Method and system for testing aging performance of multi-parameter insulating sleeve

Abstract

The invention relates to the technical field of performance testing, and discloses a multi-parameter insulating sleeve aging performance testing method and system, and the method comprises the steps: carrying out the multi-parameter detection signal collection of an insulating sleeve, and obtaining standardized signal data; inputting the standardized signal data into a coupling recognition network for feature analysis to obtain a classification recognition result; performing aging factor threshold adjustment on the rising edge slope of the discharge pulse of the insulating sleeve based on the classification identification result to obtain a dynamic separation parameter; performing feature space matching and differential identification on the partial discharge signal according to the dynamic separation parameter to obtain feature classification data; phase clustering analysis and step response characteristic analysis are carried out based on the characteristic classification data to obtain the aging state variable coefficient of the insulating sleeve, the problem that the recognition precision of a traditional fixed parameter method is reduced in different aging states is solved, and high recognition accuracy and anti-interference robustness can still be kept in a complex electromagnetic environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of performance testing, in particular to a multi-parameter insulation sleeve aging performance testing method and system. BACKGROUND

[0002] With the increase of the running time of power equipment and the deterioration of the running environment, the internal insulation material of the insulation sleeve will age, including the deterioration of dielectric performance, the decrease of mechanical strength and the decrease of thermal stability, etc. These aging processes are often accompanied by the generation and development of partial discharge. As an important characterization parameter of insulation sleeve aging, the accurate detection and identification of partial discharge are of great significance for preventing equipment failure and ensuring power grid safety.

[0003] However, the existing partial discharge detection technology of the insulation sleeve faces many challenges in complex electromagnetic environments. The traditional detection method mainly relies on single sensor or independent multi-sensor detection mode, and there is a lack of correlation analysis between detection parameters, making it difficult to fully reflect the real aging state of the insulation sleeve. At the same time, there are a large number of electromagnetic interference sources in the substation site, including switch operation of adjacent equipment, corona discharge and other multi-source interference signals. These interference signals and the real partial discharge signals overlap in time-frequency domain features, and it is difficult to achieve effective separation using traditional band-pass filter pretreatment methods. SUMMARY

[0004] The present application provides a multi-parameter insulation sleeve aging performance testing method and system, which solves the problem of decreased recognition accuracy of traditional fixed parameter method under different aging states, and still maintains high recognition accuracy and anti-interference robustness in complex electromagnetic environments.

[0005] The first aspect of the present application provides a multi-parameter insulation sleeve aging performance testing method, which comprises: Performing multi-parameter detection signal collection on the insulation sleeve to obtain standardized signal data; Inputting the standardized signal data into a coupling recognition network for feature analysis to obtain classification recognition results; Based on the classification recognition results, adjusting the aging factor threshold value of the discharge pulse rising edge slope of the insulation sleeve to obtain dynamic separation parameters; According to the dynamic separation parameters, performing feature space matching and differential identification on the partial discharge signal to obtain feature classification data; Based on the feature classification data, performing phase clustering analysis and step response feature analysis to obtain the aging state variation coefficient of the insulation sleeve.

[0006] In combination with the first aspect, in a first implementation manner of the first aspect of the present application, the multi-parameter detection signal collection on the insulating sleeve is performed to obtain standardized signal data, which comprises: The four-dimensional parameter signals of the insulating sleeve are synchronously collected by the current sensor, the ultrasonic sensor, the voltage sensor and the infrared temperature sensor to obtain original multi-parameter detection signals; The original multi-parameter detection signals are input into a signal conditioning circuit to perform amplitude linear transformation and 16-bit precision quantization processing to obtain conditioned signal data; The conditioned signal data is subjected to digital filtering processing according to the dielectric constant range and the inner diameter geometric size parameters of the insulating sleeve to obtain filtered signal data, and the filtered signal data is subjected to time reference alignment to obtain standardized signal data.

[0007] In combination with the first aspect, in a second implementation manner of the first aspect of the present application, the standardized signal data is input into a coupling recognition network to perform feature analysis to obtain a classification recognition result, which comprises: The standardized signal data is input into a feature preprocessing layer of the coupling recognition network to perform matrix conversion to obtain a multi-dimensional input feature matrix; The multi-dimensional input feature matrix is respectively input into a time domain feature extraction sub-network, a frequency domain feature extraction sub-network and a time-frequency joint feature extraction sub-network of the coupling recognition network to perform parallel feature extraction to obtain three parallel feature extraction results; The three parallel feature extraction results are subjected to cross-channel attention weight calculation and multi-head self-attention mechanism fusion by a feature fusion coupling layer of the coupling recognition network to obtain a fusion coupling feature vector; The fusion coupling feature vector is input into a classification decision layer to perform multi-level classification to obtain a classification recognition result containing real discharge, switching operation interference and corona discharge interference.

[0008] In combination with the first aspect, in a third implementation manner of the first aspect of the present application, the fusion coupling feature vector is input into a classification decision layer to perform multi-level classification to obtain a classification recognition result containing real discharge, switching operation interference and corona discharge interference, which comprises: The fusion coupling feature vector is input into a first classification branch of the classification decision layer to perform real discharge and interference signal two-classification processing to obtain a two-classification intermediate result of real discharge signal probability value and comprehensive interference signal probability value; The comprehensive interference signal probability value in the two-classification intermediate result is input into a second classification branch of the classification decision layer to perform switching operation interference and corona discharge interference subdivision type recognition to obtain a fine classification result of switching operation interference probability value and corona discharge interference probability value; The probability distribution is recalculated and the confidence threshold is judged based on the binary classification intermediate result and the fine classification result, to obtain a classification recognition result containing real discharge, switching operation interference and corona discharge interference.

[0009] In combination with the first aspect, in a fourth implementation manner of the first aspect of the application, the aging factor threshold adjustment on the rising slope of the discharge pulse of the insulating sleeve based on the classification recognition result to obtain the dynamic separation parameter comprises: The rising slope of the discharge pulse of the insulating sleeve is calculated based on the classification recognition result to obtain the rising slope of the discharge pulse; The aging factor is determined according to the running years of the insulating sleeve and the type of insulating material, and the aging degree related threshold dynamic adjustment is performed on the rising slope of the discharge pulse to obtain the threshold parameter after aging correction; The pulse width of the discharge pulse is measured based on the threshold parameter after aging correction, and a two-dimensional discrimination space coordinate system is established to obtain a two-dimensional space discrimination region division result; The phase and amplitude correlation detection is performed through the two-dimensional space discrimination region division result to obtain the dynamic separation parameter.

[0010] In combination with the first aspect, in a fifth implementation manner of the first aspect of the application, the phase and amplitude correlation detection is performed through the two-dimensional space discrimination region division result to obtain the dynamic separation parameter, which comprises: The discharge pulse in the two-dimensional space discrimination region division result is used for voltage phase angle distribution analysis to obtain phase angle distribution characteristic data; The amplitude and phase angle of the discharge pulse are related analyzed based on the phase angle distribution characteristic data to obtain a phase-amplitude correlation calculation result; The signal statistical characteristic analysis is performed according to the phase-amplitude correlation calculation result, and the detection sensitivity parameter is dynamically updated by using an adaptive algorithm to obtain a sensitivity optimized detection parameter; The sensitivity optimized detection parameter is integrated with the threshold parameter after aging correction and the two-dimensional space discrimination region parameter to obtain the dynamic separation parameter containing the slope threshold, the pulse width range and the phase distribution characteristic.

[0011] In combination with the first aspect, in a sixth implementation manner of the first aspect of the application, the feature space matching and differential recognition of the partial discharge signal according to the dynamic separation parameter to obtain the feature classification data comprises: The partial discharge signal feature library of the insulating sleeve is constructed according to the dynamic separation parameter, and is divided into five levels according to the aging degree and four types of internal bubble discharge, delamination discharge, surface discharge and suspension discharge according to the defect type, to obtain a multi-level feature library classification structure. extracting a pulse rise time, a pulse decay time, a spectral center frequency, a spectral bandwidth, a phase concentration degree, a pulse polarity, a pulse interval statistical feature, a harmonic distortion degree, a signal energy distribution, a time domain waveform similarity, a frequency domain power spectrum density and a time-frequency joint feature fingerprint from the partial discharge signal to obtain a partial discharge signal feature parameter set; differentially identifying the partial discharge signal feature parameter set and the multi-level feature library classification structure by using a support vector machine algorithm and establishing a discharge signal and interference signal separation hyperplane by using a radial basis function based on the feature classification data.

[0012] In a seventh implementation manner of the first aspect, based on the feature classification data, phase clustering analysis and step response feature analysis are performed to obtain an aging state variation coefficient of the insulating sleeve, including: signal grouping is performed based on the feature classification data to obtain a signal grouping result, and clustering processing is performed on the partial discharge signals in each group based on the signal grouping result to obtain a phase clustering grouping result; step response characteristic analysis is performed on the discharge pulse based on the phase clustering grouping result, and a discharge pulse step response feature parameter including an initial amplitude, a decay time constant, an oscillation frequency and an initial phase is established; geometric structure correction is performed on the discharge pulse step response feature parameter according to an inner-to-outer diameter ratio of the insulating sleeve to obtain a geometrically corrected step response parameter, and the aging state variation coefficient of the insulating sleeve is generated according to the geometrically corrected step response parameter.

