Cable accessory partial discharge detection method

Abstract

The invention discloses a cable accessory partial discharge detection method, and belongs to the technical field of electrical measurement, and the method comprises the steps: injecting a preset detection signal into a cable accessory insulation system, synchronously collecting an induction signal through a built-in sensing array, and generating a multi-channel mixed induction signal; the method comprises the following steps of: performing coherent demodulation on a multi-channel mixed induction signal by taking a detection signal as a reference, separating and extracting a detection signal modulation component, and constructing an initial modulation component matrix; analyzing multi-dimensional modulation features of the initial modulation component matrix, and generating a spatial modulation feature tensor; extracting a native discharge pulse waveform feature set from the multi-channel mixed induction signal; and fusing the spatial modulation feature tensor and the native discharge pulse waveform feature set to output a partial discharge diagnosis report. Active injection of detection signals and passive monitoring of native pulses are combined, and spatial modulation features and waveform features are fused for conjoint analysis, so that sensitive and reliable detection and accurate positioning of partial discharge of cable accessories can be realized.

Description

Technical Field

[0001] This invention relates to the field of electrical measurement technology, and in particular to a method for detecting partial discharge in cable accessories. Background Technology

[0002] Cable accessories are critical connecting components in power cable lines, and their insulation condition directly affects the operational reliability of the entire cable system. Partial discharge is an important early sign of insulation degradation in cable accessories; therefore, partial discharge detection of cable accessories is an effective technical means to assess their insulation condition and prevent faults.

[0003] Existing partial discharge detection technologies for cable accessories mainly rely on passive monitoring methods such as pulse current method, ultra-high frequency method, or ultrasonic method. These methods typically capture transient pulse current, electromagnetic wave, or acoustic wave signals generated by the partial discharge activity itself through external or internal sensors, and then infer the intensity and type of discharge by analyzing the amplitude, phase, frequency, and other characteristics of these passive signals.

[0004] However, these passive monitoring methods have certain limitations. They are sensitive to external electromagnetic interference and on-site noise, which can easily lead to false alarms or missed detections. For weak partial discharges occurring early within the insulating medium, the resulting signals are often buried in background noise and are difficult to reliably capture and identify. Furthermore, based on the characteristic analysis of a single type of passive signal, it is sometimes difficult to accurately distinguish between discharge types with different mechanisms, and the spatial location capability of the discharge point is usually quite limited. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a method for detecting partial discharge in cable accessories. This method combines actively injected detection signals with passively monitored native pulses, and integrates spatial modulation characteristics and waveform characteristics for joint analysis. This enables sensitive, reliable detection and precise location of partial discharge in cable accessories.

[0006] The above objectives can be achieved through the following approach: A method for detecting partial discharge in cable accessories includes injecting a preset detection signal into the insulation system of the cable accessory and synchronously acquiring the induced signal through a built-in sensor array to generate a multi-channel hybrid induced signal; using the detection signal as a reference, coherently demodulating the multi-channel hybrid induced signal to separate and extract the modulation components of the detection signal and construct an initial modulation component matrix; analyzing the multi-dimensional modulation characteristics of the initial modulation component matrix to generate a spatial modulation feature tensor; extracting a native discharge pulse waveform feature set from the multi-channel hybrid induced signal; fusing the spatial modulation feature tensor and the native discharge pulse waveform feature set for joint analysis and decision-making, and outputting a partial discharge diagnostic report.

[0007] Optionally, the step of injecting a preset detection signal into the cable accessory insulation system and synchronously acquiring the induced signal through a built-in sensor array to generate a multi-channel hybrid induced signal includes: deploying a built-in sensor array inside the insulation layer of the cable accessory or at a preset key interface; superimposing the detection signal onto the operating power frequency voltage between the high-voltage conductor and the grounding layer of the cable accessory through a perturbation signal coupling unit; controlling the frequency, amplitude, and phase of the detection signal so that the amplitude is lower than the operating power frequency voltage and the frequency is within a preset accessory structural characteristic frequency band; synchronously acquiring the voltage or current signal sensed by the built-in sensor array, amplifying and digitizing it to generate a time-aligned multi-channel hybrid induced signal.

[0008] Optionally, the perturbation signal coupling unit includes a high-pass filter and an isolation transformer to ensure that the detection signal can be injected into the high-voltage side while blocking the power frequency current.

[0009] Optionally, the step of using the detection signal as a reference to coherently demodulate the multi-channel hybrid sensing signal, separating and extracting the modulation components of the detection signal, and constructing an initial modulation component matrix includes: generating a local reference signal with the same frequency and phase as the detection signal, as a reference; multiplying each signal in the multi-channel hybrid sensing signal with the local reference signal to obtain a mixed signal; performing low-pass filtering on each mixed signal to filter out high-frequency carrier components and extracting a low-frequency baseband signal containing amplitude and phase change information as the detection signal modulation component of the corresponding channel; and arranging the detection signal modulation components of all channels in channel order to form an initial modulation component matrix.

[0010] Optionally, the step of analyzing the multi-dimensional modulation characteristics of the initial modulation component matrix to generate a spatial modulation feature tensor includes: performing time-domain analysis on the initial modulation component matrix to calculate the amplitude fluctuation variance and phase jump number of the modulation components of the probe signal in each channel, and generating an amplitude-phase perturbation vector; performing frequency-domain analysis on the initial modulation component matrix to calculate the energy proportion of a preset subharmonic component in the modulation components of the probe signal in each channel, and generating a harmonic distortion vector; and combining the spatial position information of the built-in sensor array to interpolate and grid the amplitude-phase perturbation vector and the harmonic distortion vector of each channel at the same time in space to form a three-dimensional data volume that can reflect the spatial distribution of modulation characteristics, thereby obtaining a spatial modulation feature tensor.

