Switch cabinet insulation monitoring method and system based on partial discharge feature recognition

Abstract

This invention belongs to the field of partial discharge measurement technology, specifically relating to a method and system for monitoring switchgear insulation based on partial discharge feature recognition. The method includes: extracting the power frequency cycle of a high-frequency current signal to generate current analysis window data; filtering the current analysis window data to generate current waveform data; extracting the peak amplitude decay time constant; eliminating switch operation interference pulses based on the comparison between the peak amplitude decay time constant and the standard decay time constant of the partial discharge pulse to generate a partial discharge pulse sequence; statistically analyzing the distribution quantity within each phase interval to construct a phase distribution feature vector and generate an insulation state deviation coefficient; comparing the insulation state deviation coefficient with a deviation threshold; and generating an insulation abnormality alarm signal when the insulation state deviation coefficient exceeds the deviation threshold. This invention achieves accurate, reliable, and adaptive online monitoring of switchgear insulation status under multi-source interference environments.

Description

Technical Field

[0001] This invention relates to the field of partial discharge measurement technology, and in particular to a switchgear insulation monitoring method and system based on partial discharge feature identification. Background Technology

[0002] Currently, online monitoring of switchgear insulation status commonly employs Phase-Resolved Partial Discharge Mapping (PRPD) technology. Its basic implementation involves coupling transient pulse current signals generated by partial discharge using a high-frequency current sensor installed on the switchgear grounding wire, while simultaneously acquiring the voltage signal from the feeder terminal as a power frequency phase reference. The detected pulse signals are categorized according to their corresponding power frequency phases, and the amplitude and frequency distribution of pulses within each phase window are statistically analyzed to construct a phase-resolved partial discharge map, which serves as the basis for assessing the type and degree of insulation defects. To suppress field interference, existing solutions typically incorporate a fixed-parameter filtering stage at the signal processing front end, filtering detected pulses based on a preset pulse discrimination threshold. Their effective operation relies on the assumption that field interference pulses and partial discharge pulses possess stable and clearly distinguishable characteristics in the time or frequency domain. In relatively clean electromagnetic environments, this method effectively reflects insulation status and has been widely applied.

[0003] However, the actual operating environment of switchgear is generally subject to multi-source interference. The most representative are periodic narrowband noise introduced by power line carrier communication and transient pulses generated by circuit breaker opening and closing operations. The former manifests as a concentrated narrowband component in the frequency domain, and its frequency position and bandwidth drift with the operating state of the carrier communication, making it difficult for fixed-parameter filters with preset center frequency and bandwidth to achieve continuous and effective suppression. The latter presents a sudden transient waveform in the time domain similar to the shape of partial discharge pulses, which cannot be reliably distinguished from real discharge pulses based solely on amplitude or fixed discrimination thresholds. When the spectral or time-domain characteristics of the above interferences overlap with the partial discharge signal, interference components similar to the shape of partial discharge pulses will still remain in the filtered waveform. These non-discharge-source pulse events are included in the phase distribution spectrum, causing the spectrum characteristics to fail to accurately reflect the insulation status. Furthermore, residual carrier interference can raise the base level of pulse detection, masking and missing real discharge pulses with lower amplitudes. These factors collectively lead to insufficient adaptability and reliability of existing monitoring methods in complex field environments, easily resulting in false alarms or missed alarms. Summary of the Invention

[0004] In view of the shortcomings of the prior art, the purpose of this invention is to provide a switchgear insulation monitoring method and system based on partial discharge feature recognition, which can realize accurate, reliable and adaptive online monitoring of switchgear insulation status under multi-source interference environment.

[0005] To achieve the above objectives, the present invention provides a switchgear insulation monitoring method based on partial discharge feature identification, comprising: S1. Obtain the voltage signal of the switchgear feeder terminal and the high-frequency current signal of the grounding wire. Use the power frequency zero-crossing point of the voltage signal to extract the power frequency period of the synchronously sampled high-frequency current signal and generate current analysis window data. S2. Extract the periodic narrowband spectral components from the current analysis window data. Construct a notch filter based on the center frequency and bandwidth of the periodic narrowband spectral components to filter the current analysis window data and generate current waveform data after carrier interference suppression. S3. Extract the peak amplitude decay time constant of transient pulses from the current waveform data after carrier interference suppression. Based on the comparison between the peak amplitude decay time constant and the standard decay time constant of partial discharge pulses, remove the switching operation interference pulses and generate a partial discharge pulse sequence. S4. Obtain the power frequency phase position of the pulse in the partial discharge pulse sequence, count the distribution number in each phase interval to construct the phase distribution feature vector, and compare the phase distribution feature vector with the standard feature vector to generate the insulation state deviation coefficient. S5. Compare the insulation condition deviation coefficient with the deviation threshold. When the insulation condition deviation coefficient exceeds the deviation threshold, generate an insulation abnormality alarm signal.

[0006] Preferably, in S1, the current analysis window data is generated, including: The waveform amplitude of the voltage signal after synchronous sampling is detected, and the moment when the amplitude of the voltage signal waveform is zero is identified as the zero-crossing point of the power frequency. The time interval between two adjacent zero-crossing points of the power frequency in the same direction is taken as a complete power frequency cycle intercept window; According to the capture window, the current waveform segment corresponding to the power frequency cycle is captured from the high-frequency current signal after synchronous sampling, and the current waveform segment is used as the current analysis window data.

[0007] Preferably, in S2, extracting the periodic narrowband spectral components from the current analysis window data includes: A frequency domain transformation is performed on the current analysis window data to map the current analysis window data from the time domain to the frequency domain, generating current frequency domain spectrum data; Amplitude peak search is performed on the current frequency domain spectrum data to identify frequency domain components whose amplitude exceeds the preset spectrum amplitude threshold and whose bandwidth is less than the preset bandwidth threshold. The identified frequency domain components are then used as periodic narrowband spectrum components. Record the center frequency and bandwidth of the periodic narrowband spectral components, and use the center frequency and bandwidth as carrier interference characteristic parameters.

[0008] Preferably, in S2, the current waveform data after carrier interference suppression is generated, including: Obtain the center frequency and bandwidth from the carrier interference characteristic parameters, use the center frequency as the notch filter center frequency and the bandwidth as the notch filter bandwidth, and construct a notch filter according to the notch filter center frequency and notch filter bandwidth. The notch filter is applied to the current analysis window data, so that the frequency domain components in the current analysis window data located within the notch bandwidth are attenuated to below the preset attenuation threshold, while the frequency domain components in the current analysis window data located outside the notch bandwidth are preserved. The current analysis window data after notch filtering is used as the current waveform data after carrier interference suppression.

[0009] Preferably, in S3, the peak amplitude attenuation time constant of the transient pulse in the current waveform data after carrier interference suppression is extracted, including: The current waveform data after carrier interference suppression is pulse searched in chronological order to identify time-domain waveform segments whose amplitude exceeds the preset pulse amplitude threshold. The identified time-domain waveform segments are taken as transient pulses, and the start time, peak time and end time of the transient pulses are recorded. Extract the signal amplitude variation curve of the transient pulse from the peak time to the end time, and use the signal amplitude variation curve as the attenuation curve of the transient pulse; The decay curve is subjected to exponential fitting, and the resulting exponential decay time constant is used as the peak amplitude decay time constant of the transient pulse.

[0010] Preferably, in S3, generating a partial discharge pulse sequence includes: Obtain the peak amplitude decay time constant of the transient pulse, and compare the peak amplitude decay time constant with the pre-calibrated standard decay time constant of the partial discharge pulse; When the ratio of the peak amplitude decay time constant to the standard decay time constant is less than the preset threshold, the transient pulse is identified as a switching operation interference pulse, and the switching operation interference pulse is removed from the current waveform data after carrier interference suppression. When the ratio is not less than the preset threshold, the transient pulse is identified as a partial discharge candidate pulse and the partial discharge candidate pulse is retained. All the retained partial discharge candidate pulses are arranged in chronological order to generate a partial discharge pulse sequence.

[0011] Preferably, in S4, the phase distribution feature vector is constructed by statistically analyzing the distribution quantity within each phase interval, including: The peak point time of the pulse in the partial discharge pulse sequence is extracted. The relative position of the peak point time within the power frequency period of the current analysis window data is taken as the power frequency phase position of the pulse. The power frequency phase position is associated with the pulse to generate a partial discharge pulse sequence with phase label. The power frequency cycle is divided into phase intervals according to equal phase intervals. The pulses in the partial discharge pulse sequence with phase marking are assigned to the corresponding phase intervals according to the power frequency phase position. The number of pulses assigned to each phase interval is counted to generate the pulse distribution number of each phase interval. The number of phase intervals is dynamically determined based on the total number of pulses during the acquisition period, ensuring that the statistical fluctuations in the number of pulses in each interval do not exceed a preset ratio. The phase interval division method for the current monitoring period is consistent with the phase interval division method used when constructing the standard feature vector, so that the phase distribution feature vector has the same dimension as the standard feature vector. Using the number of pulses distributed in each phase interval as components, construct a vector according to the arrangement order of the phase intervals, and use the vector as the phase distribution characteristic vector.

