique is used to convert a one-dimensional waveform signal into a two-dimensional time-frequency spectrum, and then denoise it.

[0025] The DCN method is used to train a DCN network by taking the u(t) spectral set as input and the corresponding e(t) spectral set as output, and then obtaining a trained deep network model.

[0026] The output voltage signal of the ultra-high frequency sensor is input into the trained deep network model to obtain the corresponding pulse electric field time-domain waveform.

[0027] According to another aspect of the present invention, a system for recovering the time-domain waveform of a partial discharge pulse electric field is provided, the system comprising:

[0028] The signal decomposition unit is used to decompose the broadband electric field signal and the ultra-high frequency sensor output voltage signal into multiple first sub-band signals and multiple second sub-band signals with mutually non-overlapping frequency bands, respectively.

[0029] The flatness calculation unit is used to calculate the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal, respectively.

[0030] The judgment unit is used to determine whether the ultra-high frequency sensor is a linear system based on the amplitude flatness and group delay flatness, and to obtain the judgment result.

[0031] The recovery unit is used to recover the time-domain waveform based on the judgment result and according to a preset recovery strategy.

[0032] Preferably, the system further includes:

[0033] The noise reduction unit is used to perform noise judgment on the first sub-band signal and the second sub-band signal respectively before calculating the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal respectively, and to remove the first sub-band signal and the second sub-band signal determined to be noise.

[0034] Preferably, the recovery unit performs time-domain waveform recovery based on the judgment result according to a preset recovery strategy, including:

[0035] If the judgment result indicates that it belongs to a linear system, then the phase is recovered by using the minimum phase method, and then the time domain waveform of the pulse electric field is recovered.

[0036] If the judgment result indicates that it does not belong to a linear system, then the pulse electric field time domain waveform is recovered based on a deep convolutional network.

[0037] Preferably, the recovery unit, after recovering the phase using the minimum phase method, performs recovery of the pulse electric field time-domain waveform, including:

[0038] The output voltage signal of the ultra-high frequency sensor is subjected to a Fast Fourier Transform (FFT) to obtain the complex spectrum sequence V(k);

[0039] Based on the frequency information of V(k), the correction coefficient ln|H(k)| is interpolated to obtain the number of points N of the correction coefficient; where H(k) is the discretized system transfer function, which is obtained by discretizing the Fourier transform of the output signal by the Fourier transform of the input signal.

[0040] Based on the scaling and extension of N on ln|H(k)|, we obtain G(k);

[0041] Perform an inverse Fourier transform on G(k) and then window it to obtain H(k) that satisfies the minimum phase condition.

[0042] Multiply V(k) and H(k) in complex spectral form and perform a Fast Fourier Transform (IFFT) operation to obtain the corresponding pulse electric field time-domain waveform e(t).

[0043] Preferably, the recovery unit, which recovers the time-domain waveform of the pulsed electric field based on a deep convolutional network, includes:

[0044] Collect the pulse electric field waveform e(t) at the front end of the UHF sensor and the output voltage waveform u(t) at the back end;

[0045] The time-frequency transformation technique is used to convert a one-dimensional waveform signal into a two-dimensional time-frequency spectrum, and then denoise it.

[0046] Using the DCN system, the u(t) spectral set is taken as input and the corresponding e(t) spectral set is taken as output to train the DCN network and obtain the trained deep network model.

[0047] The output voltage signal of the ultra-high frequency sensor is input into the trained deep network model to obtain the corresponding pulse electric field time-domain waveform.

[0048] This invention provides a method and system for recovering the time-domain waveform of a partial discharge pulse electric field, comprising: decomposing a broadband electric field signal and an ultra-high frequency (UHF) sensor output voltage signal into multiple first sub-band signals and multiple second sub-band signals with non-overlapping frequency bands; calculating the amplitude flatness and group delay flatness of each sub-band signal based on the first and second sub-band signals; determining whether the UHF sensor is a linear system based on the amplitude flatness and group delay flatness, and obtaining the determination result; and recovering the time-domain waveform according to a preset recovery strategy based on the determination result. This invention addresses two cases: whether the transfer function of the UHF sensor is linear or not. In the case of linearity without distortion, it uses the minimum phase method to recover the phase; in the case of nonlinear distortion, it uses a deep convolutional network (DCN) to directly recover the time-domain electric field waveform. This solves the problem of incomplete phase information leading to inaccurate subsequent signal processing in traditional methods; after calculating the phase information, the time-domain waveform can be recovered from the amplitude spectrum data, solving the problem of the difficulty in accurately recovering the time-domain waveform of the pulse electric field excited by UHF partial discharge. Attached Figure Description

