Method and System for Extracting Edge Feature Parameters of Time-Frequency Image of Partial Discharge Pulse Waveform

Abstract

The present invention relates to a method and system for extracting edge feature parameters of a time-frequency image of a partial discharge pulse waveform. This method is directed to a hybrid pulse waveform-time series, i.e., a pulse group, formed by a pulse source obtained through ultra-wideband detection. The method includes the following steps: Step 101, using wavelet denoising and amplitude normalization for a single pulse waveform in the pulse group to form a time-domain waveform; Step 102, performing time-frequency transformation on the time-domain waveform to form a time-frequency image; Step 103, performing edge calculations for time and frequency on the time-frequency image, and then obtaining the standard deviations of the time edge and the frequency edge; Step 104, obtaining the corresponding standard deviations of the time edge and the frequency edge for all single pulse waveforms in the pulse group to obtain a plane distribution of time-frequency edge standard deviation feature parameters. Compared with the prior art, the present invention has the advantages of stronger robustness and simple algorithm that can be easily implemented, etc.

Description

Technical Field

[0001] The present invention relates to partial discharge detection technology, and in particular to a method and system for extracting edge feature parameters of partial discharge pulse waveform time-frequency images. Background Art

[0002] Professor G.C. Montanari of the University of Bologna in Italy and others started in 2002, and Professor Li Yanming of Xi'an Jiaotong University in China and others started in 2008, and successively proposed partial discharge (PD) pulse source detection technology based on ultra-wideband detection (the frequency band can cover or involve the conventional electrical detection (30 kHz - 1 MHz), high frequency (3 MHz - 30 MHz), and very high frequency (30 MHz - 300 MHz) defined by GB / T7354, GB / T 20833.1, GB / T 20833.2, and GB / T 23642). That is, replacing the traditional pulse peak-time series detection with pulse waveform-time series detection, recording a single pulse waveform and its acquisition time (phase information can also be obtained under AC voltage); since ultra-wideband detection retains relatively "complete" waveform information of the pulse, and the waveforms generated by different pulse sources have "self-similarity", a certain "method" can be used to quickly classify the acquired mixed original pulse group to achieve the separation between pulse sources, that is, to separate the PD source and the noise source or multiple PD sources.

[0003] The key to implementing the above-mentioned ultra-wideband technology is to quickly classify the acquired mixed original pulse group by a certain "method", which is divided into two parts: ① is the method for extracting pulse waveform feature parameters; ② is the clustering technology based on the distribution of feature parameters. Among them, for the first part, Si Wenrong et al. published "Fast Classification of Multiple Partial Discharge Pulse Groups Based on Waveform Nonlinear Mapping" in the Journal of Electrical Engineering in March 2009, and proposed a nonlinear mapping method for pulse waveforms. That is, the extraction results are displayed in a 2D plane or 3D space, and then intelligent clustering analysis and other means (the second part of the technology) are used to separate the pulse group to form sub-pulse groups with their own characteristics, so as to achieve the separation of multiple PD sources and noise sources. The enterprise standard of the State Grid Corporation of China, "Q / GDW11400-2015 Field Application Guide for High-Frequency Partial Discharge Live Testing Technology of Power Equipment", gives a method for extracting pulse waveform feature parameters by the equivalent time-frequency method, and uses corresponding instruments to carry out the application of live detection of power equipment.

[0004] However, in actual engineering applications, due to the influence of the propagation path, there are working conditions where the pulse waveforms obtained by ultra-wideband detection of noise sources or PD sources are very similar in time-domain waveform representation. This makes the existing method for extracting pulse waveform feature parameters based on the equivalent time-frequency method unable to fully meet the separation of multiple PD sources and noise sources by means of intelligent clustering analysis and other means in the later stage. Summary of the Invention

[0005] The object of the present invention is to overcome the defects existing in the above-mentioned prior art and provide one kind.