[0011] Optionally, the step of extracting the native discharge pulse waveform feature set from the multi-channel mixed sensing signal includes: performing digital filtering on the multi-channel mixed sensing signal to separate signal components with frequency bands higher than the frequency of the detection signal, thereby obtaining a high-pass filtered signal; performing threshold comparison and pulse identification on the high-pass filtered signal to capture transient pulse waveforms exceeding a preset noise baseline; and extracting the peak amplitude, pulse width, rise time, and pulse repetition rate from each captured transient pulse waveform to form a native discharge pulse waveform feature set.

[0012] Optionally, the step of fusing the spatial modulation feature tensor and the original discharge pulse waveform feature set for joint analysis and decision-making, and outputting a partial discharge diagnostic report, includes: establishing a physical association rule base based on the expected disturbance patterns of different types of partial discharge events on the dielectric properties and their correspondence with the original discharge pulse waveform features; matching the spatial distribution pattern of the spatial modulation feature tensor at the current moment in the physical association rule base to preliminarily infer the discharge type and possible region, and generating a preliminary inference result; matching the preliminary inference result in the physical association rule base to obtain the inferred waveform feature set; verifying whether the pulse features of the original discharge pulse waveform feature set match the pulse features of the inferred waveform feature set; if the verification passes, combining the original discharge pulse waveform feature set and the spatial modulation feature tensor to generate a partial discharge diagnostic report containing the discharge type, confidence level, intensity level, and approximate region; if the verification fails, initiating a conflict resolution mechanism.

[0013] Optionally, if the verification fails, the conflict resolution mechanism is activated as follows: if the verification fails, the data acquisition time is increased, the frequency or amplitude of the detection signal is adjusted, and the original discharge pulse waveform feature set is re-extracted; the feature changes of the re-extracted original discharge pulse waveform feature set and the previously extracted original discharge pulse waveform feature set are analyzed to obtain conflict features; the conflict features are compared with a preset historical case library to obtain candidate diagnostic results and corresponding confidence levels.

[0014] Optionally, the method further includes: injecting a detection signal after the cable accessory is installed and in a non-discharge state, and acquiring a background response signal through a built-in sensor array; comparing the modulation component of the detection signal with the background response signal to eliminate the inherent deviation caused by the structural asymmetry of the accessory itself.

[0015] Based on the same inventive concept, this invention also provides a partial discharge detection system for cable accessories. The system includes: an active excitation module for injecting a preset detection signal into the insulation system of the cable accessory and synchronously acquiring the induced signal through a built-in sensor array to generate a multi-channel hybrid induced signal; a signal demodulation module for coherently demodulating the multi-channel hybrid induced signal using the detection signal as a reference, separating and extracting the modulation components of the detection signal, and constructing an initial modulation component matrix; a signal processing module for analyzing the multi-dimensional modulation characteristics of the initial modulation component matrix and generating a spatial modulation feature tensor; a signal extraction module for extracting a native discharge pulse waveform feature set from the multi-channel hybrid induced signal; and a decision module for fusing the spatial modulation feature tensor and the native discharge pulse waveform feature set for joint analysis and decision-making, and outputting a partial discharge diagnostic report.

[0016] Compared with the prior art, the present invention has the following advantages: This invention improves the ability to detect early, weak discharges by injecting specific detection signals into the insulation system of cable accessories and simultaneously acquiring multi-channel induction signals. It also utilizes coherent demodulation technology to actively extract signal components modulated by partial discharges. This is because the actively injected detection signals provide a stable reference for detection, enabling the separation of modulation information reflecting minute changes in the properties of the insulation medium, even under strong power frequency background noise. This results in highly sensitive partial discharge detection.

[0017] This invention integrates the modulation characteristics of actively detected signals with the original discharge pulse characteristics of passively monitored signals for joint analysis. The spatial modulation feature tensor constructed from the demodulated signal reflects the disturbance mode and spatial distribution of the discharge on the insulating medium's properties, while the original pulse waveform characteristics extracted directly from the mixed signal retain the original morphological information of the discharge event. This multi-dimensional data fusion analysis mechanism comprehensively utilizes features from different physical sources, improving the accuracy of discharge type identification and the reliability of diagnostic conclusions.

[0018] This invention generates a three-dimensional tensor that can characterize the spatial distribution of features by deploying a sensor array at key locations inside the cable accessories and combining it with spatial interpolation analysis of the modulation components of the detection signal. This allows the method to not only determine whether a discharge exists, but also to locate the possible location of the discharge based on the abnormal spatial clustering area of ​​the modulation features, providing a clear spatial direction for subsequent inspection and maintenance.

[0019] This invention makes a preliminary inference based on spatial modulation characteristics, and then performs reverse verification using native pulse waveform characteristics. This step-by-step verification and conflict resolution mechanism forms a closed-loop diagnostic logic, effectively avoiding diagnostic errors caused by misjudgment of a single signal feature or environmental interference, thereby outputting a more comprehensive and reliable partial discharge diagnostic report. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating a partial discharge detection method for cable accessories according to an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the structure of a partial discharge detection system for cable accessories according to an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Reference Figure 1 One embodiment of the present invention proposes a method for detecting partial discharge in cable accessories. It combines active injection of detection signals with passive monitoring of native pulses, and integrates spatial modulation features and waveform features for joint analysis, which can achieve sensitive, reliable detection and accurate location of partial discharge in cable accessories.

[0025] The method described in this embodiment specifically includes: S1. Inject a preset detection signal into the cable accessory insulation system, and synchronously collect the induction signal through the built-in sensor array to generate a multi-channel mixed induction signal; In one embodiment of the present invention, step S1 includes the following steps: Built-in sensor arrays are deployed inside the insulation layer of cable accessories or at pre-designated key interfaces. Between the high-voltage conductor and the grounding layer of the cable accessory, a perturbation signal coupling unit is used to superimpose the detection signal onto the operating power frequency voltage; the perturbation signal coupling unit includes a high-pass filter and an isolation transformer to ensure that the detection signal can be injected into the high-voltage side while blocking the power frequency power current; The frequency, amplitude, and phase of the detection signal are controlled so that the amplitude is lower than the operating power frequency voltage and the frequency is within the preset characteristic frequency band of the accessory structure. It synchronously acquires voltage or current signals sensed by the built-in sensor array, amplifies and digitizes them, and generates time-aligned multi-channel hybrid sensing signals.