[0013] In an eighth implementation manner of the first aspect, the geometric structure correction is performed on the discharge pulse step response feature parameter according to the inner-to-outer diameter ratio of the insulating sleeve to obtain a geometrically corrected step response parameter, and the aging state variation coefficient of the insulating sleeve is generated according to the geometrically corrected step response parameter, including: a correction coefficient value for different geometric structures is determined according to the inner-to-outer diameter ratio of the insulating sleeve; a product operation correction is performed on the initial amplitude, the decay time constant, the oscillation frequency and the initial phase in the discharge pulse step response feature parameter by using the correction coefficient value to obtain a geometrically corrected step response parameter; a standard deviation to mean ratio calculation of signal feature parameters is performed based on the geometrically corrected step response parameter to obtain a variation coefficient value; determining that the discharge mode is stable when the coefficient of variation value is less than the first target value, determining that the discharge mode is in a transition state when the coefficient of variation value is between the first target value and the second target value, and determining that the discharge mode is unstable when the coefficient of variation value is greater than the second target value, and finally generating the coefficient of variation of the aging state of the insulating sleeve.

[0014] The second aspect of the application provides a multi-parameter insulating sleeve aging performance test system, which comprises: A signal acquisition module is configured to acquire multi-parameter detection signals of the insulating sleeve and obtain standardized signal data. A feature analysis module is configured to input the standardized signal data into a coupling recognition network for feature analysis and obtain classification recognition results. A threshold adjustment module is configured to adjust an aging factor threshold of the rising slope of the discharge pulse of the insulating sleeve based on the classification recognition results and obtain dynamic separation parameters. A difference recognition module is configured to perform feature space matching and difference recognition on the partial discharge signal based on the dynamic separation parameters and obtain feature classification data. A clustering analysis module is configured to perform phase clustering analysis and step response feature analysis based on the feature classification data and obtain the coefficient of variation of the aging state of the insulating sleeve.

[0015] Compared with the prior art, the application has the following beneficial effects: a four-dimensional parameter detection matrix is constructed to realize multi-sensor collaborative detection, overcoming the limitation of incomplete information of traditional single sensor. A correlation weight matrix between parameters is established by using a coupling recognition network, breaking through the limitation of independent analysis of each detection parameter in the prior art, and realizing intelligent cross-coupling analysis of time-frequency domain features. An adaptive feature extraction controller based on the rising slope of the discharge pulse can dynamically adjust the discrimination threshold according to the aging degree of the insulating sleeve, effectively distinguish the real discharge signal from the multi-source interference signal, and solve the problem of decreased recognition accuracy of the traditional fixed parameter method under different aging states. The support vector machine algorithm difference recognition model combined with the Mahalanobis distance multi-dimensional feature space evaluation realizes accurate separation of the discharge signal and the interference signal. The introduction of the geometric correction coefficient of the insulating sleeve enhances the universality of the method, and the use of the coefficient of variation quantitative evaluation index realizes the technical leap from qualitative analysis to quantitative evaluation, and still maintains high recognition accuracy and anti-interference robustness in a complex electromagnetic environment. BRIEF DESCRIPTION OF DRAWINGS

[0016] 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] The structures, proportions, sizes, etc., shown in the accompanying drawings of this specification are only for the purpose of assisting those skilled in the art in understanding and reading the content disclosed in the specification, and are not intended to limit the conditions under which the present invention can be implemented. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0018] Figure 1 This is a flowchart illustrating the multi-parameter insulating sleeve aging performance testing method provided in this embodiment of the invention. Figure 2 This is a schematic block diagram of the structure of the multi-parameter insulating sleeve aging performance testing system provided in the embodiment of the present invention. Detailed Implementation

[0019] 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, not all, of the embodiments of the present invention. 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.

[0020] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0021] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items, and all possible combinations, and includes such combinations. See also Figure 1One embodiment of the multi-parameter insulating bushing aging performance testing method of the present invention includes: Step 100: Acquire multi-parameter detection signals from the insulating bushing to obtain standardized signal data; It is understood that the executing entity of this invention can be a multi-parameter insulating sleeve aging performance testing system, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0023] Specifically, a multi-channel signal synchronous acquisition system adapted to the structural characteristics and operating environment of insulating bushings is constructed. Current sensors, ultrasonic sensors, voltage sensors, and infrared temperature sensors are deployed to acquire instantaneous current pulses, high-frequency mechanical fluctuations, insulation voltage changes, and thermal distribution changes generated by partial discharge activities, forming raw multi-parameter detection signal data in four dimensions. The sensor signals are synchronously acquired through a high-bandwidth analog channel, with the center frequency and sampling frequency set according to the differences in physical mechanisms. For example, the sampling frequency of the high-frequency current sensor is set to 200MHz, the operating frequency of the ultrasonic sensor is set to 150kHz, the bandwidth of the voltage sensor reaches 10MHz, and the infrared temperature sensor supports a wide range response with an upper limit of 85°C. The sampling data of each channel is synchronized at the nanosecond level through a GPS time synchronization device to ensure consistent timing. The raw analog signals from all channels are input into a dedicated signal conditioning circuit. This circuit incorporates a high-linearity amplifier and a low-noise level conversion module, allowing the amplitude range of different source signals to be compressed or expanded to a unified dynamic range of ±5V. Based on this amplitude standardization, a 16-bit analog-to-digital converter (ADC) performs high-precision digitization of the signal amplitude, ensuring sufficient quantization accuracy and detail retention during computation and model training. The conditioned signal data output at this stage maintains a uniform numerical scale while preserving the original physical characteristics. To suppress power frequency noise, electromagnetic interference, and information mixed in with non-target frequency bands, and to improve the extraction efficiency of target features such as partial discharge signals, hot spot anomalies, or dielectric disturbances, adaptive digital filtering is applied to the conditioned signal data based on the structural properties of the insulating bushing under test. Structural properties include the dielectric constant of the insulating material (between 2.1 and 4.5) and the inner diameter (between 200mm and 800mm). These two parameters are used to dynamically adjust the cutoff frequency and passband characteristics of the filter to optimize the frequency domain response for different bushing configurations. Simultaneously, the filtered signal data is aligned with the time base, and the GPS synchronization signal is used as the master clock source to rearrange and correct the time of each channel data to meet the timing consistency requirements. Finally, standardized signal data with unified time base, amplitude normalization, frequency band limitation and high-precision quantization are output.

[0024] Step 200: Input the standardized signal data into the coupled recognition network for feature analysis to obtain the classification and recognition results; Specifically, the standardized signal data is input into the feature preprocessing layer of the coupled recognition network and subjected to matrix transformation. This layer merges the signals from four types of sensors—current, voltage, ultrasound, and infrared—using a unified sampling time reference through matrix transformation operations, and reconstructs them into an input feature matrix containing multiple channels, multiple time steps, and multidimensional physical meaning. The multidimensional feature matrix is ​​input in parallel to three sub-network modules in the coupled recognition network structure: the time-domain feature extraction sub-network, the frequency-domain feature extraction sub-network, and the time-frequency joint feature extraction sub-network. The time-domain sub-network uses a one-dimensional convolutional neural network structure to mine key dynamic information such as pulse rising edge, pulse width, amplitude change rate, and interval distribution. The frequency-domain sub-network extracts frequency-domain energy characteristics such as spectral center frequency, bandwidth, harmonic components, and frequency drift through Fourier transform interface and frequency domain convolution. The time-frequency joint feature sub-network uses short-time Fourier transform or wavelet transform structure to map the signal to the joint time-frequency plane, from which it identifies spectral abrupt change positions, instantaneous energy concentration, and time-frequency synchronous fluctuation characteristics. Through a multi-angle parallel extraction mechanism, three structured feature sequences are obtained, representing the signal representation results in the time domain, frequency domain, and joint domain, respectively. The results of the three parallel feature extractions are input into the feature fusion coupling layer of the coupled recognition network. Within this layer, a cross-channel attention weighting mechanism is deployed. The importance of different sensor source data channels is dynamically weighted by learning a parameter matrix. Simultaneously, a multi-head self-attention mechanism is introduced to compute the correlation between different feature subspaces in parallel. The outputs of multiple attention subspaces are integrated into a unified fused coupled feature vector. This fused coupled feature vector is then input into the classification decision layer. The decision layer employs a fully connected neural network structure, setting multiple levels of nonlinear activation units and a cross-entropy loss function. Combined with weight preferences set during training, such as setting a higher sensitivity coefficient for the true discharge category, three classification results are generated at the output, corresponding to true discharge, switching operation interference, and corona discharge interference, respectively. Confidence is determined based on the output probabilities.