[0049] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures:

[0050] Figure 1This is a flowchart of a partial discharge pulse electric field time-domain waveform recovery method 100 according to an embodiment of the present invention;

[0051] Figure 2 The flowchart for solving the minimum phase sequence according to an embodiment of the present invention is shown below;

[0052] Figure 3 This is a flowchart of a DCN-based broadband pulse recovery method according to an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of the structure of a partial discharge pulse electric field time-domain waveform recovery system 400 according to an embodiment of the present invention. Detailed Implementation

[0054] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0055] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0056] To address the challenge of accurately calculating the pulse electric field waveform at the sensor input end during UHF partial discharge measurement and calibration, where the voltage waveform at the sensor output end is readily obtainable, a time-domain waveform recovery method for UHF partial discharge pulse electric fields based on signal processing technology is proposed. First, the time-domain signal is decomposed into a series of non-overlapping sub-band signals, and noise is removed. Then, the linearity of the system is determined. If it is a linear system (i.e., without distortion), the phase is recovered using the minimum phase method, and the pulse electric field time-domain waveform is reconstructed. If it is not a linear system (i.e., with distortion), a deep convolutional network (DCN) from deep learning theory is used to recover the broadband pulse electric field time-domain signal.

[0057] Figure 1 This is a flowchart of a partial discharge pulse electric field time-domain waveform recovery method 100 according to an embodiment of the present invention. Figure 1As shown, the partial discharge pulse electric field time-domain waveform recovery method provided by this invention addresses two cases: linearity and non-linearity of the UHF sensor's transfer function. In the linear, distortion-free case, the minimum phase method is used to recover the phase; in the non-linear, distortion-free case, a deep convolutional network (DCN) is used to directly recover the time-domain electric field waveform. This solves the problem of incomplete phase information leading to inaccurate subsequent signal processing in traditional methods. After calculating the phase information, the time-domain waveform can be recovered from the amplitude spectrum data, solving the problem of accurately recovering the time-domain waveform of the pulse electric field excited by UHF partial discharge. The partial discharge pulse electric field time-domain waveform recovery method 100 provided by this invention, starting from step 101, decomposes the broadband electric field signal and the UHF sensor output voltage signal into multiple first sub-band signals and multiple second sub-band signals with non-overlapping frequency bands.

[0058] In step 102, the amplitude flatness and group delay flatness of each sub-band signal are calculated based on the first sub-band signal and the second sub-band signal, respectively.

[0059] Preferably, the method further includes:

[0060] Before calculating the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal respectively, noise judgment is performed on the first sub-band signal and the second sub-band signal respectively, and the first sub-band signal and the second sub-band signal determined to be noise are removed.

[0061] In this invention, firstly, since broadband pulse signals often contain multiple characteristic frequency bands, time-domain signal decomposition technology is used to decompose the broadband electric field signal e(t) and the UHF sensor output voltage signal u(t) into a series of non-overlapping sub-band signals, and noise components are removed. Then, by calculating the amplitude flatness and group delay flatness of the corresponding sub-band signals in e(t) and u(t), it is determined whether the UHF sensor is a linear system. Finally, if the UHF sensor satisfies the linearity condition, the minimum phase method is used to recover the e(t) waveform; otherwise, a deep convolutional network with excellent nonlinear approximation capability is used to recover the e(t) waveform.

[0062] In step 103, based on the amplitude flatness and group delay flatness, it is determined whether the UHF sensor is a linear system, and the determination result is obtained.

[0063] In step 104, the time-domain waveform is restored according to the preset recovery strategy based on the judgment result.

[0064] Preferably, the recovery of the time-domain waveform based on the judgment result according to a preset recovery strategy includes:

[0065] If the judgment result indicates that it belongs to a linear system, then the phase is recovered by using the minimum phase method, and then the time domain waveform of the pulse electric field is recovered.

[0066] If the judgment result indicates that it does not belong to a linear system, then the pulse electric field time domain waveform is recovered based on a deep convolutional network.