[0012] Preferably, in S4, the phase distribution feature vector is compared with the standard feature vector to generate an insulation state deviation coefficient, including: Obtain multiple historical phase distribution feature vectors of the switchgear under normal insulation conditions, take the arithmetic mean of the components in the same phase interval position in the historical phase distribution feature vector, use the arithmetic mean as the standard component of the corresponding phase interval position, and construct the standard feature vector from the standard components of all phase interval positions. The insulation state deviation coefficient is obtained by comparing the distance between the normalized phase distribution eigenvector and the normalized standard eigenvector using the following formula: ; Where D is the insulation state deviation coefficient, and m is the total number of phase intervals. v represents the total number of partial discharge pulses during the current monitoring period. j This refers to the component in the phase distribution feature vector corresponding to position j in the phase interval, i.e., the number of pulses distributed within the j-th phase interval. u is the sum of the historical average pulse counts for each phase interval under normal conditions. j It is the component in the standard feature vector corresponding to the position j of the phase interval, that is, the average number of pulses distributed under the historical normal state in the j-th phase interval; Distance comparison includes normalized Euclidean distance or weighted normalized Euclidean distance; The insulation condition deviation coefficient is used as the criterion for insulation condition monitoring.

[0013] Preferably, in S5, generating an insulation fault alarm signal includes: Obtain the historical insulation state deviation coefficient of the switchgear under normal insulation conditions, and set the deviation threshold according to the upper quantile of the statistical distribution of the historical insulation state deviation coefficient. The insulation state deviation coefficient is numerically compared with the deviation threshold. When the insulation condition deviation coefficient exceeds the deviation threshold, it is determined that the insulation condition of the switchgear is abnormal, an insulation abnormality alarm signal is generated, and the insulation abnormality alarm signal is output to the monitoring terminal.

[0014] A switchgear insulation monitoring system based on partial discharge feature recognition, used to implement the above method, includes: The signal window acquisition module is used to acquire the voltage signal of the switch cabinet feeder terminal and the high-frequency current signal of the grounding wire. It uses the power frequency zero crossing point of the voltage signal to extract the power frequency period of the synchronously sampled high-frequency current signal and generate current analysis window data. The carrier filtering module is used to extract the periodic narrowband spectral components in the current analysis window data. Based on the center frequency and bandwidth of the periodic narrowband spectral components, a notch filter is constructed to filter the current analysis window data and generate current waveform data after carrier interference suppression. The pulse discrimination module is used to extract the peak amplitude decay time constant of transient pulses in the current waveform data after carrier interference suppression. Based on the comparison between the peak amplitude decay time constant and the standard decay time constant of partial discharge pulses, the switching operation interference pulses are eliminated, and a partial discharge pulse sequence is generated. The feature measurement module is used to obtain the power frequency phase position of the pulse in the partial discharge pulse sequence, count the distribution quantity in each phase interval to construct the phase distribution feature vector, and compare the phase distribution feature vector with the standard feature vector to generate the insulation state deviation coefficient. The abnormal alarm module is used to compare the insulation state deviation coefficient with the deviation threshold. When the insulation state deviation coefficient exceeds the deviation threshold, an insulation abnormal alarm signal is generated.

[0015] The present invention has the following beneficial effects: This invention constructs a multi-stage interference suppression link consisting of power frequency cycle truncation, frequency domain notch filtering, and time domain attenuation time constant discrimination. It applies targeted processing to two typical types of interference with drastically different properties in the field: for narrow-band frequency domain and continuous time domain power carrier interference, a notch filter is dynamically constructed based on the real-time detected center frequency and bandwidth for directional attenuation; for transient time domain and wide-band frequency domain switching operation pulses, they are discriminated against and eliminated based on their essential physical difference with partial discharge pulses in peak amplitude attenuation time constant. The two-stage processing is implemented collaboratively in the order of frequency domain filtering followed by time domain discrimination. Notch filtering first suppresses the interference floor, ensuring the detectability of weak discharge pulses; attenuation discrimination then purifies the pulse sequence, ensuring the sample purity of phase statistics. This overcomes the limitation of single-domain processing methods in simultaneously handling two types of interference, achieving high-fidelity extraction of partial discharge characteristic signals in multi-source interference environments.

[0016] All key parameters of this invention are constructed using an adaptive method. The notch filter is dynamically generated based on the spectral characteristics of each detected carrier interference, and can automatically adjust to follow the drift of carrier frequency and bandwidth. The attenuation time constant discrimination benchmark is determined individually through laboratory calibration or field measurement based on the insulation type of the switchgear under test. The standard feature vector is constructed by averaging the historical phase distribution data of the monitored switchgear under normal insulation conditions, accurately matching the individualized benchmark distribution pattern formed by the equipment due to its insulation structure, service life, and load characteristics. The deviation threshold is independently set based on the upper quantile of the statistical distribution of the historical deviation coefficient of the equipment, so that its alarm boundary is adapted to its normal fluctuation range. Compared with the existing technology that uses uniform fixed parameters, this invention avoids adaptability problems such as system deviation, frequent false alarms, or insufficient sensitivity caused by individual differences in equipment and environmental changes, making the monitoring link widely applicable to switchgear with different insulation types and different operating conditions.

[0017] This invention transforms the traditional phase distribution map comparison, which relies on manual visual interpretation, into a normalized Euclidean distance calculation with a clearly defined mathematical range (where the value range is bounded). The insulation state deviation coefficient continuously and quantitatively characterizes the degree of deviation of the current phase distribution from the normal reference, eliminating the influence of the difference in the total pulse amount at different acquisition periods, making the evaluation results objective, comparable, and reproducible. On this basis, combined with threshold setting and automatic comparison judgment based on historical data statistical distribution, a complete monitoring closed loop from signal acquisition, interference suppression, feature extraction to abnormal alarm output is realized. The alarm signal can carry the deviation value and the proportion of exceeding the threshold and output to local and remote monitoring terminals, providing a reliable quantitative decision basis for online evaluation of the insulation state of switchgear, and significantly improving the accuracy, objectivity and automation level of monitoring. Attached Figure Description

[0018] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 This is a functional block diagram of the system of the present invention. Detailed Implementation

[0019] The embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] Example 1: As Figure 1 As shown, the switchgear insulation monitoring method based on partial discharge feature identification includes the following steps: S1. Obtain the voltage signal of the switchgear feeder terminal and the high-frequency current signal of the grounding wire. Use the power frequency zero-crossing point of the voltage signal to extract the power frequency period of the synchronously sampled high-frequency current signal and generate current analysis window data. In the field of switchgear operation, the partial discharge pulse superimposed on the high-frequency current signal is difficult to distinguish directly from the field interference signal in the time domain. The power frequency voltage signal serves as a reference, and its zero-crossing point corresponds to a fixed position of the system voltage phase. Since the phase distribution of the partial discharge pulse relative to the power frequency voltage is a key criterion for identifying the discharge type and determining the insulation condition, it is necessary to precisely synchronize the high-frequency current signal with the power frequency phase. Therefore, this step first obtains the voltage signal from the switchgear feeder terminal through a voltage transformer, and simultaneously obtains the high-frequency current signal from the switchgear grounding wire through a high-frequency current transformer.

[0021] Synchronous sampling is performed on the acquired voltage and high-frequency current signals. The sampling frequency is determined according to the Nyquist sampling theorem, meaning the sampling frequency should not be less than twice the highest frequency component of the sampled signal. The time-domain width of a partial discharge pulse is typically in the nanosecond to microsecond range, and its spectral energy is mainly distributed in the frequency band from several MHz to hundreds of MHz. The sampling frequency should be determined such that the Nyquist frequency is not lower than the highest effective frequency component of the partial discharge pulse. For typical switchgear partial discharge signals, the energy is concentrated in the frequency band below 50MHz. Therefore, a sampling frequency of 100MHz (Nyquist frequency of 50MHz) is sufficient to capture the main waveform characteristics of the pulse. When the monitored object is a high-voltage switchgear with severe insulation aging and a steeper rising edge of the discharge pulse, the high-frequency components of the signal may extend to the 100MHz band. In this case, the sampling frequency can be increased to 200MHz to ensure the integrity of the waveform sampling. For monitoring scenarios with a relatively clean electromagnetic environment where the focus is mainly on low-frequency discharge components, a sampling frequency of 50MHz can also meet basic detection requirements. The specific value of the above sampling frequency can be adaptively adjusted by those skilled in the art based on the voltage level of the switchgear, the type of insulation structure, and the spectral distribution characteristics of the background noise at the site.

[0022] Based on synchronous sampling, the sampled voltage signal undergoes power frequency zero-crossing detection. The detection method involves continuously monitoring the waveform amplitude of the voltage signal. When the voltage amplitude changes from positive to negative or from negative to positive and crosses the zero amplitude line, this moment is marked as the power frequency zero-crossing moment. The accuracy of the power frequency zero-crossing moment detection directly affects the accuracy of subsequent phase positioning. In actual detection, a linear interpolation method can be used to interpolate the sampling points near the zero-crossing point. The moment when the interpolated voltage amplitude is exactly zero is taken as the zero-crossing moment, thereby improving the zero-crossing positioning accuracy to the sub-microsecond level.

[0023] After detecting the zero-crossing point of the power frequency, the time interval between two adjacent zero-crossing points in the same direction is used as a complete power frequency cycle intercept window. For a 50Hz power system, the time interval between two adjacent zero-crossing points in the same direction is 20 milliseconds, corresponding to a complete power frequency cycle. The current waveform within this intercept window covers the complete 0° to 360° phase range, which can comprehensively reflect the complete distribution pattern of the partial discharge pulse in the positive and negative half-cycles of the power frequency.

