Partial discharge signal identification method, device, equipment, medium and program product

Abstract

The invention relates to a partial discharge signal identification method, device and equipment, a medium and a program product. The method comprises the following steps: acquiring original partial discharge pulse data of to-be-detected equipment; inputting the original partial discharge pulse data into a pre-trained partial discharge signal enhancement model to obtain enhanced partial discharge pulse data; matching the enhanced partial discharge pulse data with a standard pulse signal to obtain candidate partial discharge pulse data; acquiring response feature information of the to-be-detected equipment; and inputting the candidate partial discharge pulse data and the response feature information into a partial discharge identification model, respectively executing feature extraction to obtain a to-be-detected partial discharge pulse feature and a response distribution feature, and obtaining a target response identification result of the to-be-detected device based on the to-be-detected partial discharge pulse feature and the response distribution feature. By adopting the method, multi-dimensional features such as a topological structure and an operation state of the equipment can be fused, the identification accuracy is high, and the anti-interference capability is strong.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, medium, and program product for partial discharge signal identification. Background Technology

[0002] Partial discharge is a significant indicator of insulation degradation in power equipment, and accurate detection and identification of partial discharge signals are crucial for safe equipment operation. Partial discharge signals acquired on-site are often interfered with by strong background noise. Traditional signal enhancement methods, such as filtering techniques, struggle to effectively separate weak partial discharge pulses from complex noise, resulting in limited improvement in the signal-to-noise ratio. Furthermore, partial discharge identification often relies on manual experience or simple pattern recognition algorithms, making it difficult to integrate multi-dimensional features such as equipment topology and operating status, leading to low accuracy and weak anti-interference capabilities. Summary of the Invention

[0003] Therefore, it is necessary to provide a partial discharge signal identification method, device, equipment, medium, and program product that can integrate multi-dimensional features such as device topology and operating status, and has high identification accuracy and strong anti-interference capability to address the above-mentioned technical problems.

[0004] Firstly, this application provides a method for identifying partial discharge signals. The method includes:

[0005] Acquire the raw partial discharge pulse data of the device under test;

[0006] The original partial discharge pulse data is input into a pre-trained partial discharge signal enhancement model to obtain enhanced partial discharge pulse data;

[0007] The enhanced partial discharge pulse data is matched with a standard pulse signal to obtain candidate partial discharge pulse data;

[0008] Obtain the response feature information of the device under test;

[0009] The candidate partial discharge pulse data and the response feature information are input into the partial discharge identification model, and feature extraction is performed to obtain the partial discharge pulse features to be detected and the response distribution features. Based on the partial discharge pulse features to be detected and the response distribution features, the target response identification result of the device to be detected is obtained.

[0010] In some embodiments of the method, matching the enhanced partial discharge pulse data with a standard pulse signal to obtain candidate partial discharge pulse data includes:

[0011] The original partial discharge pulse data and the standard pulse signal are time-domain aligned and matched to obtain the target signal segment in the candidate sampling point set of the original partial discharge pulse data;

[0012] By comparing each pulse on the target signal segment with the corresponding pulse on the standard pulse signal, the background pulse and partial discharge pulse in the target signal segment are determined.

[0013] Based on the standard pulse signal, the background pulse, and the partial discharge pulse, construct the discharge type time-series distribution and pulse time stamp distribution of the device under test;

[0014] Based on the standard pulse timestamp in the standard pulse signal, the discharge type time sequence distribution and the pulse time stamp distribution are mapped in the time domain to obtain the candidate partial discharge pulse data.

[0015] In some embodiments of the method, the partial discharge pulse includes at least one of ionizing discharge characteristic pulses, intermittent interruption characteristic pulses, and phase distortion characteristic pulses; the discharge type time series distribution includes a standard power frequency pulse time series distribution and a discharge characteristic topology; the pulse time stamp distribution includes a set of periodic phase coordinates and a standard pulse time stamp distribution; and constructing the discharge type time series distribution and pulse time stamp distribution of the device under test based on the standard pulse signal, the background pulse, and the partial discharge pulse includes:

[0016] The discharge spectrum identifier of each pulse in the standard pulse signal corresponding to the target signal segment and the free event marker of each free discharge characteristic pulse are integrated by pulse timing to obtain the standard power frequency pulse timing distribution.

[0017] The discharge characteristic topology is obtained by integrating the discharge power spectrum identifier of the background pulse, the discharge power spectrum identifier of the phase distortion characteristic pulse, the discharge power spectrum identifier of the ionization discharge characteristic pulse, and the oscillation missing flag of each intermittent interruption characteristic pulse through pulse timing.

[0018] The periodic phase coordinates corresponding to each pulse in the discharge characteristic topology and the periodic phase coordinates of each intermittent interruption characteristic pulse are integrated by pulse timing to obtain the set of periodic phase coordinates. In addition, the standard pulse timestamps of each pulse in the discharge characteristic topology and the standard pulse timestamps of each intermittent interruption characteristic pulse are integrated by pulse timing to obtain the standard pulse timestamp distribution.

[0019] In some embodiments of the method, obtaining the response feature information of the device to be detected includes:

[0020] Obtain the device topology description, dynamic measurement point parameters, and infrared thermal image features of the device under test;

[0021] According to the preset topology coding mapping, the topology description of the device is parametrically coded to obtain the topology feature index of the device to be detected;

[0022] The infrared thermal image features are extracted to obtain the temperature field distribution spectrum of the infrared thermal image features. The dynamic measurement point parameters and the temperature field distribution spectrum are then fused using multi-source features to obtain the state feature vector of the device to be detected.

[0023] The response feature information is obtained based on the topological feature index and the state feature vector.

[0024] In some embodiments of the method, the partial discharge identification model includes a feature extraction module, a feature fusion module, and a state classification module. The state classification module includes a feature aggregation component and a state predictor. The step of inputting the candidate partial discharge pulse data and the response feature information into the partial discharge identification model, performing feature extraction to obtain the partial discharge pulse features to be detected and the response distribution features, and obtaining the target response identification result of the device to be detected based on the partial discharge pulse features to be detected and the response distribution features includes:

[0025] Based on the feature extraction module, the detection partial discharge pulse features of the candidate partial discharge pulse data are extracted, and the response distribution features of the response feature information are extracted;

[0026] Based on the feature fusion module, the partial discharge pulse feature to be detected and the response distribution feature are concatenated, and the feature space of the concatenated feature is expanded to obtain a multi-source discharge feature field.

[0027] Based on the feature aggregation component, the feature space of the multi-source discharge feature field is compressed to obtain the one-dimensional partial discharge pulse feature of the device under test.

[0028] Based on the state predictor, the one-dimensional partial discharge pulse features to be detected are refined to predict the state attribution confidence of each candidate response feature in the set of candidate response feature information of the device to be detected. The candidate response feature information whose state attribution confidence is at the state determination threshold is determined as the target response identification result of the device to be detected.

[0029] In some embodiments of the method, the partial discharge signal enhancement model is trained through the following steps:

[0030] A high signal-to-noise ratio (SNR) clean partial discharge pulse sample set is obtained, which includes clean partial discharge pulse samples of various known discharge types and corresponding intensity levels.

[0031] Obtain a real background noise sample set, which includes background noise samples without partial discharge activity collected from a complex acoustic environment similar to the device under test;

[0032] The pure partial discharge pulse samples in the high signal-to-noise ratio pure partial discharge pulse sample set are scaled according to different preset amplitude ratios to obtain scaled partial discharge pulse samples of different intensities.

[0033] Each of the scaled partial discharge pulse samples is mixed with a background noise sample randomly selected from the real background noise sample set at different preset superposition ratios to generate a mixed sample of partial discharge pulses simulating different signal-to-noise ratio conditions.

[0034] The partial discharge pulse mixture sample is used as the input sample, and the corresponding unscaled high signal-to-noise ratio pure partial discharge pulse sample is used as the target output sample to form the partial discharge signal enhancement training sample set.

[0035] The original partial discharge signal enhancement model is trained using the aforementioned partial discharge signal enhancement training sample set. The model parameters are adjusted by optimizing a preset loss function, enabling the original partial discharge signal enhancement model to learn the mapping relationship from noisy partial discharge pulses to clean partial discharge pulses, thereby achieving effective enhancement of input low signal-to-noise ratio partial discharge pulses, and obtaining the trained partial discharge signal enhancement model.

[0036] According to a second aspect of the present disclosure, a partial discharge signal identification device is provided. The device includes:

[0037] The first module is used to acquire the raw partial discharge pulse data of the device under test;

[0038] The second module is used to input the original partial discharge pulse data into a pre-trained partial discharge signal enhancement model to obtain enhanced partial discharge pulse data;

[0039] The third module is used to match the enhanced partial discharge pulse data with a standard pulse signal to obtain candidate partial discharge pulse data.

