Converter substation partial discharge on-line monitoring signal processing method and system

Abstract

The application discloses a converter transformer partial discharge online monitoring signal processing method and system, and relates to the technical field of power equipment insulation state online monitoring. The method comprises the following steps: collecting an electromagnetic induction signal, and performing sampling digitization to obtain an original sampling digital stream; establishing a power frequency phase reference to generate a continuous power frequency phase stamp; pre-processing the original sampling digital stream and sending it to a fast and slow double link for parallel digital signal processing; the fast link performs full-rate transient partial discharge waveform capture and commutation dead zone gating; the slow link performs multi-rate statistical feature analysis to generate a phase resolution partial discharge spectrum; and the event waveform frame and the statistical results are output in stages. The application can realize real-time capture of high-fidelity transient waveforms and generation of continuous phase statistical spectrum under a complex time-varying and strong interference electromagnetic environment of a converter transformer, effectively suppress false alarms caused by commutation interference, reduce the pressure on edge processing and communication resources, and is suitable for online monitoring under a complex electromagnetic environment of a converter transformer.

Description

Technical Field

[0001] This application relates to the field of power equipment insulation condition monitoring technology, and in particular to a method and system for online monitoring signal processing of partial discharge in converter transformers. Background Technology

[0002] Converter transformers are key equipment in ultra-high voltage direct current (UHVDC) transmission systems, and their insulation condition directly affects the system's operational safety. Partial discharge is an important characteristic in the early development of insulation defects; therefore, conducting online monitoring of partial discharge in converter transformers is of great significance.

[0003] However, compared to ordinary AC transformers, the converter transformer's valve-side winding is connected to the converter valve. During operation, it is affected by the switching on and off of the converter devices and the commutation process, generating broadband transient interference pulses characterized by significant periodicity, a strong correlation with the power frequency phase, large amplitude, and wide bandwidth. The time-domain morphology and frequency-domain distribution of these commutation interferences may resemble real partial discharge signals, posing a significant challenge to partial discharge monitoring.

[0004] Existing online monitoring technologies for partial discharge in converter transformers mainly suffer from the following problems: (1) It is difficult to upload the full high-speed sampling data: The high-frequency sensor has a high monitoring bandwidth and the high-speed ADC has a high sampling rate. If the original sampling stream is uploaded in full, it will easily cause excessive pressure on the station network and the back-end storage.

[0005] (2) Simple threshold capture is easily affected by strong interference: If only a fixed threshold or simple event triggering method is used, continuous false alarms are easily generated when the commutation broadband group pulses appear densely, and even the cache resources and bus resources are filled with invalid data.

[0006] (3) Single capture mode is difficult to form continuous phase statistical spectrum: Although simple triggering device can capture discrete events, it is usually difficult to continuously output PRPD / PRPS statistical spectrum with power frequency phase information, which is not conducive to long-term evolution analysis of insulation state.

[0007] (4) Existing solutions often fail to balance real-time waveforms and long-term statistics: if high-speed waveforms are the main focus, the system data volume will be too large; if compressed statistics are the main focus, event details may be lost.

[0008] Therefore, existing solutions struggle to simultaneously achieve waveform fidelity, continuous statistics, real-time processing, and engineering deployability. There is an urgent need for a technical solution capable of reliably capturing partial discharge event waveform frames and continuously generating phase statistical maps in environments with strong interference from converter transformers. Summary of the Invention

[0009] The purpose of this application is to provide a method and system for online monitoring signal processing of partial discharge in converter transformers, so as to solve the problems in the prior art of difficult identification, high false alarm rate, high data upload pressure, and difficulty in continuously generating statistical graphs in the monitoring of partial discharge in converter transformers under complex time-varying electromagnetic environments.

[0010] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for processing online monitoring signals of partial discharge in a converter transformer, including: Collect electromagnetic induction signals output by the partial discharge monitoring sensor of the converter transformer; The electromagnetic induction signal is sent to the analog front-end unit for analog signal conditioning; The conditioned analog signal is sampled and digitized to obtain the original sampled digital stream; Establish a power frequency phase reference, and generate continuous power frequency phase stamps that correspond point-by-point to the original sampled digital stream based on the power frequency synchronization signal; The original sampled digital stream is preprocessed to obtain a preprocessed digital stream; The preprocessed digital stream is fed into a fast link and a slow link for dual-link parallel signal processing. The fast link is used to perform full-rate transient partial discharge waveform capture, including initial judgment based on thresholds, re-judgment based on waveform morphology, and commutation dead-zone gating based on the continuous power frequency phase stamp. When the trigger condition is met, the event waveform frame is captured. The slow link is used to perform multi-rate statistical feature analysis, including downsampling the data stream, mapping effective pulses to the power frequency phase interval, and statistically analyzing discharge parameters within the phase window to generate a phase-resolved partial discharge spectrum. The event waveform frame and the phase-resolved partial discharge map are encapsulated and output in a hierarchical manner.