[0006] The object of the present invention can be achieved by the following technical solutions:

[0007] According to one aspect of the present invention, a method for extracting edge feature parameters of a time-frequency image of a partial discharge pulse waveform is provided, characterized in that the method aims at a mixed pulse waveform-time series, i.e., a pulse group, formed by a pulse source obtained through ultra-wideband detection, and the method comprises the following steps:

[0008] Step 101, perform wavelet denoising and amplitude normalization on a single pulse waveform in the pulse group to form a time-domain waveform;

[0009] Step 102, perform time-frequency transformation on the time-domain waveform to form a time-frequency image;

[0010] Step 103, perform edge calculations on time and frequency of the time-frequency image, and then obtain the standard deviations of the time edge and the frequency edge;

[0011] Step 104, obtain the corresponding time-edge standard deviation and frequency-edge standard deviation for all single pulse waveforms in the pulse group to obtain a plane distribution of time-frequency edge standard deviation characteristic parameters.

[0012] According to another aspect of the present invention, a system for the method for extracting edge feature parameters of a time-frequency image of a partial discharge pulse waveform is provided, including a pulse waveform-time series module of a PD source and a noise source mixture obtained through ultra-wideband detection, a pulse time-domain waveform module, a time-frequency transformation module, a time edge module, a frequency edge module, a time-edge standard deviation module, a frequency-edge standard deviation module, and a plane distribution module of time-frequency edge standard deviation characteristic parameters;

[0013] The pulse waveform-time series module of the PD source and the noise source mixture obtained through ultra-wideband detection is connected to the time-frequency transformation module through the pulse time-domain waveform module. The time-frequency transformation module is respectively connected to the time edge module and the frequency edge module. The time edge module is connected to the time-edge standard deviation module. The frequency edge module is connected to the frequency-edge standard deviation module. The time-edge standard deviation module and the frequency-edge standard deviation module are respectively connected to the plane distribution module of time-frequency edge standard deviation characteristic parameters.

[0014] As a preferred technical solution, the pulse waveform-time series module for the mixed PD source and noise source obtained by the ultra-wideband detection is a data acquisition device with an analog bandwidth of dozens of MHz and a sampling rate of 100 MS / s or more. Through a coupling device with a frequency response of dozens of MHz or more that meets the nanosecond-level PD ultra-wideband detection, the single-pulse time-domain waveform and the pulse waveform-time series at the corresponding trigger moment, that is, the pulse group p j (t), is specifically defined as follows:

[0015]

[0016] In the formula:

[0017] j is the jth pulse, where j = 1, 2, … N, and N is the total number of pulse waveforms included in the pulse group;

[0018] k indicates that the pulse waveform consists of k points, and the number of points is determined by the sampling rate f s * sampling duration.

[0019] As a preferred technical solution, the system further includes a wavelet denoising module provided between the pulse waveform-time series module for the mixed PD source and noise source obtained by the ultra-wideband detection and the time-frequency transformation module. The wavelet denoising module uses wavelet denoising and amplitude normalization for the single original pulse waveform in the pulse group to form a time-domain waveform.

[0020] As a preferred technical solution, the pulse time-domain waveform module is used to store the time-domain waveform, that is, p j (t), as the object to be processed by subsequent modules.

[0021] As a preferred technical solution, the time-frequency transformation module uses the Page time-frequency distribution algorithm to perform time-frequency transformation on p j (t) to obtain the tfr j (t, f) time-frequency image.

[0022] As a preferred technical solution, the processing formula of the time edge module is as follows:

[0023]

[0024] Where tfr j (t, f) is the time-frequency image information processed by the time-frequency transformation module, and m f (t) j is the result after the time edge calculation.

[0025] As a preferred technical solution, the processing formula of the frequency edge module is as follows:

[0026]

[0027] where tfr j (t, f) is the time-frequency image information processed by the time-frequency transformation module, m t (f) j is the result after time-frequency calculation.

[0028] As a preferred technical solution, the calculation of the time edge standard deviation module is as follows:

[0029]

[0030] where is the time edge mean value;

[0031] is the energy of the time edge.

[0032] As a preferred technical solution, the calculation of the frequency edge standard deviation module is as follows:

[0033]

[0034] where is the frequency edge mean value;

[0035] is the energy of the frequency edge.