[0025] Step 300: Based on the classification and recognition results, adjust the aging factor threshold of the rising edge of the discharge pulse of the insulating bushing to obtain dynamic separation parameters; It should be noted that the signal subset identified as a true discharge is extracted, and the rising edge slope is numerically calculated based on its time-domain waveform. This process uses the derivative form of the time period corresponding to the amplitude variation interval, that is, the rising edge slope of the discharge pulse is calculated based on the waveform steepness from the pulse start point to 90% peak value. To make the discharge slope parameter more realistically reflect the current aging state of the insulation system, an aging factor related to the insulation service time and material characteristics is introduced, and an adaptive discrimination threshold adjustment mechanism is constructed accordingly. This aging factor comprehensively considers the service life of the insulating bushing (set to a range of 1 to 30 years) and the type of insulation material (such as paper insulation or composite insulation), and is numerically defined between 0.1 and 1.0, where lower values ​​represent slight aging and higher values ​​represent severe deterioration. Based on this, the traditional fixed threshold range is modified to a dynamic range associated with the aging factor. For example, when the aging factor α is less than 0.5, indicating relatively new insulation, the system modifies the lower limit of the slope threshold to α multiplied by the standard lower limit value to enhance sensitivity to early subtle discharge behavior. When α is greater than or equal to 0.5, a standard slope discrimination interval is used to maintain recognition stability, thus generating the aging-corrected threshold parameters. Based on the aging-corrected threshold parameters, pulse width is measured on the discharge pulse, calculating the time span between 10% and 90% of its amplitude. The resulting rising slope value and pulse width value are mapped together to a two-dimensional discrimination space. In this space, the vertical axis represents the unit voltage rise rate, and the horizontal axis represents the pulse width. The two-dimensional coordinates formed by these two axes allow different types of signals to be naturally distributed in their respective regions. For example, real discharge signals are concentrated in regions with higher slopes and smaller pulse widths, while switching operation interference and corona discharge interference are distributed in other regions due to their different characteristics. Based on this two-dimensional coordinate system, multiple discrimination regions are constructed, and signal classification rules are formed through region boundaries to clarify the spatial division of different discharge types. In the discrimination region, the phase angle and amplitude characteristics of each signal are jointly input into the phase-amplitude analysis module to analyze its distribution position within the power frequency cycle, paying particular attention to whether it is concentrated within ±30° of the positive and negative voltage peaks. This feature is used to distinguish between discharges caused by internal insulation defects and environmental interference signals, the latter exhibiting a random phase distribution. The results of the two-dimensional discrimination space division are fused with the phase-amplitude correlation analysis results to comprehensively determine whether each signal conforms to the target discharge characteristics, and then output dynamic separation parameters.

[0026] Step 400: Perform feature space matching and differential identification on the partial discharge signal based on the dynamic separation parameters to obtain feature classification data; Specifically, a partial discharge signal feature library is constructed based on dynamic separation parameters. This feature library fully considers the differences in partial discharge behavior exhibited by insulating bushings at different aging stages and classifies them into five levels according to aging factors, corresponding to mild aging, moderate aging, severe aging, serious aging, and extreme aging states, respectively. At the same time, based on the defect mechanism, the discharge type is further subdivided into four categories: internal bubble discharge, layered discharge, surface discharge, and suspended discharge, forming a multi-level feature library with a two-dimensional classification structure. For each signal identified as a real discharge through dynamic separation, a high-dimensional feature extraction operation is performed. Specifically, the extracted features include: pulse rise time (the time it takes for the signal to rise from its initial point to 90% of its peak); pulse decay time (the time it takes for the signal to fall back to its reference amplitude from its peak); spectral center frequency and spectral bandwidth, representing the location and distribution range of the main energy concentration in the frequency domain, respectively; phase concentration, measuring the phase distribution pattern of the discharge pulse within the power frequency cycle; pulse polarity, reflecting the positive and negative polarity of the signal; statistical features of the pulse interval, such as the average interval and variability, reflecting the repetition pattern of the discharge; and also harmonic distortion, overall energy distribution, waveform similarity, frequency domain power spectral density, and time-frequency joint feature fingerprints obtained through short-time Fourier transform or wavelet analysis. These features together constitute a high-dimensional set of feature parameters for partial discharge signals. To achieve signal category determination, this set of feature parameters is input into a differential recognition module based on a support vector machine algorithm. High-dimensional spatial mapping is performed based on each classification entry in the feature library, and a radial basis function is used as the kernel function to construct a nonlinear classification boundary. By finding the optimal margin hyperplane in the feature space, Support Vector Machines (SVMs) can effectively distinguish between two major categories: discharge signals and interference signals. Furthermore, they can achieve detailed classification of multiple labels within each discharge category. During model recognition, the Euclidean or Mahalanobis distances between test samples and the centers of various feature types are calculated, and a covariance matrix is ​​introduced for standardization to enhance the model's ability to capture correlations between feature dimensions. When the matching degree between the target signal and a certain category in the feature library reaches a high confidence interval, it is determined to belong to that discharge type, and feature classification data containing aging level and discharge category is output.

[0027] Step 500: Perform phase clustering analysis and step response feature analysis based on feature classification data to obtain the aging state variation coefficient of the insulating bushing.

[0028] Specifically, based on phase angle and timing attributes, all signals identified as actual discharges are grouped. This grouping process is based on a fixed time window (e.g., signals sampled within 5 minutes form a group), initially aggregating the signals according to the sampling period to form signal grouping results. Within each time window, cluster analysis is performed based on the phase angle distribution of the discharge pulse within the voltage cycle. The K-means clustering algorithm is applied to classify and identify the phase characteristics of the discharge signals within each group. The selection of cluster centers is based on minimizing the sum of squares within the group, so that the signals are divided into different subclasses according to their phase concentration characteristics, thereby obtaining phase clustering grouping results. It can also be inferred whether the discharge defect sources corresponding to each cluster are consistent. After completing the phase clustering, step response characteristic analysis is performed on the discharge pulse signals corresponding to each cluster result. This analysis employs an oscillation decay model to fit the dynamic changes of the signal after the rising edge, constructing a typical step response function. This function contains four core parameters: the initial amplitude represents the maximum discharge intensity, the decay time constant reflects the energy dissipation rate of the discharge signal, the oscillation frequency reveals the reflection and resonance characteristics of the internal insulation structure, and the initial phase characterizes the waveform's starting position and phase shift. These parameters comprehensively characterize the dynamic behavior and response speed of the discharge pulse in the time domain. To ensure that the extracted step response parameters accurately reflect the response differences caused by different types of insulating bushings, a geometric correction mechanism for the insulating bushing is introduced. This mechanism, based on the inner-outer diameter ratio, corrects for differences in wave propagation characteristics caused by different structural thicknesses and volume ratios. The correction factor is set according to the following rules: when the inner-outer diameter ratio r is less than 0.3, the correction factor is 1.2, indicating that the relatively thick sleeve structure leads to a long signal propagation path and fast energy dissipation; when r is between 0.3 and 0.6, it is set to 1.0, representing the default response under the standard sleeve structure; when r is greater than 0.6, it is set to 0.8, corresponding to enhanced high-frequency response characteristics under a relatively thin-walled structure. This correction factor is used to geometrically adjust the initial amplitude, attenuation constant, and oscillation frequency to generate geometrically corrected step response parameters. Based on each time window, each cluster group, and a complete set of geometrically corrected response parameters, statistical analysis is performed. The mean and standard deviation of these parameters over continuous time periods are calculated, and the stability of the parameters is quantitatively evaluated using the coefficient of variation CV = σ / μ. The coefficient of variation (CV) is a key indicator of the aging state fluctuation. When the CV value is less than 0.15, it indicates that the discharge behavior is stable and the aging state is slow. When the CV is between 0.15 and 0.3, it is a transition period, indicating that there is local instability in the aging state. When the CV is greater than 0.3, it is considered a typical aging fluctuation stage, which requires high attention.

[0029] In one specific embodiment, the process of performing step 100 may specifically include the following steps: The insulating bushing is simultaneously acquired using a current sensor, an ultrasonic sensor, a voltage sensor, and an infrared temperature sensor to obtain the original multi-parameter detection signal. The original multi-parameter detection signal input signal conditioning circuit is subjected to amplitude linear transformation and 16-bit precision quantization to obtain the conditioned signal data. The conditioned signal data is digitally filtered based on the dielectric constant range and inner diameter geometry of the insulating bushing to obtain filtered signal data. The filtered signal data is then time-referenced to obtain standardized signal data.