[0026] Specifically, after the cable accessories are installed, a signal injection and synchronous acquisition process is executed. This process begins by deploying a built-in sensor array inside the insulation layer of the cable accessories, or at a predetermined critical interface where the semiconductor shielding layer and the insulation layer meet. This built-in sensor array consists of multiple piezoelectric ceramic sensors or capacitive electric field sensors arranged in a predetermined spatial configuration. This arrangement is based on statistical analysis of 300 historical failure cases of power cable accessories, ensuring optimal spatial coverage and sensitivity for electric field or mechanical vibration disturbances in critical areas inside the insulation. A perturbation signal coupling unit is connected between the high-voltage conductor and the grounding layer of the cable accessories. This perturbation signal coupling unit includes a high-pass filter with a cutoff frequency 50 Hz higher than the power system frequency but lower than the predetermined detection signal frequency, and an isolation transformer with an electrostatic shielding layer between the primary and secondary windings. Its function is to couple and superimpose the subsequently generated detection signal onto the high-voltage power frequency of the cable during operation, while using the series impedance of the high-pass filter and the isolation transformer to block the power frequency power current from flowing into the signal source, ensuring that the detection signal can be effectively injected into the high-voltage side conductor without interfering with the operation of the power grid.

[0027] A preset detection signal is generated by a signal generator. The frequency of this detection signal is controlled. Amplitude With initial phase .frequency The setting is based on the structural resonance characteristics of the cable accessories themselves, and its characteristic frequency band is obtained through finite element simulation calculation. The frequency range is set between 1 kHz and 1 MHz to avoid power frequency harmonic interference and to excite the characteristic response of the insulation system. Amplitude The setting must meet safety and sensitivity requirements, and its value must be significantly lower than the peak value of the operating power frequency voltage. To avoid introducing additional insulation stress, it is usually based on the relationship... Set, where the scaling factor Based on engineering experience, the value range is between 0.001 and 0.05; in this embodiment, k is taken as 0.01. Initial phase It can be set to 0 radians or modulated with a pseudo-random sequence to facilitate subsequent signal separation.

[0028] The high-precision data acquisition system, triggered synchronously with the signal generator, is activated. This system synchronously acquires the voltage or current signals sensed by each sensor in the built-in sensor array. Each sensed signal is first amplified by a preamplifier, with the amplification factor set according to the sensor sensitivity and the expected signal dynamic range, for example, 100 times. The amplified analog signal is then digitized by a multi-channel synchronous analog-to-digital converter, with the sampling clock allocated from the same clock source to ensure time alignment of all channel data. The resulting discrete signal sequence, arranged in chronological and channel order, is thus generated as a multi-channel hybrid sensed signal. This signal contains both the injected probe signal and... The response component generated after modulation by the insulation system also includes the induced component of the operating power frequency voltage. And noise or interference components such as possible partial discharge pulses. Its mathematical model can be expressed as the signal of the i-th channel. ,in This represents the transmission and modulation function of the i-th channel of the probe signal in the insulation system.

[0029] For example, for a 110 kV cross-linked polyethylene cable termination, step S1 is performed. The signal frequency is detected. Based on the simulation characteristics of this type of terminal, the frequency band was selected as 50 kHz. Peak operating power frequency voltage. It is 89.8 kV, based on Volt, setting the amplitude of the detection signal The voltage rating is 898 volts. The high-pass filter cutoff frequency of the perturbation signal coupling unit is set to 500 Hz, the isolation transformer turns ratio is 1:1, and the withstand voltage level is higher than 110 kV. The built-in sensor array consists of 8 piezoelectric ceramic sensors, which are arranged in a two-layer, four-point grid at the interface between the terminal insulating core and the stress cone, based on the high-risk area of ​​electric field distortion determined by simulation. The sampling rate of the synchronous acquisition system is set according to the Nyquist criterion. Megahertz, meaning 500,000 samples per second, with a synchronization clock jitter of less than 100 picoseconds. After the above process, a multi-channel hybrid induction signal data matrix containing 8 channels, each with 10,000 sampling points, is finally output for subsequent processing steps.

[0030] S2. Using the detection signal as a reference, coherently demodulate the multi-channel hybrid sensing signal, separate and extract the modulation components of the detection signal, and construct an initial modulation component matrix; In one embodiment of the present invention, step S2 includes the following steps: A local reference signal with the same frequency and phase as the detection signal is generated and used as a reference. Each signal in the multi-channel mixed sensing signal is multiplied by the local reference signal to obtain the mixed signal; The signal after each mixing is low-pass filtered to remove the high-frequency carrier component and extract the low-frequency baseband signal containing amplitude and phase change information, which is used as the detection signal modulation component of the corresponding channel. Arrange the modulation components of the detection signals from all channels in channel order to form an initial modulation component matrix.

[0031] In one embodiment of the present invention, step S2 further includes the following steps: After the cable accessories are installed and in a state of no discharge, a detection signal is injected, and the background response signal is collected through the built-in sensor array. The modulation component of the detection signal is compared with the background response signal to eliminate the inherent bias caused by the structural asymmetry of the accessory itself.

[0032] Specifically, using the preset detection signal in step S1 as a reference, coherent demodulation is performed on the multi-channel mixed sensing signal. This process employs an orthogonal demodulation method to obtain complete complex modulation information. First, the digital signal processor determines the frequency of the detection signal based on the known frequency of the detection signal. With initial phase In the digital domain, a pair of orthogonal local reference signals are generated: in-phase reference signals. Orthogonal reference signal These two signals together serve as the phase reference for subsequent demodulation.