[0036] As a preferred technical solution, the calculation of the time-frequency edge standard deviation feature parameter plane distribution module is as follows:

[0037]

[0038] where j is the jth pulse, j = 1, 2, … N, and N is the total number of pulse waveforms included in the pulse group.

[0039] Compared with the prior art, the present invention has the following advantages:

[0040] 1. The present invention makes full use of the joint time-frequency distribution characteristics of the non-stationary characterization of the pulse waveform signal in the time domain and the frequency domain, and obtains the characteristic parameters by using the time and frequency moment distributions of the time-frequency image.

[0041] 2. The present invention has stronger robustness than the existing methods for obtaining characteristic parameters only based on the time domain and frequency domain waveforms of the pulse (such as the equivalent time-frequency method), and has the advantages of simple algorithm and convenient implementation.

[0042] 3. The most important thing of the present invention is that it can process pulse sources with similar time domain characteristics of the pulse waveform. The equivalent time-frequency moment characteristic parameter plane distribution has the characteristics of "intra-class cohesion" and "inter-class separation", and is suitable for constructing a fast classification technology for pulse groups in ultra-wideband PD detection. Description of the Drawings

[0043] Figure 1 Main modules constituting the method shown in the present invention

[0044] Figure 2 Flow chart for processing pulse data in the method of the present invention

[0045] Figure 3 Data diagram of the original waveform of a typical single PD pulse for detecting discharge of a tip-plate defect in oil and the data during its processing in the method of the present invention under DC withstand voltage ultra-wideband detection; where (a) is the original time-domain waveform; (b) is the time-frequency transformation of the original time-domain waveform - time-frequency image; (c) is the time marginal distribution of the time-frequency image; (d) is the frequency marginal distribution of the time-frequency image

[0046] Figure 4 Data diagram of the original waveform of a typical single PD pulse for detecting discharge of an internal air-gap defect in oil-impregnated paperboard and the data during its processing in the method of the present invention under DC withstand voltage ultra-wideband detection; where (a) is the original time-domain waveform; (b) is the time-frequency transformation of the original time-domain waveform - time-frequency image; (c) is the time marginal distribution of the time-frequency image; (d) is the frequency marginal distribution of the time-frequency image

[0047] Figure 5 Data diagram of the original waveform of a typical single PD pulse for detecting discharge of an oil-paper surface defect and the data during its processing in the method of the present invention under DC withstand voltage ultra-wideband detection; where (a) is the original time-domain waveform; (b) is the time-frequency transformation of the original time-domain waveform - time-frequency image; (c) is the time marginal distribution of the time-frequency image; (d) is the frequency marginal distribution of the time-frequency image

[0048] Figure 6 Data diagram of the original waveform of a typical single PD pulse for detecting discharge of a suspended defect in oil and the data during its processing in the method of the present invention under DC withstand voltage ultra-wideband detection; where (a) is the original time-domain waveform; (b) is the time-frequency transformation of the original time-domain waveform - time-frequency image; (c) is the time marginal distribution of the time-frequency image; (d) is the frequency marginal distribution of the time-frequency image

[0049] Figure 7 Time-frequency marginal standard deviation feature plane formed by the method of the present invention, parameter distributions corresponding to pulse waveform-time series (each containing 100 pulse waveforms) for detecting discharges of tip-plate in oil, internal air-gap in oil-impregnated paperboard, oil-paper surface, and suspended defect in oil under DC withstand voltage ultra-wideband detection Detailed Description of the Invention

[0050] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0051] The present invention makes full use of the joint time-frequency distribution characteristics of the non-stationary representation of the pulse waveform signal in the time domain and the frequency domain, and has stronger robustness than the current method of extracting characteristic parameters only based on the pulse time-domain and frequency-domain waveforms. Moreover, the algorithm is simple and easy to implement, etc. It is suitable for constructing a fast classification technology for pulse groups in ultra-wideband PD detection.