[0067] Preferably, the recovery of the pulse electric field time-domain waveform after recovering the phase using the minimum phase method includes:

[0068] The output voltage signal of the ultra-high frequency sensor is subjected to a Fast Fourier Transform (FFT) to obtain the complex spectrum sequence V(k);

[0069] Based on the frequency information of V(k), the correction coefficient ln|H(k)| is interpolated to obtain the number of points N of the correction coefficient; where H(k) is the discretized system transfer function, which is obtained by discretizing the Fourier transform of the output signal by the Fourier transform of the input signal.

[0070] Based on the scaling and extension of N on ln|H)k)|, we obtain G(k);

[0071] Perform an inverse Fourier transform on G(k) and then window it to obtain H(k) that satisfies the minimum phase condition.

[0072] Multiply V(k) and H(k) in complex spectral form and perform a Fast Fourier Transform (IFFT) operation to obtain the corresponding pulse electric field time-domain waveform e(t).

[0073] Preferably, the recovery of the pulse electric field time-domain waveform based on a deep convolutional network includes:

[0074] Collect the pulse electric field waveform e(t) at the front end of the UHF sensor and the output voltage waveform u(t) at the back end;

[0075] The time-frequency transformation technique is used to convert a one-dimensional waveform signal into a two-dimensional time-frequency spectrum, and then denoise it.

[0076] The DCN method is used to train a DCN network by taking the u(t) spectral set as input and the corresponding e(t) spectral set as output, and then obtaining a trained deep network model.

[0077] The output voltage signal of the ultra-high frequency sensor is input into the trained deep network model to obtain the corresponding pulse electric field time-domain waveform.

[0078] (1) Pulse time-domain waveform recovery when it is determined to be linear

[0079] There are many methods for recovering the phase frequency curve from the amplitude-frequency response curve, and the minimum phase method is one of the more effective ones. Assuming a signal transmission system satisfies the minimum phase condition, the amplitude and phase of its transfer function satisfy a Hilbert transform relationship, allowing the other to be solved from one. A minimum phase system is one whose discrete transfer function has all zeros and poles within the unit circle. If a linear time-invariant system and its inverse are both causal and stable, then this system is a minimum phase system. Although actual signal transmission systems may not necessarily satisfy the minimum phase condition, any causal rational system can be represented as a cascade of an all-pass system and a minimum phase system. Since the all-pass system does not affect the signal's amplitude characteristics, in most cases, the original system can be approximated by its corresponding minimum phase system, i.e., assuming the system satisfies the minimum phase condition.

[0080] Assuming the system satisfies minimum phase, the phase frequency information can be recovered from the system's amplitude frequency information using the Hilbert transform. The FFT can conveniently convert data between the time and frequency domains, thus enabling the use of... Figure 2 The method shown yields the minimum phase sequence corresponding to the amplitude spectrum. Where X... c (n) is the real cepstral. The complex cepstral spectrum, x(n), is a minimum phase sequence. First, the natural logarithm of the amplitude spectrum is taken, then the Hilbert transform is performed using IFFT and FFT, and the exponent is taken to obtain the Fourier transform X(ω) of x(n), which is then obtained after IFFT.

[0081] By treating the pulse measurement system as a signal transmission system and assuming it satisfies the minimum phase condition, we can utilize the properties of minimum phase systems to derive its phase frequency characteristics from the system's amplitude-frequency response (i.e., correction coefficients), thus supplementing the missing phase information and obtaining the system's transfer function. Since the transfer function and correction coefficients are inversely related, for convenience, we can invert the system's input and output, making the amplitude spectrum of the transfer function equal to the correction coefficients. Let the transfer function of the minimum phase system be:

[0082] H(ω)=|H(ω)|·e jθ(ω) (45)

[0083] In the above formula, ω is the signal frequency, and |H(ω)| and θ(ω) represent the amplitude frequency response and phase frequency response of the system at frequency ω, respectively.

[0084] Taking the logarithm of both sides of the above equation, we get

[0085] ln(H(ω))=ln(|H(ω)|)+jθ(ω)=A(ω)+jP(ω) (46)

[0086] A(ω) and P(ω) reflect the frequency amplitude and phase frequency characteristics of the system, respectively, and they satisfy the Hilbert transform.

[0087]

[0088]

[0089] In the formula: and The Hilbert transforms of A(ω) and P(ω) are respectively. Thus, by performing a Hilbert transform on A(ω), we can obtain P(ω), and then the system transfer function H(ω). After that, we multiply the complex spectrum of the measured voltage signal with it to obtain the complex spectrum of the pulse to be measured. Then, we perform an inverse Fourier transform to transform the result to the time domain, and obtain the recovered waveform.