[0040] The fourth module is used to acquire the response characteristic information of the device under test;

[0041] The fifth module is used to input the candidate partial discharge pulse data and the response feature information into the partial discharge identification model, perform feature extraction respectively to obtain the partial discharge pulse features to be detected and the response distribution features, and obtain the target response identification result of the device to be detected based on the partial discharge pulse features to be detected and the response distribution features.

[0042] According to a third aspect of the present disclosure, a computer device is provided. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the partial discharge signal identification method described above.

[0043] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the partial discharge signal identification method described above.

[0044] According to a fifth aspect of the present disclosure, a computer program product is provided. The computer program product includes a computer program that, when executed by a processor, implements the partial discharge signal identification method described above.

[0045] The partial discharge (PD) signal identification scheme provided in this application acquires the original PD pulse data of the device under test, improves the signal-to-noise ratio (SNR) using a pre-trained PD signal enhancement model, and obtains enhanced PD pulse data. Subsequently, the enhanced signal is matched with a standard pulse signal to extract candidate PD pulse data containing pulse timestamps and discharge types. Simultaneously, the device's response feature information is collected. Finally, the candidate PD pulse data and response feature information are input into the PD identification model, and the target response identification result is obtained through feature extraction and fusion. This scheme effectively enhances and accurately identifies PD signals under low SNR conditions using deep learning technology, improving the reliability of power equipment insulation status assessment.

[0046] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0047] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0048] Figure 1 This is a flowchart illustrating a partial discharge signal identification method according to an exemplary embodiment;

[0049] Figure 2 This is a schematic flowchart illustrating a partial discharge signal identification method according to an exemplary embodiment;

[0050] Figure 3 This is a structural block diagram of a partial discharge signal identification device according to an exemplary embodiment;

[0051] Figure 4 This is a diagram illustrating the internal structure of a computer device according to an exemplary embodiment. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0053] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, product, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in a process, method, product, or apparatus that includes said elements is not excluded. For example, the use of terms such as "first," "second," etc., to denote names does not indicate any specific order.

[0054] In some embodiments provided in this disclosure, the execution of the partial discharge signal identification method can be controlled by a unified controller or by multiple controllers. These controllers may include controllers of local terminals or controllers of remote servers. In some embodiments, the controllers of local terminals and the controllers of servers may jointly assist in completing the partial discharge signal identification and control processing. The local terminal mentioned in this disclosure may include, but is not limited to, various robotic devices, vehicle-mounted devices, personal computers, laptops, smartphones, tablets, wearable devices, medical devices, VR (Virtual Reality) devices, etc. The server may also be a server, server cluster, distributed subsystem, cloud processing platform, server containing blockchain nodes, and combinations thereof. The controllers described in this disclosure may include various control units capable of implementing logic processing functions, including but not limited to CPU (Central Processing Unit), PLC (Programmable Logic Controller), ECU (Electronic Control Unit), MCU (Microcontroller Unit), FPGA (Field Programmable Gate Array), and CPLD (Complex Programmable Logic Device), as well as controllers composed of one or more logic function units, chips, etc.

[0055] In some embodiments of this disclosure, a method for identifying partial discharge signals is provided, such as... Figure 1 As shown, it includes the following steps:

[0056] S20. Acquire the raw partial discharge pulse data of the device under test.

[0057] In some implementations, the server first performs the acquisition and preprocessing of raw partial discharge pulse data from the device under test. In some examples, taking a 10kV high-voltage switchgear (the device under test) as an example, this device may experience partial discharge (hereinafter referred to as partial discharge) due to insulation defects during operation. The server needs to acquire raw pulse signals using a high-frequency current sensor (bandwidth 500kHz-20MHz) and an ultrasonic sensor (center frequency 40kHz) deployed inside the switchgear. Specifically, the server receives the raw pulse acquisition signal uploaded by the sensor array via an industrial Ethernet network. This signal contains a continuously sampled pulse sequence (sampling frequency 100MHz, single acquisition duration 10 seconds, total 1×10^6 pulses). 9(Sampling points), which include possible partial discharge pulses, as well as mixed equipment operating noise (such as mechanical vibration, electromagnetic interference) and environmental noise (such as cooling fan airflow noise).

[0058] S22. Input the original partial discharge pulse data into the pre-trained partial discharge signal enhancement model to obtain the enhanced partial discharge pulse data.

[0059] The server inputs raw partial discharge (PD) pulse data into a pre-trained PD signal enhancement model to improve the signal-to-noise ratio (SNR) and highlight PD pulse characteristics. This model can be a deep learning model based on the U-Net architecture, and its training process is as follows: The server first constructs a training dataset, consisting of two parts: a high SNR clean PD pulse sample set and a real background noise sample set. During training, the server scales the clean PD pulse samples according to a preset amplitude ratio (e.g., 0.2-1.0 times) to simulate PD signals of different intensities. These samples are then mixed with randomly selected background noise samples according to the SNR (e.g., -10dB to 10dB), generating 100,000 sets of mixed PD pulse samples (input) and corresponding clean samples (labels). The model is trained by minimizing the mean square error loss function, enabling the model to learn the mapping relationship from noisy signals to clean signals.

[0060] In the application phase, the server receives the raw partial discharge pulse data and inputs it into the trained enhancement model. The model extracts low-frequency features of the noise signal through the encoder, and the decoder reconstructs the pulse details, outputting the enhanced partial discharge pulse data. For example, the signal-to-noise ratio of the enhanced data is improved to 15dB, the bipolar characteristics of the ionizing discharge pulse (positive peak followed by negative peak, peak-to-peak value of 8mV, pulse width of 50μs) are clearly visible, and the background noise amplitude is reduced to below 0.5mV, providing high-quality input for subsequent matching and recognition.

[0061] S24. Match the enhanced partial discharge pulse data with the standard pulse signal to obtain candidate partial discharge pulse data.

[0062] The server can perform time-domain matching between the enhanced partial discharge pulse data and the standard pulse signal to locate the partial discharge pulse and extract its timestamp and discharge type. The standard pulse signal is a pre-stored partial discharge pulse template library on the server, containing time-domain characteristic parameters and time-series distributions (based on a 50Hz power frequency cycle, 20ms per cycle, containing 1000 sampling points) for three types: ionizing discharge (standard waveform: bipolar pulse, peak-to-peak value 10mV, pulse width 40-60μs), intermittent interruption discharge (standard waveform: unipolar pulse, amplitude 8mV, interrupted after 2-3 cycles of oscillation, pulse width 80-120μs), and phase distortion discharge (standard waveform: polarity correlated with power frequency phase, positive half-cycle amplitude 12mV, negative half-cycle amplitude 5mV, pulse width 60-100μs).

[0063] S26. Obtain the response feature information of the device to be detected.

[0064] The server needs to combine the physical state characteristics of the device under test to improve the accuracy of partial discharge identification. Therefore, it needs to obtain its response characteristic information, which can specifically include three parts: device topology description, dynamic measurement point parameters, and infrared thermal image characteristics.

[0065] The equipment topology description can include the structural composition information of the switch cabinet. The server encodes it according to the preset topology coding mapping table to obtain the topology feature index. The dynamic measurement point parameters are real-time monitoring data, including the temperature and humidity inside the cabinet collected by temperature and humidity sensors, and the load current collected by current sensors. The infrared thermal image features are the temperature field images of the switch cabinet surface captured by infrared cameras. The server uses a thermal image feature extraction network based on ResNet-50 to output the temperature field distribution spectrum.

[0066] S28. Input the candidate partial discharge pulse data and the response feature information into the partial discharge identification model, perform feature extraction respectively to obtain the partial discharge pulse features to be detected and the response distribution features, and obtain the target response identification result of the device to be detected based on the partial discharge pulse features to be detected and the response distribution features.

[0067] The server inputs candidate partial discharge pulse data and response characteristic information into the partial discharge identification model to determine the partial discharge status of the equipment. In this way, from raw signal acquisition to status identification, the partial discharge signal of the switchgear can be enhanced and accurately identified, providing a basis for equipment operation and maintenance decisions.

[0068] In some embodiments of this disclosure, the original partial discharge (PD) pulse data of the device under test is acquired, and the signal-to-noise ratio (SNR) is improved using a pre-trained PD signal enhancement model to obtain enhanced PD pulse data. Subsequently, the enhanced signal is matched with a standard pulse signal to extract candidate PD pulse data containing pulse timestamps and discharge types. Simultaneously, the device's response feature information is collected. Finally, the candidate PD pulse data and response feature information are input into a PD identification model, and the target response identification result is obtained through feature extraction and fusion. This method enables effective enhancement and accurate identification of PD signals under low SNR conditions through deep learning technology, improving the reliability of insulation status assessment for power equipment.

[0069] In some embodiments of this disclosure, S20 includes:

[0070] Acquire the raw pulse acquisition signal of the device under test, wherein the raw pulse acquisition signal includes pulses from multiple sampling points;

[0071] The original pulse acquisition signal is preprocessed to obtain the original partial discharge pulse data.