[0033] The first in the multi-channel mixed sensing signal Digital signals of each channel Quadrature mixing is achieved by multiplying the signals with two local reference signals. This operation produces two mixed signals: and .because It contains the probe signal component modulated by the insulation system. This orthogonal multiplication operation can shift the sideband spectrum carrying the modulation information to the baseband.

[0034] For each channel of the mixed signal and Perform the same low-pass filtering process. The cutoff frequency of the low-pass filter. Based on the modulation bandwidth of the detection signal The settings must meet the following requirements. and It is usually set to To ensure complete filtering of high-frequency carrier components generated by multiplication operations, whose frequency is located at... Nearby. Power frequency voltage induced component. Also because its frequency is much lower than the cutoff frequency This is effectively filtered out. After low-pass filtering, two baseband signals are output. and These are combined to form the complex form of the probe signal modulation components for the corresponding channel. It is a complex baseband signal, and the changes in its magnitude and argument correspond to the modulation of amplitude and phase, respectively.

[0035] Then all Complex probe signal modulation components of each channel ,in From 1 to Arranged in time according to the physical numbering order of the channels, they form a two-dimensional complex initial modulation component matrix. The rows of this matrix correspond to channel indices, and the columns correspond to discretized time sample points.

[0036] After the cable accessories are installed and confirmed to be in a normal, discharge-free operating state, an additional background calibration procedure is performed. This procedure repeats the injection and acquisition operations of step S1, and through the same orthogonal demodulation processing described above, a set of complex background response signals is obtained, i.e., the modulation components of the background detection signal. To eliminate the detection signal response deviation caused by the inherent structural asymmetry of the cable accessories due to the manufacturing process, the complex detection signal obtained from subsequent online monitoring is modulated. Modulation component of the corresponding background complex detection signal Perform element-wise complex division to obtain the calibrated modulation components. This eliminates inherent biases. The initial modulation component matrix used in subsequent steps is constructed from this calibrated complex data.

[0037] For example, following the example of step S1, the acquired 8-channel multi-channel hybrid sensing signal is processed in step S2. The frequency of the detected signal is known. 50 kHz, initial phase Since it is 0 radians, the generated orthogonal local reference signal is: and Sampling rate It is 2 MHz. Eight signals. to respectively with and Digital multiplication is performed to obtain 16-channel mixed signals. The low-pass filter is set to a finite-length unit impulse response filter of order 128, with a cutoff frequency of... Set as The frequency band of kilohertz is much larger than the expected modulation bandwidth, which allows for the complete preservation of modulation information while filtering out high-frequency components near 100 kilohertz. The filtered components are then combined to obtain eight complex-form probe signal modulation components. to Each signal channel remains 10,000 points long. These eight complex signals are arranged in channel order to form an 8-row, 10,000-column complex initial modulation component matrix. During the installation phase, the background response was acquired by injecting the same detection signal and using the same quadrature demodulation process to obtain the background complex modulation components. to During online monitoring, the data obtained in real time will be... and Perform point-by-point complex division The calculation results are used as elements of the initial modulation component matrix after eliminating inherent biases. In this example, after background calibration, the mean magnitude of the complex modulation component of channel 1 is calibrated from 0.85 volts to 1.02 volts, and the mean phase offset is calibrated from 0.12 radians to 0.01 radians, verifying the effectiveness of orthogonal demodulation and complex division correction in eliminating inherent biases of structural asymmetry.

[0038] S3. Analyze the multi-dimensional modulation features of the initial modulation component matrix and generate a spatial modulation feature tensor; In one embodiment of the present invention, step S3 includes the following steps: Time-domain analysis is performed on the initial modulation component matrix to calculate the amplitude fluctuation variance and phase jump number of the modulation components of the detection signal in each channel, and an amplitude-phase perturbation vector is generated. Frequency domain analysis is performed on the initial modulation component matrix to calculate the energy proportion of the preset subharmonic components in the modulation components of the detection signal of each channel, and a harmonic distortion vector is generated. By combining the spatial position information of the built-in sensor array, the amplitude and phase perturbation vectors and harmonic distortion vectors of each channel at the same time are interpolated and gridded in space to form a three-dimensional data volume that can reflect the spatial distribution of modulation features, thus obtaining the spatial modulation feature tensor.

[0039] Specifically, the multi-dimensional modulation features of the initial modulation component matrix output in step S2 are analyzed to generate a spatial modulation feature tensor. This process extracts features from the initial modulation component matrix in both the time and frequency domains and maps the extracted features to the spatial location of the built-in sensing array.

[0040] Time-domain analysis focuses on the complex probe signal modulation components of each channel in the initial modulation component matrix. First, calculate the instantaneous amplitude sequence of the complex signal. With instantaneous phase sequence Before calculating the phase characteristics, it is necessary to... Phase dewinding is performed to eliminate the interference caused by the calculation of principal values ​​between adjacent sampling points. Jumps in radians that are multiples of each other yield a continuous sequence of instantaneous phases. Calculate the variance of amplitude fluctuation. To quantify amplitude stability, the formula is: ,in It is the length of the sliding analysis window. It is the mean of the amplitudes within that window, i.e. , The starting index of the window. Number of phase transitions. The detection is performed by analyzing the unwound phase sequence. Calculate the difference To achieve, when The timer jumps once. It is a preset phase transition threshold, which is usually set based on the statistical characteristics of phase noise. Radius. For each channel, the calculated amplitude fluctuation variance will be... Number of phase transitions Combined into a two-dimensional vector This is called the amplitude and phase perturbation vector of the channel.

[0041] Frequency domain analysis also targets the complex probe signal modulation components for each channel. Performing a discrete Fourier transform on the time-domain sequence yields its complex spectrum. The analysis focuses on the preset subharmonic components, i.e., the frequency of the detected signal. Frequency components that are integer fractions of a frequency. Calculate the second harmonic. With third harmonic The proportion of energy within a certain bandwidth relative to the total signal energy. Harmonic energy proportion. The calculation formula is ,in This represents the harmonic frequency of interest. The allowable frequency offset bandwidth is typically set to a value of [value to be filled in]. , The sampling frequency is used. For each channel, the proportions of the second and third harmonics are calculated, generating a harmonic distortion vector. .