[0024] As an alternative, the time interval between two adjacent zero-crossing points in opposite directions can be used as a 10-millisecond intercept window for a half-cycle of the power frequency signal. The corresponding current waveform segment covers a half-cycle phase range of 0° to 180° or 180° to 360°. This method is advantageous in rapid monitoring scenarios requiring higher time resolution. The pulse phase extracted within the half-cycle window can be used for subsequent phase analysis through mirror mapping or direct application. Alternatively, the time interval between four adjacent zero-crossing points in the same direction can be used as a 40-millisecond intercept window, corresponding to two complete power frequency cycles. This method allows for the acquisition of current waveform data over a longer period, which is beneficial for increasing the statistical sample size and improving the statistical stability of phase distribution analysis when the number of pulses is small. Based on any of the above intercept windows, a current waveform segment corresponding to the power frequency cycle is intercepted from the synchronously sampled high-frequency current signal. This intercepted current waveform segment is used as the current analysis window data for subsequent analysis and processing steps. Regardless of whether full-cycle or half-cycle interception is used, the synchronization alignment between the high-frequency current signal and the power frequency phase is ensured by precise positioning at the zero-crossing moment, providing a reliable data foundation for accurately extracting the power frequency phase position of the pulse in subsequent steps.

[0025] By utilizing the zero-crossing point of the power frequency to extract the power frequency cycle of the synchronously sampled high-frequency current signal, the precise synchronization and alignment of the high-frequency current signal with the power frequency phase is achieved, providing a data foundation for accurately extracting the power frequency phase position of the pulse in subsequent steps.

[0026] S2. Extract the periodic narrowband spectral components from the current analysis window data. Construct a notch filter based on the center frequency and bandwidth of the periodic narrowband spectral components to filter the current analysis window data and generate current waveform data after carrier interference suppression. In switchgear operation environments where power line carrier communication is widely used, after the carrier communication signal couples to the grounding wire, it appears in the current analysis window data as a narrow-band spectral component with concentrated energy, narrow spectral width, and periodic repetition. This type of component is fundamentally different in the frequency domain from the wide-spectrum, transient, single-shot characteristics of partial discharge pulses. However, its amplitude is usually much higher than that of partial discharge pulses, making it easy to mask partial discharge pulses when directly performing time-domain threshold detection. Therefore, this step extracts the periodic narrow-band spectral component from the current analysis window data and constructs a targeted notch filter accordingly to suppress carrier interference.

[0027] A frequency domain transformation is performed on the current analysis window data, mapping it from the time domain to the frequency domain to generate current frequency domain spectrum data. This frequency domain transformation can be implemented using the Fast Fourier Transform (FFT) algorithm. The number of points in the FFT determines the frequency resolution, which is equal to the sampling frequency divided by the number of transformation points. When the sampling frequency is 100MHz and the number of transformation points is 1024, the frequency resolution is approximately 97.7kHz, which is sufficient to distinguish the spectral peak positions of typical carrier signals. To distinguish multiple carrier signals that are close to each other, a transformation point count of 2048 can be used to increase the frequency resolution to approximately 48.8kHz. However, increasing the number of transformation points will correspondingly increase the computational load, and a trade-off must be made between frequency resolution and real-time performance in practical applications. Therefore, those skilled in the art can set the number of transformation points according to the density of the carrier signals in the field and the computing power of the processor. By searching for amplitude peaks in the current frequency domain spectrum data, frequency domain components whose amplitude exceeds a preset spectral amplitude threshold and whose bandwidth is less than a preset bandwidth threshold are identified. These identified frequency domain components are then used as periodic narrowband spectral components.

[0028] Carrier interference manifests in the frequency domain as narrow-band spikes with energy significantly higher than the background spectrum, while the spectrum of partial discharge pulses is a wide-band continuous spectrum, and does not form such narrow-band spikes in the frequency domain. This difference in frequency domain morphology is the physical basis for identifying carrier interference using amplitude thresholds. The preset spectral amplitude threshold is set based on the average amplitude of the current frequency domain spectrum data. When the carrier interference signal is strong, its spectral peak value is much higher than the average spectral level, and a higher multiple threshold can still be effectively identified; when the carrier signal is weak or the background noise is high, a lower multiple threshold is required to ensure detection sensitivity. The bandwidth threshold is set based on the main lobe width of the carrier signal spectrum. Typically, the main lobe width of the carrier signal spectrum is in the range of tens of kHz to hundreds of kHz. Setting the bandwidth threshold to within three times the main lobe width ensures that only narrow-band components that conform to the carrier characteristics are filtered out, while wide-band discharge components are excluded.

[0029] The bandwidth threshold is set with reference to the main lobe width of the current frequency domain spectrum data. A frequency domain width not exceeding a specific multiple of this main lobe width is used as the bandwidth threshold. This multiple is based on the fact that the spectral energy of the carrier signal is concentrated within the main lobe and a small number of nearby side lobes. Three times the main lobe width essentially covers more than 99% of the carrier signal's spectral energy, sufficient to include the complete carrier interference component within the notch filter range. Simultaneously, this width is much smaller than the continuous spectrum width of the partial discharge pulse, and will not cause substantial damage to the effective frequency band of the discharge signal. When the spectral purity of the carrier signal in the field is high, this multiple can be appropriately reduced to decrease the notch filter bandwidth and reduce the impact on adjacent frequency bands; when the carrier signal exhibits frequency modulation or spectral broadening, this multiple needs to be appropriately increased to ensure complete interference coverage.

[0030] Record the center frequency and bandwidth of each identified periodic narrowband spectral component, and use the center frequency and bandwidth as carrier interference characteristic parameters. For cases with multiple periodic narrowband spectral components, record the center frequency and corresponding bandwidth of each component separately.

[0031] The center frequency and bandwidth of the carrier interference are obtained. The center frequency is used as the notch filter center frequency, and the bandwidth is used as the notch filter bandwidth. A notch filter is constructed according to the notch filter center frequency and bandwidth. The frequency domain response characteristics of this notch filter are as follows: it has maximum attenuation at the notch filter center frequency, and the attenuation gradually decreases as the frequency deviates from the notch filter center frequency. At the boundary frequencies of the notch filter bandwidth, the attenuation decreases to below a preset attenuation threshold. The setting of the attenuation threshold needs to consider both carrier suppression depth and discharge signal fidelity: when the carrier interference intensity is high, a deeper attenuation threshold is needed to fully suppress the interference, but excessive attenuation may lead to widening of the notch filter bandwidth, thereby damaging the spectral components of the discharge signal near the carrier frequency band; when the carrier interference and the discharge signal frequency band are close, a shallower attenuation threshold needs to be selected to control the extent of notch filter bandwidth expansion and protect the high-frequency components of the discharge signal.

[0032] The constructed notch filter is applied to the current analysis window data, attenuating the frequency components within the notch bandwidth to below a preset attenuation threshold, while preserving the frequency components outside the notch bandwidth. For cases with multiple periodic narrowband spectral components, multiple notch filters can be cascaded onto the current analysis window data, or a multi-notch filter containing multiple notch bands can be constructed and applied to the current analysis window data all at once. The current analysis window data after notch filtering is then used as the current waveform data after carrier interference suppression.

[0033] By performing frequency domain transformation on the current analysis window data and searching for frequency domain components with amplitudes exceeding the threshold and bandwidths less than the threshold, the frequency band location of carrier interference can be accurately located and its center frequency and bandwidth parameters can be extracted. Then, by constructing a notch filter, the periodic narrowband carrier interference can be accurately suppressed, while the wideband components of the partial discharge pulse can be preserved, providing current waveform data with a higher signal-to-noise ratio for subsequent pulse discrimination steps.

[0034] S3. Extract the peak amplitude decay time constant of transient pulses from the current waveform data after carrier interference suppression. Based on the comparison between the peak amplitude decay time constant and the standard decay time constant of partial discharge pulses, remove the switching operation interference pulses and generate a partial discharge pulse sequence. After notch filtering, carrier interference is effectively suppressed. However, the current waveform data after carrier interference suppression may still contain transient pulses generated by circuit breaker opening and closing operations. These pulses, like partial discharge pulses, exhibit sudden transient waveforms in the time domain, making them difficult to distinguish based solely on amplitude or spectral characteristics. However, the energy of switching operation pulses originates from the traveling wave process of external power transmission through the line, and its attenuation characteristics are dominated by line impedance and load parameters, typically exhibiting a rapidly decaying oscillating waveform. In contrast, the energy of partial discharge pulses originates from the charge released by local breakdown at insulation defects, and its attenuation is determined by the equivalent resistance and distributed capacitance of the discharge channel, exhibiting a relatively stable exponential decay characteristic. The two types of pulses differ fundamentally in their peak amplitude attenuation time constants; this step utilizes this physical difference to differentiate between them.

[0035] The current waveform data after carrier interference suppression is subjected to pulse search in chronological order to identify time-domain waveform segments whose amplitude exceeds a preset pulse amplitude threshold. These identified time-domain waveform segments are then identified as transient pulses. The preset pulse amplitude threshold is set based on the root mean square (RMS) value of the background noise in the current waveform data after carrier interference suppression, and multiples of the RMS value of the background noise are used as the preset pulse amplitude threshold. Background noise typically originates from sensor thermal noise and circuit noise, and its amplitude follows a normal distribution. When a threshold of three times the RMS value is used, the probability of noise being misidentified as a pulse is less than 0.3%, ensuring a high detection confidence level. In low-background-noise environments, a threshold of 2.5 times the RMS value can be used to improve the capture capability of weak discharge pulses; while in high-background-noise environments, a threshold of five times the RMS value is required to reduce the false triggering rate caused by noise. For each identified transient pulse, its start time, peak time, and end time are recorded. The start time is the moment when the amplitude of the waveform segment in the time domain first exceeds the preset pulse amplitude threshold continuously, the peak time is the moment when the amplitude in the waveform segment in the time domain reaches the maximum value, and the end time is the moment when the amplitude of the waveform segment in the time domain continuously falls below the preset pulse amplitude threshold.