[0072] In some examples, the server first acquires the raw pulse acquisition signal from the device under test. Taking a 10kV high-voltage switchgear as an example, the server acquires signals using a high-frequency current sensor (bandwidth 500kHz-20MHz) and an ultrasonic sensor (center frequency 40kHz) deployed inside the cabinet. The sampling frequency is set to 100MHz, and each acquisition lasts for 10 seconds, generating a total of 1×102 pulses. 9 The original pulse acquisition signal of each sampling point. The signal is a time series array, and each element corresponds to the pulse amplitude (unit: mV) of the sampling point, for example [0.15,0.22,-0.08,...,4.9,5.3,...]. It contains possible partial discharge pulses (such as bipolar pulses with an amplitude of about 5mV), as well as equipment operating noise (such as continuous fluctuations below 0.3mV caused by mechanical vibration) and electromagnetic interference (such as the 0.5mV periodic component of 50Hz power frequency harmonics).

[0073] Subsequently, the server preprocesses the raw pulse acquisition signal to obtain the raw partial discharge pulse data. This involves two steps: First, electromagnetic interference is removed using wavelet threshold filtering. The signal is decomposed into 5 levels using a db4 wavelet basis, and high-frequency coefficients are processed with a soft threshold (threshold set to 0.2mV). After reconstruction, 50Hz and its harmonics are filtered out. Second, burst noise is removed using adaptive threshold denoising. Based on the 3σ criterion of signal amplitude distribution (the mean of the signal before preprocessing is calculated to be 0.1mV, the standard deviation is 0.05mV, and the threshold is set to 0.1 + 3 × 0.05 = 0.25mV), noise components with amplitudes below 0.25mV and spike pulses caused by poor sensor contact (such as abrupt amplitude jumps to above 10mV) are removed. After preprocessing, the server obtains the raw partial discharge pulse data, which retains the effective pulse characteristics with an amplitude ≥ 0.25mV. For example, the time series array is presented as [0.3,0.4,-0.2,...,5.1,4.7,...], in which the bipolar characteristics of the partial discharge pulse (positive peak followed by negative peak, pulse width of about 50μs) are initially revealed, and the background noise amplitude is suppressed to below 0.2mV.

[0074] In some embodiments of this disclosure, reference is made to Figure 2 S24 includes:

[0075] S242. Perform time-domain alignment matching between the original partial discharge pulse data and the standard pulse signal to obtain the target signal segment in the candidate sampling point set where the original partial discharge pulse data is located.

[0076] S244. Compare each pulse on the target signal segment with the corresponding pulse on the standard pulse signal to determine the background pulse and partial discharge pulse in the target signal segment;

[0077] S246. Based on the standard pulse signal, the background pulse, and the partial discharge pulse, construct the discharge type time-series distribution and pulse time marker distribution of the device under test;

[0078] S248. According to the standard pulse timestamp of the pulse in the standard pulse signal, perform time-domain feature mapping on the discharge type time sequence distribution and the pulse time stamp distribution to obtain the candidate partial discharge pulse data.

[0079] In some examples, the server first performs time-domain alignment and matching between the raw partial discharge pulse data and the standard pulse signal to obtain the target signal segment. The standard pulse signal is a pre-stored partial discharge template library on the server, containing three types: ionizing discharge (bipolar, peak-to-peak 10mV, pulse width 40-60μs), intermittent interruption discharge (unipolar, amplitude 8mV, pulse width 80-120μs), and phase distortion discharge (polarity varies with power frequency phase, pulse width 60-100μs). The time series distribution is constructed based on a 50Hz power frequency cycle (20ms per cycle, containing 1000 sampling points). The raw partial discharge pulse data is a pre-processed 10-second time series (1×10⁻⁶). 9 (Sampling points), the server aligns them with the standard pulse signal using a dynamic time warping algorithm: taking the starting point of the standard signal's power frequency cycle as a reference, it locates 5 consecutive power frequency cycles (100ms, 5×10) in the original data. 4The target signal segment (sampling points) is selected to ensure it contains complete pulse period characteristics (e.g., pulse distribution with a power frequency phase of 0-360°). Next, the server compares the target signal segment with the corresponding pulses of the standard pulse signal to distinguish between background pulses and partial discharge pulses. For each pulse in the target signal segment (a total of 20 pulses were detected, each defined as a continuous sampling point sequence with an amplitude exceeding 0.5mV and a duration ≥10μs), the similarity to the pulses at the same position in the standard pulse signal is calculated using the waveform cross-correlation coefficient (value range 0-1, higher values ​​indicate greater waveform similarity). A similarity threshold of 0.3 is set, with 12 pulses having a similarity <0.3 (e.g., irregular fluctuations with an amplitude of 0.6mV and a pulse width of 20μs, judged as background pulses), and 8 pulses having a similarity ≥0.3 (e.g., bipolar pulses with a similarity of 0.82 and unipolar oscillating pulses with a similarity of 0.75, judged as partial discharge pulses). Subsequently, the server constructs a discharge type time-series distribution and a pulse time-stamped distribution. The discharge type time-series distribution includes a standard power frequency pulse time-series distribution and a discharge feature topology: The standard power frequency pulse time-series distribution is based on the timestamp of the standard pulse signal (with the power frequency cycle starting at 0ms), integrating the type of partial discharge pulse and event markers. For example, the partial discharge pulse at 2.5ms matches the standard ionizing discharge template (similarity 0.82), recording its "ionizing discharge type" and "ionizing event marker quantity = 1", forming the entry {timestamp: 2.5ms, type: ionizing discharge, marker quantity: 1}; The discharge feature topology integrates the spectral identifiers of the background pulse and the partial discharge pulse. For example, the partial discharge pulse at 15.7ms matches the standard intermittent discharge template (similarity 0.75), recording its "intermittent discharge type" and "oscillation missing flag bit = 1", forming the entry {timestamp: 15.7ms, type: intermittent discharge, flag bit: 1}. The pulse time stamp distribution comprises a set of periodic phase coordinates and a standard pulse time stamp distribution. The periodic phase coordinate set converts the partial discharge pulse timestamps into power frequency phases (e.g., 2.5ms corresponds to 12.5% ​​of the power frequency period (20ms), i.e., phase 2.5 / 20×360=45°), and records it as {timestamp: 2.5ms, phase: 45°}. The standard pulse time stamp distribution integrates the standard timestamps of all partial discharge pulses, forming a time series of [2.5ms, 15.7ms, 22.3ms,...]. Finally, the server performs time-domain feature mapping on the discharge type time series distribution and the pulse time stamp distribution according to the standard pulse timestamps of the standard pulse signals. For example, {2.5ms, ionizing discharge, marker = 1} in the standard power frequency pulse time series distribution is associated with {2.5ms, 45°} in the periodic phase coordinate set through timestamps, integrating them into a candidate pulse entry containing pulse timestamp: 2.5ms, discharge type: ionizing discharge, periodic phase coordinate: 45°, and ionizing event marker: 1.The same mapping is performed on the 8 partial discharge pulses to finally obtain candidate partial discharge pulse data, which is an array containing 8 entries. Each entry contains the pulse timestamp, discharge type, period phase coordinates and corresponding event marker.

[0080] In some embodiments of this disclosure, the partial discharge pulse includes at least one of ionizing discharge characteristic pulse, intermittent interruption characteristic pulse, and phase distortion characteristic pulse; the discharge type timing distribution includes a standard power frequency pulse timing distribution and a discharge characteristic topology; the pulse time stamp distribution includes a set of periodic phase coordinates and a standard pulse time stamp distribution; S246 includes:

[0081] The discharge spectrum identifier of each pulse in the standard pulse signal corresponding to the target signal segment and the free event marker of each free discharge characteristic pulse are integrated by pulse timing to obtain the standard power frequency pulse timing distribution.

[0082] The discharge characteristic topology is obtained by integrating the discharge power spectrum identifier of the background pulse, the discharge power spectrum identifier of the phase distortion characteristic pulse, the discharge power spectrum identifier of the ionization discharge characteristic pulse, and the oscillation missing flag of each intermittent interruption characteristic pulse through pulse timing.

[0083] The periodic phase coordinates corresponding to each pulse in the discharge characteristic topology and the periodic phase coordinates of each intermittent interruption characteristic pulse are integrated by pulse timing to obtain the set of periodic phase coordinates. In addition, the standard pulse timestamps of each pulse in the discharge characteristic topology and the standard pulse timestamps of each intermittent interruption characteristic pulse are integrated by pulse timing to obtain the standard pulse timestamp distribution.