[0042] It integrates time-domain and frequency-domain features and assigns them spatial attributes. This is combined with the pre-calibrated three-dimensional spatial coordinates of each sensing unit in the built-in sensing array. For each channel corresponding to the same analysis time window, the calculated amplitude and phase perturbation vector is... Harmonic distortion vector These are considered as discrete feature points at their spatial locations within the channel. Spatial interpolation methods, such as Kriging interpolation or radial basis function interpolation, are used to obtain the four-dimensional feature values ​​at these discrete points. Interpolation is performed onto a regular three-dimensional volumetric mesh. This mesh covers the entire target monitoring area of ​​the cable accessory insulation system, with the mesh resolution set according to diagnostic accuracy requirements. After interpolation, each mesh node contains complete four-dimensional feature information. This data structure, composed of the three-dimensional spatial mesh and the four-dimensional feature vectors at each node, is the spatial modulation feature tensor. .

[0043] For example, following the example of step S2, step S3 analyzes the calibrated complex initial modulation component matrix of the 8 channels. First, the instantaneous amplitude of the signal in each channel is calculated. With instantaneous phase And the phase sequence is unwound to obtain In time-domain analysis, the sliding window length is taken. Each sampling point, phase transition threshold Radius. The amplitude and phase perturbation vectors of the eight channels are calculated. Taking channel 1 as an example, the mean of the amplitude sequence within the first analysis window is... Volts, calculate the variance The phase sequence difference after unwinding exceeds The number of times is 3, therefore the amplitude and phase perturbation vector of channel 1 is... In frequency domain analysis, the frequency of the probe signal... Considering the low-frequency envelope characteristic of this complex modulation component, the local energy proportion at the second modulation frequency (set to 100 Hz) and the third modulation frequency (set to 150 Hz), which characterize the low-frequency nonlinear effect of partial discharge defects on the structure, is analyzed. The frequency offset bandwidth is adjusted to... Hertz. The proportion of second harmonic energy was calculated by performing a discrete Fourier transform on the complex modulation components of channel 1. Third harmonic energy percentage Therefore, the harmonic distortion vector is Eight sensors are known to be distributed in two layers at four points inside the terminal, possessing known three-dimensional coordinates. A three-dimensional interpolation mesh with a resolution of 5 mm is generated to cover the stress cone region of the terminal, with a size of... The grid. The four-dimensional feature vectors at 8 discrete points. Radial basis functions are interpolated onto all nodes of the mesh, ultimately forming a mesh with dimension [missing value]. Spatial Modulation Feature Tensor Spatially, this tensor shows that features with amplitude variance greater than 0.04 are concentrated in the grid coordinates. The nearby region, while the feature of phase transitions greater than 2 is in The high values ​​in the vicinity enable a visual mapping of modulation features from discrete channels to a continuous spatial distribution.

[0044] S4. Extract the original discharge pulse waveform feature set from the multi-channel mixed induction signal; In one embodiment of the present invention, step S4 includes the following steps: The multi-channel hybrid sensing signal is digitally filtered to separate the signal components with a frequency band higher than that of the detection signal, thus obtaining a high-pass filtered signal. The high-pass filtered signal is subjected to threshold comparison and pulse identification to capture transient pulse waveforms that exceed a preset noise baseline; For each captured transient pulse waveform, the peak amplitude, pulse width, rise time, and pulse repetition rate are extracted to form a feature set of the original discharge pulse waveform.

[0045] Specifically, the native discharge pulse waveform feature set is extracted from the multi-channel hybrid induction signal generated in step S1. This process is independent of the analysis path of the probe signal and focuses on capturing possible partial discharge transient pulses in the signal. First, the multi-channel hybrid induction signal is digitally filtered, and a cutoff frequency of [missing value] is designed. The high-pass digital filter is designed as a finite-length unit impulse response filter using the window function method. Cutoff frequency. The setting is based on the preset frequency of the detection signal. Their relationship is , where the coefficient The value ranges from 2 to 10 to ensure effective filtering of the detection signal and the power frequency fundamental component, while retaining the higher-frequency discharge pulse signal. The order of this filter... The stopband attenuation is set according to requirements, typically not lower than 64th order. This is used to mix the sensing signals of each channel. The corresponding high-pass filtered signal is obtained through this filter. .

[0046] High-pass filtered signal Perform pulse identification and set a dynamic noise baseline. As a benchmark for threshold comparison, this baseline is determined by comparing... Statistical calculations were performed to obtain the data. Specifically, the signal was divided into segments of length [missing information]. For consecutive non-overlapping segments, calculate the root mean square value of each segment of the signal. ,in For segment indexes. Retrieve from all segments. minimum value The noise baseline is defined as proportionality coefficient Based on the noise statistical characteristics under numerous no-discharge operating conditions, the value is typically set to 3 to 5. When the signal... The instantaneous absolute value exceeds At that time, the pulse capture process is triggered. This exceeds the limit point. Centered on, looking back Each sampling point continues backwards. A complete transient pulse waveform is extracted from each sampling point. ,in From 1 to .length and The setting is based on statistical analysis of the width distribution of historical discharge pulse waveforms, and it needs to ensure that the rising edge and decay process of the pulse can be completely included.

[0047] For each captured transient pulse waveform Four quantization features are extracted sequentially. Peak amplitude Defined as the absolute maximum value of the waveform sequence, i.e. Pulse width Defined as waveform amplitude exceeding The duration, through calculation, satisfies Number of consecutive sampling points Multiply by the sampling interval To obtain, that is Rise time Defined as waveform from Rise to The time elapsed. Specifically, by finding the first crossover in the waveform sequence. point and the first crossing point ,calculate Pulse repetition rate Defined as within the current analysis time window The total number of pulses identified from the same channel. The ratio to the length of the time window, i.e. .