[0030] Specifically, based on the functional requirements of insulating bushing aging performance testing and the coupling characteristics of electrical, thermal, and acoustic fields, a multi-sensor joint acquisition architecture is constructed. This architecture uses four types of sensors as the core components of the acquisition unit, respectively covering partial discharge current waveforms, ultrasonic information generated by local oscillations, voltage conduction state inside the insulation layer, and surface temperature field changes due to heat distribution. The current sensor has a high bandwidth and high sensitivity of over 200MHz, enabling real-time capture of transient current pulse signals generated by insulation breakdown or partial discharge; the ultrasonic sensor is equipped with a piezoelectric sensing head with a center frequency of 150kHz to ensure the response capability to high-frequency acoustic energy at local defects; the voltage sensor adopts a capacitive high-frequency sampling structure with a bandwidth greater than 10MHz, which can reproduce the complex transient voltage change process in the insulating medium; the infrared temperature sensor supports wide temperature range measurement from -40°C to 85°C and has high thermal resolution to achieve early detection of hot spots or overheating trends caused by insulation degradation. After connecting the sensors for these four types of physical quantities in parallel to a unified sampling platform, a synchronous trigger mechanism is used to uniformly acquire the start signal. All channels rely on a high-precision ADC synchronous acquisition module for parallel data sampling. The raw output signal of each channel enters the corresponding signal conditioning circuit for amplitude linear transformation and 16-bit precision quantization. Amplitude linear transformation uses a programmable gain amplifier and a limiting and shaping circuit to compress the amplitude range of various signals to a unified ±5V dynamic range, thereby avoiding quantization unevenness caused by amplitude differences. 16-bit precision quantization uses a high-resolution analog-to-digital converter to convert analog signals into digital signals, ensuring that the original pulses, periodic changes, or instantaneous disturbances are not weakened or lost during the conversion process. Through this step, the conditioned signal data is output, retaining the correspondence of the four dimensions. To ensure that the data has the effectiveness and analyzability required for target recognition, the conditioned signal data undergoes digital filtering. A structure-adaptive digital filtering strategy is adopted, using the structural parameters of the insulating sleeve as input conditions to dynamically adjust the response range of the filter. When the dielectric constant of the insulating bushing is known to be between 2.1 and 4.5, and its inner diameter is between 200 mm and 800 mm, the filtering system models the dielectric response frequency and wave propagation characteristics through its internal calculation module, thereby deriving the optimal upper and lower limits of the filter passband. For example, bushings with a larger dielectric constant will produce a stronger capacitive coupling effect, resulting in a more prominent high-frequency component of the discharge signal. In this case, the filter needs to increase the lower limit of the high-pass frequency to eliminate the influence of the voltage fundamental wave. On the other hand, bushings with a smaller inner diameter, due to their compact structure, have a shorter propagation path and more concentrated signal energy, requiring the preservation of the main energy distribution in the mid-to-high frequency range. Therefore, when setting the filter bandwidth, it is necessary to simultaneously balance structural dimensions and electrical characteristics to ensure that the processed filtered signal data retains the effective components related to insulation aging to the maximum extent, while eliminating unnecessary loads caused by power frequency interference, harmonic mixing, and electromagnetic noise. The filtered signal data is then time-aligned.Using GPS timing technology as the clock source, current, ultrasonic, voltage, and temperature signals are assigned corresponding timestamps at the sampling start point and within each time step. Specifically, an FPGA timing controller or time synchronization module extracts the time stamps and corrects for inter-channel deviations, aligning all signal data within nanosecond-level precision to construct a multi-dimensional time-series dataset. Within this dataset structure, the system can perform collaborative analysis of multi-sensor responses at any time point, using time as the primary axis. This includes operations such as synchronously comparing whether thermal rise and current anomalies coexist, and detecting whether ultrasonic pulses and voltage spikes originate from the same source. Standardized signal data is output.

[0031] In one specific embodiment, the process of performing step 200 may specifically include the following steps: Standardized signal data is input into the feature preprocessing layer of the coupled recognition network and subjected to matrix transformation to obtain a multi-dimensional input feature matrix. The multi-dimensional input feature matrix is ​​respectively input into the time-domain feature extraction subnetwork, frequency-domain feature extraction subnetwork and time-frequency joint feature extraction subnetwork of the coupled recognition network for parallel feature extraction, resulting in three-way parallel feature extraction results; By using the feature fusion coupling layer of the coupled recognition network, the cross-channel attention weight calculation and multi-head self-attention mechanism are performed on the three parallel feature extraction results to obtain the fused coupled feature vector. The fused coupled feature vectors are input into the classification decision layer for multi-level classification, resulting in classification and identification results that include real discharge, switching operation interference, and corona discharge interference.

[0032] Specifically, standardized signal data is input into the feature preprocessing layer of the coupled recognition network. In the preprocessing stage, the four-dimensional sensor signals are converted into a unified format tensor matrix through time window sliding and channel mapping operations, forming a multi-dimensional input feature matrix. This matrix includes time series, channel, and amplitude dimensions, reflecting the dynamic evolution of physical signals such as current, voltage, ultrasound, and temperature within the same time window. During this process, the feature preprocessing layer normalizes the input data to ensure the stability of the feature distribution and slices and encodes long-sequence data using short-time block partitioning techniques, ensuring that the time window length of the feature matrix matches the network parameter configuration and maintaining synchronization across channels. The multi-dimensional input feature matrix is ​​then input into three independently constructed feature extraction sub-network modules: a time-domain feature extraction sub-network, a frequency-domain feature extraction sub-network, and a time-frequency joint feature extraction sub-network. The time-domain feature subnetwork combines a one-dimensional convolutional structure with a Long Short-Term Memory (LSTM) network to mine the dynamic behavior of the input signal in the time domain, such as waveform trends, rising slope, pulse width variations, and signal delays, and extracts the instantaneous value distribution, mean shift, and abrupt change features. The frequency-domain feature subnetwork first uses a Fast Fourier Transform (FFT) module to map the time series to the frequency domain, and then uses a two-dimensional convolutional neural network to extract spectral energy, center frequency, bandwidth characteristics, and distribution patterns of various frequency components, showing particular advantages in analyzing harmonic interference and periodic discharge characteristics. The time-frequency joint feature subnetwork integrates Short-Time Fourier Transform (SFT) and wavelet transform, constructing a time-frequency image and inputting it into a deep residual network structure to identify sudden pulse clusters, time drift characteristics, and frequency jump behaviors in partial discharge signals. This module is suitable for identifying complex partial discharge characteristics affected by environmental interference or signal masking. The three subnetworks extract features from the input matrix in different forms and generate three parallel output data streams. Each stream represents the representative information of its dimension using a fixed-length feature vector or tensor. The results of these three parallel feature extractions are input into the feature fusion coupling layer of the coupled recognition network. This layer uses cross-channel attention and multi-head self-attention mechanisms as its core components. By constructing a trainable weight matrix, the importance of features between different channels is weighted and adjusted. The system determines the influence weight of different channels on the overall classification result based on the feature similarity, time synchronization, and frequency crossover between them, thus achieving adaptive feature coupling based on physical properties. At the same time, the introduction of the multi-head self-attention mechanism allows the network to model the interaction relationships of different feature subspaces in parallel in different attention heads. This enables the system to not only capture local salient features but also retain global long-distance dependencies. Through the normalization of attention weights and output reconstruction, a fused coupled feature vector is finally formed. This feature vector centrally reflects the discriminative information of multi-dimensional signals in the time domain, frequency domain, and joint domain.The fused coupled feature vectors are input into the classification decision layer, which consists of two or three fully connected neural networks. The softmax function is then used to output the final multi-class classification probabilities. To improve recognition performance, the classification decision layer employs weighted cross-entropy as the loss function during training, introducing higher weights for the true discharge category to enhance the network's ability to identify a small number of high-risk discharge events. Simultaneously, Dropout and L2 regularization are incorporated during network training to suppress overfitting, resulting in good generalization ability when facing new samples. The final classification results include three clearly defined labels: true discharge, switching operation interference, and corona discharge interference, with each result accompanied by a confidence score.

[0033] In one specific embodiment, the process of performing the step of inputting the fused coupled feature vector into the classification decision layer for multi-level classification to obtain classification and identification results including real discharge, switching operation interference, and corona discharge interference can specifically include the following steps: The fused coupling feature vector is input into the first classification branch of the classification decision layer to perform binary classification of real discharge signals and interference signals, and the intermediate binary classification results of the probability values ​​of real discharge signals and comprehensive interference signals are obtained. The probability value of the comprehensive interference signal in the intermediate results of the binary classification is input into the second classification branch of the classification decision layer to identify the sub-types of switching operation interference and corona discharge interference, and the sub-classification results of the probability values ​​of switching operation interference and corona discharge interference are obtained. Based on the intermediate and fine classification results of the binary classification, the probability distribution is recalculated and the confidence threshold is judged to obtain the classification and identification results that include real discharge, switching operation interference and corona discharge interference.