[0036] The signal amplitude variation curve of each transient pulse from its peak to its end is extracted, and this curve is used as the attenuation curve of the transient pulse. An exponential fit is performed on this attenuation curve, and the resulting exponential attenuation time constant is used as the peak amplitude attenuation time constant of the transient pulse. In this embodiment, the attenuation curve is preprocessed, including envelope extraction or low-pass filtering, to obtain the envelope curve of the oscillation waveform before exponential fitting. The attenuation model expression used for exponential fitting is: ; Where A(t) is the signal amplitude at time t, and A0 is the peak time t. p The corresponding peak amplitude, where t is a time variable based on the start time of the current analysis window data. p At peak time, Let be the peak amplitude decay time constant. The solution that minimizes the sum of squared fitting errors is found by performing least-squares fitting on the sampling points of the decay curve. The value, will The value is used as the peak amplitude decay time constant of the transient pulse. The choice of the number of sampling points during the fitting process will affect the fitting accuracy: using all sampling points of the decay curve can make full use of the waveform information, but when the tail of the decay curve is dominated by noise, adding sampling points at the tail will reduce the fitting accuracy; in this case, only using the sampling points of the beginning segment of the decay curve that contain the main energy and have a high signal-to-noise ratio can obtain a more accurate time constant value that reflects the characteristics of the main decay stage.

[0037] Obtain the pre-calibrated standard decay time constant of the partial discharge pulse. The standard decay time constant is determined as follows: Under laboratory conditions, for a switchgear test sample with the same insulation structure and material system as the switchgear under test in the field, partial discharge is excited using a step-up voltage method. Partial discharge pulse waveforms are acquired under conditions that eliminate external interference. The decay curves of multiple acquired partial discharge pulses are subjected to exponential fitting to obtain their respective decay time constants. The arithmetic mean of these multiple decay time constants is taken as the standard decay time constant for the partial discharge pulse of this type of switchgear. It should be noted that the standard decay time constant of the partial discharge pulse may differ for switchgear with different insulation structures and material systems. For air-insulated switchgear, due to the low dielectric constant of the insulating medium and the small equivalent capacitance of the discharge channel, its standard decay time constant is typically between 0.5 microseconds and 2 microseconds. For solid-insulated or gas-insulated switchgear, due to the high dielectric constant of the insulating material, the large equivalent capacitance of the discharge channel, and the relatively long charge discharge path, its standard decay time constant is typically between 2 microseconds and 5 microseconds.

[0038] Therefore, the specific value of the standard decay time constant should be determined individually based on the insulation type and structural dimensions of the switchgear under test, through laboratory calibration or on-site measurement. The standard decay time constant for different switchgears may fall at different positions within the aforementioned range. Once the specific value of the standard decay time constant for a particular switchgear under test is determined through calibration or measurement, the judgment threshold in the subsequent monitoring process will be based on this calibration value, rather than using the entire 0.5 to 5 microsecond range for judgment. For the same type of switchgear, the decay time constant of its partial discharge pulse is relatively stably distributed around the calibration value, and there will be no arbitrary distribution spanning the entire range. Therefore, using a single threshold for judgment is physically reasonable.

[0039] As an alternative, the standard decay time constant can also be determined by on-site measurement: when the switchgear is put into operation and the insulation is in a healthy state, multiple partial discharge events are collected and the decay time constant of each discharge pulse is extracted. The average value of the multiple extraction results is used as the standard decay time constant for subsequent monitoring.

[0040] The peak amplitude decay time constant of the extracted transient pulse is numerically compared with the pre-calibrated standard decay time constant of a partial discharge pulse. A ratio determination method is used to calculate the ratio of the peak amplitude decay time constant to the standard decay time constant. When the ratio is less than a preset threshold, it indicates that the decay rate of the transient pulse is significantly faster than the typical decay rate of a partial discharge pulse, consistent with the rapid decay characteristics of a switching operation pulse. This transient pulse is then classified as a switching operation interference pulse and removed from the current waveform data after carrier interference suppression. When the ratio is not less than the preset threshold, it indicates that the decay characteristics of the transient pulse are consistent with the typical decay characteristics of a partial discharge pulse. This transient pulse is then classified as a partial discharge candidate pulse and retained. The physical meaning of this threshold is that only when the decay time constant of a transient pulse significantly deviates from the standard value, and the degree of deviation exceeds the normal fluctuation range, is it classified as interference.

[0041] The selection of the threshold value needs to balance the sensitivity of the discrimination and the risk of false rejection: if the value is too low, it is easy to misjudge genuine partial discharge pulses as interference and reject them; if the value is too high, it may miss switching operation pulses with slightly faster decay. For most application scenarios, setting the threshold between 0.6 and 0.8 can achieve a balance between the two. The specific value can be adjusted by those skilled in the art based on the actual separation degree of the decay time constant between the switching operation pulse and the partial discharge pulse in the field. All the retained partial discharge candidate pulses are arranged in chronological order to generate a partial discharge pulse sequence.

[0042] By extracting the peak amplitude decay time constant of transient pulses and comparing it with the standard decay time constant of partial discharge pulses, the essential physical difference in decay characteristics between the two types of pulses is utilized to accurately distinguish between switching operation interference pulses and partial discharge pulses. This eliminates interference events with time-domain characteristics similar to partial discharge pulses from the root, providing a high-purity pulse sequence containing only partial discharge pulses for subsequent phase distribution feature extraction.

[0043] The frequency-domain notch filtering in step S2 and the time-domain attenuation time constant discrimination in step S3 are not simply two independent processing steps, but rather two organically integrated components that work together and progress in a multi-level interference suppression chain in this embodiment. Step S2's notch filtering suppresses periodic narrowband carrier interference in the frequency domain, targeting continuous wave interference with concentrated energy, a defined center frequency, and narrow bandwidth. Step S3's attenuation time constant discrimination targets transient pulse interference in the time domain, targeting transient waveforms that appear abruptly in the time domain and have specific attenuation characteristics. The physical properties of these two types of interference are fundamentally different: carrier interference is a narrowband signal in the frequency domain and a continuous signal in the time domain, while switching operation interference is a transient signal in the time domain and a wideband pulse in the frequency domain. A single-domain processing method cannot effectively address these two interference sources with such drastically different properties simultaneously. Through the coordinated action of steps S2 and S3, interference suppression in both the frequency and time domains is achieved: periodic carrier interference is first attenuated in the frequency domain by a notch filter, and its residual energy is significantly reduced; subsequently, the transient pulses of switching operations that still exist in the time domain are further identified and eliminated due to their physical difference from the partial discharge pulses in terms of attenuation time constant.

[0044] The combined effect of the frequency and time domain dual interference suppression architecture is significantly better than either notch filtering or attenuation time constant discrimination alone. While notch filtering alone can suppress carrier interference, it is ineffective against switching pulses with time-domain characteristics similar to partial discharge pulses. The filtered waveform still retains residual switching pulses, which are subsequently included in the phase distribution feature vector, leading to distortion in insulation condition assessment. Similarly, while attenuation time constant discrimination can distinguish transient pulse types, the presence of carrier interference raises the amplitude threshold for pulse detection, causing low-amplitude partial discharge pulses to be masked by the carrier interference's noise floor and thus undetectable. Only by implementing both methods in a coordinated manner, with step S2 preceding step S3, can we achieve high-fidelity extraction of partial discharge pulse sequences by simultaneously using frequency domain filtering to suppress carrier interference, reduce the detection floor, and ensure the detectability of weak discharge pulses, and using time domain discrimination to eliminate switching pulses and ensure the purity of the pulse sequence.

[0045] If notch filtering is used alone (only step S2 without step S3), although carrier interference is effectively suppressed, the residual transient pulses from switching operations in the current waveform data after carrier interference suppression are similar in shape to partial discharge pulses in the time domain. Their amplitude may exceed the pulse detection threshold, thus being misjudged as partial discharge pulses and included in subsequent phase distribution statistics. At this time, a large number of pulse events from non-discharge sources are mixed into each phase interval of the phase distribution feature vector, causing the phase distribution spectrum to be distorted. The insulation state deviation coefficient calculated accordingly will deviate from the true value, which may trigger false alarms or missed alarms.

[0046] Conversely, if the attenuation time constant is used alone for discrimination (only step S3 without step S2), the periodic narrowband carrier signal will appear as continuous high-amplitude base noise in the time domain due to the lack of frequency domain suppression of carrier interference. This forces the amplitude threshold of pulse detection to be raised above the carrier interference level; otherwise, a large number of false triggers will overwhelm the effective detection. After the threshold is raised, partial discharge pulses with lower amplitudes will be missed because they are below the threshold, resulting in a large number of missing low-amplitude discharge events in the partial discharge pulse sequence. The sample size for phase distribution statistics is severely insufficient, which also makes the calculation results of the insulation state deviation coefficient unrepresentative. Therefore, this embodiment adopts a technical solution that implements steps S2 and S3 sequentially. First, notch filtering is used to reduce the base level of carrier interference in the frequency domain to ensure that weak discharge pulses can be reliably detected. Then, attenuation time constant discrimination is used to accurately eliminate switching operation pulses in the time domain. The two work together and are indispensable to achieve high-fidelity extraction of the partial discharge pulse sequence and accurate calculation of the insulation state deviation coefficient.

