A partial discharge feature extraction method based on phase-energy-frequency three-dimensional spectrum

Abstract

The application relates to a partial discharge feature extraction method based on a phase-energy-frequency three-dimensional spectrum, which is suitable for insulation state monitoring and defect type identification in electric power equipment. The method comprises the following steps: collecting a partial discharge current signal in electric power equipment, combining phase information of a power frequency voltage signal, an amplitude of the partial discharge current signal and a single discharge main frequency, constructing a phase-energy-frequency three-dimensional spectrum, and realizing accurate differentiation of three typical discharge modes of corona discharge, air gap discharge and surface discharge. The application can significantly improve the recognition accuracy and robustness of the discharge mode by comprehensively utilizing information in different dimensions of the three-dimensional spectrum, and provides a more accurate technical means for insulation aging monitoring, defect identification and equipment operation safety of the electric power equipment.

Description

Technical Field

[0001] This invention belongs to the field of power equipment insulation condition monitoring and partial discharge feature extraction technology, specifically involving a partial discharge feature extraction method based on phase-energy-frequency three-dimensional spectrum. Background Technology

[0002] Partial discharge is an electrical phenomenon in power equipment, typically occurring in localized areas of insulation due to excessively strong electric fields or insulation defects, resulting in arcing. Partial discharge is not only a significant indicator of insulation aging and performance degradation, but it also often causes equipment damage in its early stages. Failure to detect and address it promptly can lead to serious equipment failures or even major accidents such as fires. As power equipment ages, especially under high-voltage and high-current operating conditions, partial discharge becomes increasingly prominent, posing a significant threat to the reliability and safety of power equipment. In power systems, partial discharge is a major cause of equipment failure, particularly in the insulation systems of high-voltage equipment such as transformers, switchgear, and generators. With the development of power equipment towards higher voltage levels and greater power, accurately monitoring and analyzing partial discharge current signals and promptly detecting potential insulation defects has become a key technology for ensuring safe equipment operation and extending its service life.

[0003] In existing technologies, partial discharge measurement and analysis often employ methods such as phase-resolved partial discharge mapping (PRPD) and phase-resolved pulse sequence mapping (PRPS). PRPD typically uses discharge phase and amplitude as coordinates, providing a two-dimensional representation by counting the number of discharges in each phase interval. PRPS, on the other hand, further introduces a pulse sequence or time dimension, representing each discharge pulse in three dimensions—its power frequency phase, amplitude, and sequence of occurrence—to characterize the evolution of partial discharge over time. These methods, by capturing the phase and amplitude of the discharge, provide preliminary identification tools for partial discharge and are widely used in insulation condition monitoring. However, PRPD and PRPS are essentially projections of multi-dimensional discharge information, primarily relying on discharge amplitude and counts. They struggle to quantitatively reflect the energy magnitude and distribution patterns of a single discharge, and their characterization of discharge frequency is relatively limited. In partial discharge analysis of power equipment, map interpretation still heavily relies on experience, resulting in insufficient accuracy and robustness in discharge pattern recognition, making it difficult to meet the needs of refined partial discharge classification.

[0004] To address the aforementioned issues, it is necessary to introduce key physical quantities such as discharge energy and main frequency while maintaining power frequency phase information, in order to comprehensively characterize partial discharge in a higher dimension and at a deeper level. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings and deficiencies of existing technologies by providing a partial discharge feature extraction method based on a phase-energy-frequency three-dimensional spectrum. Taking voltage signals and partial discharge current signals collected from a discharge model test of PP / PBE mixed insulation of AC cables as an example, the method constructs an instantaneous power sequence by calculating the actual voltage and current on the primary side, and performs power integration on each discharge event using an adaptive time window to obtain the energy of a single discharge. Simultaneously, the discharge phase is calculated by combining zero-crossing point location, and the discharge signal is windowed and subjected to high-resolution spectral analysis to obtain the dominant frequency of each discharge. Based on this, a phase-energy-frequency three-dimensional spectrum is constructed using discharge phase, energy, and frequency as three coordinates to characterize the differentiated distribution features of typical discharges such as corona discharge, air gap discharge, and surface discharge. This overcomes the problems of insufficient information in the energy and frequency dimensions of traditional PRPD and PRPS methods, providing a more informative and discriminative method for partial discharge feature extraction in new power equipment.