[0034] Specifically, the fused coupled feature vector is input into the first classification branch of the classification decision layer. This branch is a typical binary classification neural network structure, employing two fully connected layers to extract low-dimensional discriminative information and generate the final output node. A softmax activation function generates two probability outputs, corresponding to the mutually exclusive categories of real discharge and integrated interference. The main task of this branch is to first classify all samples in the full category space into whether they are real partial discharges. The integrated interference category is temporarily merged, containing multiple sub-class signals. During this process, a weighted cross-entropy function is used as the loss function, introducing a weight coefficient for the real discharge category to improve the model's sensitivity to high-risk events with small samples. Simultaneously, during training, the recognition boundary is continuously optimized through backpropagation gradients, enabling the model to accurately capture the subtle but crucial feature differences between discharge and interference signals. The output real discharge probability value and integrated interference probability value constitute the intermediate binary classification result of this layer. The system uses the integrated interference probability value identified in the first classification branch as the input weight factor for the second classification branch, which is then fed into a substructure dedicated to refining the interference type determination—the second classification branch. This branch also employs a fully connected neural network structure, but its target category no longer includes actual discharges. Instead, it further distinguishes between two different mechanisms of disturbance signals: switching operation interference and corona discharge interference, within the interference category. In terms of network structure design, an attention-guided mechanism is appropriately introduced to enhance the recognition of high-frequency oscillation characteristics and periodic amplitude variations, modeling the difference between the sharp frequency spectrum peaks of corona discharge and the wide-amplitude low-frequency pulses in switching operation interference. The output of this branch is also normalized using a softmax function, generating two independent probability values: the probability of switching operation interference and the probability of corona discharge interference. These are used as the refined identification result of the comprehensive interference signal. This refined classification result, together with the actual discharge probability output from the first classification branch, constitutes the initial probability distribution for the three target categories. Based on the intermediate results of the binary classification and the refined classification results, the probability distribution is recalculated, i.e., a probability distribution remapping and confidence threshold determination mechanism is performed. During the probability distribution reshaping process, the true discharge probability value in the first category branch is directly used as the confidence output of the final true discharge category. For the sub-classification probability values ​​of switching operation interference and corona discharge interference obtained in the second category branch, they are multiplied by the normalized weight from the comprehensive interference category. The sub-classification probability is then multiplied by the comprehensive interference probability to obtain the true and comparable final probabilities of switching operation interference and corona discharge interference.For each final output category, its corresponding probability value is compared with a preset confidence threshold. For example, the confidence threshold for true discharge is set to 0.85, and the thresholds for switching operation interference and corona interference are both 0.80. When the probability of a certain category exceeds its set threshold and is the maximum value among the three categories, that category is output as the final recognition result and marked as the "high confidence" category in the output. If the maximum value is less than all thresholds, a "low confidence" mark is output, indicating that there is interference that has not been identified or the signal is blurred.

[0035] In one specific embodiment, the process of performing step 300 may specifically include the following steps: Based on the classification and recognition results, the rising edge slope of the discharge pulse of the insulating bushing is calculated to obtain the rising edge slope of the discharge pulse. The aging factor is determined based on the service life of the insulating bushing and the type of insulating material, and the threshold for judging the degree of aging is dynamically adjusted by correlating the rise edge slope of the discharge pulse, so as to obtain the threshold parameter after aging correction. Based on the threshold parameter after aging correction, the pulse width of the discharge pulse is measured and a two-dimensional discrimination space coordinate system is established to obtain the two-dimensional spatial discrimination region division result. Phase and amplitude correlation are detected by using the two-dimensional spatial discrimination region division results to obtain dynamic separation parameters.

[0036] Specifically, time-domain feature extraction is performed on each effective pulse segment of the real discharge signal output by the classification and recognition module. Numerical differentiation algorithms or interpolation fitting methods are used to refine the rising segment of the discharge pulse. The signal change interval is extracted from the starting point to 90% of the peak value, and the ratio of voltage increment to time increment is calculated to obtain the rising edge slope, i.e., the rate of voltage change per unit time. Its physical meaning is the steepness of local ionization discharge or the intensity of the discharge source, characterized in V / μs. In multiple discharge pulses, the rising edge slope exhibits high dispersion and is related to the aging process. To combine the rising slope with the historical operating status of the insulating bushing and thus obtain a dynamic discrimination benchmark with engineering guidance, an aging factor modeling process is introduced. This aging factor comprehensively considers two dimensions: the service life of the insulating bushing and the material type. The service life is extracted from the field operation and maintenance database and ranges from 1 to 30 years, while the material type is extracted from the insulation process list, such as paper insulation, oil-paper composite, epoxy resin, or polymer materials. The corresponding aging response weights are determined based on their performance evolution models under electrical and thermal aging. By setting empirical or fitting weight functions, the service life and material response coefficients are weighted and combined to form a normalized aging factor α. The value of this factor ranges from 0.1 to 1.0, where α close to 1 indicates a high aging state, while a smaller α indicates relatively good insulation performance. Based on this aging factor, the system dynamically corrects the standard discharge rise edge slope threshold range. For example, if the original lower threshold is set to 8V / μs, it is corrected to α×8V / μs when α<0.5, thus widening the recognition range to improve early aging sensitivity. When α≥0.5, the standard threshold is still used to maintain discrimination accuracy, resulting in a corrected threshold parameter that adapts to aging conditions. Furthermore, the system measures the pulse width of the discharge pulse, selecting the time interval from 10% to 90% of the amplitude as the definition standard for the pulse width τ to enhance the perception of actual waveform changes. Using the aforementioned rise edge slope K and pulse width τ as two coordinate axes, a two-dimensional spatial coordinate system is constructed, where the vertical axis represents the rise edge slope and the horizontal axis represents the pulse width. All discharge pulses are then mapped onto this two-dimensional discrimination space to form a dot matrix distribution. In this two-dimensional space, different types of signals exhibit clear clustering trends. For example, the real discharge signal is located in the region with a high K value (8-25V / μs) and a small τ value (0.1-2.5μs), while the switching interference signal, due to the slow action of its mechanical contacts, exhibits a distribution with a small K value (0.5-3V / μs) and a large τ value (10-100μs). The corona discharge signal has a relatively high K value (15~40V / μs) but its τ value is in the middle transition range.By constructing a boundary model or training a two-dimensional boundary surface using supervised learning algorithms, the system achieves region partitioning based on K-τ joint features, forming discrimination intervals for real discharge, switching interference, and corona discharge. The classification effect is verified by using graphical visualization tools to check cluster density and region integrity. After the two-dimensional spatial partitioning is completed, to improve the physical consistency and engineering interpretability of signal classification, the system introduces a coupled analysis mechanism of phase and amplitude on the basis of two-dimensional discrimination. The system maps the occurrence time of each discharge pulse to the phase position of the power frequency cycle, constructing a phase angle distribution map from 0° to 360°, and combining it with its amplitude peak value to form a joint phase-amplitude feature map. In partial discharges caused by internal defects in power equipment, the signal is concentrated near the positive and negative voltage peaks, i.e., around 0°±30° and 180°±30°, showing a certain phase aggregation trend; while environmental interference signals show random phase distribution and lack periodic synchronization. Alternatively, by statistically analyzing the phase interval density corresponding to different amplitudes, a multi-dimensional discrimination rule is formed to distinguish signals with amplitudes concentrated in the peak interval (real discharge) or signals with smaller amplitudes and random phase distribution (corona interference). By detecting the correlation between phase and amplitude as described above, the system can further filter signals with blurred or overlapping boundaries in two-dimensional space, thereby enhancing the overall robustness of recognition. The system performs joint logical judgments on the two-dimensional discrimination region division results and the phase-amplitude detection results to generate dynamic separation parameters for signal separation and classification labeling.

[0037] In one specific embodiment, the process of performing phase and amplitude correlation detection based on the two-dimensional spatial discrimination region division results to obtain dynamic separation parameters can specifically include the following steps: Voltage phase angle distribution analysis was performed using the discharge pulses in the two-dimensional spatial discrimination region division results to obtain phase angle distribution characteristic data; Based on the phase angle distribution characteristic data, the amplitude and phase angle of the discharge pulse are correlated and analyzed to obtain the phase-amplitude correlation calculation results; Based on the phase-amplitude correlation calculation results, the signal statistical characteristics are analyzed and the detection sensitivity parameters are dynamically updated using an adaptive algorithm to obtain the detection parameters after sensitivity optimization. By integrating the sensitivity-optimized detection parameters with the aging-corrected threshold parameters and the two-dimensional spatial discrimination region parameters, dynamic separation parameters that include slope threshold, pulse width range, and phase distribution characteristics are obtained.