[0052] The present invention provides a method for extracting edge characteristic parameters of the time-frequency image of the partial discharge pulse waveform. That is, for the mixed pulse waveform-time series, namely the pulse group, formed by PD sources and noise sources and other pulse sources obtained by ultra-wideband detection, first, the single pulse waveform in the pulse group is denoised by wavelet (for example, harmonic wavelet transform) and amplitude-normalized to form a time-domain waveform (or filtering processing may not be performed); secondly, the time-frequency transformation (for example, Page time-frequency transformation) is performed on the time-domain waveform to form a time-frequency image; thirdly, the edge calculations of time and frequency are performed on the time-frequency image; and the standard deviations of the time edge and the frequency edge are obtained respectively; finally, the corresponding time edge standard deviation and frequency edge standard deviation are obtained for all single pulse waveforms in the pulse group, and the plane distribution of the time-frequency edge standard deviation characteristic parameters is obtained. Based on the plane distribution of the time-frequency edge standard deviation characteristic parameters, a fast classification technology for pulse groups in ultra-wideband PD detection can be constructed to achieve the separation of multiple PD sources and noise sources through the clustering of "self-similar" sub-pulse groups.

[0053] As Figure 1 shown, the system for the method of extracting edge characteristic parameters of the time-frequency image of the partial discharge pulse waveform according to the present invention includes a pulse waveform-time series module 10 mixed with PD sources and noise sources obtained by ultra-wideband detection, a pulse time-domain waveform module 12, a time-frequency transformation module 13, a time edge module 14, a frequency edge module 15, a time edge standard deviation module 16, a frequency edge standard deviation module 17, and a plane distribution module 18 of time-frequency edge standard deviation characteristic parameters;

[0054] The pulse waveform-time series module 10 that mixes the PD source and the noise source obtained by the ultra-wideband detection is connected to the time-frequency transformation module 13 through the pulse time-domain waveform module 12. The time-frequency transformation module 13 is respectively connected to the time edge module 14 and the frequency edge module 15. The time edge module 14 is connected to the time edge standard deviation module 16. The frequency edge module 15 is connected to the frequency edge standard deviation module 17. The time edge standard deviation module 16 and the frequency edge standard deviation module 17 are respectively connected to the time-frequency edge standard deviation characteristic parameter plane distribution module 18.

[0055] The pulse waveform-time series module 10 that mixes the PD source and the noise source obtained by the ultra-wideband detection is a data acquisition device with an analog bandwidth of dozens of MHz and a sampling rate of 100 MS / s or more (a high-pass filter of 10 kHz or more needs to be installed to filter out background noise signals such as high-order harmonics). Through a coupling device with a frequency response of dozens of MHz or more that meets the nanosecond-level PD ultra-wideband detection (which can be used in the PD withstand voltage test circuits under AC, DC, and impulse voltages), based on the pulse waveform triggering technology, the single-pulse time-domain waveform and the pulse waveform-time series at the corresponding trigger moment, that is, the pulse group p j (t) is defined as follows:

[0056]

[0057] In the formula:

[0058] j —— the jth pulse (j = 1, 2, … N, N is the total number of pulse waveforms included in the pulse group);

[0059] k —— the pulse waveform consists of k points, and the number of points is determined by the sampling rate f s * sampling duration.

[0060] Figure 3 (a), Figure 4 (a), Figure 5 (a) and Figure 6 (a) are examples of the single original pulse time-domain waveforms included in the pulse group, which are the pulse waveform-time series obtained by the ultra-wideband detection with a sampling rate of 2.5 GS / s and an analog bandwidth of 1 GHz.

[0061] The system also includes a wavelet denoising module 11 arranged between the pulse waveform-time series module 10 that mixes the PD source and the noise source obtained by the ultra-wideband detection and the time-frequency transformation module 13. The wavelet denoising module 11 uses wavelet denoising and amplitude normalization on the single original pulse waveform in the pulse group to form a time-domain waveform.

[0062] The harmonic wavelet transform of the discrete signal p j (t) is defined as follows:

[0063] There is a real even function w e (t) and a real odd function w o (t), and their Fourier transforms are respectively:

[0064]

[0065]

[0066] Among them,

[0067] Then for there is

[0068]

[0069] The corresponding function w(t) = w e (t) + iw o (t) is obtained from the Fourier transform of

[0070] w(t) = [exp(i4πt) - exp(i2πt)] / (i2πt) (5)

[0071] It is called a harmonic wavelet.