[0084] In some examples, the server uses the standard pulse timestamp of a standard pulse signal (based on a 50Hz power frequency cycle, 20ms per cycle, timestamp range 0-100ms, accurate to 0.1ms) as the timing reference, and constructs the discharge type timing distribution (including standard power frequency pulse timing distribution and discharge characteristic topology) and pulse time stamp distribution (including periodic phase coordinate set and standard pulse time stamp distribution) through the following steps:

[0085] The server first extracts the discharge spectrum identifier of each pulse in the standard pulse signal (a predefined unique code for the pulse type: for example, "FD" for ionizing discharge characteristic pulses, "ID" for intermittent interruption characteristic pulses, "PD" for phase distortion characteristic pulses, and "BN" for background pulses), and obtains the ionization event marker (a binary value representing whether an ionizing discharge event has occurred: 1 indicates the presence of an ionizing event, and 0 indicates that it has not occurred) for the three ionizing discharge characteristic pulses within the target signal segment. Within the target signal segment, the standard pulse signal contains 20 standard timestamps (4 pulses per cycle, 20 for 5 cycles, such as 2.5ms, 5.3ms, 10.2ms, and 15.7ms for cycle 1 (0-20ms); and 22.5ms, 25.3ms, 30.2ms, and 35.7ms for cycle 2 (20-40ms)). Among them, the standard pulse timestamps of the three ionizing discharge characteristic pulses are 2.5ms (period 1), 22.5ms (period 2), and 42.5ms (period 3), and their ionizing event markers are all 1 (after waveform feature detection, the pulses exhibit typical bipolarity of ionizing discharge, peak-to-peak value of 8-10mV, and pulse width of 50μs); the ionizing event markers of the remaining 17 pulses (including background pulses and non-ionizing partial discharge pulses) are 0. The server performs pulse timing integration in ascending order of timestamps: associating the standard timestamp, discharge spectrum identifier, and ionizing event marker of each pulse to generate a standard power frequency pulse timing distribution. For example, some entries are: [{timestamp: 2.5ms, discharge spectrum identifier: "FD", ionization event marker: 1}, {timestamp: 5.3ms, discharge spectrum identifier: "BN", ionization event marker: 0}, {timestamp: 10.2ms, discharge spectrum identifier: "ID", ionization event marker: 0}, {timestamp: 15.7ms, discharge spectrum identifier: "PD", ionization event marker: 0}, {timestamp: 22.5ms, discharge spectrum identifier: "FD", ionization event marker: 1},...]. The server extracts the discharge spectrum identifiers of all pulses within the target signal segment (including 12 background pulses, 3 ionization discharges, 3 intermittent interruptions, and 2 phase distortion characteristic pulses), and obtains the oscillation missing flag bit of the 3 intermittent interruption characteristic pulses (a binary quantity characterizing whether the pulse has an oscillation interruption: 1 indicates the presence of oscillation missing, and 0 indicates complete oscillation).Among them, the discharge spectrum of 12 background pulses is identified as "BN" (e.g., pulses with timestamps of 5.3ms and 25.3ms, similarity <0.3, amplitude 0.6-0.8mV); the discharge spectrum of 2 phase distortion characteristic pulses is identified as "PD" (timestamps of 15.7ms and 35.7ms, similarity 0.78 and 0.76, showing phase asymmetry characteristics of 8mV amplitude in the positive half-cycle and 3mV amplitude in the negative half-cycle); the discharge spectrum of 3 ionization discharge characteristic pulses is identified as "FD" (same as step 1); the discharge spectrum of 3 intermittent interruption characteristic pulses is identified as "ID" (timestamps of 10.2ms, 30.2ms and 50.2ms, similarity 0.75, 0.73 and 0.77, two periodic oscillation interruptions occur after the unipolar pulse), and their oscillation missing flag bits are all 1 (a 20μs amplitude drop was detected at the pulse trailing edge, which is consistent with the oscillation interruption characteristics).

[0086] The server performs pulse timing integration in ascending order of timestamps: it associates the standard timestamp, discharge spectrum identifier (background "BN", ionization "FD", intermittent interruption "ID", phase distortion "PD") of each pulse with the oscillation missing flag bit of the intermittent interruption pulse (this field is 0 for non-intermittent interruption pulses) to generate a discharge characteristic topology. For example, some entries are: [{timestamp: 2.5ms, discharge spectrum identifier: "FD", oscillation missing flag bit: 0}, {timestamp: 5.3ms, discharge spectrum identifier: "BN", oscillation missing flag bit: 0}, {timestamp: 10.2ms, discharge spectrum identifier: "ID", oscillation missing flag bit: 1}, {timestamp: 15.7ms, discharge spectrum identifier: "PD", oscillation missing flag bit: 0}, {timestamp: 22.5ms, discharge spectrum identifier: "FD", oscillation missing flag bit: 0}, {timestamp: 30.2ms, discharge spectrum identifier: "ID", oscillation missing flag bit: 1},...]. The server converts the standard pulse timestamp of each pulse in the discharge characteristic topology into periodic phase coordinates (phase angle within the power frequency cycle, formula: phase coordinate = (timestamp - cycle start time) / cycle duration × 360°), and integrates the periodic phase coordinates of 3 intermittent interruption characteristic pulses (consistent with the phase corresponding to their own timestamps). For example, a pulse with a timestamp of 2.5ms is located in period 1 (0-20ms), and its period phase coordinate is (2.5-0) / 20×360=45°; a pulse with a timestamp of 10.2ms (intermittent interruption characteristic pulse) is located in period 1, and its phase coordinate is (10.2-0) / 20×360=183.6°; the phase coordinate of a pulse with a timestamp of 22.5ms (period 2, 20-40ms) is (22.5-20) / 20×360=45°; the phase coordinate of a pulse with a timestamp of 30.2ms (intermittent interruption, period 2) is (30.2-20) / 20×360=183.6°; the phase coordinate of a pulse with a timestamp of 15.7ms (phase distortion, period 1) is (15.7-0) / 20×360=282.6°... The server associates the standard timestamp of each pulse with its corresponding period phase coordinate according to the temporal order of the discharge characteristic topology, generating a set of period phase coordinates. For example, some entries are: [{timestamp: 2.5ms, periodic phase coordinate: 45°}, {timestamp: 5.3ms, periodic phase coordinate: 95.4°}, {timestamp: 10.2ms, periodic phase coordinate: 183.6°}, {timestamp: 15.7ms, periodic phase coordinate: 282.6°}, {timestamp: 22.5ms, periodic phase coordinate: 45°},...]. The server extracts the standard pulse timestamps (20, 0-100ms) of all pulses in the discharge characteristic topology and supplements them with the standard pulse timestamps of 3 intermittent interruption characteristic pulses (to ensure no omissions).Since the discharge characteristic topology already contains the timestamps of all pulses (including intermittent interruptions), the server directly integrates them in ascending order of timestamps to generate a standard pulse timestamp distribution, i.e., the timestamp sequence of all pulses: [2.5ms, 5.3ms, 10.2ms, 15.7ms, 22.5ms, 25.3ms, 30.2ms, 35.7ms, 42.5ms, 45.3ms, 50.2ms, 55.7ms, 62.5ms, 65.3ms, 70.2ms, 75.7ms, 82.5ms, 85.3ms, 90.2ms, 95.7ms].

[0087] Through the above steps, the server constructs the discharge type time sequence distribution (including standard power frequency pulse time sequence distribution and discharge feature topology) and pulse time marker distribution (including periodic phase coordinate set and standard pulse time marker distribution), providing structured pulse time sequence and feature data for subsequent partial discharge identification.

[0088] In some embodiments of this disclosure, S26 includes:

[0089] Obtain the device topology description, dynamic measurement point parameters, and infrared thermal image features of the device under test;

[0090] According to the preset topology coding mapping, the topology description of the device is parametrically coded to obtain the topology feature index of the device to be detected;

[0091] The infrared thermal image features are extracted to obtain the temperature field distribution spectrum of the infrared thermal image features. The dynamic measurement point parameters and the temperature field distribution spectrum are then fused using multi-source features to obtain the state feature vector of the device to be detected.

[0092] The response feature information is obtained based on the topological feature index and the state feature vector.

[0093] In some examples, the server acquires response characteristic information for a 10kV high-voltage switchgear (the device under test) through the following steps: The server first collects three types of basic information: For example, the equipment topology is described as a series structure of "circuit breaker, current transformer, and disconnector" (characterizing the connection relationship of internal components); dynamic measurement parameters are acquired in real time by sensors inside the cabinet, including temperature 35℃ (temperature and humidity sensor, accuracy ±0.5℃), humidity 60%RH (same as above), and load current 80A (current sensor, range 0-630A); the infrared thermal image feature is an image of the switchgear surface temperature field taken by a 640×512 pixel infrared camera (temperature range -20-150℃, accuracy ±1℃), with the circuit breaker area (upper left corner) temperature at 38℃, the current transformer area (middle) at 36℃, the disconnector area (lower right corner) at 34℃, and the background temperature at 32℃. The server calls a preset topology coding mapping table (component type corresponds one-to-one with coding: circuit breaker "001", current transformer "002", disconnector "003", grounding switch "004") to perform parametric feature coding on the equipment topology description "circuit breaker-current transformer-disconnector". The coding is concatenated according to the component connection order to obtain the topology feature index "001-002-003" (string format, length 11 characters), which uniquely identifies the structural topology of the switchgear.