[0038] Specifically, based on the two-dimensional spatial discrimination region division model, all discharge pulse signals that conform to the true discharge discrimination characteristics are mapped to the phase angle coordinate system of the power frequency cycle. By periodically synchronizing the pulse occurrence time with the power frequency voltage cycle, the phase position of each discharge pulse within the voltage cycle from 0° to 360° is calculated, thereby forming a phase angle distribution dataset. The phase position, corresponding amplitude, and the two-dimensional spatial region number of each signal are uniformly incorporated into the phase angle distribution feature matrix. This matrix, through statistical analysis, yields parameters such as the frequency distribution, density peak, and extreme value concentration of discharge pulses in different phase intervals, characterizing the synchronous behavior and aggregation trend of discharge events within the voltage cycle. Based on phase angle distribution characteristic data, a joint analysis of discharge pulse amplitude and phase angle is performed. A two-dimensional joint density estimation method or a kernel function-based distribution fitting algorithm is used to construct a two-dimensional amplitude-phase mapping. By analyzing the signal density, offset direction, and uniformity in the phase space at different amplitude levels, the system identifies whether high-amplitude discharges are concentrated in the peak region of the voltage waveform (0°±30° or 180°±30°). This characteristic provides important evidence for genuine discharges caused by internal insulation defects. Conversely, if the discharge pulse amplitude is generally small and uniformly distributed throughout the phase space, it is inferred that the signal belongs to non-structural discharge signals such as corona discharge or environmental interference. A statistical model of the amplitude-phase joint distribution is established to quantify key indicators such as the phase concentration coefficient, peak position offset, and phase-amplitude correlation coefficient. Based on the phase-amplitude correlation calculation results, signal statistical characteristics are analyzed, and an adaptive algorithm is used to dynamically update the detection sensitivity parameters. The system analyzes the distribution of discharge pulses in a multi-dimensional feature space within the current monitoring period, including parameters such as signal quantity, average amplitude, standard deviation, phase concentration, and the proportion of abnormal slopes. Based on these statistical indicators, the system determines whether the current detection environment is in a stage of high interference, high-frequency discharge, or increased signal fluctuation. Accordingly, it calls an adaptive algorithm module to dynamically update the detection sensitivity parameters. This algorithm records the characteristic statistical change trend of the signal over a continuous time period using a sliding window structure, and automatically corrects the threshold enhancement coefficient, boundary stretching coefficient, or feature enhancement weight in the current discharge identification model by combining the coefficient of variation (CV), the proportion of outliers, and the variance level of the abnormal signal distribution. For example, if a large number of signals with low amplitude but highly concentrated phases are detected in a certain period, the system will increase the weight of the low-amplitude signal identification channel to prevent missed detections; conversely, when the interference signal fluctuates violently or the slope characteristics show a randomized trend, the sensitivity will be reduced to prevent false judgments. The adaptive module outputs a set of sensitivity-optimized detection parameters, including threshold correction factors, identification boundary adjustment coefficients, and low-amplitude compensation factors.After all the basic modules have been run, the system integrates the sensitivity-optimized detection parameters, the aging-corrected threshold parameters, and the two-dimensional spatial discrimination region parameters to construct a dynamic separation parameter set for signal discrimination and feature classification. The slope threshold is generated by the aging factor correction model, the pulse width range is output by the two-dimensional spatial modeling, the phase distribution characteristics are calculated by the phase density statistics, and the sensitivity control parameters are derived from the dynamic adjustment results of the adaptive algorithm.

[0039] In one specific embodiment, the process of performing step 400 may specifically include the following steps: A partial discharge signal feature library for insulating bushings was constructed based on dynamic separation parameters and classified into five levels according to the degree of aging. It was also classified into four categories according to the defect type: internal bubble discharge, layered discharge, surface discharge, and floating discharge, resulting in a multi-level feature library classification structure. The pulse rise time, pulse decay time, spectral center frequency, spectral bandwidth, phase concentration, pulse polarity, pulse interval statistical characteristics, harmonic distortion, signal energy distribution, time-domain waveform similarity, frequency-domain power spectral density, and time-frequency joint feature fingerprints are extracted from the partial discharge signal to obtain the set of characteristic parameters of the partial discharge signal. The feature parameter set of partial discharge signal is combined with the multi-level feature library classification structure and the support vector machine algorithm is used for differential identification. Radial basis functions are used to establish a separation hyperplane between discharge signal and interference signal to obtain feature classification data.

[0040] Specifically, based on dynamic separation parameters, separation criteria are established with slope threshold, pulse width range, and phase concentration distribution as the core. Based on these criteria, representative partial discharge signal samples are selected from historical monitoring samples and high-confidence labeled signals to construct a standardized feature library system for insulation aging states and defect modes. During the feature library construction phase, the system uses the aging factor α as an index variable to classify all partial discharge signals into five levels according to their aging degree: mild aging (α = 0.1–0.3), moderate aging (α = 0.3–0.5), severe aging (α = 0.5–0.7), critical aging (α = 0.7–0.9), and extreme aging (α = 0.9–1.0). Simultaneously, based on the combined characteristics of high-frequency current and ultrasonic signals, the symmetry of discharge location, the consistency of discharge polarity, and signal clustering distribution, the system classifies partial discharge signals into four major categories according to defect type: internal bubble discharge, layered discharge, surface discharge, and suspended discharge. Internal bubble discharge exhibits high amplitude, strong phase concentration, and fast rise time characteristics; layered discharge is accompanied by frequency drift and multi-peak characteristics; surface discharge signals show periodic slippage and strong amplitude fluctuations; and suspended discharge has unstable phase distribution and random discharge intervals. Through cross-organization in both horizontal and vertical dimensions, a multi-level feature library classification structure containing 5×4 subcategories is formed. Each subcategory includes a large number of standard signal samples, each sample is accompanied by label information, and also includes multi-dimensional feature parameters and source data indexes. The system performs refined feature extraction on the partial discharge signals identified within the current monitoring period. The extracted features include twelve categories of parameters across three domains: time domain, frequency domain, and combined time-frequency domain. Among these, the pulse rise time characterizes the steepness change during the discharge signal triggering process and has significant discriminative power for identifying the actual discharge type; the pulse decay time reflects the time required for the signal to recover to a static state after discharge; the spectral center frequency and spectral bandwidth are obtained through Fast Fourier Transform (FFT) to determine the location and distribution range of the main energy concentration in the frequency domain; the phase concentration is obtained by fitting the signal density function within the power frequency cycle and is used to measure whether the signal is associated with the power frequency voltage peak; and the pulse polarity is used to determine whether the discharge is positive or negative. Polarity breakdown; pulse interval statistical characteristics such as mean, variance, and range reflect the stability of discharge behavior; harmonic distortion describes whether the signal is affected by power supply harmonics or whether there is secondary ionization activity; signal energy distribution reflects the overall density of discharge intensity; time-domain waveform similarity is used to match and compare with feature library samples to determine whether it belongs to a typical pattern; frequency-domain power spectral density reflects the distribution of signal energy in the frequency dimension; and time-frequency joint feature fingerprint is constructed by wavelet packet decomposition, short-time Fourier or Hilbert transform to construct a three-dimensional feature template. This feature has strong uniqueness and high confidence matching ability. The above twelve dimensions of parameters together constitute the feature parameter set of partial discharge signal.The signal feature parameter set is input into a differential recognition module based on the support vector machine algorithm. This module loads a multi-level feature library classification structure and completes the classification boundary fitting process through sample training. The recognition model uses radial basis functions as kernel functions to ensure the boundary construction capability of nonlinearly separable data in high-dimensional space. A multi-hyperplane model is jointly constructed by maximizing inter-class margin and minimizing classification error rate. During model training, confidence enhancement strategies and Mahalanobis distance are used as auxiliary screening mechanisms to ensure that signals near the boundary can be finely labeled. In the final classification stage, the system calculates the high-dimensional spatial distance between the input signal and the center samples of all subclasses. The similarity between the signal and each class in the feature library is comprehensively evaluated by using support vector distance and radial basis function output weights, and then the most matching class label is selected as the final output. Meanwhile, to achieve effective signal filtering, an interference removal mechanism is set up in the model, that is, a separation hyperplane is constructed between the discharge signal and the interference signal. During the model training phase, high-confidence noise samples and typical interference samples are used as negative class inputs to construct a binary classification adversarial sample set with real discharge samples, so that the RBF kernel function can form a high-dimensional separation boundary in the feature space. In the actual inference process, the distance metric and kernel value output are used to determine whether the signal falls into the discharge area or the interference area, thereby achieving reliable identification and removal of discharge-interference. Finally, the feature classification data output by the system includes discharge type label, aging level, signal strength index and confidence score, and generates two-dimensional mapping coordinates, phase distribution map and power spectral density map for visualization analysis.

[0041] In one specific embodiment, the process of performing step 500 may specifically include the following steps: Signals are grouped based on feature classification data to obtain signal grouping results. Then, the partial discharge signals in each group are clustered according to the signal grouping results to obtain phase clustering grouping results. Based on the phase clustering grouping results, the step response characteristics of the discharge pulse are analyzed and the characteristic parameters of the discharge pulse step response, including the initial amplitude, decay time constant, oscillation frequency and initial phase, are established. The discharge pulse step response characteristic parameters are geometrically corrected based on the ratio of the inner and outer diameters of the insulating bushing to obtain the geometrically corrected step response parameters. The aging state variation coefficient of the insulating bushing is then generated based on the geometrically corrected step response parameters.