[0090] Therefore, in this invention, when the transmission is linear, i.e. undistorted, the pulsed electric field time-domain waveform is recovered using the following method:

[0091] 1) Perform FFT on the measured voltage signal to obtain its complex spectrum sequence V(k).

[0092] 2) To facilitate the direct multiplication of V(k) and H(k) later, the correction coefficient ln|H(k)| is interpolated based on the frequency information of V(k), and the number of points N of the correction coefficient is recorded. It is not required that the lengths of both be 2. N The form is because the FFT operation automatically pads the number of points; where H(k) is the discretized system transfer function, which is obtained by discretizing the Fourier transform of the output signal by the Fourier transform of the input signal.

[0093] 3) Scale and extend ln|H(k)| according to N to obtain G(k), which should be in the form of the discrete Fourier transform of h(t). This step is very important because the amplitude spectrum cannot be directly inverted Fourier transformed; it must be transformed into a discrete time sequence that meets the requirements.

[0094] 4) Perform an inverse Fourier transform on G(k) and then window it to obtain H(k) that satisfies the minimum phase condition.

[0095] 5) Multiply V(k) and H(k) in complex spectrum form and perform Fast Fourier Transform (FFT) to obtain the waveform e(t) of the pulse to be measured.

[0096] (2) Pulse time-domain waveform recovery when it is determined to be nonlinear

[0097] When the transfer function of a UHF sensor is nonlinear, it is difficult to recover the pulse electric field time domain signal using the traditional minimum phase method.

[0098] Deep learning methods possess excellent feature learning capabilities, automatically searching for the most suitable feature information from massive amounts of data, avoiding the subjectivity of manual feature selection. This invention applies deep learning theory to the recovery of UHF pulsed electric field time-domain signals, providing a new approach to improving the accuracy of pulse field recovery under nonlinear conditions. A "deep convolutional network" (DCN) from deep learning theory is used to realize the recovery of broadband pulsed electric field time-domain signals. The technical route is as follows: Figure 3 As shown.

[0099] Specifically, the following steps are included:

[0100] 1) Collect the pulse electric field waveform e(t) at the front end of the UHF sensor and the output voltage waveform u(t) at the back end;

[0101] 2) The one-dimensional waveform signal is transformed into a two-dimensional time-frequency spectrum using time-frequency transformation technology and then denoised. Specifically, the spectral components in the two-dimensional amplitude spectrum that are not in the ultra-high frequency band (0.3GHz-3GHz) are directly set to zero to obtain the denoised time-frequency spectrum.

[0102] 3) Use the DCN method to train the DCN network by taking the u(t) spectrum set as input and the corresponding e(t) spectrum set as output;

[0103] 4) Adjust network parameters to improve the accuracy of cross-validation experiments.

[0104] Finally, using the trained deep network, the time-domain waveform of the pulse electric field corresponding to the voltage waveform output by the UHF sensor can be predicted.

[0105] This invention proposes a technical approach that addresses the need for signal denoising, sensor transfer function linearity determination, phase recovery, and waveform reconstruction. For the two cases of whether the UHF sensor's transfer function is linear, the minimum phase method is used to recover the phase in the linear, distortion-free case, while a deep convolutional network (DCN) is used to directly recover the time-domain electric field waveform in the nonlinear, distortion-free case. This solves the problem of incomplete phase information leading to inaccurate subsequent signal processing in traditional methods. After calculating the phase information, the time-domain waveform can be recovered from the amplitude spectrum data, solving the current problem of accurately recovering the time-domain waveform of pulsed electric fields excited by UHF partial discharge.

[0106] Figure 4 This is a schematic diagram of the structure of a partial discharge pulse electric field time-domain waveform recovery system 400 according to an embodiment of the present invention. Figure 4As shown, the partial discharge pulse electric field time-domain waveform recovery system 400 provided in this embodiment of the invention includes: a signal decomposition unit 401, a flatness calculation unit 402, a judgment unit 403, and a recovery unit 404.

[0107] Preferably, the signal decomposition unit 401 is used to decompose the broadband electric field signal and the ultra-high frequency sensor output voltage signal into multiple first sub-band signals and multiple second sub-band signals with non-overlapping frequency bands, respectively.

[0108] Preferably, the system further includes:

[0109] The noise reduction unit is used to perform noise judgment on the first sub-band signal and the second sub-band signal respectively before calculating the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal respectively, and to remove the first sub-band signal and the second sub-band signal determined to be noise.