[0006] The technical problem solved by this invention is achieved through the following technical solution: A method for extracting partial discharge features based on a phase-energy-frequency three-dimensional spectrum, the steps of which are as follows: S1. Platform Setup: The power frequency partial discharge test platform for feature extraction includes a voltage regulator, a step-up transformer, a protective resistor, a test sample, a high-frequency current transformer (HFCT), a voltage divider, an oscilloscope, a data acquisition unit, and a computer. The voltage regulator outputs an adjustable power frequency voltage, and its output is connected to the primary side of the step-up transformer to provide the required high-voltage power frequency electric field to the test sample. The high-voltage terminal of the secondary side of the step-up transformer is connected in series with the protective resistor and the high-voltage electrode of the test sample. The protective resistor is used to limit the fault current and isolate the influence of the high-frequency discharge pulse on the power supply side. The lower electrode of the test sample is grounded through a ground wire, and the HFCT is connected in series on the lower electrode of the test sample to sense and output the partial discharge current signal. The voltage divider is connected in parallel to both ends of the sample under test. The high voltage end reduces the high voltage signal to a safe level at a ratio of 1000:1 through the voltage divider. The low voltage end of the voltage divider is connected to the voltage channel of the oscilloscope and the data acquisition unit respectively, for real-time monitoring and acquisition of the waveform and phase of the applied voltage signal. The output of the high-frequency current transformer (HFCT) is connected to a data acquisition unit via a coaxial cable. The data acquisition unit samples the partial discharge current signal and keeps it synchronized with the voltage measurement signal of the oscilloscope. The data acquisition unit is connected to a computer via a communication interface. The computer is responsible for processing the acquired voltage and partial discharge current signal data, including voltage zero-crossing detection, discharge event identification, discharge energy integral calculation based on |u|·|i|, and spectrum analysis of the main frequency of a single discharge. Based on this, a three-dimensional spectrum of phase-energy-frequency is constructed with phase, energy, and frequency as the three coordinates to realize the visualization of the characteristic distribution of different partial discharge types and subsequent identification and analysis. S2. Data Acquisition and Preprocessing: Voltage signals and partial discharge current signals are acquired through a voltage divider and a high-frequency current transformer. The sampling rate is set to 1MHz and the number of sampling points is set to 1M. The acquired voltage signals and partial discharge current signals are preprocessed, including noise reduction, smoothing and time synchronization, to ensure the accuracy and consistency of the voltage signals and partial discharge current signals. The obtained voltage signals and partial discharge current signals are used for subsequent phase compensation and partial discharge analysis. S3. Phase compensation and zero-crossing detection: The voltage signal of S2 is fitted with a power frequency sine wave to obtain the amplitude, phase and offset of the sine wave model. The fitting result of the voltage signal is obtained by fitting the parameters using the least squares method. The fitted sine wave model is used to perform overall lag or lead compensation on the voltage signal to ensure that the voltage signal is in phase with the power frequency voltage. The zero-crossing position of the smoothed voltage signal is detected to determine the power supply frequency. The positive zero-crossing point of the voltage signal on the time axis is calculated, and the power frequency period is determined by the zero-crossing point. The power supply frequency is accurately calculated by the interval of the zero-crossing point, providing a reference for the power frequency period for subsequent analysis. S4. Partial Discharge Event Detection and Event Merging: Partial discharge current signals have pulse characteristics. The partial discharge current signal is analyzed by peak detection method to find the peak value in the partial discharge current signal, i.e., the partial discharge pulse. The pulses detected according to the set threshold are merged. If multiple peak values ​​appear in adjacent time windows and their peak amplitudes are close, these peak values ​​will be regarded as the same discharge event. The window size and threshold are set to ensure that no valid partial discharge events are missed. S5. Energy calculation and single discharge main frequency estimation; S6. Three-dimensional spectrum generation and discharge pattern recognition: Based on the phase, energy and frequency information mentioned above, the phase, energy and frequency of each partial discharge event are treated as a point in three-dimensional space to construct a phase-energy-frequency three-dimensional spectrum, ensuring that each discharge event appears as a point in the spectrum. In the three-dimensional spectrum, the X-axis represents the phase, the Y-axis represents the main frequency of a single discharge, and the Z-axis represents the energy. The shape of the three-dimensional spectrum can reflect the characteristics of different types of discharge.