[0047] S4. Obtain the power frequency phase position of the pulse in the partial discharge pulse sequence, count the distribution number in each phase interval to construct the phase distribution feature vector, and compare the phase distribution feature vector with the standard feature vector to generate the insulation state deviation coefficient. For a pure partial discharge pulse sequence, there is a definite correspondence between the power frequency phase position distribution pattern of each pulse and the type and severity of the insulation defect. Different types of insulation defects exhibit different characteristic patterns in the discharge phase intervals of the positive and negative half-cycles of the power frequency voltage. Point discharges typically occur in the phase interval with higher voltage amplitudes, internal discharges often occur near the voltage zero crossing, while surface discharges may exhibit an asymmetrical distribution in both the positive and negative half-cycles. This step quantifies and statistically analyzes the power frequency phase position of each pulse in the partial discharge pulse sequence, converting the pulse phase distribution information into a vectorized representation, and then using distance metrics to achieve a quantitative assessment of the insulation state.

[0048] The peak time of each pulse in the partial discharge pulse sequence is extracted, and the relative position of this peak time within the power frequency cycle of the corresponding current analysis window data is taken as the power frequency phase position of the pulse. This relative position is calculated as follows: taking the start time of the power frequency cycle corresponding to the current analysis window data as zero, the time offset of the pulse peak time relative to this start time is calculated. This time offset is divided by the total duration of the power frequency cycle, and the resulting dimensionless ratio is between 0 and 1. This ratio is multiplied by 360 degrees to obtain the corresponding power frequency phase angle value. For a 50Hz power system, the total duration of the power frequency cycle is 20 milliseconds, and the value of this time offset ranges from 0 milliseconds to 20 milliseconds. The calculated power frequency phase position is then associated with the pulse to generate a partial discharge pulse sequence with phase marking.

[0049] The power frequency cycle is divided into multiple phase intervals with equal phase spacing. Each pulse in the partial discharge pulse sequence marked with phase is assigned to its corresponding phase interval according to its power frequency phase position. The number of pulses assigned to each phase interval is counted to generate the pulse distribution count for each phase interval. The number of phase intervals is dynamically determined based on the total number of pulses during the acquisition period, ensuring that the statistical fluctuation of the pulse distribution count within each interval does not exceed a preset proportion. The setting of the total number of phase intervals needs to balance phase resolution and statistical stability: the more phase intervals, the narrower the phase window corresponding to each interval, and the stronger the ability to resolve details of the discharge phase distribution; however, the number of pulses falling into each interval will decrease accordingly, increasing the impact of statistical fluctuations. Conversely, the fewer phase intervals, the more sufficient the statistical sample size within each interval, and the higher the stability of the distribution pattern; however, it is less sensitive to subtle changes in phase distribution.

[0050] For applications where hundreds to thousands of pulses can be obtained within a typical acquisition period, dividing the power frequency cycle into 36 phase intervals ensures that each interval receives a sufficient number of pulses for statistical analysis, while maintaining a 10-degree phase resolution to distinguish the phase distribution characteristics of different defect types. When the number of acquired pulses is large, the number of phase intervals can be increased to obtain more refined phase distribution information; when the number of pulses is small, the number of phase intervals should be reduced to ensure the reliability of the statistics for each interval. The phase interval division method for the current monitoring period is consistent with the phase interval division method used when constructing the standard feature vector, ensuring that the phase distribution feature vector has the same dimension as the standard feature vector.

[0051] Using the number of pulses distributed in each phase interval as a component, a vector is constructed according to the arrangement order of the phase intervals. This vector is then used as the phase distribution feature vector. The dimension of this phase distribution feature vector is equal to the total number of phase intervals, and the dimensionality of the dimensionality vector is... The component corresponds to the first The number of pulse distributions within each phase interval.

[0052] Multiple historical phase distribution feature vectors of the switchgear under normal insulation conditions are obtained. The arithmetic mean of the components within the same phase interval in these historical phase distribution feature vectors is taken as the standard component for that phase interval. The standard feature vector is constructed from the standard components of all phase intervals. This standard feature vector reflects the baseline distribution pattern of partial discharge pulses on the power frequency phase of the switchgear under normal insulation conditions. The historical phase distribution feature vectors are obtained during the initial operation of the switchgear and within a period after a withstand voltage test confirms good insulation conditions. Multiple partial discharge pulse sequences are collected, and phase distribution feature vectors are constructed using the aforementioned method. These constructed feature vectors are then stored in a database as historical data.

[0053] When there are fewer than 10 sets of historical data, a common standard feature vector of the same model and insulation type of switchgear can be used as a temporary initial benchmark. Once more than 10 sets of historical data have been accumulated for this equipment, the standard feature vector constructed based on this equipment's data can be replaced. The number of historical feature vector sets directly affects the statistical representativeness of the standard feature vector: if the number of sets is too small, a few abnormal data collections will significantly distort the average value, causing the standard feature vector to deviate from its true normal distribution; if the number of sets is large enough, random fluctuations are effectively suppressed through averaging, and the standard feature vector tends to stabilize. Therefore, when data accumulation conditions permit, the number of historical data sets should be increased as much as possible, generally preferably more than 10 sets. The specific number of sets can be determined by those skilled in the art based on the stability of the switchgear operation and the data collection frequency.

[0054] The construction method of this standard feature vector allows each monitored switchgear to independently generate its own standard feature vector based on its own normal operation history data, rather than using a uniform and fixed standard reference. Different switchgears exhibit objective individual differences in the reference distribution pattern of partial discharge pulses at the power frequency phase due to variations in insulation structure (e.g., the different dielectric constants of air insulation and solid insulation lead to different propagation paths and attenuation characteristics), operating years (differences in insulation aging cause changes in the phase distribution pattern of the discharge), and load characteristics (differences in temperature rise caused by load current affect the dielectric properties of the insulation material). This embodiment, through an adaptive parameter construction method based on the equipment's own historical data, enables the standard feature vector to accurately match the individual characteristics of each switchgear, avoiding the systemic bias introduced by individual equipment differences when using a uniform fixed value as a universal reference benchmark.

[0055] For example, for two switchgear with the same insulation structure but different years of operation, the older switchgear may have slight insulation degradation inside. Even under "normal" conditions, its discharge phase distribution may differ from that of a brand new switchgear. If a uniform standard feature vector is used, the older switchgear may be misjudged as abnormal. However, in this embodiment, standard vectors are independently constructed based on their respective historical data, which can accurately identify the baseline distribution of their respective normal states. This is an individual adaptability advantage that cannot be achieved by fixed parameter schemes.

[0056] The insulation state deviation coefficient is obtained by comparing the distance between the normalized phase distribution eigenvector and the standard eigenvector using the following formula: ; Where D is the insulation state deviation coefficient, and m is the total number of phase intervals. v represents the total number of partial discharge pulses during the current monitoring period. j This refers to the component in the phase distribution feature vector corresponding to position j in the phase interval, i.e., the number of pulses distributed within the j-th phase interval. This represents the proportion of pulses in the j-th phase interval to the total number of pulses in the current state. u is the sum of the historical average pulse counts for each phase interval under normal conditions. j It is the component in the standard feature vector corresponding to the position j of the phase interval, that is, the average number of pulses distributed under the historical normal state in the j-th phase interval; This represents the proportion of pulses in the j-th phase interval under normal conditions to the total number of pulses. This normalization process ensures that each component's value is between 0 and 1, and that the sum is 1. This eliminates the influence of differences in the total number of pulses at different acquisition times on the deviation coefficient calculation, making the deviation coefficients calculated at different time points comparable. The insulation state deviation coefficient is used as a criterion for insulation state monitoring and is used in subsequent steps to determine the insulation state. In this embodiment, distance comparison includes normalized Euclidean distance or weighted normalized Euclidean distance.

[0057] In traditional partial discharge detection, the assessment of insulation status relies on maintenance personnel visually interpreting phase distribution maps. The quality of insulation is qualitatively judged by observing the clustering characteristics and changes in the distribution pattern of phase intervals within the map. This method, based on manual map interpretation, not only depends on the professional experience of maintenance personnel but also struggles to provide an objective and continuous quantitative description of the degree of insulation degradation. This embodiment, through the aforementioned Euclidean distance calculation formula, transforms the differences in phase distribution maps, which originally required manual visual interpretation, into a calculable and comparable numerical index, achieving a substantial technological advancement from qualitative map comparison to quantitative numerical judgment. This insulation status deviation coefficient accurately characterizes the degree of deviation of the current phase distribution characteristics from the normal baseline distribution in a continuous numerical manner. This ensures that the insulation status assessment results are no longer influenced by the observer's subjective judgment but have a unified mathematical definition and a repeatable calculation process, laying a numerical foundation for subsequent statistical setting of deviation thresholds and automated alarm determination.

[0058] The formula means that for each phase interval, the difference between the pulse proportion of that phase interval in the current state and the pulse proportion of that phase interval in the normal state is calculated. This difference is squared, and the square root of the sum of the squared differences of all phase interval positions is taken. The result is the insulation state deviation coefficient. This coefficient quantifies the total deviation of the current phase distribution characteristics from the normal reference distribution. The larger the deviation coefficient, the greater the difference between the current pulse phase distribution and the normal state.

[0059] Due to normalization and Both are probability distributions, satisfying and Therefore, the difference between each component The absolute value of D does not exceed 1. According to the properties of Euclidean distance, D = 0 when the two probability distributions are perfectly identical; D reaches its maximum value when the two distributions are completely separated (i.e., one distribution concentrates all probabilities in one phase interval, and the other concentrates all probabilities in a different phase interval). Therefore, the theoretical range of the insulation condition deviation coefficient D is: This finite range of values ​​gives the deviation coefficient a clear physical upper bound, providing a bounded numerical space for setting subsequent statistical thresholds and avoiding the potential numerical divergence problem that may occur with dimensionless indicators.