[0094] The server processes infrared thermal image features using a ResNet-50-based thermal image feature extraction network: it divides the temperature field image into regions (circuit breaker region, transformer region, disconnector region, and background region based on component location), extracts the mean temperature (e.g., 38℃ for the circuit breaker region), temperature gradient (e.g., 2℃ / pixel temperature difference at the edge of the circuit breaker region), and percentage of high-temperature points (15% of pixels in the circuit breaker region > 35℃), and outputs a 128-dimensional temperature field distribution spectrum (each 32 dimensions correspond to the features of one region). Subsequently, the server converts dynamic measurement point parameters (temperature 35℃, humidity 60%RH, current 80A) into 3-dimensional numerical vectors, and fuses them with the 128-dimensional temperature field distribution spectrum through an attention mechanism: assigning a feature weight of 0.6 to the circuit breaker region (high-temperature region), 0.3 to the transformer region, and 0.1 to the remaining regions. After weighted concatenation, the vectors are compressed into a 256-dimensional state feature vector (vector element values ​​range from 0-1, representing a comprehensive indicator of the equipment's operating status) through a fully connected layer. The server converts the topology feature index "001-002-003" into a 128-dimensional one-hot encoded vector (each encoded bit corresponds to a component type, with the active bit being 1 and the rest being 0), and concatenates it with the 256-dimensional state feature vector according to the dimensions to obtain 384-dimensional response feature information (vector format, stored as a floating-point array). This information simultaneously contains the device structure topology and real-time operating status features.

[0095] In some embodiments of this disclosure, the partial discharge identification model includes a feature extraction module, a feature fusion module, and a state classification module. The state classification module includes a feature aggregation component and a state predictor. S28 includes:

[0096] Based on the feature extraction module, the detection partial discharge pulse features of the candidate partial discharge pulse data are extracted, and the response distribution features of the response feature information are extracted;

[0097] Based on the feature fusion module, the partial discharge pulse feature to be detected and the response distribution feature are concatenated, and the feature space of the concatenated feature is expanded to obtain a multi-source discharge feature field.

[0098] Based on the feature aggregation component, the feature space of the multi-source discharge feature field is compressed to obtain the one-dimensional partial discharge pulse feature of the device under test.

[0099] Based on the state predictor, the one-dimensional partial discharge pulse features to be detected are refined to predict the state attribution confidence of each candidate response feature in the set of candidate response feature information of the device to be detected. The candidate response feature information whose state attribution confidence is at the state determination threshold is determined as the target response identification result of the device to be detected.

[0100] In some examples, the server performs the following steps using a partial discharge identification model (including a feature extraction module, a feature fusion module, and a state classification module) for candidate partial discharge pulse data (containing 8 partial discharge pulses: 3 ionizing discharges, 3 intermittent interruption discharges, and 2 phase distortion discharges) and 384-dimensional response feature information of 10kV high-voltage switchgear: The feature extraction module processes two types of input data: candidate partial discharge pulse data and response feature information.

[0101] For candidate partial discharge pulse data, the data includes the type features of 8 pulses (ionization / intermittent interruption / phase distortion), periodic phase coordinates (0-360°), and standard timestamps (0-100ms). The module generates a type feature tensor (8×7, 8 pulses × 7-dimensional discharge type vector: such as the ionization discharge vector [1,0,0,0,1,0,0]) and a periodic phase coordinate feature tensor (8×360, 8 pulses × 360-dimensional position encoding vector: such as the 23rd bit being 1 corresponding to a phase of 45°) through the feature encoding component, and concatenates them with the standard pulse timestamp distribution (8×1, timestamp values) to obtain the partial discharge pulse features to be detected (fused feature tensor) of 8×(7+360+1)=8×368.

[0102] For the response feature information, the 384-dimensional response feature information includes a topological feature index (“001-002-003”) and a state feature vector (256-dimensional). The module uses a structured feature extractor (BERT model) to convert the topological index into a 128-dimensional classification feature representation (capturing structural topological semantics), and a numerical feature processor (3-layer fully connected network) to convert the state feature vector into a 256-dimensional scalar feature representation (integrating numerical patterns such as temperature / current), concatenating them to obtain a 384-dimensional response distribution feature (128+256=384). The feature fusion module performs feature integration: feature concatenation: the 8×368 partial discharge pulse features to be detected are concatenated with the 384-dimensional response distribution feature by dimension—first expanding the response distribution feature to 8×384 (each pulse corresponds to the same device state feature), and then concatenating to obtain a joint feature tensor of 8×(368+384)=8×752 (each row is “pulse feature + device state feature” for one pulse). Feature space expansion: The 8×752 joint feature tensor is spatially expanded by a 1×1 convolutional layer (64 convolutional kernels, ReLU activation function), expanding the feature dimension from 752 to 752×64, generating an 8×752×64 multi-source discharge feature field (a three-dimensional tensor that represents the spatial correlation between pulse features and device state). Feature aggregation component (including global average pooling layer) spatially compresses the 8×752×64 multi-source discharge feature field: Average pooling is performed along the pulse number dimension (8) and feature depth dimension (64) to reduce the three-dimensional feature field to a 1×752 one-dimensional feature vector; then, it is refined by a 2-layer fully connected network (hidden layer dimension 512→256, tanh activation function) to obtain a 256-dimensional one-dimensional partial discharge pulse feature to be detected (vector element value range -1~1, comprehensively representing the correlation between partial discharge pulse and device state). The state predictor (including the softmax output layer) processes 256 one-dimensional features: The candidate response feature information set predefines 5 equipment states: "normal operation," "ionizing discharge," "intermittent interruption discharge," "phase distortion discharge," and "mixed discharge," with one candidate label for each state. Feature refinement and confidence prediction: The one-dimensional features are mapped to 5 state attribution confidence scores (summing to 1) through a fully connected layer (256→5), with an output of [0.03, 0.92, 0.02, 0.02, 0.01] ("ionizing discharge" confidence score 0.92, all others < 0.1). Target determination result: The state determination threshold is set to 0.8. The "ionizing discharge" confidence score 0.92 > 0.8, therefore it is determined as the target response identification result, meaning the server determines that the 10kV high-voltage switchgear has an "ionizing discharge" partial discharge state.

[0103] Through the above process, the model achieves deep fusion of partial discharge pulse characteristics and equipment status characteristics, and ultimately accurately identifies the partial discharge type of the equipment.

[0104] In some embodiments of this disclosure, the feature extraction module includes a feature encoding component, and the candidate partial discharge pulse data includes a standard power frequency pulse timing distribution, discharge feature topology, a set of periodic phase coordinates, and a standard pulse time stamp distribution. The extraction of the detectable partial discharge pulse features based on the feature extraction module of the candidate partial discharge pulse data can be performed through the following examples.

[0105] The feature encoding component identifies the standard power frequency pulse timing distribution and the discharge feature topology to determine the discharge type of each pulse in the candidate partial discharge pulse data. For any pulse, a multi-dimensional feature state vector is constructed. The multi-dimensional feature state vector includes new event markers, missing event markers, multiple types of baseline feature markers, and multiple types of abnormal feature markers.

[0106] If the pulse is an ionizing discharge characteristic pulse, then the values ​​of the newly added event markers and the abnormal feature markers that match the type of the pulse in the multidimensional feature state vector are set as active state values, and the values ​​of the remaining feature markers in the multidimensional feature state vector are set as inactive state values, thereby generating the discharge type vector of the pulse;

[0107] If the pulse is an intermittent interruption pulse, then the values ​​of the missing event markers in the multidimensional feature state vector and the values ​​of the reference feature markers that match the type of the pulse are set as the active state values, and the values ​​of the remaining feature markers in the multidimensional feature state vector are set as the inactive state values, thereby generating the discharge type vector of the pulse;

[0108] If the pulse is a phase distortion characteristic pulse, then the historical feature type of the pulse is determined, and the values ​​of the reference feature markers that match the historical feature type and the abnormal feature markers that match the pulse are set as the active state values, and the values ​​of the remaining feature markers in the multidimensional feature state vector are set as the inactive state values, thereby generating the discharge type vector of the pulse;

[0109] The discharge type vectors of each pulse are stacked as tensors to obtain the type feature tensor of the candidate partial discharge pulse data;

[0110] The periodic phase coordinate set is identified by the feature encoding component to determine the periodic phase coordinate of each pulse in the candidate partial discharge pulse data. For any pulse, a position state vector of the pulse is constructed. The position state vector includes multiple candidate position markers.