[0007] Moreover, S5 specifically refers to: The process of energy calculation using the instantaneous power calculation method based on |u|·|i| is as follows: (1) Voltage signal and partial discharge current signal conversion: First, the voltage signal is converted into a partial discharge current signal. and partial discharge current signal Perform a side restoration; voltage signal By pressure ratio Convert to primary side real voltage ; Partial discharge current signal Through the current transformer ratio and load resistance Converted to primary side real current ; (2) Instantaneous power calculation: Instantaneous power It is the actual voltage. and real current The product of absolute values: ; During this process, the voltage signal and the partial discharge current signal have been converted into the actual voltage and actual current on the primary side in order to perform power calculation; (3) Energy integration: For each partial discharge event time window, the total energy of the instantaneous power is calculated using the numerical integration method. For each event, its start time is found. and end time And power integration is performed within the time window between these two times; Numerical integration is used to calculate the energy of each partial discharge more accurately. ; The process of estimating the dominant frequency of each partial discharge event using high-resolution spectral analysis is as follows: (1) Signal window selection and de-averaging: Select a time window for the partial discharge signal. The window should cover a complete discharge event. If the duration of the event is short, the window will be automatically adjusted to a fixed length to perform de-averaging on the signal to eliminate the DC component. (2) Windowing: To reduce spectral leakage, the signal window is multiplied by a window function for windowing. ; in, It is the Hanning window function; (3) FFT analysis: Perform a Fast Fourier Transform (FFT) on the windowed signal to obtain the spectrum of the signal, and then find the frequency corresponding to the maximum amplitude in the spectrum to obtain the main frequency. .

[0008] The advantages and beneficial effects of this invention are as follows: 1. Accurately distinguish different defect types This invention constructs a three-dimensional phase-energy-frequency spectrum to fuse multi-dimensional features of three typical discharge modes: corona discharge, air gap discharge, and surface discharge. This results in independent and clearly defined feature clusters in three-dimensional space, significantly improving the accuracy of defect identification. In contrast, traditional PRPD maps rely solely on the three elements of "discharge phase, discharge amplitude, and count," lacking energy and frequency information. This makes them highly susceptible to confusion due to overlapping amplitude ranges, noise disturbances, or similar phase distributions, especially in single defect models where local overlap may occur. The three-dimensional spectrum of this invention, by incorporating energy and dominant frequency—two key physical quantities—greatly enhances the separability of different defects, fundamentally overcoming the limitation of feature confusion in PRPD and achieving more reliable and stable discharge type identification.

[0009] 2. More accurate energy calculation Traditional PRPD analysis primarily relies on discharge amplitude, failing to reflect the true energy output throughout the entire discharge process and easily overlooking the oscillating components at the pulse tail, leading to energy estimation errors. This invention employs an instantaneous power calculation method based on |u|·|i|, integrating the complete discharge current waveform without depending on single-point amplitudes, thus obtaining a more accurate and physically meaningful energy characterization for each discharge event. This method is particularly effective in weak discharges or defects with small energy differences, avoiding the omissions or miscalculations of low-energy events by traditional methods. By introducing energy as a key dimension into the three-dimensional spectrum, the energy level differences between different discharge types are visually presented. Since the discharge energy directly relates to the degree of insulation damage, the three-dimensional spectrum clearly demonstrates the differences in the damage caused to insulation materials by different defects, providing a more engineering-significant basis for insulation condition assessment and early fault diagnosis.