[0060] As an alternative, a weighted Euclidean distance method can be used for distance comparison. The principle of this method is as follows: a weighting coefficient is multiplied before the squared term. This weighting coefficient is inversely proportional to the variance of the corresponding phase interval position under normal conditions. The larger the variance of the phase interval position, the greater the fluctuation in the number of pulses under normal conditions, and the lower the reliability of the phase distribution information reflected by that position. Therefore, a smaller weight is assigned to reduce the interference of the phase interval with large fluctuations under normal conditions on the calculation of the deviation coefficient, thereby improving the sensitivity of the deviation coefficient to detecting abnormal changes in insulation condition.

[0061] The insulation condition deviation coefficient is used as the insulation condition monitoring criterion for determining the insulation condition in subsequent steps.

[0062] By classifying each pulse into the corresponding phase interval according to its power frequency phase position and counting the number of pulses distributed, a phase distribution feature vector with clear physical meaning is constructed. Then, the deviation of the current insulation state from the normal reference state is quantitatively characterized by Euclidean distance measurement, providing a calculable numerical criterion for subsequent insulation state determination.

[0063] S5. Compare the insulation condition deviation coefficient with the deviation threshold. When the insulation condition deviation coefficient exceeds the deviation threshold, generate an insulation abnormality alarm signal.

[0064] After obtaining the insulation state deviation coefficient, which quantifies the degree to which the current insulation state deviates from the normal baseline, it is necessary to establish a judgment criterion to distinguish between acceptable normal fluctuations and abnormal deviations that require attention. Since factors such as load changes in the switchgear, fluctuations in ambient temperature and humidity, and small deviations in system voltage can all cause normal fluctuations in the phase distribution characteristic vector, the deviation coefficient itself has a certain statistical fluctuation range under normal conditions. Therefore, not every non-zero value of the deviation coefficient can be considered abnormal; instead, a deviation threshold needs to be set as the judgment boundary.

[0065] Obtain the historical insulation state deviation coefficients of the switchgear under normal insulation conditions. The method for obtaining these historical insulation state deviation coefficients is as follows: during the initial operation phase of the switchgear, after the insulation state has been confirmed to be healthy by withstand voltage testing, calculate the insulation state deviation coefficients at multiple time points following the aforementioned steps. Store these calculated deviation coefficients as historical insulation state deviation coefficients in the database. The time span for collecting historical data must ensure that it covers various operating condition changes that the switchgear may experience during normal operation, including load fluctuation cycles and daily ambient temperature variation cycles, so that the statistical distribution of the historical deviation coefficients can fully reflect the fluctuation range under normal conditions.

[0066] Typically, a continuous 7-day data collection period is sufficient to cover the load differences between weekdays and rest days, as well as the diurnal temperature variation cycle, providing good statistical representativeness. For sites with significant load variations or extreme environmental conditions, the data collection period can be extended to 30 days to ensure coverage of seasonal climate fluctuations or changes in operating conditions caused by periodic maintenance. Conversely, for indoor switchgear with stable operating conditions and controllable environmental conditions, a shorter data collection period is sufficient to obtain adequate statistical samples, and a shorter data collection period can also be used when rapid deployment of the monitoring system is required. Regardless of the data collection period selected, it is essential to ensure that the insulation status of the switchgear remains normal throughout the data collection period.

[0067] The deviation threshold is set based on the upper quantile of the statistical distribution of historical insulation state deviation coefficients. In statistical process control, the upper quantile threshold method does not rely on the assumption that the data follows a specific distribution pattern, but directly selects a value corresponding to a specific cumulative probability from the empirical distribution as the control limit. After arranging the historical deviation coefficients in ascending order of value, the 95th percentile means that only 5% of the observations will exceed this threshold under normal conditions. Using this as a judgment boundary can control the false alarm rate within a statistically acceptable range. When higher alarm sensitivity is required, a lower upper quantile, such as the 90th percentile, can be selected as the threshold, in which case the false alarm probability under normal conditions increases to 10%. When the tolerance for false alarms is low and higher determinism of alarms is required, the 98th percentile can be selected.

[0068] As an alternative, the deviation threshold can also be determined as follows: calculate the arithmetic mean and standard deviation of the historical insulation state deviation coefficients, and add a certain multiple of the standard deviation to the mean as the threshold. A threshold of three times the standard deviation corresponds to a one-sided tail probability of 0.135% under a normal distribution, and a threshold of two times the standard deviation corresponds to a one-sided tail probability of approximately 2.28%. These two methods correspond to different false alarm control levels. The above-mentioned multiple threshold setting methods can be flexibly selected to adapt to the needs of different monitoring scenarios. When the historical data follows a normal distribution, the combination of mean and standard deviation has statistical theoretical support. When the historical data distribution is unknown or asymmetrical, the upper quantile method is not limited by the distribution shape. Since the statistical distribution of the insulation condition deviation coefficient under normal conditions may vary depending on the type of switchgear, operating conditions, and data acquisition period, and may not strictly follow a normal distribution, in practical engineering applications, it is preferable to use the upper quantile method to set the deviation threshold to avoid the assumption of dependence on the distribution pattern. Only when statistical tests (such as the Shapiro-Wilk test) confirm that the historical deviation coefficient follows an approximately normal distribution can the threshold be set using a combination of the mean and standard deviation.

[0069] The deviation threshold is also set adaptively by the equipment. The deviation threshold for each switchgear is determined independently based on the statistical distribution of its historical deviation coefficient under normal operating conditions, rather than using a fixed threshold uniform to all equipment. Due to the different load fluctuation amplitudes (which depend on the type and variation of the load), environmental condition variations (outdoor switchgear is greatly affected by the daily temperature range, while indoor switchgear has a relatively stable environment), and differences in sensor coupling paths (sensors with different installation locations and coupling methods have different signal transmission paths and noise coupling degrees), the statistical distribution range of their deviation coefficients under normal conditions varies.

[0070] By adopting an adaptive threshold, the alarm boundary of each switchgear is matched with its own normal fluctuation range, avoiding the insufficient adaptability problem of some devices being too sensitive (frequent false alarms when normal fluctuations are large) and some devices being too insensitive (insensitive to anomalies when normal fluctuations are small) under the fixed threshold scheme. For example, for switchgear with drastic load changes, the deviation coefficient of its normal state fluctuates within a wide range. If the fixed threshold is set too low, false alarms will occur frequently; if it is set too high, it will lose sensitivity on lightly loaded switchgear. This embodiment independently determines the threshold based on its own historical data, so that the judgment boundary accurately matches the actual fluctuation characteristics of each device.

[0071] The calculated insulation condition deviation coefficient is compared with the deviation threshold. When the insulation condition deviation coefficient is less than or equal to the deviation threshold, the insulation condition of the switchgear is determined to be within the normal fluctuation range, no alarm is triggered, and the process returns to S1 to obtain the next set of data for continuous monitoring. When the insulation condition deviation coefficient exceeds the deviation threshold, it indicates that the phase distribution pattern of the current partial discharge pulse has deviated significantly from the reference pattern under normal conditions, exceeding the normal fluctuation range. This significant deviation usually corresponds to the development of insulation defects or the deepening of deterioration. Therefore, the insulation condition of the switchgear is determined to be abnormal, and an insulation abnormality alarm signal is generated.

[0072] After generating an insulation anomaly alarm signal, the signal is output to a monitoring terminal. This terminal can be an indicator light, buzzer, LCD screen, or the host computer interface of a remote monitoring system. The insulation anomaly alarm signal can include the specific value of the insulation condition deviation coefficient and the percentage exceeding the threshold, allowing maintenance personnel to obtain a quantitative understanding of the degree of anomaly. It can also carry the current time information and the corresponding switchgear number information, facilitating rapid location of the abnormal equipment and the time of occurrence in multi-switchgear centralized monitoring scenarios. In some implementations, the insulation anomaly alarm signal can be simultaneously output to multiple monitoring terminals, including the local monitoring backend of the substation and the monitoring system of the remote maintenance center, achieving multi-point synchronous alarm and timely notification of personnel at different levels.

[0073] To verify the effectiveness of the method in this embodiment, a simulated test was conducted on a laboratory 10kV air-insulated switchgear full-scale test platform. The test platform included a switchgear with a rated voltage of 10kV and a rated current of 630A, containing a three-phase busbar and a vacuum circuit breaker. Copper wires with a diameter of approximately 2mm and a length of approximately 5mm were pasted onto the surface of the busbar insulation support in the cable compartment of the switchgear to simulate surface discharge defects. A high-frequency current sensor (bandwidth 100kHz~50MHz) was installed on the grounding wire, and the voltage signal was taken from the secondary side of the voltage transformer, with a synchronous sampling rate set to 100MHz. Simultaneously, continuous carrier interference with a frequency of 10.5MHz and a bandwidth of approximately 200kHz was injected through a carrier signal generator, and operating pulses were generated by the opening and closing of the circuit breaker to simulate a multi-source interference environment in the field.

[0074] Data acquisition lasted 30 minutes, obtaining current waveforms for 1500 complete power frequency cycles. Frequency domain analysis of data from typical cycles revealed a narrowband spike at 10.5MHz with an amplitude 35 times the background average and a bandwidth of approximately 180kHz, identified as carrier interference. Based on this, a notch filter (40dB notch depth) was constructed to filter all data. After filtering, the spectral spike at 10.5MHz was suppressed to near background levels, the carrier interference amplitude was reduced by approximately 99%, while the broadband components (2MHz~30MHz) of the partial discharge pulses were largely preserved, and the previously masked weak pulses became clear.