[0042] Specifically, based on the classified discharge signal data, the monitored signals are segmented and managed in chronological order. A fixed-duration time window is selected, for example, every 5 minutes as a unit. All discharge pulse signals collected within this time period are grouped together, thus classifying the discharge behavior by time. This facilitates the analysis of differences in the operating state of the insulation system at different times and supports continuous statistical analysis. Partial discharge signals within each time group are divided according to their phase angle information appearing in the voltage cycle. Each discharge pulse has a specific trigger time, which can be aligned with the power frequency cycle and thus mapped to a phase angle. The system uses these phase angles as the basis for cluster analysis, employing a clustering algorithm to aggregate them based on similarity, grouping pulse signals appearing at similar voltage phase positions into the same category, thereby obtaining the phase clustering grouping results for each signal group. A time group contains multiple phase clusters, each cluster representing a set of discharge signals with a common physical cause or the same discharge source characteristics. After obtaining the signal set after phase clustering, the system performs dynamic behavior modeling on the waveform of each group of discharge pulses, paying particular attention to the waveform changes after discharge triggering. By analyzing the waveform's rise and subsequent curves, the system extracts the response characteristics of each discharge pulse. These characteristics include the initial discharge amplitude, the rate of energy decay, the waveform's oscillation frequency, and the phase angle corresponding to the waveform's initial moment. The combination of these parameters describes the temporal variation of a discharge event and is a crucial indicator for assessing the stability and aging characteristics of discharge behavior. In equipment experiencing accelerated aging, this manifests as gradual fluctuations in discharge amplitude, faster or slower decay rates, and shifts in oscillation frequency. Due to differences in the structural dimensions of various insulating bushings, the system incorporates the bushing's structural parameters, particularly the ratio of its inner to outer diameter, to perform structural compensation on the aforementioned discharge response parameters. The system standardizes and corrects the characteristic parameters of the discharge signal based on the ratio of the inner and outer diameters to compensate for the influence of structural dimensions on the discharge signal's response behavior. For example, in thicker insulating bushings, the discharge signal propagation time is longer and the signal energy decays faster; conversely, the opposite is true in thinner structures. After correction, each discharge pulse possesses a set of standardized characteristic parameters that can be compared across structures. The system then performs time-series analysis on these geometrically corrected response characteristics. The system statistically calculates the discharge characteristics of clusters with the same phase within a continuous time window, focusing on analyzing the stability and trends of each characteristic over multiple time periods. The system calculates the concentration and fluctuation of each characteristic to measure the stability of the discharge behavior. For example, if the discharge amplitude, decay rate, and frequency characteristics fluctuate little within a certain time period, it indicates stable discharge behavior and minimal interference, suggesting that the insulation state is in a stable phase. Conversely, if the same signal exhibits drastic changes in different time periods, especially abrupt changes or periodic fluctuations, it indicates that the aging process of the insulation system has entered an accelerated phase.To quantify this behavior, the system generates an index representing the degree of fluctuation in discharge characteristics based on these statistical results. This index is used to describe the dynamic stability of the insulation system during the aging process; it is called the aging state variation coefficient.

[0043] In one specific embodiment, the process of performing geometric correction on the discharge pulse step response characteristic parameters based on the ratio of the inner and outer diameters of the insulating bushing to obtain geometrically corrected step response parameters, and generating the aging state variation coefficient of the insulating bushing based on the geometrically corrected step response parameters, can specifically include the following steps: The correction factor value for different geometries is determined based on the ratio of the inner and outer diameters of the insulating sleeve. The correction coefficient is multiplied with the initial amplitude, decay time constant, oscillation frequency and initial phase in the characteristic parameters of the discharge pulse step response to obtain the geometrically corrected step response parameters. The ratio of the standard deviation to the mean of the signal characteristic parameters is calculated based on the geometrically corrected step response parameters to obtain the coefficient of variation. When the coefficient of variation is less than the first target value, it is determined to be a stable discharge mode; when the coefficient of variation is between the first and second target values, it is determined to be a transitional state; when the coefficient of variation is greater than the second target value, it is determined to be an unstable mode, and finally the aging state coefficient of variation of the insulating sleeve is generated.

[0044] Specifically, the system collects the geometric data of each insulating sleeve, extracts its inner and outer diameter values, and calculates its inner-outer diameter ratio. Based on empirical classification and signal propagation model analysis, this ratio is divided into several structural intervals. For example, when the ratio is less than a certain fixed value, it is considered a thick-walled structure, indicating a longer signal propagation path, greater loss, and significant reflection interference. When the ratio is in the middle interval, it is defined as a standard structure, whose response behavior is close to the typical propagation model. When the ratio is higher than a certain threshold, it is considered a thin-walled structure, with a short signal transmission distance, fast response speed, and more complete preservation of high-frequency components in the discharge signal. Based on this classification result, the system configures a dedicated correction coefficient for each structural interval. The system couples the above structural correction coefficient with step response characteristic parameters. These response parameters include four items: initial amplitude, decay time constant, oscillation frequency, and initial phase. Each parameter reflects the signal response characteristics in the time or frequency dimension and is related to the physical behavior of partial discharge. The initial amplitude reflects the level of energy released during discharge, the decay time represents the rate at which the signal strength diminishes over time, the oscillation frequency reflects the inductance-capacitance effect present in the system, and the initial phase represents the starting angle of the signal response within the power frequency cycle. These parameters exhibit inconsistent amplitude and frequency characteristics in different bushing structures due to factors such as propagation path, impedance matching, and boundary reflections. Therefore, the system obtains the corrected parameter values ​​by multiplying each parameter by its corresponding structural interval correction coefficient. Using the geometrically corrected parameters as the analysis object, statistical characteristics are extracted over multiple continuous time windows. For each type of corrected parameter, if multiple response data points are recorded within a specific time period, the system calculates its mean and standard deviation, thereby obtaining the fluctuation intensity and concentration of that type of parameter. By comparing the standard deviation with the mean, the relative variation level of each type of response parameter is obtained; this ratio is the coefficient of variation. The coefficient of variation (COP) reflects the stability of the discharge pulse within a specific time interval. A small COP indicates that the signal variation within that period is small and concentrated, suggesting stable discharge behavior, which is related to the early stage of structural aging or stable discharge source. A moderate COP indicates that the signal begins to fluctuate, with the possibility of structural stress changes, electric field disturbances, or localized dielectric degradation. A significantly increased COP indicates that the discharge signal has entered a highly unstable state, with high-risk situations such as multi-source superposition, defect propagation, and dielectric breakdown edge. A hierarchical judgment rule is introduced into the COP calculation system to clearly classify the discharge state within the current time period. The judgment rule sets two target values ​​as discrimination boundaries: the first target value represents the upper limit between stable and transitional states, and the second target value represents the boundary between transitional and unstable states.When the coefficient of variation within a certain time period is lower than the first target value, the system determines that the discharge behavior of that data segment is in a stable mode, meaning that the insulation structure is in good working condition, the signal is highly concentrated, and the changes are controllable. When the coefficient of variation is between the two target values, the system identifies it as a transitional state, indicating that the discharge signal fluctuation is rising but has not yet shown abnormal divergence. At this time, the equipment is in a stage of local stress redistribution or pre-aging criticality. When the coefficient of variation exceeds the second target value, the system issues an unstable mode flag, indicating that the current discharge behavior is in a stage of abnormal diffusion, high-frequency fluctuation, or multi-point triggering. Based on the determination results of all time periods, the system summarizes and generates the coefficient of variation of aging status for the current insulating bushing, and outputs the corresponding aging level classification, trend indicators, and risk level recommendations.

[0045] The above describes the multi-parameter insulating bushing aging performance testing method in the embodiments of the present invention. The following describes the multi-parameter insulating bushing aging performance testing system in the embodiments of the present invention. Please refer to [link / reference]. Figure 2 One embodiment of the multi-parameter insulating sleeve aging performance testing system of the present invention includes: The signal acquisition module is used to acquire multi-parameter detection signals from the insulating bushing and obtain standardized signal data. The feature analysis module is used to input standardized signal data into the coupled recognition network for feature analysis to obtain classification and recognition results; The threshold adjustment module is used to adjust the aging factor threshold of the discharge pulse rising edge slope of the insulating bushing based on the classification and recognition results, so as to obtain dynamic separation parameters. The differential recognition module is used to perform feature space matching and differential recognition on partial discharge signals based on dynamic separation parameters to obtain feature classification data; The clustering analysis module is used to perform phase clustering analysis and step response feature analysis based on feature classification data to obtain the aging state variation coefficient of the insulating bushing.

[0046] Through the collaborative efforts of the aforementioned components, a four-dimensional parameter detection matrix is ​​constructed using a high-frequency current sensor, an ultrasonic sensor, a high-frequency voltage sensor, and an infrared temperature sensor. This enables the synchronous acquisition and correlation analysis of multiple physical quantities, overcoming the limitations of incomplete information from traditional single-sensor detection. A coupled recognition network is employed to achieve cross-coupling analysis of time-frequency domain features. By establishing a correlation weight matrix between parameters, the limitations of independent analysis of each detection parameter in existing technologies are overcome, allowing the system to automatically learn and recognize complex signal feature patterns, significantly enhancing its ability to identify real discharge signals. An adaptive feature extraction controller based on the discharge pulse rise edge slope dynamically adjusts the discrimination threshold according to the aging degree of the insulating bushing, effectively distinguishing real discharge signals from switching operation interference and corona discharge interference, solving the problem of decreased recognition accuracy under different aging conditions using traditional fixed-parameter methods. A differential recognition model established through a support vector machine algorithm, combined with Mahalanobis distance multi-dimensional feature space evaluation, achieves accurate separation of discharge signals and interference signals, providing more reliable recognition results than traditional single-feature matching and effectively reducing the false positive rate. By employing PD-Regroup grouping identification and Step-PD-Function step response characteristic function analysis, a quantitative evaluation index based on the coefficient of variation was established, achieving a technological leap from qualitative analysis to quantitative evaluation and providing an accurate numerical characterization of the aging state of insulating bushings. By introducing a geometric correction coefficient for the ratio of the inner and outer diameters of the insulating bushing, the detection method can be adapted to insulating bushings of different specifications and structures, enhancing its versatility and applicability, and solving the problem of the lack of specificity in existing technologies. High identification accuracy is maintained even in complex electromagnetic environments. Through genetic algorithm parameter optimization and multiple verification mechanisms, the stability and reliability of the system under strong interference conditions are ensured, significantly outperforming the anti-interference capabilities of existing technologies.