[0010] 3. High-resolution frequency analysis This invention employs spectral analysis combined with three-point interpolation to extract the main frequency of discharge signals with high precision. This method further improves frequency resolution based on FFT, effectively revealing subtle differences in oscillation frequencies among different types of discharges. It significantly overcomes the limitations of PRPD in characterizing frequency characteristics and addressing the lack of frequency information, thereby enhancing sensitivity to differences in discharge mechanisms.

[0011] 4. Simple and efficient implementation method This invention, through precise analysis of voltage and partial discharge current signals, combined with existing experimental platforms and acquisition equipment, enables rapid and accurate detection and pattern recognition of partial discharge events. It is simple to operate and highly adaptable. This method requires no complex equipment modifications or additional hardware investment and can be directly applied to existing partial discharge detection systems for power equipment.

[0012] In summary, the partial discharge detection method based on phase-energy-frequency three-dimensional spectrum provided by this invention can effectively solve the problems of insufficient accuracy in partial discharge pattern recognition and energy calculation in the prior art, and has significant engineering application value, especially in the accurate detection and evaluation of cable insulation defects, showing strong application potential. Attached Figure Description

[0013] Figure 1 The present invention provides phase-energy-frequency three-dimensional spectra of three different defects at 6 kV based on a discharge model of a polypropylene and elastomer PBE blend. Figure 2 The present invention provides phase-energy-frequency three-dimensional spectra of three different defects at 8 kV based on a discharge model of a polypropylene and elastomer PBE blend. Figure 3 The present invention provides phase-energy-frequency three-dimensional spectra of three different defects at 10 kV based on a discharge model of a polypropylene and elastomer PBE blend. Figure 4 This is a schematic diagram of the power frequency partial discharge test platform of the present invention. Detailed Implementation