[0110] Preferably, the flatness calculation unit 402 is used to calculate the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal, respectively.

[0111] Preferably, the judgment unit 403 is used to determine whether the UHF sensor is a linear system based on the amplitude flatness and group delay flatness, and to obtain the judgment result.

[0112] Preferably, the recovery unit 404 is used to recover the time-domain waveform based on the judgment result and according to a preset recovery strategy.

[0113] Preferably, the recovery unit 404 performs time-domain waveform recovery based on the judgment result according to a preset recovery strategy, including:

[0114] If the judgment result indicates that it belongs to a linear system, then the phase is recovered by using the minimum phase method, and then the time domain waveform of the pulse electric field is recovered.

[0115] If the judgment result indicates that it does not belong to a linear system, then the pulse electric field time domain waveform is recovered based on a deep convolutional network.

[0116] Preferably, the recovery unit 404, after recovering the phase using the minimum phase method, performs recovery of the pulse electric field time-domain waveform, including:

[0117] The output voltage signal of the ultra-high frequency sensor is subjected to a Fast Fourier Transform (FFT) to obtain the complex spectrum sequence V(k);

[0118] Based on the frequency information of V(k), the correction coefficient ln|H(k)| is interpolated to obtain the number of points N of the correction coefficient; where H(k) is the discretized system transfer function, which is obtained by discretizing the Fourier transform of the output signal by the Fourier transform of the input signal.

[0119] Based on the scaling and extension of N on ln|H(k)|, we obtain G(k);

[0120] Perform an inverse Fourier transform on G(k) and then window it to obtain H(k) that satisfies the minimum phase condition.

[0121] Multiply V(k) and H(k) in complex spectral form and perform a Fast Fourier Transform (IFFT) operation to obtain the corresponding pulse electric field time-domain waveform e(t).

[0122] Preferably, the recovery unit 404, which performs time-domain waveform recovery of the pulse electric field based on a deep convolutional network, includes:

[0123] Collect the pulse electric field waveform e(t) at the front end of the UHF sensor and the output voltage waveform u(t) at the back end;

[0124] The time-frequency transformation technique is used to convert a one-dimensional waveform signal into a two-dimensional time-frequency spectrum, and then denoise it.

[0125] Using the DCN system, the u(t) spectral set is taken as input and the corresponding e(t) spectral set is taken as output to train the DCN network and obtain the trained deep network model.

[0126] The output voltage signal of the ultra-high frequency sensor is input into the trained deep network model to obtain the corresponding pulse electric field time-domain waveform.

[0127] The partial discharge pulse electric field time-domain waveform recovery system 400 of the present invention corresponds to the partial discharge pulse electric field time-domain waveform recovery method 100 of another embodiment of the present invention, and will not be described again here.

[0128] The invention has been described with reference to a few embodiments. However, as will be known to those skilled in the art, and as defined in the appended claims, other embodiments besides those disclosed above fall equivalently within the scope of the invention.

[0129] Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the art, unless otherwise expressly defined herein. All references to “a / the / the [device, component, etc.]” ​​are openly interpreted as at least one instance of said device, component, etc., unless otherwise expressly stated. The steps of any method disclosed herein need not be performed in the exact order disclosed unless explicitly stated otherwise.

[0130] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0131] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0132] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0133] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for recovering the time-domain waveform of a partial discharge pulse electric field, characterized in that, The method includes: The broadband electric field signal and the ultra-high frequency sensor output voltage signal are respectively decomposed into multiple first sub-band signals and multiple second sub-band signals with non-overlapping frequency bands; Calculate the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal, respectively. Based on the amplitude flatness and group delay flatness, determine whether the UHF sensor is a linear system and obtain the determination result; Based on the judgment result, the time-domain waveform is restored according to the preset recovery strategy; The step of restoring the time-domain waveform according to a preset recovery strategy based on the judgment result includes: If the judgment result indicates that it belongs to a linear system, then the phase is recovered by using the minimum phase method, and then the time domain waveform of the pulse electric field is recovered. If the judgment result indicates that it does not belong to a linear system, then the pulse electric field time domain waveform is recovered based on the deep convolutional network model; The recovery of the time-domain waveform of a pulsed electric field based on a deep convolutional network model includes: Collect the pulse electric field waveform e(t) at the front end of the UHF sensor and the output voltage waveform u(t) at the back end; The time-frequency transformation technique is used to convert a one-dimensional waveform signal into a two-dimensional time-frequency spectrum, and then denoise it. The deep convolutional network model is trained by taking the u(t) spectral set as input and the corresponding e(t) spectral set as output, and the trained deep convolutional network model is obtained. The output voltage signal of the ultra-high frequency sensor is input into the trained deep convolutional network model to obtain the corresponding pulse electric field time-domain waveform.