[0072] The pulse time-domain waveform module 12 mentioned above is mainly used to store the time-domain waveform, that is, p j (t), as the object to be processed by subsequent modules. If wavelet denoising is not performed, p j (t) is Figure 3 (a), Figure 4 (a), Figure 5 (a) and Figure 6 (a) shown as a single original pulse time-domain waveform, all stored in the pulse time-domain waveform module 12.

[0073] The time-frequency transformation module 13 mentioned above uses the Page time-frequency distribution algorithm to perform time-frequency transformation on p j (t) to obtain the time-frequency image tfr j (t, f). The algorithm is defined as follows:

[0074]

[0075] Equation (6) is actually the differential of the energy spectral density before time t.

[0076] Figure 3 (a), Figure 4 (a), Figure 5 (a) and Figure 6(a) The time-frequency image of the original time-domain waveform shown is as Figure 3 (b), Figure 4 (b), Figure 5 (b) and Figure 6 (b) shown.

[0077] The time edge module 14 and the frequency edge module 15 mentioned above are defined in sequence as follows:

[0078]

[0079]

[0080] Figure 3 (b), Figure 4 (b), Figure 5 (b) and Figure 6 The time edge distribution of the time-frequency image tfr j (t, f) shown in (c), Figure 3 (c), Figure 4 (c), Figure 5 (c) and Figure 6 (c) shown; the frequency edge distribution is as Figure 3 (d), Figure 4 (d), Figure 5 (d) and Figure 6 (d) shown.

[0081] The algorithms of the time edge standard deviation module 16 and the frequency edge standard deviation module 17 mentioned above are as follows respectively:

[0082] Time edge standard deviation:

[0083] Frequency edge standard deviation:

[0084] In the formula:

[0085] is the time edge mean;

[0086] is the frequency edge mean;

[0087] is the energy of the time edge;

[0088] is the energy of the frequency edge.

[0089] Figure 3 (c), Figure 4 (c), Figure 5 (c) and Figure 6 (c) shown in the time edge distribution of the time-frequency graph, Figure 3 (d),Figure 4 (d), Figure 5 (d) and Figure 6 The frequency marginal distribution shown in (d), and the corresponding values of the time marginal standard deviation and the frequency marginal standard deviation are shown in Table 1 below:

[0090] Table 1

[0091]

[0092] The time-frequency marginal standard deviation characteristic parameter plane distribution module 18 described above is defined as follows:

[0093]

[0094] j —— the jth pulse (j = 1, 2, … N, where N is the total number of pulse waveforms included in the pulse group).

[0095] Figure 7 The time-frequency marginal standard deviation characteristic parameter plane distribution formed by the method of the present invention is for the parameter distribution corresponding to the pulse waveform-time series (each containing 100 pulse waveforms) obtained by DC withstand voltage ultra-wideband detection for detecting discharges in sharp plates in oil, internal air gaps in oil-impregnated paperboards, surface discharges along oil-paper interfaces, and suspended defects in oil. It can be seen that the individual pulse waveforms corresponding to internal air gaps in oil-impregnated paperboards and suspended defects in oil are very similar in the time domain, but Figure 7 The formed time-frequency marginal standard deviation characteristic parameter plane distribution still shows the characteristics of "intra-class cohesion" and "inter-class separation", which is convenient for later use of means such as intelligent clustering analysis to separate the pulse group to form sub-pulse groups with their respective characteristics, thereby realizing the separation of multiple PD sources and noise sources.

[0096] The flowchart of the characteristic parameter processing of the pulse waveform data in the method of the present invention is as Figure 2 shown.

[0097] Figures 3 to 7 Shown are the single original pulse waveform, the time-frequency image of the Page time-frequency distribution, the time marginal distribution diagram, the frequency marginal distribution diagram, and the time-frequency marginal standard deviation characteristic parameter plane distribution diagram corresponding to the typical single pulse waveform in the pulse waveform-time series obtained by ultra-wideband detection with a sampling rate of 2.5 GS / s and an analog bandwidth of 1 GHz according to the Figure 2 processing flow shown.