[0014] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0015] like Figure 4 As shown, a method for extracting partial discharge features based on a phase-energy-frequency three-dimensional spectrum is innovative in that the method comprises the following steps: S1. Platform Setup: The power frequency partial discharge test platform for feature extraction includes a voltage regulator, a step-up transformer, a protective resistor, a test sample, a high-frequency current transformer (HFCT), a voltage divider, an oscilloscope, a data acquisition unit, and a computer. The voltage regulator outputs an adjustable power frequency voltage, and its output is connected to the primary side of the step-up transformer to provide the required high-voltage power frequency electric field to the test sample. The high-voltage terminal of the secondary side of the step-up transformer is connected in series with the protective resistor and the high-voltage electrode of the test sample. The protective resistor is used to limit the fault current and isolate the influence of the high-frequency discharge pulse on the power supply side. The lower electrode of the test sample is grounded through a ground wire, and the HFCT is connected in series on the lower electrode of the test sample to sense and output the partial discharge current signal. The voltage divider is connected in parallel to both ends of the sample under test. The high voltage end reduces the high voltage signal to a safe level at a ratio of 1000:1 through the voltage divider. The low voltage end of the voltage divider is connected to the voltage channel of the oscilloscope and the data acquisition unit respectively, for real-time monitoring and acquisition of the waveform and phase of the applied voltage signal. The output of the high-frequency current transformer (HFCT) is connected to a data acquisition unit via a coaxial cable. The data acquisition unit samples the partial discharge current signal and keeps it synchronized with the voltage measurement signal of the oscilloscope. The data acquisition unit is connected to a computer via a communication interface. The computer is responsible for processing the acquired voltage and partial discharge current signal data, including voltage zero-crossing detection, discharge event identification, discharge energy integral calculation based on |u|·|i|, and spectrum analysis of the main frequency of a single discharge. Based on this, a three-dimensional spectrum of phase-energy-frequency is constructed with phase, energy, and frequency as the three coordinates to realize the visualization of the characteristic distribution of different partial discharge types and subsequent identification and analysis. S2. Data Acquisition and Preprocessing: Voltage and partial discharge current signals are acquired through a voltage divider and a high-frequency current transformer. The sampling rate is set to 1MHz and the number of sampling points is set to 1M. The acquired voltage and partial discharge current signals are preprocessed, including noise reduction, smoothing and time synchronization, to ensure the accuracy and consistency of the voltage and partial discharge current signals. The obtained voltage and partial discharge current signals are used for subsequent phase compensation and partial discharge analysis. S3. Phase compensation and zero-crossing detection: The voltage signal of S2 is fitted with a power frequency sine wave to obtain the amplitude, phase and offset of the sine wave model. The fitting result of the voltage signal is obtained by fitting the parameters using the least squares method. The fitted sine wave model is used to perform overall lag or lead compensation on the voltage signal to ensure that the voltage signal is in phase with the power frequency voltage. The zero-crossing position of the smoothed voltage signal is detected to determine the power supply frequency. The positive zero-crossing point of the voltage signal on the time axis is calculated, and the power frequency period is determined by the zero-crossing point. The power supply frequency is accurately calculated by the interval of the zero-crossing point, providing a reference for the power frequency period for subsequent analysis. S4. Partial Discharge Event Detection and Event Merging: Partial discharge current signals have pulse characteristics. The partial discharge current signal is analyzed by peak detection method to find the peak value in the partial discharge current signal, i.e., the partial discharge pulse. The pulses detected according to the set threshold are merged. If multiple peak values ​​appear in adjacent time windows and their peak amplitudes are close, these peak values ​​will be regarded as the same discharge event. The window size and threshold are set to ensure that no valid partial discharge events are missed. S5. Energy Calculation and Main Frequency Estimation: The process of energy calculation using the instantaneous power calculation method based on |u|·|i| is as follows: (1) Voltage signal and partial discharge current signal conversion: First, the voltage signal is converted into a partial discharge current signal. and partial discharge current signal Perform a side restoration; voltage signal By pressure ratio Convert to primary side real voltage ; Partial discharge current signal Through the current transformer ratio and load resistance Converted to primary side real current ; (2) Instantaneous power calculation: Instantaneous power It is the actual voltage. and real current The product of absolute values: ; During this process, the voltage signal and the partial discharge current signal have been converted into the actual voltage and actual current on the primary side in order to perform power calculation; (3) Energy integration: For each partial discharge event time window, the total energy of the instantaneous power is calculated using the numerical integration method. For each event, its start time is found. and end time And power integration is performed within the time window between these two times; Numerical integration is used to calculate the energy of each partial discharge more accurately. ; The process of estimating the dominant frequency of each partial discharge event using high-resolution spectral analysis is as follows: (1) Signal window selection and de-averaging: Select a time window for the partial discharge signal. The window should cover a complete discharge event. If the duration of the event is short, the window will be automatically adjusted to a fixed length to perform de-averaging on the signal to eliminate the DC component. (2) Windowing: To reduce spectral leakage, the signal window is multiplied by a window function for windowing. ; in, It is the Hanning window function; (3) FFT analysis: Perform a Fast Fourier Transform (FFT) on the windowed signal to obtain the spectrum of the signal, and then find the frequency corresponding to the maximum amplitude in the spectrum to obtain the main frequency. ; S6. Three-dimensional spectrum generation and discharge pattern recognition: Based on the phase, energy and frequency information mentioned above, the phase, energy and frequency of each partial discharge event are treated as a point in three-dimensional space to construct a phase-energy-frequency three-dimensional spectrum, ensuring that each discharge event appears as a point in the spectrum. In the three-dimensional spectrum, the X-axis represents the phase, the Y-axis represents the dominant frequency, and the Z-axis represents the energy. The shape of the three-dimensional spectrum can reflect the characteristics of different types of discharge.

[0016] This invention is based on a discharge model of a polypropylene / elastomer PBE blend. The phase-energy-frequency three-dimensional spectra of three different defects at 6, 8, and 10 kV are shown below. Figures 1-3 As shown in the figure, (a) is corona discharge, (b) is surface discharge, and (c) is air gap discharge.