[0111] Among the multiple candidate position markers, the target position marker corresponding to the periodic phase coordinate of the pulse is determined, and the value of the target position marker is set as the active state value. The values ​​of the remaining position markers in the position state vector are set as inactive state values ​​to obtain the position encoding vector of the pulse. The position encoding vectors of each pulse are stacked as tensors to obtain the periodic phase coordinate feature tensor of the candidate partial discharge pulse data.

[0112] The type feature tensor, the periodic phase coordinate feature tensor, and the standard pulse time stamp distribution are concatenated to obtain a fused feature tensor, which is then used as the partial discharge pulse feature to be detected.

[0113] In some examples, the server takes candidate partial discharge pulse data (containing three typical pulses: ionizing discharge characteristic pulse P1, intermittent interruption characteristic pulse P2, and phase distortion characteristic pulse P3) from a 10kV high-voltage switchgear as an example. The server then performs the following steps through the feature encoding component of the feature extraction module to generate the characteristics of the partial discharge pulse to be detected: Constructing a multi-dimensional feature state vector and generating a discharge type vector: Candidate partial discharge pulse data details: P1 (ionizing discharge characteristic pulse): Standard pulse timestamp 2.5ms, periodic phase coordinate 45°, discharge spectrum identifier "FD", ionization event marker 1; P2 (intermittent interruption characteristic pulse): Standard pulse timestamp 10.2ms, periodic phase coordinate 183.6°, discharge spectrum identifier "ID", oscillation missing flag 1; P3 (phase distortion characteristic pulse): Standard pulse timestamp 15.7ms, periodic phase coordinate 282.6°, discharge spectrum identifier "PD", historical feature type is "ionizing discharge baseline type" (based on the preceding pulse type determination in the discharge feature topology).

[0114] The multidimensional feature state vector is defined as follows: each pulse's multidimensional feature state vector is 7-dimensional, including: 2 basic markers: new event marker (specific to ionizing discharge) and missing event marker (specific to intermittent interruption); 3 baseline feature markers: ionizing baseline marker (corresponding to the standard waveform of ionizing discharge), intermittent interruption baseline marker (corresponding to the standard waveform of intermittent interruption), and phase distortion baseline marker (corresponding to the standard waveform of phase distortion); and 3 abnormal feature markers: ionizing abnormal marker (abnormal ionizing discharge feature), intermittent interruption abnormal marker (abnormal intermittent interruption feature), and phase distortion abnormal marker (abnormal phase distortion feature). The activation state value is 1 (feature exists), and the deactivation value is 0 (feature does not exist). A discharge type vector is generated according to the pulse type: P1 (ionizing discharge characteristic pulse): "New event marker" and "Ionizing abnormal marker" need to be activated. In the multidimensional feature state vector, new event marker = 1, ionizing abnormal marker = 1, and the other 5 markers = 0, generating a discharge type vector: [1,0,0,0,1,0,0] (7×1 dimensions). P2 (Intermittent Interruption Characteristic Pulse): Requires activation of "Missing Event Flag" and "Intermittent Interruption Baseline Flag".

[0115] In the multidimensional feature state vector, the missing event marker = 1, the intermittent interruption reference marker = 1, and the other 5 markers = 0, generating the discharge type vector: [0,1,0,1,0,0,0] (7×1 dimension). P3 (phase distortion characteristic pulse): The historical feature type is "ionization discharge reference type", and the "ionization reference marker" (matching the historical type) and "phase distortion anomaly marker" (matching the current type) need to be activated. In the multidimensional feature state vector, the ionization reference marker = 1, the phase distortion anomaly marker = 1, and the other 5 markers = 0, generating the discharge type vector: [0,0,1,0,0,0,1] (7×1 dimension). Stacking the discharge type vectors to obtain the type feature tensor: The server stacks the discharge type vectors of the 3 pulses in ascending order of timestamp (P1→P2→P3) to generate the type feature tensor. Each vector is 7-dimensional, and the 3 pulses correspond to a 3×7 two-dimensional tensor, as shown in the following formula (1):

[0116]

[0117] A position state vector can be constructed and a position encoding vector can be generated. The position state vector is defined as follows: the position state vector of each pulse is 360-dimensional (corresponding to a phase of 0-359°, with a candidate position marker for every 1°, where index 0 corresponds to 0°, index 1 corresponds to 1°, ..., index 359 corresponds to 359°). The target position marker is the index corresponding to the pulse period phase coordinate, with an activation value of 1 and the rest being 0. The position encoding vector is generated according to the period phase coordinate: P1 (phase 45°): the period phase coordinate 45° corresponds to index 45 (0°→0, 45°→45), and the target position marker is index 45. In the position state vector, index 45=1, and the remaining 359 indices=0, generating the position encoding vector: [0,...,1,...,0] (360×1 dimension, only the 46th bit is 1). P2 (phase 183.6°): the period phase coordinate 183.6° is rounded to 184°, corresponding to index 184. In the position state vector, index 184=1, the rest=0, generating a position encoding vector: [0,...,1,...,0] (360×1 dimension, the 185th bit is 1). P3 (phase 282.6°): the periodic phase coordinate 282.6° is rounded to 283°, corresponding to index 283. In the position state vector, index 283=1, the rest=0, generating a position encoding vector: [0,...,1,...,0] (360×1 dimension, the 284th bit is 1). Stacking position encoding vectors to obtain the periodic phase coordinate feature tensor: the server stacks the position encoding vectors of the 3 pulses in ascending order of timestamps to generate the periodic phase coordinate feature tensor. Each vector is 360-dimensional, and the 3 pulses correspond to a 3×360 two-dimensional tensor, which can be referred to in the following formula (2):

[0118] Periodic phase coordinate characteristic tensor =

[0119]

[0120] Feature concatenation yields a fused feature tensor (features of the partial discharge pulse to be detected): The server concatenates the type feature tensor (3×7), the periodic phase coordinate feature tensor (3×360), and the standard pulse timestamp distribution (3×1, containing the timestamp values ​​of 3 pulses: [2.5, 10.2, 15.7]) by column dimension: After concatenating the type feature tensor (3×7) and the periodic phase coordinate feature tensor (3×360), the dimension is 3×(7+360)=3×367; then, the standard pulse timestamp distribution (3×1) is concatenated, finally yielding a fused feature tensor of 3×(367+1)=3×368. This fused feature tensor is thus identified as the feature of the partial discharge pulse to be detected, containing comprehensive features of pulse type, phase position, and timestamp. Through the above steps, the server completes feature extraction from the candidate partial discharge pulse data, generating 3×368 features of the partial discharge pulse to be detected, providing input for subsequent feature fusion and state recognition.

[0121] In some embodiments of this disclosure, the feature extraction module includes a structured feature extractor and a numerical feature processor, and the extraction of response distribution features of the response feature information can be implemented through the following examples.

[0122] The structured feature extractor extracts the classification feature representation from the response feature information, and the numerical feature processor extracts the scalar feature representation from the response feature information.

[0123] The response distribution features are determined based on the classification feature representation and the scalar feature representation.

[0124] In some examples, the server extracts response distribution features from the 384-dimensional response feature information (including topological feature index "001-002-003" and 256-dimensional state feature vector) of a 10kV high-voltage switchgear through the following steps: The server calls a structured feature extractor (a text embedding module based on a pre-trained BERT model) to process the topological feature index "001-002-003" (string format, representing the equipment structural topology). The model inputs the string as a character sequence (length 11 characters), learns the element connection semantics through a 12-layer Transformer encoder, and outputs a 128-dimensional classification feature representation (vector element values ​​range from -1 to 1, such as [0.2, -0.1, 0.3, ...]), which encodes the topological association rules of "circuit breaker, current transformer, disconnector". The server calls a numerical feature processor (2-layer fully connected network, hidden layer dimension 512→256, activation function ReLU) to process the 256-dimensional state feature vector (including temperature 35℃, humidity 60%RH, current 80A, and temperature field distribution spectrum fusion features). The network extracts scalar feature patterns (such as the coupling relationship between temperature and current) through nonlinear transformations, outputting a 256-dimensional scalar feature representation (vector element values ​​range from 0 to 1, such as [0.6, 0.4, 0.5, ...]). This representation integrates the numerical features of the device's dynamic operating status. The server concatenates the 128-dimensional classification feature representation with the 256-dimensional scalar feature representation according to their dimensions, obtaining a 128 + 256 = 384-dimensional response distribution feature (floating-point vector). This feature retains the structural semantics of the device topology while incorporating the numerical patterns of the real-time operating status. For example, the first 128 dimensions of the vector represent the topological association between "circuit breaker, transformer, and disconnector," while the last 256 dimensions represent the coordinated change characteristics of the temperature field gradient and the load current.

[0125] In some embodiments of this disclosure, the partial discharge signal enhancement model is trained through the following steps:

[0126] A high signal-to-noise ratio (SNR) clean partial discharge pulse sample set is obtained, which includes clean partial discharge pulse samples of various known discharge types and corresponding intensity levels.