[0075] Subsequently, a pulse search was performed on the filtered waveform (with the pulse amplitude threshold set at three times the root mean square value of the noise), detecting a total of 237 transient pulses. The peak amplitude decay time constant was extracted for each pulse. Taking a typical pulse as an example, its peak amplitude was 2.7mV, the peak time was 0.32μs, and the decay time to the end time was 1.85μs. An exponential fitting yielded a decay time constant of 0.52μs. The standard decay time constant for the partial discharge pulse in this cabinet was previously calibrated to 1.8μs. The ratio of the two, 0.29, is far below the preset threshold of 0.7, therefore this pulse was judged as switching operation interference and eliminated. After all screenings, a total of 68 interfering pulses were eliminated, retaining 169 partial discharge candidate pulses, generating a clean pulse sequence.

[0076] The retained pulse sequence was used for power frequency phase positioning. The entire power frequency cycle was divided into 36 equal phase intervals (each interval being 10°), and the number of pulses in each interval was counted. The count of each interval was normalized using the total number of pulses (169) to obtain the current phase distribution. Simultaneously, under normal conditions before the defect was set, 10 sets of historical data were collected for 7 consecutive days, and a normal baseline phase distribution was constructed using the same method. The current distribution was compared with the baseline distribution using normalized Euclidean distance, and the deviation coefficient was calculated to be 0.387.

[0077] Then, 30 days of historical operating data under normal conditions were collected for the cabinet. The deviation coefficient was calculated daily, resulting in 30 values ​​ranging from 0.051 to 0.179, with an average of 0.112. An alarm threshold of 0.167 was set based on the 95th percentile statistical method. The current deviation coefficient of 0.387 exceeds the threshold by more than 130%, generating an insulation abnormality alarm signal. Maintenance personnel subsequently shut down the power and inspected the cabinet, finding obvious discharge marks on the surface of the insulation support components, matching the preset defect location, confirming the alarm was correct.

[0078] To verify the contribution of each processing step, a comparative experiment was conducted: When only notch filtering was used without attenuation time constant discrimination, 68 residual operational interference pulses resulted in a deviation coefficient of 0.312 (although exceeding the threshold, the phase distribution characteristics did not match the actual discharge); when only attenuation time constant discrimination was used without filtering, 43 (approximately 25%) weak discharge pulses were masked and not detected due to carrier interference raising the detection substrate, resulting in a deviation coefficient of 0.215 (although exceeding the threshold, it was too low, posing a risk of missed alarms); when neither processing was used, the deviation coefficient was only 0.142, below the threshold, causing missed alarms. This embodiment, however, combined both processing methods, detecting all valid discharge pulses and eliminating all interference, with a deviation coefficient of 0.387 accurately reflecting insulation degradation and achieving accurate alarms.

[0079] All data presented in this embodiment originates from actual measurements taken on a laboratory-scale switchgear test platform and subsequent computer processing of the measurement data. To verify the effectiveness of the method in a real switchgear operating environment, a complete 10kV air-insulated switchgear test system was built in the laboratory. This voltage level is the most widely used medium-voltage level in distribution networks, ensuring that the verification conclusions are representative of mainstream application scenarios. On the test platform, a 2mm diameter, 5mm long copper wire was attached to the surface of the insulating support of the busbar inside the cabinet. This is a classic model used to simulate surface discharge defects. Using this classic model ensures the comparability and reproducibility of the verification results. Furthermore, the defect is visually visible during power outage inspections, facilitating the final verification of the alarm's accuracy.

[0080] To simulate the complex electromagnetic interference environment on-site, a continuous carrier signal with a frequency of 10.5 MHz was injected through a signal generator to simulate power line carrier communication interference. Transient pulse interference was generated by actually operating circuit breakers to open and close them. These two types of interference are the most common and difficult-to-handle multi-source interference types in switchgear environments. Subsequently, current and voltage signals were synchronously acquired at a sampling rate of 100 MHz using a high-frequency current sensor installed on the grounding wire and a voltage transformer at the feeder terminal. This sampling rate was determined according to the Nyquist sampling theorem. Since the energy of partial discharge pulses is mainly distributed in the frequency band below 50 MHz, a sampling rate of 100 MHz can capture the pulse waveform details completely without placing excessive data storage and computational pressure on the hardware system. Therefore, the acquired raw data, such as the 10.5 MHz narrowband spike with an amplitude exceeding 35 times the background average and the transient pulse waveform with a peak amplitude of 2.7 mV, are all actual voltage values ​​measured by the sensors and converted from analog to digital.

[0081] The characteristic parameters used in the experiment, such as the peak amplitude decay time constant of 0.52 microseconds and the deviation coefficient of 0.387, are results obtained by the computer through mathematical fitting and statistical calculation based on the collected raw waveform data. Taking the decay time constant as an example, when the computer detects a transient pulse, it extracts the waveform segment from the peak moment to the end moment of the pulse, and uses the least squares exponential fitting algorithm to fit the decay curve. The fitted exponential decay time constant is the decay time constant of the pulse. In this embodiment, 0.52 microseconds is the specific value obtained after fitting the falling edge of a typical pulse. The standard decay time constant of 1.8 microseconds for partial discharge pulses is the calibration value obtained by taking the average value after multiple partial discharge measurements of this type of switchgear under clean conditions. The ratio of the two, 0.29, is much lower than the preset threshold of 0.7, so the pulse is accurately judged as interference and eliminated. The deviation coefficient of 0.387 is the quantitative result obtained by calculating the normalized Euclidean distance between the current phase distribution and the normal reference, reflecting the total deviation of the current phase distribution from the reference. These characteristic parameters all have clear physical meanings and traceable calculation processes, and are verifiable and repeatable.

[0082] In establishing the normal baseline and alarm threshold, the data used also originates from real historical data collection and statistical summarization. The standard feature vector is obtained by taking the arithmetic mean of each component after collecting 10 sets of historical phase distribution data for 7 consecutive days, without setting any insulation defects and confirming that the insulation condition is completely healthy through withstand voltage tests. The 7-day time span covers different load conditions on weekdays and weekends, and 10 sets of data are sufficient to effectively eliminate the influence of random fluctuations by averaging, ensuring the stability and reliability of the baseline. The deviation threshold is set based on 30 deviation coefficient values ​​calculated from 30 consecutive days of historical data under normal conditions. These values ​​range from 0.051 to 0.179, with an average value of 0.112. The threshold is determined to be 0.167 using the 95th percentile method in statistics (i.e., only 5% of the observations under normal conditions will exceed this limit). The 30-day historical data can cover seasonal climate fluctuations and a more complete load change cycle, so that the threshold fully includes various accidental fluctuations under normal conditions, avoiding frequent false alarms caused by instantaneous load changes or short-term environmental fluctuations.

[0083] The selection of experimental parameters was based on practical engineering specifications and scientific principles. The choice of a 10 kV voltage level stemmed from its dominant position in the distribution network; the 100 MHz sampling rate followed the Nyquist sampling theorem; the division into 36 phase intervals (each 10 degrees) was based on the physical fact that the phase distribution characteristics of partial discharge under different defect types mostly differed within the range of 10 to 20 degrees; the attenuation time constant ratio threshold of 0.7 was the optimal balance value obtained through extensive experimental statistics, which can both prevent the accidental deletion of genuine discharge pulses and effectively eliminate interference from switching operations; the 10.5 MHz carrier frequency is the typical operating frequency band for power line carrier communication. These well-founded parameter selections ensured that the verification experiment was both close to engineering practice and scientifically rigorous. Finally, when the current deviation coefficient calculated by the computer is 0.387, which exceeds the statistical threshold of 0.167, an insulation abnormality alarm signal is automatically generated. After the maintenance personnel shut down the power and inspected the equipment, they found obvious discharge traces on the surface of the insulation support, which completely matched the preset defect location. This fully verified the effectiveness, accuracy and reliability of the method in this embodiment under real multi-source interference environment.

[0084] By comparing the insulation condition deviation coefficient with the deviation threshold set based on the statistical distribution of historical data, the problem of insufficient field adaptability caused by using a fixed threshold is avoided. At the same time, the automatic judgment and alarm output of insulation condition anomalies are realized, completing the complete monitoring closed loop from signal acquisition to insulation condition judgment, and providing a reliable decision basis for online monitoring of switchgear insulation condition.

[0085] Example 2: Figure 2As shown, the switchgear insulation monitoring system based on partial discharge feature recognition can be installed in electronic equipment. It includes a signal windowing module, a carrier filtering module, a pulse discrimination module, a feature measurement module, and an anomaly alarm module connected in sequence. A module, also called a unit, refers to a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function; these segments are stored in the memory of the electronic device. The functions of each module are as follows: The signal window acquisition module is used to acquire the voltage signal of the switch cabinet feeder terminal and the high-frequency current signal of the grounding wire. It uses the power frequency zero crossing point of the voltage signal to extract the power frequency period of the synchronously sampled high-frequency current signal and generate current analysis window data. The carrier filtering module is used to extract the periodic narrowband spectral components in the current analysis window data. Based on the center frequency and bandwidth of the periodic narrowband spectral components, a notch filter is constructed to filter the current analysis window data and generate current waveform data after carrier interference suppression. The pulse discrimination module is used to extract the peak amplitude decay time constant of transient pulses in the current waveform data after carrier interference suppression. Based on the comparison between the peak amplitude decay time constant and the standard decay time constant of partial discharge pulses, the switching operation interference pulses are eliminated, and a partial discharge pulse sequence is generated. The feature measurement module is used to obtain the power frequency phase position of the pulse in the partial discharge pulse sequence, count the distribution quantity in each phase interval to construct the phase distribution feature vector, and compare the phase distribution feature vector with the standard feature vector to generate the insulation state deviation coefficient. The abnormal alarm module is used to compare the insulation state deviation coefficient with the deviation threshold. When the insulation state deviation coefficient exceeds the deviation threshold, an insulation abnormal alarm signal is generated.