[0017] Depend on Figure 1 It can be seen that the dominant frequency of corona discharge is concentrated between 0 and 150 kHz, with the discharge point mainly concentrated in the negative half-cycle. The single discharge energy in the positive half-cycle is relatively large, reaching a maximum of 8 μJ, while the discharge energy values ​​in the negative half-cycle are relatively small, all below 2 μJ. The dominant frequency of surface discharge is concentrated between 0 and 250 kHz, a significantly wider frequency band than corona discharge. The distribution of discharge points in the positive and negative half-cycles is relatively even, with significant high-energy discharges in both. The peak value in the positive half-cycle can reach 7 μJ, and the high-energy discharges are mainly concentrated below 50 kHz, with smaller discharge energy at frequencies above 50 kHz. The dominant frequency of air gap discharge is concentrated between 200 kHz and 350 kHz, with a relatively even distribution of discharge points in the positive and negative half-cycles. The maximum discharge energy value is 4 μJ, which is smaller than the other two types of defects. There are significant differences in the phase-energy-frequency diagrams of the three types of defects.

[0018] Depend on Figure 2It can be seen that the phase and frequency distributions of the three discharge points are basically the same as those at 6 kV, but the maximum discharge energy increases slightly. The maximum energy of corona discharge is 12 μJ, the maximum energy of surface discharge is 9 μJ, and the maximum energy of air gap discharge is 6 μJ. There are significant differences in the phase-energy-frequency diagrams of the three defects.

[0019] Depend on Figure 3 It can be seen that the phase and frequency distributions of the three discharge points are similar to... Figure 1 , Figure 2 The results are largely consistent, but the maximum discharge energy increases slightly: the maximum energy for corona discharge is 17 μJ, for surface discharge it is 14 μJ, and for air gap discharge it is 9 μJ. With increasing voltage levels, the number of discharge points increases significantly for all three types of defects, and higher voltage levels are often accompanied by higher single-discharge energies. However, the phase and frequency of the discharge points remain unchanged, and the phase-energy-frequency diagrams for the three types of defects show significant differences.

[0020] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for extracting partial discharge features based on a three-dimensional phase-energy-frequency spectrum, characterized in that: The steps of the method are as follows: S1. Platform Setup: The power frequency partial discharge test platform for feature extraction includes a voltage regulator, a step-up transformer, a protective resistor, a sample under test, a high-frequency current transformer (HFCT), a voltage divider, an oscilloscope, a data acquisition unit, and a computer. The voltage regulator outputs an adjustable power frequency voltage, and the output terminal of the voltage regulator is connected to the primary side of the step-up transformer to provide the required high-voltage power frequency electric field to the sample under test. The high-voltage terminal of the secondary side of the step-up transformer is connected in series with a protection resistor and the high-voltage electrode of the sample under test. The protection resistor is used to limit the fault current and isolate the influence of the high-frequency discharge pulse on the power supply side. The lower electrode of the sample under test is grounded through a ground wire. The lower electrode of the sample under test is connected in series with the high-frequency current transformer (HFCT) to sense and output the partial discharge current signal. The voltage divider is connected in parallel to both ends of the sample under test. The high voltage end of the voltage divider reduces the high voltage signal to a safe level at a ratio of 1000:

1. The low voltage end of the voltage divider is connected to the voltage channel of the oscilloscope and the data acquisition unit respectively. The oscilloscope is used to monitor the waveform of the applied voltage signal in real time, and the data acquisition unit is used to acquire the waveform and phase of the applied voltage signal. The output of the high-frequency current transformer (HFCT) is connected to a data acquisition unit via a coaxial cable. The data acquisition unit samples the partial discharge current signal and keeps it synchronized with the applied voltage signal acquired by the data acquisition unit. The data acquisition unit is connected to a computer via a communication interface. The computer is responsible for processing the acquired applied voltage signal and partial discharge current signal data, including voltage zero-crossing detection, discharge event identification, discharge energy integral calculation based on |u|·|i|, and spectrum analysis of the main frequency of a single discharge. Based on this, a three-dimensional spectrum of phase-energy-frequency is constructed with phase, energy, and frequency as the three coordinates to realize the visualization of the characteristic distribution of different partial discharge types and subsequent identification and analysis. S2. Data Acquisition and Preprocessing: The sampling rate is set to 1MHz and the number of sampling points is set to 1M. The acquired applied voltage signal and partial discharge current signal are preprocessed, including noise reduction, smoothing and time synchronization, to ensure the accuracy and consistency of the applied voltage signal and partial discharge current signal. The obtained applied voltage signal and partial discharge current signal are used for subsequent phase compensation and partial discharge analysis. S3. Phase compensation and zero-crossing detection: The applied voltage signal of S2 is fitted with a power frequency sine wave to obtain the amplitude, phase and offset of the sine wave model. The fitting result of the applied voltage signal is obtained by fitting the parameters using the least squares method. The fitted sine wave model is used to perform overall lag or lead compensation on the applied voltage signal to ensure that the applied voltage signal is phase synchronized with the power frequency voltage. The zero-crossing position of the smoothed applied voltage signal is detected to determine the power supply frequency. The positive zero-crossing point of the applied voltage signal on the time axis is calculated, and the power frequency period is determined by the zero-crossing point. The interval of the zero-crossing point is used to accurately calculate the power supply frequency, providing a reference for the power frequency period for subsequent analysis. S4. Partial Discharge Event Detection and Event Merging: Partial discharge current signals have pulse characteristics. The partial discharge current signal is analyzed by peak detection method to find the peak value in the partial discharge current signal, i.e., the partial discharge pulse. The pulses detected according to the set threshold are merged. If multiple peak values ​​appear in adjacent time windows and their peak amplitudes are close, these peak values ​​will be regarded as the same discharge event. The window size and threshold are set to ensure that no valid partial discharge events are missed. S5. Energy calculation and single discharge main frequency estimation; S6. Three-dimensional spectrum generation and discharge pattern recognition: Based on the phase, energy and frequency information mentioned above, the phase, energy and frequency of each partial discharge event are treated as a point in three-dimensional space to construct a phase-energy-frequency three-dimensional spectrum, ensuring that each discharge event appears as a point in the spectrum. In the three-dimensional spectrum, the X-axis represents the phase, the Y-axis represents the main frequency of a single discharge, and the Z-axis represents the energy. The shape of the three-dimensional spectrum can reflect the characteristics of different types of discharge.

2. The partial discharge feature extraction method based on phase-energy-frequency three-dimensional spectrum according to claim 1, characterized in that: Specifically, S5 is: The process of energy calculation using the instantaneous power calculation method based on |u|·|i| is as follows: (1) Conversion between applied voltage signal and partial discharge current signal: First, convert the applied voltage signal... and partial discharge current signal Perform a side restoration; Apply voltage signal By pressure ratio Convert to primary side real voltage ； Partial discharge current signal Through the current transformer ratio and load resistance Converted to primary side real current ； (2) Instantaneous power calculation: Instantaneous power It is the actual voltage. and real current The product of absolute values: ； During this process, the applied voltage signal and partial discharge current signal have been converted into the actual voltage and actual current on the primary side in order to perform instantaneous power calculation; (3) Energy integration: For each partial discharge event time window, the total energy of the instantaneous power is calculated using the numerical integration method. For each event, its start time is found. and end time And power integration is performed within the time window between these two times; Numerical integration is used to calculate the energy of each partial discharge more accurately. ； The process of estimating the main frequency of a single discharge using high-resolution spectrum analysis is as follows: (1) Signal window selection and de-averaging: Select a time window of the partial discharge current signal. The window should cover a complete discharge event. If the duration of the event is short, the window will be automatically adjusted to a fixed length. The signal is de-averaged to eliminate the DC component. (2) Windowing: To reduce spectral leakage, the signal window is multiplied by a window function for windowing. ； in, It is the Hanning window function; (3) FFT analysis: Perform a Fast Fourier Transform (FFT) on the windowed signal to obtain the spectrum of the signal, and then find the frequency corresponding to the maximum amplitude in the spectrum to obtain the main frequency. .

Images