[0127] Obtain a real background noise sample set, which includes background noise samples without partial discharge activity collected from a complex acoustic environment similar to the device under test;

[0128] The pure partial discharge pulse samples in the high signal-to-noise ratio pure partial discharge pulse sample set are scaled according to different preset amplitude ratios to obtain scaled partial discharge pulse samples of different intensities.

[0129] Each of the scaled partial discharge pulse samples is mixed with a background noise sample randomly selected from the real background noise sample set at different preset superposition ratios to generate a mixed sample of partial discharge pulses simulating different signal-to-noise ratio conditions.

[0130] The partial discharge pulse mixture sample is used as the input sample, and the corresponding unscaled high signal-to-noise ratio pure partial discharge pulse sample is used as the target output sample to form the partial discharge signal enhancement training sample set.

[0131] The original partial discharge signal enhancement model is trained using the aforementioned partial discharge signal enhancement training sample set. The model parameters are adjusted by optimizing a preset loss function, enabling the original partial discharge signal enhancement model to learn the mapping relationship from noisy partial discharge pulses to clean partial discharge pulses, thereby achieving effective enhancement of input low signal-to-noise ratio partial discharge pulses, and obtaining the trained partial discharge signal enhancement model.

[0132] In some implementations, the server targets the partial discharge signal enhancement requirements of a 10kV high-voltage switchgear and trains a partial discharge signal enhancement model (a deep learning model based on the U-Net architecture) through the following steps:

[0133] The server collects high signal-to-noise ratio (SNR) clean partial discharge pulse samples from a laboratory partial discharge simulation platform. The platform simulates a typical partial discharge scenario in a 10kV switchgear. Three types of samples are generated by adjusting the insulation defect type (e.g., needle-point-plate electrode simulating ionizing discharge, air gap simulating intermittent discharge, and surface contamination simulating phase distortion discharge). Specific parameters are as follows: Ionizing discharge characteristic pulse: bipolar waveform, peak-to-peak value 5-25mV (5 intensity levels: 5mV / 10mV / 15mV / 20mV / 25mV). The sample set consists of 3 types × 5 levels × 200 groups = 3000 groups, with a signal-to-noise ratio (SNR) ≥ 30dB (measured by a spectrum analyzer). The samples are stored as time series arrays (sampling frequency 100MHz, single sample duration 10ms, containing 1×10^6 mV, pulse width 40-60μs, 200 samples per level); intermittent interruption characteristic pulses: unipolar waveform, amplitude 4-20mV (5 intensity levels), interrupted after 2-3 cycles, pulse width 80-120μs, 200 samples per level; phase distortion characteristic pulses: polarity changes with power frequency phase, positive half-cycle amplitude 6-30mV (5 intensity levels), negative half-cycle amplitude 3-15mV, pulse width 60-100μs, 200 samples per level). 6 (Sampling points).

[0134] The server collected real background noise samples from a 10kV switchgear operating site: Five identical switchgear units without partial discharge activity were selected, and noise signals were collected using a high-frequency current sensor (bandwidth 500kHz-20MHz). The noise included: mechanical vibration noise (circuit breaker operating mechanism vibration, amplitude 0.1-0.5mV, frequency 50-200Hz); electromagnetic interference noise (50Hz power frequency harmonics, amplitude 0.3-0.8mV, period 20ms); and environmental noise (cooling fan airflow noise, amplitude 0.2-0.4mV, randomly distributed). A total of 500 samples were collected, each lasting 10ms (1×10⁻⁶). 6 Sampling points), SNR≤-5dB, are stored as a time series array consistent with the format of the clean samples. The server scales 3000 sets of clean samples according to a preset amplitude ratio to simulate partial discharge signals of different intensities. The preset ratio is 0.2-1.0 times (intervals of 0.1 times, a total of 9 ratios). For example, the original ionizing discharge sample (intensity level 3, amplitude 15mV) is scaled by 0.4 times to obtain a weak signal sample with an amplitude of 6mV; the original intermittent interruption sample (intensity level 5, amplitude 20mV) is scaled by 0.8 times to obtain a medium intensity sample with an amplitude of 16mV.

[0135] Each clean sample generates 9 scaled samples, for a total of 3000×9=27000 scaled samples, covering the intensity fluctuation range of partial discharge signals in real-world scenarios (1-25mV).

[0136] The server mixes 27,000 scaled samples with 500 noise samples at different signal-to-noise ratios (SNR) to simulate low SNR scenarios. The SNR range is set from -10dB to 10dB (5dB intervals, for a total of 5 SNRs), and the calculation formula is: SNR = 10lg(signal power / noise power), where the power is proportional to the square of the amplitude. For example, when a scaled sample (ionization discharge, amplitude 6mV, power 36mV²) is mixed with a noise sample (amplitude 2mV, power 4mV²), the SNR is 10lg(36 / 4) = 10lg9 ≈ 9.5dB (close to 10dB); when a scaled sample (phase distortion, amplitude 2mV, power 4mV²) is mixed with a noise sample (amplitude 3mV, power 9mV²), the SNR is 10lg(4 / 9) ≈ -3.5dB (close to -5dB). A total of 27,000 × 5 = 135,000 mixed samples (input) were generated, with the corresponding labels being the unscaled, original, clean samples (target output).

[0137] The server divides 135,000 input-label pairs into a training set (100,000 pairs), a validation set (20,000 pairs), and a test set (15,000 pairs). It trains a raw partial discharge signal enhancement model (U-Net architecture, 4-layer downsampling encoder, 4-layer upsampling decoder, ReLU activation function, and Tanh output layer): Loss function: Mean Squared Error (MSE), calculating the amplitude difference between the model output and the label; Optimizer: Adam, initial learning rate 0.001, decaying by 10% every 10 epochs; Training parameters: batchsize=32, epochs=50, trained on an NVIDIA A100 GPU. The validation set MSE loss converges from an initial 0.12 to 0.005 (corresponding to a 15-20dB SNR improvement). After training, the model can enhance the input low signal-to-noise ratio (SNR) (e.g., -5dB) mixed samples into a high SNR (e.g., 15dB) pure signal. For example, if the input is a noisy ionizing discharge pulse (SNR=3dB, peak-to-peak value of 8mV masked by noise), the output enhanced pulse has a peak-to-peak value of 7.8mV and a pulse width of 52μs, and the waveform similarity with the original pure sample reaches 0.92, which meets the requirements for subsequent recognition.

[0138] Through the above process, the server obtains a trained partial discharge signal enhancement model, which has the ability to extract clean partial discharge pulse features from noisy signals with low signal-to-noise ratio.

[0139] This disclosure provides several methods for partial discharge (PD) signal identification. These methods acquire raw PD pulse data from the device under test, enhance the signal-to-noise ratio (SNR) using a pre-trained PD signal enhancement model, and obtain enhanced PD pulse data. The enhanced signal is then matched with a standard pulse signal to extract candidate PD pulse data containing pulse timestamps and discharge types. Simultaneously, the device's response characteristic information is collected. Finally, the candidate PD pulse data and response characteristic information are input into the PD identification model, and the target response identification result is obtained through feature extraction and fusion. This approach utilizes deep learning technology to effectively enhance and accurately identify PD signals under low SNR conditions, improving the reliability of power equipment insulation condition assessment.

[0140] It is understood that the various embodiments of the methods described in this specification are presented in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. Related details can be found in the descriptions of other method embodiments.

[0141] It should be understood that although the steps in the flowcharts shown in the accompanying drawings are displayed sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some of the steps in the accompanying drawings may include multiple steps or stages, which are not necessarily completed at the same time, but may be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but may be performed alternately or in turn with other steps or at least a portion of the steps or stages of other steps.

[0142] Based on the description of the partial discharge signal identification method embodiments described above, this disclosure also provides a partial discharge signal identification device for implementing the aforementioned partial discharge signal identification method. The device may include a system (including a distributed system), software (application), module, component, controller, server, terminal, etc., using the method described in the embodiments of this specification, combined with necessary hardware implementation. Based on the same innovative concept, the devices in one or more embodiments provided by this disclosure are as described in the following embodiments. Since the implementation schemes and methods for solving the problem by the devices are similar, the implementation of specific devices in the embodiments of this specification can refer to the implementation of the aforementioned method, and repeated details will not be repeated. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0143] Figure 3This is a schematic block diagram illustrating a partial discharge signal identification device according to an exemplary embodiment. The device can be the aforementioned terminal, a server, or a module, component, device, control unit, etc., integrated into the terminal. For details, please refer to... Figure 3 The device 100 may include: a first module 120, a second module 140, a third module 160, a fourth module 180, and a fifth module 190. The first module 120 is used to acquire the original partial discharge pulse data of the device under test. The second module 140 is used to input the original partial discharge pulse data into a pre-trained partial discharge signal enhancement model to obtain enhanced partial discharge pulse data. The third module 160 is used to match the enhanced partial discharge pulse data with a standard pulse signal to obtain candidate partial discharge pulse data. The fourth module 180 is used to acquire the response feature information of the device under test. The fifth module 190 is used to input the candidate partial discharge pulse data and the response feature information into a partial discharge recognition model, perform feature extraction respectively, obtain the partial discharge pulse features to be detected and the response distribution features, and obtain the target response recognition result of the device under test based on the partial discharge pulse features to be detected and the response distribution features.