[0086] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can make equivalent substitutions or modifications based on the technical solution and concept of the present invention within the scope of the technology disclosed in the present invention, and such modifications should also be considered to fall within the scope of protection of the present invention.

Claims

1. A switchgear insulation monitoring method based on partial discharge feature identification, characterized in that the steps are as follows: include: S1. Obtain the voltage signal of the switchgear feeder terminal and the high-frequency current signal of the grounding wire. Use the power frequency zero-crossing point of the voltage signal to extract the power frequency period of the synchronously sampled high-frequency current signal and generate current analysis window data. S2. Extract the periodic narrowband spectral components from the current analysis window data. Construct a notch filter based on the center frequency and bandwidth of the periodic narrowband spectral components to filter the current analysis window data and generate current waveform data after carrier interference suppression. S3. Extract the peak amplitude decay time constant of transient pulses from the current waveform data after carrier interference suppression. Based on the comparison between the peak amplitude decay time constant and the standard decay time constant of partial discharge pulses, remove the switching operation interference pulses and generate a partial discharge pulse sequence. S4. Obtain the power frequency phase position of the pulse in the partial discharge pulse sequence, count the distribution number in each phase interval to construct the phase distribution feature vector, and compare the phase distribution feature vector with the standard feature vector to generate the insulation state deviation coefficient. S5. Compare the insulation condition deviation coefficient with the deviation threshold. When the insulation condition deviation coefficient exceeds the deviation threshold, generate an insulation abnormality alarm signal.

2. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 1, characterized in that, In S1, current analysis window data is generated, including: The waveform amplitude of the voltage signal after synchronous sampling is detected, and the moment when the amplitude of the voltage signal waveform is zero is identified as the zero-crossing point of the power frequency. The time interval between two adjacent zero-crossing points of the power frequency in the same direction is taken as a complete power frequency cycle intercept window; According to the capture window, the current waveform segment corresponding to the power frequency cycle is captured from the high-frequency current signal after synchronous sampling, and the current waveform segment is used as the current analysis window data.

3. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 1, characterized in that, In S2, the periodic narrowband spectral components are extracted from the current analysis window data, including: A frequency domain transformation is performed on the current analysis window data to map the current analysis window data from the time domain to the frequency domain, generating current frequency domain spectrum data; Amplitude peak search is performed on the current frequency domain spectrum data to identify frequency domain components whose amplitude exceeds the preset spectrum amplitude threshold and whose bandwidth is less than the preset bandwidth threshold. The identified frequency domain components are then used as periodic narrowband spectrum components. Record the center frequency and bandwidth of the periodic narrowband spectral components, and use the center frequency and bandwidth as carrier interference characteristic parameters.

4. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 3, characterized in that, In S2, the current waveform data after carrier interference suppression is generated, including: Obtain the center frequency and bandwidth from the carrier interference characteristic parameters, use the center frequency as the notch filter center frequency and the bandwidth as the notch filter bandwidth, and construct a notch filter according to the notch filter center frequency and notch filter bandwidth. The notch filter is applied to the current analysis window data, so that the frequency domain components in the current analysis window data located within the notch bandwidth are attenuated to below the preset attenuation threshold, while the frequency domain components in the current analysis window data located outside the notch bandwidth are preserved. The current analysis window data after notch filtering is used as the current waveform data after carrier interference suppression.

5. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 1, characterized in that, In S3, the peak amplitude attenuation time constant of transient pulses in the current waveform data after carrier interference suppression is extracted, including: The current waveform data after carrier interference suppression is pulse searched in chronological order to identify time-domain waveform segments whose amplitude exceeds the preset pulse amplitude threshold. The identified time-domain waveform segments are taken as transient pulses, and the start time, peak time and end time of the transient pulses are recorded. Extract the signal amplitude variation curve of the transient pulse from the peak time to the end time, and use the signal amplitude variation curve as the attenuation curve of the transient pulse; The decay curve is subjected to exponential fitting, and the resulting exponential decay time constant is used as the peak amplitude decay time constant of the transient pulse.

6. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 1, characterized in that, In S3, a partial discharge pulse sequence is generated, including: Obtain the peak amplitude decay time constant of the transient pulse, and compare the peak amplitude decay time constant with the pre-calibrated standard decay time constant of the partial discharge pulse; When the ratio of the peak amplitude decay time constant to the standard decay time constant is less than the preset threshold, the transient pulse is identified as a switching operation interference pulse, and the switching operation interference pulse is removed from the current waveform data after carrier interference suppression. When the ratio is not less than the preset threshold, the transient pulse is identified as a partial discharge candidate pulse and the partial discharge candidate pulse is retained. All the retained partial discharge candidate pulses are arranged in chronological order to generate a partial discharge pulse sequence.

7. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 1, characterized in that, In S4, the distribution quantity within each phase interval is counted to construct the phase distribution feature vector, including: The peak point time of the pulse in the partial discharge pulse sequence is extracted. The relative position of the peak point time within the power frequency period of the current analysis window data is taken as the power frequency phase position of the pulse. The power frequency phase position is associated with the pulse to generate a partial discharge pulse sequence with phase label. The power frequency cycle is divided into phase intervals according to equal phase intervals. The pulses in the partial discharge pulse sequence with phase marking are assigned to the corresponding phase intervals according to the power frequency phase position. The number of pulses assigned to each phase interval is counted to generate the pulse distribution number of each phase interval. The number of phase intervals is dynamically determined based on the total number of pulses during the acquisition period, ensuring that the statistical fluctuations in the number of pulses in each interval do not exceed a preset ratio. The phase interval division method for the current monitoring period is consistent with the phase interval division method used when constructing the standard feature vector, so that the phase distribution feature vector has the same dimension as the standard feature vector. Using the number of pulses distributed in each phase interval as components, construct a vector according to the arrangement order of the phase intervals, and use the vector as the phase distribution characteristic vector.

8. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 1, characterized in that, In S4, the phase distribution feature vector is compared with the standard feature vector by distance to generate the insulation state deviation coefficient, including: Obtain multiple historical phase distribution feature vectors of the switchgear under normal insulation conditions, take the arithmetic mean of the components in the same phase interval position in the historical phase distribution feature vector, use the arithmetic mean as the standard component of the corresponding phase interval position, and construct the standard feature vector from the standard components of all phase interval positions. The insulation state deviation coefficient is obtained by comparing the distance between the normalized phase distribution eigenvector and the normalized standard eigenvector using the following formula: ； Where D is the insulation state deviation coefficient, and m is the total number of phase intervals. v represents the total number of partial discharge pulses during the current monitoring period. j This refers to the component in the phase distribution feature vector corresponding to position j in the phase interval, i.e., the number of pulses distributed within the j-th phase interval. u is the sum of the historical average pulse counts for each phase interval under normal conditions. j It is the component in the standard feature vector corresponding to the position j of the phase interval, that is, the average number of pulses distributed under the historical normal state in the j-th phase interval; Distance comparison includes normalized Euclidean distance or weighted normalized Euclidean distance; The insulation condition deviation coefficient is used as the criterion for insulation condition monitoring.

9. The switchgear insulation monitoring method based on partial discharge feature recognition as described in claim 1, characterized in that, In S5, an insulation fault alarm signal is generated, including: Obtain the historical insulation state deviation coefficient of the switchgear under normal insulation conditions, and set the deviation threshold according to the upper quantile of the statistical distribution of the historical insulation state deviation coefficient. The insulation state deviation coefficient is numerically compared with the deviation threshold. When the insulation condition deviation coefficient exceeds the deviation threshold, it is determined that the insulation condition of the switchgear is abnormal, an insulation abnormality alarm signal is generated, and the insulation abnormality alarm signal is output to the monitoring terminal.

10. A switchgear insulation monitoring system based on partial discharge feature recognition, used to implement the method according to any one of claims 1 to 9, characterized in that, include: The signal window acquisition module is used to acquire the voltage signal of the switch cabinet feeder terminal and the high-frequency current signal of the grounding wire. It uses the power frequency zero crossing point of the voltage signal to extract the power frequency period of the synchronously sampled high-frequency current signal and generate current analysis window data. The carrier filtering module is used to extract the periodic narrowband spectral components in the current analysis window data. Based on the center frequency and bandwidth of the periodic narrowband spectral components, a notch filter is constructed to filter the current analysis window data and generate current waveform data after carrier interference suppression. The pulse discrimination module is used to extract the peak amplitude decay time constant of transient pulses in the current waveform data after carrier interference suppression. Based on the comparison between the peak amplitude decay time constant and the standard decay time constant of partial discharge pulses, the switching operation interference pulses are eliminated, and a partial discharge pulse sequence is generated. The feature measurement module is used to obtain the power frequency phase position of the pulse in the partial discharge pulse sequence, count the distribution quantity in each phase interval to construct the phase distribution feature vector, and compare the phase distribution feature vector with the standard feature vector to generate the insulation state deviation coefficient. The abnormal alarm module is used to compare the insulation state deviation coefficient with the deviation threshold. When the insulation state deviation coefficient exceeds the deviation threshold, an insulation abnormal alarm signal is generated.

Images