[0047] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0048] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0049] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for testing the aging performance of multi-parameter insulating bushings, characterized in that, include: Multi-parameter detection signals were acquired from the insulating bushing to obtain standardized signal data; The standardized signal data is input into a coupled recognition network for feature analysis to obtain classification and recognition results. Based on the classification and identification results, the aging factor threshold is adjusted for the rising edge slope of the discharge pulse of the insulating bushing to obtain dynamic separation parameters; Based on the dynamic separation parameters, feature space matching and differential identification are performed on the partial discharge signal to obtain feature classification data; Phase clustering analysis and step response feature analysis are performed based on the feature classification data to obtain the aging state variation coefficient of the insulating bushing.

2. The multi-parameter insulating bushing aging performance testing method according to claim 1, characterized in that, The process of acquiring multi-parameter detection signals from the insulating bushing to obtain standardized signal data includes: The insulating bushing is simultaneously acquired using a current sensor, an ultrasonic sensor, a voltage sensor, and an infrared temperature sensor to obtain the original multi-parameter detection signal. The original multi-parameter detection signal input signal conditioning circuit is subjected to amplitude linear transformation and 16-bit precision quantization to obtain the conditioned signal data. The conditioned signal data is digitally filtered based on the dielectric constant range and inner diameter geometry parameters of the insulating sleeve to obtain filtered signal data. The filtered signal data is then time-referenced to obtain standardized signal data.

3. The multi-parameter insulating sleeve aging performance testing method according to claim 1, characterized in that, The step of inputting the standardized signal data into a coupled recognition network for feature analysis to obtain classification and recognition results includes: The standardized signal data is input into the feature preprocessing layer of the coupled recognition network and subjected to matrix transformation to obtain a multi-dimensional input feature matrix. The multi-dimensional input feature matrix is ​​respectively input into the time-domain feature extraction subnetwork, frequency-domain feature extraction subnetwork and time-frequency joint feature extraction subnetwork of the coupled recognition network for parallel feature extraction, resulting in three-way parallel feature extraction results. Through the feature fusion coupling layer of the coupled recognition network, the three parallel feature extraction results are fused with cross-channel attention weight calculation and multi-head self-attention mechanism to obtain a fused coupled feature vector. The fused coupled feature vector is input into the classification decision layer for multi-level classification to obtain classification and recognition results that include real discharge, switching operation interference and corona discharge interference.

4. The multi-parameter insulating sleeve aging performance test method according to claim 3, characterized in that, The process involves inputting the fused coupled feature vector into a classification decision layer for multi-level classification to obtain classification and recognition results that include real discharge, switching operation interference, and corona discharge interference, including: The fused coupling feature vector is input into the first classification branch of the classification decision layer to perform binary classification of real discharge signals and interference signals, and the intermediate binary classification results of the probability values ​​of real discharge signals and comprehensive interference signals are obtained. The comprehensive interference signal probability value in the intermediate binary classification result is input into the second classification branch of the classification decision layer to identify the sub-types of switching operation interference and corona discharge interference, thereby obtaining the sub-classification results of the probability values ​​of switching operation interference and corona discharge interference. Based on the intermediate results of the binary classification and the results of the fine classification, the probability distribution is recalculated and the confidence threshold is judged to obtain the classification and identification results that include real discharge, switching operation interference and corona discharge interference.

5. The multi-parameter insulating bushing aging performance test method according to claim 1, characterized in that, The dynamic separation parameters are obtained by adjusting the aging factor threshold of the discharge pulse rising edge slope of the insulating bushing based on the classification and identification results, including: Based on the classification and identification results, the rising edge slope of the discharge pulse of the insulating sleeve is calculated to obtain the rising edge slope of the discharge pulse. The aging factor is determined based on the service life of the insulating bushing and the type of insulating material, and the threshold for judging the degree of aging is dynamically adjusted by relating the rising edge slope of the discharge pulse to the degree of aging, so as to obtain the threshold parameter after aging correction. Based on the aging-corrected threshold parameters, the pulse width of the discharge pulse is measured and a two-dimensional discrimination space coordinate system is established to obtain the two-dimensional spatial discrimination region division result. Phase and amplitude correlation are detected using the two-dimensional spatial discrimination region division results to obtain dynamic separation parameters.

6. The multi-parameter insulating sleeve aging performance test method according to claim 5, characterized in that, The step of performing phase and amplitude correlation detection based on the two-dimensional spatial discrimination region division results to obtain dynamic separation parameters includes: Voltage phase angle distribution analysis is performed using the discharge pulses in the two-dimensional spatial discrimination region division results to obtain phase angle distribution characteristic data; Based on the phase angle distribution characteristic data, the amplitude and phase angle of the discharge pulse are correlated and analyzed to obtain the phase-amplitude correlation calculation results. Based on the phase-amplitude correlation calculation results, signal statistical characteristics are analyzed and an adaptive algorithm is used to dynamically update the detection sensitivity parameters to obtain the detection parameters after sensitivity optimization. The sensitivity-optimized detection parameters are integrated with the aging-corrected threshold parameters and the two-dimensional spatial discrimination region parameters to obtain dynamic separation parameters that include slope threshold, pulse width range, and phase distribution characteristics.

7. The multi-parameter insulating bushing aging performance test method according to claim 1, characterized in that, The step of performing feature space matching and differential recognition on the partial discharge signal based on the dynamic separation parameters to obtain feature classification data includes: Based on the dynamic separation parameters, a partial discharge signal feature library of the insulating bushing is constructed and divided into five levels according to the degree of aging. At the same time, it is divided into four categories according to the defect type: internal bubble discharge, layered discharge, surface discharge and floating discharge, resulting in a multi-level feature library classification structure. The pulse rise time, pulse decay time, spectral center frequency, spectral bandwidth, phase concentration, pulse polarity, pulse interval statistical characteristics, harmonic distortion, signal energy distribution, time-domain waveform similarity, frequency-domain power spectral density, and time-frequency joint feature fingerprint are extracted from the partial discharge signal to obtain a set of characteristic parameters of the partial discharge signal. The set of feature parameters of the partial discharge signal is subjected to differential identification using the support vector machine algorithm and the multi-level feature library classification structure. A radial basis function is then used to establish a hyperplane for separating the discharge signal and the interference signal to obtain feature classification data.

8. The multi-parameter insulating bushing aging performance test method according to claim 1, characterized in that, The phase clustering analysis and step response feature analysis based on the feature classification data are used to obtain the aging state variation coefficient of the insulating bushing, including: Based on the feature classification data, the signals are grouped to obtain the signal grouping results. Then, based on the signal grouping results, the partial discharge signals in each group are clustered to obtain the phase clustering grouping results. Based on the phase clustering grouping results, the step response characteristics of the discharge pulse are analyzed and the characteristic parameters of the discharge pulse step response, including the initial amplitude, decay time constant, oscillation frequency and initial phase, are established. The discharge pulse step response characteristic parameters are geometrically corrected based on the ratio of the inner and outer diameters of the insulating sleeve to obtain geometrically corrected step response parameters, and the aging state variation coefficient of the insulating sleeve is generated based on the geometrically corrected step response parameters.

9. The multi-parameter insulating bushing aging performance test method according to claim 8, characterized in that, The step of geometrically correcting the discharge pulse step response characteristic parameters based on the ratio of the inner and outer diameters of the insulating sleeve to obtain geometrically corrected step response parameters, and generating the aging state variation coefficient of the insulating sleeve based on the geometrically corrected step response parameters, includes: The correction factor value for different geometries is determined based on the ratio of the inner and outer diameters of the insulating sleeve. The correction coefficient value is multiplied by the initial amplitude, decay time constant, oscillation frequency and initial phase in the discharge pulse step response characteristic parameters to obtain the geometrically corrected step response parameters. Based on the geometrically corrected step response parameters, the ratio of the standard deviation to the mean of the signal characteristic parameters is calculated to obtain the coefficient of variation. When the coefficient of variation is less than the first target value, it is determined to be a stable discharge mode; when the coefficient of variation is between the first target value and the second target value, it is determined to be a transitional state; when the coefficient of variation is greater than the second target value, it is determined to be an unstable mode, and finally the aging state coefficient of variation of the insulating sleeve is generated.

10. A multi-parameter insulating bushing aging performance testing system, characterized in that, For performing the multi-parameter insulating bushing aging performance test method as described in any one of claims 1-9, the multi-parameter insulating bushing aging performance test system comprises: The signal acquisition module is used to acquire multi-parameter detection signals from the insulating bushing and obtain standardized signal data. The feature analysis module is used to input the standardized signal data into the coupled recognition network for feature analysis to obtain classification and recognition results; The threshold adjustment module is used to adjust the aging factor threshold of the discharge pulse rising edge slope of the insulating bushing based on the classification and recognition results, so as to obtain dynamic separation parameters. The differential recognition module is used to perform feature space matching and differential recognition on the partial discharge signal according to the dynamic separation parameters to obtain feature classification data; The clustering analysis module is used to perform phase clustering analysis and step response feature analysis based on the feature classification data to obtain the aging state variation coefficient of the insulating bushing.

Images