[0048] The four eigenvalues ​​of each pulse are combined into an eigenvector. The feature vectors of all pulses captured by all channels within all time windows are aggregated to form the original discharge pulse waveform feature set. ,in This represents the total number of pulses detected.

[0049] For example, following the example of step S1, the signal frequency is detected. kilohertz, setting coefficient The cutoff frequency of the high-pass filter is... kilohertz. Design a 128th-order finite-length unit impulse response high-pass filter using a Hanning window, with passband ripple less than 0.1 dB and stopband attenuation greater than 60 dB. Pass the 8-channel mixed induced signal through this filter to obtain 8 high-pass filtered signals. to Set the signal segment length. Each sampling point corresponds to 10 milliseconds. Calculate the values ​​for each segment of channel 1. Assuming the minimum value is 0.003 volts, take... Then the noise baseline Volts. In the analysis time window Within seconds, the channel 1 signal in The amplitude reached 0.05 volts, exceeding... .set up , The pulse waveform was captured. Calculate its characteristics: peak amplitude Volts; the number of consecutive points exceeding 0.025 volts is 85, sampling interval microseconds, therefore pulse width microseconds; found Corresponding points , Corresponding points Rising time Microseconds; within a 0.1-second time window, channel 1 identified a total of 8 pulses, with a pulse repetition rate of... Hertz. The eigenvector of this pulse is Assuming that the 8 channels capture a total of 65 valid pulses within 0.1 seconds, the final generated native discharge pulse waveform feature set... It contains 65 such four-dimensional feature vectors.

[0050] S5. The spatial modulation feature tensor and the original discharge pulse waveform feature set are fused together for joint analysis and decision-making, and a partial discharge diagnosis report is output.

[0051] In one embodiment of the present invention, step S5 includes the following steps: A physical association rule base is established based on the expected perturbation patterns of different types of partial discharge events on the dielectric properties and their correspondence with the waveform characteristics of the original discharge pulse. Based on the spatial distribution pattern of the spatial modulation feature tensor at the current moment, a match is made in the physical association rule base to preliminarily infer the discharge type and possible region, and generate a preliminary inference result. Based on the preliminary inference results, matching is performed in the physical association rule base to obtain the inferred waveform feature set; Verify whether the pulse characteristics of the original discharge pulse waveform feature set match the pulse characteristics of the inferred waveform feature set; If the verification passes, a partial discharge diagnostic report is generated by combining the original discharge pulse waveform feature set with the spatial modulation feature tensor, which includes the discharge type, confidence level, intensity level, and approximate area. If the verification fails, a conflict resolution mechanism is initiated, including: increasing the data acquisition time, adjusting the frequency or amplitude of the detection signal, and re-extracting the original discharge pulse waveform feature set; analyzing the feature changes between the re-extracted original discharge pulse waveform feature set and the previously extracted original discharge pulse waveform feature set to obtain conflict features; and comparing the conflict features with a preset historical case library to obtain candidate diagnostic results and corresponding confidence levels.

[0052] Specifically, the spatial modulation feature tensor generated in step S3 is combined with the native discharge pulse waveform feature set extracted in step S4 for joint analysis and decision-making. This process begins with the establishment of a physical association rule base. This rule base is constructed based on the physical mechanism analysis of various known types of partial discharge events, which are derived from the summarization of a large amount of experimental data and finite element simulation results. Each rule describes a specific discharge type, such as internal air gap discharge or surface creepage, and the expected perturbation mode it produces on the properties of the insulating medium. This mode is manifested as a specific spatial distribution of amplitude and phase perturbation and harmonic distortion features in the spatial modulation feature tensor. At the same time, the rules also record the typical native discharge pulse waveform features corresponding to this discharge type, including the typical numerical range or statistical distribution of peak amplitude, pulse width, rise time, and pulse repetition rate. The rule base is stored in the form of a data table or knowledge graph and serves as the benchmark for subsequent matching and diagnosis.

[0053] Based on the spatial modulation feature tensor obtained at the current moment The presented spatial distribution pattern is matched against a physical association rule base. The matching process calculates the similarity between the current tensor feature and the expected spatial distribution pattern recorded in each rule of the rule base. Similarity This can be quantified by calculating the distance between feature vectors in a multidimensional space, for example, by using the reciprocal of the normalized Euclidean distance: .here, This represents the extraction of the first feature from the current spatial modulation feature tensor within the suspected region (a set of grid points whose feature values ​​exceed a certain proportion of the global mean). Statistics for each feature dimension, such as the mean; Indicates the first rule in the rule base The first rule corresponding to the Typical values ​​for each feature dimension; These are weights assigned to different feature dimensions, set based on their discriminative power against different discharge types. They can be determined through feature selection methods, and their sum is 1. Similarity is then selected. The highest rule, along with the corresponding discharge type and the typical occurrence area recorded in the rule, serves as the preliminary inference result.

[0054] It should be noted that in step S5, similarity is calculated. hour, Represents the current spatial modulation feature tensor The extracted first The statistical measure of each feature dimension within the suspected region. The suspected region is determined by selecting all grid points whose feature values ​​exceed a certain proportion (e.g., 1.5 times) of the global mean. Statistical measure The average of the feature values ​​of these points is usually taken. Weights These are preset values, reflecting the importance of different feature dimensions in distinguishing discharge types. They can be preset through feature importance analysis based on historical data and must meet certain requirements. .

[0055] Based on this preliminary inference, a set of inferred waveform features that perfectly corresponds to this discharge type was retrieved from the physical association rule base. This feature set is defined in the form of feature value intervals. For example, for peak amplitude, it may be recorded as an interval in the rule base. .