2. The method according to claim 1, characterized in that, The method further includes: Before calculating the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal respectively, noise judgment is performed on the first sub-band signal and the second sub-band signal respectively, and the first sub-band signal and the second sub-band signal determined to be noise are removed.

3. The method according to claim 1, characterized in that, The process of restoring the phase using the minimum phase method and then recovering the time-domain waveform of the pulsed electric field includes: The output voltage signal of the ultra-high frequency sensor is subjected to a Fast Fourier Transform (FFT) to obtain the complex spectrum sequence V(k). The correction coefficient is adjusted based on the frequency information of V(k). Interpolation calculations are performed to obtain the number of correction coefficients N; where H(k) is the discretized system transfer function, which is obtained by discretizing the Fourier transform of the output signal by the Fourier transform of the input signal. According to N pairs After scaling and extension, we obtain G(k); Perform an inverse Fourier transform on G(k) and then window it to obtain H(k) that satisfies the minimum phase condition. Multiply V(k) and H(k) in complex spectral form and perform an inverse fast Fourier transform (IFFT) operation to obtain the corresponding pulse electric field time-domain waveform e(t).

4. A partial discharge pulse electric field time-domain waveform recovery system, characterized in that, The system includes: The signal decomposition unit is used to decompose the broadband electric field signal and the ultra-high frequency sensor output voltage signal into multiple first sub-band signals and multiple second sub-band signals with mutually non-overlapping frequency bands, respectively. The flatness calculation unit is used to calculate the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal, respectively. The judgment unit is used to determine whether the ultra-high frequency sensor is a linear system based on the amplitude flatness and group delay flatness, and to obtain the judgment result. The recovery unit is used to recover the time-domain waveform based on the judgment result and according to a preset recovery strategy; The recovery unit, based on the judgment result, performs time-domain waveform recovery according to a preset recovery strategy, including: If the judgment result indicates that it belongs to a linear system, then the phase is recovered by using the minimum phase method, and then the time domain waveform of the pulse electric field is recovered. If the judgment result indicates that it does not belong to a linear system, then the pulse electric field time domain waveform is recovered based on the deep convolutional network model; The recovery unit, based on a deep convolutional network model, recovers the time-domain waveform of the pulsed electric field, including: Collect the pulse electric field waveform e(t) at the front end of the UHF sensor and the output voltage waveform u(t) at the back end; The time-frequency transformation technique is used to convert a one-dimensional waveform signal into a two-dimensional time-frequency spectrum, and then denoise it. The deep convolutional network model is trained by taking the u(t) spectral set as input and the corresponding e(t) spectral set as output, and the trained deep convolutional network model is obtained. The output voltage signal of the ultra-high frequency sensor is input into the trained deep convolutional network model to obtain the corresponding pulse electric field time-domain waveform.

5. The system according to claim 4, characterized in that, The system also includes: The noise reduction unit is used to perform noise judgment on the first sub-band signal and the second sub-band signal respectively before calculating the amplitude flatness and group delay flatness of each sub-band signal based on the first sub-band signal and the second sub-band signal respectively, and to remove the first sub-band signal and the second sub-band signal determined to be noise.

6. The system according to claim 4, characterized in that, The recovery unit, after recovering the phase using the minimum phase method, performs recovery of the pulse electric field time-domain waveform, including: The output voltage signal of the ultra-high frequency sensor is subjected to a Fast Fourier Transform (FFT) to obtain the complex spectrum sequence V(k). The correction coefficient is adjusted based on the frequency information of V(k). Interpolation calculations are performed to obtain the number of correction coefficients N; where H(k) is the discretized system transfer function, which is obtained by discretizing the Fourier transform of the output signal by the Fourier transform of the input signal. According to N pairs After scaling and extension, we obtain G(k); Perform an inverse Fourier transform on G(k) and then window it to obtain H(k) that satisfies the minimum phase condition. Multiply V(k) and H(k) in complex spectral form and perform an inverse fast Fourier transform (IFFT) operation to obtain the corresponding pulse electric field time-domain waveform e(t).

Images