[0144] Each module in the aforementioned partial discharge signal identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0145] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a partial discharge signal identification method.

[0146] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0147] Based on the foregoing description of the relevant methods and apparatus embodiments, this disclosure also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the partial discharge signal identification method described in any embodiment of this specification.

[0148] Based on the foregoing description of the relevant methods and apparatus embodiments, this disclosure also provides a computer-readable storage medium that, when the instructions in the computer-readable storage medium are executed by the processor of a computer device, enables the computer device to implement the partial discharge signal identification method as described in any embodiment of this disclosure.

[0149] Based on the foregoing description of the relevant methods and apparatus embodiments, this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the partial discharge signal identification method described in any embodiment of this specification.

[0150] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, hardware + program embodiments are relatively simple in description because they are fundamentally similar to method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0151] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0152] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0153] It should be noted that the apparatus, computer equipment, storage medium, and computer program products described above may also include other implementation methods according to the description of the method embodiments. Specific implementation methods can be found in the description of the relevant method embodiments. Furthermore, new embodiments formed by combinations of features from various methods, apparatuses, devices, and server embodiments still fall within the scope of this disclosure and will not be elaborated upon here.

[0154] For ease of description, the above devices are described in terms of function, divided into various modules. Of course, when implementing one or more of these specifications, the functions of each module can be implemented in the same or different software and / or hardware, or a module that performs the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling and communication connections between the devices or units shown or described can be implemented through direct and / or indirect coupling / connection, through standard or custom interfaces or protocols, and can be implemented electrically, mechanically, or in other forms.

[0155] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0156] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A method for identifying partial discharge signals, characterized in that, The method includes: Acquire the raw partial discharge pulse data of the device under test; The original partial discharge pulse data is input into a pre-trained partial discharge signal enhancement model to obtain enhanced partial discharge pulse data; The enhanced partial discharge pulse data is matched with a standard pulse signal to obtain candidate partial discharge pulse data; Obtain the response feature information of the device under test; The candidate partial discharge pulse data and the response feature information are input into the partial discharge identification model, and feature extraction is performed to obtain the partial discharge pulse features to be detected and the response distribution features. Based on the partial discharge pulse features to be detected and the response distribution features, the target response identification result of the device to be detected is obtained.

2. The method according to claim 1, characterized in that, The step of matching the enhanced partial discharge pulse data with a standard pulse signal to obtain candidate partial discharge pulse data includes: The original partial discharge pulse data and the standard pulse signal are time-domain aligned and matched to obtain the target signal segment in the candidate sampling point set of the original partial discharge pulse data; By comparing each pulse on the target signal segment with the corresponding pulse on the standard pulse signal, the background pulse and partial discharge pulse in the target signal segment are determined. Based on the standard pulse signal, the background pulse, and the partial discharge pulse, construct the discharge type time-series distribution and pulse time stamp distribution of the device under test; Based on the standard pulse timestamp in the standard pulse signal, the discharge type time sequence distribution and the pulse time stamp distribution are mapped in the time domain to obtain the candidate partial discharge pulse data.

3. The method according to claim 2, characterized in that, The partial discharge pulse includes at least one of ionizing discharge characteristic pulses, intermittent interruption characteristic pulses, and phase distortion characteristic pulses; the discharge type time sequence distribution includes a standard power frequency pulse time sequence distribution and a discharge characteristic topology; the pulse time stamp distribution includes a set of periodic phase coordinates and a standard pulse time stamp distribution; and constructing the discharge type time sequence distribution and pulse time stamp distribution of the device under test based on the standard pulse signal, the background pulse, and the partial discharge pulse includes: The discharge spectrum identifier of each pulse in the standard pulse signal corresponding to the target signal segment and the free event marker of each free discharge characteristic pulse are integrated by pulse timing to obtain the standard power frequency pulse timing distribution. The discharge characteristic topology is obtained by integrating the discharge power spectrum identifier of the background pulse, the discharge power spectrum identifier of the phase distortion characteristic pulse, the discharge power spectrum identifier of the ionization discharge characteristic pulse, and the oscillation missing flag of each intermittent interruption characteristic pulse through pulse timing. The periodic phase coordinates corresponding to each pulse in the discharge characteristic topology and the periodic phase coordinates of each intermittent interruption characteristic pulse are integrated by pulse timing to obtain the set of periodic phase coordinates. In addition, the standard pulse timestamps of each pulse in the discharge characteristic topology and the standard pulse timestamps of each intermittent interruption characteristic pulse are integrated by pulse timing to obtain the standard pulse timestamp distribution.

4. The method according to claim 1, characterized in that, The step of obtaining the response feature information of the device under test includes: Obtain the device topology description, dynamic measurement point parameters, and infrared thermal image features of the device under test; According to the preset topology coding mapping, the topology description of the device is parametrically coded to obtain the topology feature index of the device to be detected; The infrared thermal image features are extracted to obtain the temperature field distribution spectrum of the infrared thermal image features. The dynamic measurement point parameters and the temperature field distribution spectrum are then fused using multi-source features to obtain the state feature vector of the device to be detected. The response feature information is obtained based on the topological feature index and the state feature vector.

5. The method according to claim 1, characterized in that, The partial discharge (PD) identification model includes a feature extraction module, a feature fusion module, and a state classification module. The state classification module includes a feature aggregation component and a state predictor. The candidate PD pulse data and the response feature information are input into the PD identification model, and feature extraction is performed to obtain the PD pulse features to be detected and the response distribution features. Based on the PD pulse features to be detected and the response distribution features, the target response identification result of the device to be detected is obtained, including: Based on the feature extraction module, the detection partial discharge pulse features of the candidate partial discharge pulse data are extracted, and the response distribution features of the response feature information are extracted; Based on the feature fusion module, the partial discharge pulse feature to be detected and the response distribution feature are concatenated, and the feature space of the concatenated feature is expanded to obtain a multi-source discharge feature field. Based on the feature aggregation component, the feature space of the multi-source discharge feature field is compressed to obtain the one-dimensional partial discharge pulse feature of the device under test. Based on the state predictor, the one-dimensional partial discharge pulse features to be detected are refined to predict the state attribution confidence of each candidate response feature in the set of candidate response feature information of the device to be detected. The candidate response feature information whose state attribution confidence is at the state determination threshold is determined as the target response identification result of the device to be detected.

6. The method according to claim 1, characterized in that, The partial discharge signal enhancement model is trained through the following steps: A high signal-to-noise ratio (SNR) clean partial discharge pulse sample set is obtained, which includes clean partial discharge pulse samples of various known discharge types and corresponding intensity levels. Obtain a real background noise sample set, which includes background noise samples without partial discharge activity collected from a complex acoustic environment similar to the device under test; The pure partial discharge pulse samples in the high signal-to-noise ratio pure partial discharge pulse sample set are scaled according to different preset amplitude ratios to obtain scaled partial discharge pulse samples of different intensities. Each of the scaled partial discharge pulse samples is mixed with a background noise sample randomly selected from the real background noise sample set at different preset superposition ratios to generate a mixed sample of partial discharge pulses simulating different signal-to-noise ratio conditions. The partial discharge pulse mixture sample is used as the input sample, and the corresponding unscaled high signal-to-noise ratio pure partial discharge pulse sample is used as the target output sample to form the partial discharge signal enhancement training sample set. The original partial discharge signal enhancement model is trained using the aforementioned partial discharge signal enhancement training sample set. The model parameters are adjusted by optimizing a preset loss function, enabling the original partial discharge signal enhancement model to learn the mapping relationship from noisy partial discharge pulses to clean partial discharge pulses, thereby achieving effective enhancement of input low signal-to-noise ratio partial discharge pulses, and obtaining the trained partial discharge signal enhancement model.

7. A partial discharge signal identification device, characterized in that, The device includes: The first module is used to acquire the raw partial discharge pulse data of the device under test; The second module is used to input the original partial discharge pulse data into a pre-trained partial discharge signal enhancement model to obtain enhanced partial discharge pulse data; The third module is used to match the enhanced partial discharge pulse data with a standard pulse signal to obtain candidate partial discharge pulse data. The fourth module is used to acquire the response characteristic information of the device under test; The fifth module is used to input the candidate partial discharge pulse data and the response feature information into the partial discharge identification model, perform feature extraction respectively to obtain the partial discharge pulse features to be detected and the response distribution features, and obtain the target response identification result of the device to be detected based on the partial discharge pulse features to be detected and the response distribution features.

8. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, It stores a computer program thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images