[0056] Next, we will verify the actual extracted native discharge pulse waveform feature set from step S4. Does the pulse characteristic match the inferred waveform feature set? The pulse characteristics match. Verification is performed on the feature set. Each feature vector in Perform the following: For each pulse's four features, check if its value falls within the range of the corresponding feature in the rule base. Set a matching threshold. For example, 0.7, if the feature set China Super League If all the pulse characteristics of the ratio pass the interval check, the check is considered passed. The specific criterion is: if... If the result is positive, the verification passes. It is the total number of pulses. It is an indicator function, when the pulse... The value is 1 when all features are within their corresponding rule intervals, and 0 otherwise.

[0057] If the verification passes, a partial discharge diagnostic report is generated by combining the original discharge pulse waveform feature set and the spatial modulation feature tensor. The report content must include at least: the discharge type, whose confidence level is directly calculated using the highest similarity obtained from the matching process. Discharge intensity level, based on feature set The statistical distribution of the peak amplitude of the medium pulse is classified, for example, the average peak amplitude is mapped to a preset level such as "weak", "medium" and "strong"; the approximate discharge area is the spatial grid coordinate range from which the feature anomaly is most significant, extracted from the spatial modulation feature tensor.

[0058] If the verification fails, a conflict resolution mechanism is activated. This mechanism first performs a re-analysis process, which may include increasing the data acquisition time to twice the original time, or fine-tuning the frequency of the detection signal based on preliminary inferences. or amplitude Then, steps S1 to S4 are repeated to obtain a new set of native discharge pulse waveform features. Analyze the re-extracted feature set. Compared with the previously extracted feature set The characteristic changes between them are calculated, and the relative change in each characteristic dimension (such as average pulse width) is calculated. The system defines the feature with the largest change and its direction of change as the conflict feature. This conflict feature is then compared with a pre-defined historical case database. The historical case database stores detailed data on past conflict events, including the conflict features at the time, the final confirmed discharge type, and the confirmation method. The comparison uses a similarity calculation method to find several historical cases in the database with the most similar conflict features. The final diagnostic results of these cases are used as candidate diagnostic results, and each candidate result is assigned a confidence level, which is inversely proportional to the difference between the current conflict feature and the conflict features of historical cases. The system outputs these candidate diagnostic results and their confidence levels for operator review, or marks them as events to be investigated for further processing.

[0059] For example, following steps S3 and S4, the spatial modulation feature tensor Displayed in coordinates The amplitude variance characteristic of the surrounding area is significantly higher. This can be determined by calculating the average amplitude variance characteristic of this area. Average value of phase jump characteristics The average proportion of second harmonics The average proportion of the third harmonic This distribution pattern corresponds to the typical feature vector recorded in the "internal air gap discharge" rule in the physical association rule base. Highly similar. Define a weight vector. Calculate similarity The highest value was found, and the preliminary inference is that the discharge type is internal air gap discharge, possibly located in the coordinate region. Nearby. The inferred waveform feature set for this type of discharge was retrieved from the rule base. For: peak amplitude interval Volts, pulse width interval microseconds, rise time interval microseconds, pulse repetition rate interval Hertz. Verification step S4 extracts the original discharge pulse waveform feature set. Of the 65 pulses, 58 pulses have all their feature values ​​falling within the corresponding intervals mentioned above, resulting in a matching ratio. The verification passed. Calculate the feature set. The average peak amplitude of the medium-pulse was 0.048 volts, belonging to the "medium" intensity level. Combining these two factors, a diagnostic report was generated: the discharge type was "internal air gap discharge," with a confidence level of 0.82; the intensity level was "medium"; and the approximate discharge area was defined by grid coordinates. If the verification fails, assuming only 30 pulses conform to the rules, conflict resolution is initiated. The acquisition time is increased from 0.1 seconds to 0.2 seconds, and the feature set is re-extracted. The average pulse width was found to have decreased from 170 microseconds to 15 microseconds, a relative change. The similarity is 0.912, and the conflict feature is "pulse width narrows sharply". Comparison with the historical case database reveals the historical case most similar to "pulse width narrows sharply", ultimately diagnosed as "corona discharge of a metal protrusion". Its conflict feature similarity reaches 0.85, so the output candidate diagnosis result is "corona discharge of a metal protrusion", with a confidence level of 0.85.

[0060] Based on the same inventive concept, such as Figure 2 As shown, the present invention also provides a partial discharge detection system for cable accessories, the system comprising: The active excitation module is used to inject a preset detection signal into the cable accessory insulation system and synchronously collect the induction signal through the built-in sensor array to generate a multi-channel hybrid induction signal. The signal demodulation module is used to coherently demodulate the multi-channel hybrid sensing signal with the detection signal as a reference, separate and extract the modulation components of the detection signal, and construct an initial modulation component matrix. The signal processing module is used to analyze the multi-dimensional modulation features of the initial modulation component matrix and generate a spatial modulation feature tensor. The signal extraction module is used to extract the original discharge pulse waveform feature set from the multi-channel mixed induction signal; The decision module is used to fuse the spatial modulation feature tensor and the original discharge pulse waveform feature set, perform joint analysis and decision-making, and output a partial discharge diagnostic report.

[0061] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0062] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for detecting partial discharge in cable accessories, characterized in that, The method includes: A preset detection signal is injected into the cable accessory insulation system, and the induction signal is synchronously acquired through the built-in sensor array to generate a multi-channel hybrid induction signal; Using the detection signal as a reference, the multi-channel hybrid sensing signal is coherently demodulated to separate and extract the modulation components of the detection signal, and an initial modulation component matrix is ​​constructed. Analyze the multi-dimensional modulation features of the initial modulation component matrix to generate a spatial modulation feature tensor; Extract the original discharge pulse waveform feature set from the multi-channel mixed induction signal; By fusing the spatial modulation feature tensor with the original discharge pulse waveform feature set, joint analysis and decision-making are performed to output a partial discharge diagnostic report.