[0098] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for extracting edge feature parameters of the time-frequency image of partial discharge pulse waveforms, characterized in that This method is for obtaining the hybrid pulse waveform-time series, i.e., pulse group, formed by a pulse source in ultra-wideband detection. The method includes the following steps: Step 101: Using wavelet denoising and amplitude normalization for a single pulse waveform in the pulse group to form a time-domain waveform; Step 102: Performing time-frequency transformation on the time-domain waveform to form a time-frequency image; Step 103: Performing edge calculations on time and frequency for the time-frequency image, and then obtaining the standard deviations of the time edge and frequency edge; Step 104: Obtaining the corresponding time-edge standard deviation and frequency-edge standard deviation for all single pulse waveforms in the pulse group to obtain the planar distribution of time-frequency edge standard deviation characteristic parameters; where tfr j (t, f) is the time-frequency image information processed by the time-frequency transformation module (13), m f (t) j is the result after time edge calculation, j is the j-th pulse, where j = 1, 2, … N, and N is the total number of pulse waveforms included in the pulse group; where tfr j (t, f) is the time-frequency image information processed by the time-frequency transformation module (13), m t (f) j is the result after frequency edge calculation; Standard deviation at the time edge: Among them is the time edge mean value; is the energy at the edge of time; Frequency edge standard deviation: wherein is the frequency edge mean value; is the energy at the frequency edge.

2. A system for the method of extracting edge feature parameters of the time-frequency image of partial discharge pulse waveforms described in claim 1, characterized in that, Including a pulse waveform-time series module (10) of the hybrid of PD source and noise source obtained by ultra-wideband detection, a pulse time-domain waveform module (12), a time-frequency transformation module (13), a time edge module (14), a frequency edge module (15), a time-edge standard deviation module (16), a frequency-edge standard deviation module (17), and a planar distribution module (18) of time-frequency edge standard deviation characteristic parameters; The pulse waveform-time series module (10) of the hybrid of PD source and noise source obtained by the ultra-wideband detection is connected to the time-frequency transformation module (13) through the pulse time-domain waveform module (12). The time-frequency transformation module (13) is respectively connected to the time edge module (14) and the frequency edge module (15). The time edge module (14) is connected to the time-edge standard deviation module (16). The frequency edge module (15) is connected to the frequency-edge standard deviation module (17). The time-edge standard deviation module (16) and the frequency-edge standard deviation module (17) are respectively connected to the planar distribution module (18) of time-frequency edge standard deviation characteristic parameters.

3. The system according to claim 2, wherein The pulse waveform-time series module (10) for the PD source and noise source mixture obtained by ultra-wideband detection uses a data acquisition device with an analog bandwidth of dozens of MHz and a sampling rate of over 100 MS / s. Through a coupling device with a frequency response of over dozens of MHz that meets the requirements of nanosecond-level PD ultra-wideband detection, the single-pulse time-domain waveform and the pulse waveform-time series, i.e., the pulse group p j (t), are recorded based on the pulse waveform triggering technique, and the specific definition is as follows: In the formula: j is the jth pulse, where j = 1, 2, … N, and N is the total number of pulse waveforms included in the pulse group; The pulse waveform consists of k points, and the number of points is determined by the sampling rate f s * and the sampling duration.

4. The system according to claim 2, wherein The system further includes a wavelet denoising module (11) provided between the pulse waveform-time series module (10) of the hybrid of PD source and noise source obtained by ultra-wideband detection and the time-frequency transformation module (13). The wavelet denoising module (11) uses wavelet denoising and amplitude normalization for a single original pulse waveform in the pulse group to form a time-domain waveform; The described pulse time-domain waveform module (12) is used to store the time-domain waveform, i.e., p j (t), which serves as the object to be processed by subsequent modules.

5. The system according to claim 2, characterized in that The time-frequency transformation module (13) performs time-frequency transformation on p j (t) using the Page time-frequency distribution algorithm to obtain the time-frequency image of tfr j (t, f).

6. The system according to claim 2, characterized in that, The calculation of the planar distribution module (18) of time-frequency edge standard deviation characteristic parameters is as follows: where j is the jth pulse, j = 1, 2, … N, and N is the total number of pulse waveforms included in the pulse group.

Images