2. The method for detecting partial discharge in cable accessories according to claim 1, characterized in that, The process of injecting a preset detection signal into the cable accessory insulation system and synchronously acquiring the induced signal through a built-in sensor array to generate a multi-channel hybrid induced signal includes: Built-in sensor arrays are deployed inside the insulation layer of cable accessories or at pre-designated key interfaces. Between the high-voltage conductor and the grounding layer of the cable accessory, the detection signal is superimposed on the operating power frequency voltage through a perturbation signal coupling unit; The frequency, amplitude, and phase of the detection signal are controlled so that the amplitude is lower than the operating power frequency voltage and the frequency is within the preset characteristic frequency band of the accessory structure. It synchronously acquires voltage or current signals sensed by the built-in sensor array, amplifies and digitizes them, and generates time-aligned multi-channel hybrid sensing signals.

3. The method for detecting partial discharge in cable accessories according to claim 2, characterized in that, The perturbation signal coupling unit includes a high-pass filter and an isolation transformer to ensure that the detection signal can be injected into the high-voltage side while blocking the power frequency current.

4. The method for detecting partial discharge in cable accessories according to claim 2, characterized in that, The step of using the detected signal as a reference to coherently demodulate the multi-channel hybrid sensing signal, separating and extracting the modulation components of the detected signal, and constructing an initial modulation component matrix includes: A local reference signal with the same frequency and phase as the detection signal is generated and used as a reference. Each signal in the multi-channel mixed sensing signal is multiplied by the local reference signal to obtain the mixed signal; The signal after each mixing is low-pass filtered to remove the high-frequency carrier component and extract the low-frequency baseband signal containing amplitude and phase change information, which is used as the detection signal modulation component of the corresponding channel. Arrange the modulation components of the detection signals from all channels in channel order to form an initial modulation component matrix.

5. The method for detecting partial discharge in cable accessories according to claim 4, characterized in that, The analysis of the multi-dimensional modulation features of the initial modulation component matrix to generate a spatial modulation feature tensor includes: Time-domain analysis is performed on the initial modulation component matrix to calculate the amplitude fluctuation variance and phase jump number of the modulation components of the detection signal in each channel, and an amplitude-phase perturbation vector is generated. Frequency domain analysis is performed on the initial modulation component matrix to calculate the energy proportion of the preset subharmonic components in the modulation components of the detection signal of each channel, and a harmonic distortion vector is generated. By combining the spatial position information of the built-in sensor array, the amplitude and phase perturbation vectors and harmonic distortion vectors of each channel at the same time are interpolated and gridded in space to form a three-dimensional data volume that can reflect the spatial distribution of modulation features, thus obtaining the spatial modulation feature tensor.

6. The method for detecting partial discharge in cable accessories according to claim 5, characterized in that, The extraction of the native discharge pulse waveform feature set from the multi-channel mixed induction signal includes: The multi-channel hybrid sensing signal is digitally filtered to separate the signal components with a frequency band higher than that of the detection signal, thus obtaining a high-pass filtered signal. The high-pass filtered signal is subjected to threshold comparison and pulse identification to capture transient pulse waveforms that exceed a preset noise baseline; For each captured transient pulse waveform, the peak amplitude, pulse width, rise time, and pulse repetition rate are extracted to form a feature set of the original discharge pulse waveform.

7. The method for detecting partial discharge in cable accessories according to claim 6, characterized in that, The process of fusing the spatial modulation feature tensor with the original discharge pulse waveform feature set for joint analysis and decision-making, and outputting a partial discharge diagnostic report, includes: A physical association rule base is established based on the expected perturbation patterns of different types of partial discharge events on the dielectric properties and their correspondence with the waveform characteristics of the original discharge pulse. Based on the spatial distribution pattern of the spatial modulation feature tensor at the current moment, a match is made in the physical association rule base to preliminarily infer the discharge type and possible region, and generate a preliminary inference result. Based on the preliminary inference results, matching is performed in the physical association rule base to obtain the inferred waveform feature set; Verify whether the pulse characteristics of the original discharge pulse waveform feature set match the pulse characteristics of the inferred waveform feature set; If the verification passes, a partial discharge diagnostic report is generated by combining the original discharge pulse waveform feature set with the spatial modulation feature tensor, which includes the discharge type, confidence level, intensity level, and approximate area. If the verification fails, the conflict resolution mechanism will be activated.

8. The method for detecting partial discharge in cable accessories according to claim 7, characterized in that, If the verification fails, the conflict resolution mechanism will be activated, including: If the verification fails, the data acquisition time is increased, the frequency or amplitude of the detection signal is adjusted, and the original discharge pulse waveform feature set is re-extracted. By analyzing the feature changes between the newly extracted original discharge pulse waveform feature set and the previously extracted original discharge pulse waveform feature set, conflict features are obtained. The conflict features are compared with a preset historical case database to obtain candidate diagnostic results and corresponding confidence levels.

9. A method for detecting partial discharge in cable accessories according to claim 4, characterized in that, The method further includes: After the cable accessories are installed and in a state of no discharge, a detection signal is injected, and the background response signal is collected through the built-in sensor array. The modulation component of the detection signal is compared with the background response signal to eliminate the inherent bias caused by the structural asymmetry of the accessory itself.

10. A partial discharge detection system for cable accessories, characterized in that, The system includes: The active excitation module is used to inject a preset detection signal into the cable accessory insulation system and synchronously collect the induction signal through the built-in sensor array to generate a multi-channel hybrid induction signal. The signal demodulation module is used to coherently demodulate the multi-channel hybrid sensing signal with the detection signal as a reference, separate and extract the modulation components of the detection signal, and construct an initial modulation component matrix. The signal processing module is used to analyze the multi-dimensional modulation features of the initial modulation component matrix and generate a spatial modulation feature tensor. The signal extraction module is used to extract the original discharge pulse waveform feature set from the multi-channel mixed induction signal; The decision module is used to fuse the spatial modulation feature tensor and the original discharge pulse waveform feature set, perform joint analysis and decision-making, and output a partial discharge diagnostic report.

Images