[0011] Optionally, establishing the power frequency phase reference specifically includes: The AC voltage signal on the secondary side of the voltage transformer is acquired and zero-crossing detection is performed to obtain the digital synchronization pulse signal; Edge detection is performed on the digital synchronization pulse signal to extract the zero-crossing time sequence; A fully digital phase-locked loop is used to track the zero-crossing time sequence, generate a phase error signal, and correct the phase step. Based on the corrected phase step, the power frequency phase accumulation value is continuously output in each high-speed sampling clock cycle; The continuous power frequency phase stamp is obtained by performing fixed phase shift compensation on the accumulated power frequency phase value.

[0012] Optionally, the original digital sample stream undergoes signal preprocessing, specifically including: The original digital sample stream is bandpass filtered to suppress low-frequency drift and low-frequency noise. Notch filters are used to notch-process the bandpass filtered signal in order to suppress known narrowband communication interference frequencies.

[0013] Optionally, the initial threshold-based judgment includes: using an adaptive threshold algorithm to perform threshold judgment on the preprocessed digital stream signal; when the amplitude or energy of the digital stream signal exceeds the adaptive threshold, the initial threshold judgment is deemed to have passed; The adaptive threshold algorithm is an improved Otsu threshold algorithm, including: Let the candidate threshold be T, and divide the samples into background class and impulse class according to the candidate threshold; Calculate the probability of the background class, the probability of the impulse class, the mean of the background class, the mean of the impulse class, and the skewness; The objective function is determined based on the background class probability, impulse class probability, background class mean, impulse class mean, and skewness. Determine the optimal threshold based on the objective function; When the signal amplitude or energy exceeds the optimal threshold, the threshold is initially determined to be passed.

[0014] Optionally, the re-judgment based on waveform morphology specifically includes: Perform kurtosis morphological verification on suspected pulses that pass the initial threshold judgment, and calculate the kurtosis of the signal within the sliding window; When the kurtosis is greater than the preset kurtosis threshold, the suspected pulse is determined to have a spike non-Gaussian characteristic, and the re-evaluation is passed.

[0015] Optionally, the commutation dead-time gating based on the continuous power frequency phase stamp specifically includes: Based on the continuous power frequency phase stamp, determine whether the current sampling point or the current suspected event falls into the preset high-incidence phase interval of commutation interference; If it is located in the commutation dead zone, i.e., the preset high-incidence phase interval of commutation interference, but the initial threshold judgment and waveform morphology re-judgment are both passed, then it is specially marked: the partial discharge event is located in the commutation dead zone.

[0016] If it is not located in the commutation dead zone, and both the initial threshold judgment and waveform morphology re-judgment pass, then formal triggering is allowed.

[0017] Optionally, the step of intercepting the event waveform frame when the triggering condition is met specifically includes: Freeze the circular cache when the trigger condition is met; Extract data from the pre-triggered window, data near the trigger point, and subsequent triggered window; The pre-trigger window data, the data near the trigger point, and the subsequent trigger window data are spliced ​​together to form an event waveform frame, which is then written to the event storage area via DMA burst.

[0018] Optionally, the downsampling process on the data stream specifically includes: A cascaded integrator comb filter (CIC) and a half-band filter (HBF) are used to perform multi-stage decimation processing on the preprocessed digital stream.

[0019] Optionally, generating a phase-resolved partial discharge map specifically includes: Divide one power frequency cycle into multiple power frequency phase windows; Based on the continuous power frequency phase stamp, the extracted signal samples or identified valid pulses are mapped into the corresponding power frequency phase window; Within each phase window, at least one of the following is statistically analyzed: pulse count, peak amplitude, average energy, pulse density, or period trend value, to obtain the phase window statistical results. PRPD or PRPS maps are generated based on the phase window statistics.

[0020] Secondly, this application provides an online monitoring system for partial discharge of a converter transformer, comprising: The signal sensing unit is used to collect the electromagnetic induction signal output by the partial discharge monitoring sensor of the converter transformer; An analog front-end unit, connected to the signal sensing unit, is used to send electromagnetic induction signals into the analog front-end unit for analog signal conditioning; A sampling digitization unit, connected to the analog front-end unit, is used to sample and digitize the conditioned analog signal to obtain the original sampled digital stream; A power frequency synchronization unit is used to generate continuous power frequency phase stamps that correspond point-by-point to the original sampled digital stream based on the power frequency synchronization signal. A signal preprocessing unit, connected to the high-speed sampling unit, is used to preprocess the original sampled digital stream. The edge heterogeneous processing unit is used to perform dual-link parallel digital signal processing (fast link and slow link) on the preprocessed digital stream. The fast link is used to perform full-rate transient partial discharge waveform capture, including initial judgment based on a threshold, re-judgment based on waveform morphology, and commutation dead-zone gating based on the continuous power frequency phase stamp, capturing event waveform frames when trigger conditions are met. The slow link is used to perform multi-rate statistical feature analysis, including downsampling the data stream, mapping effective pulses to the power frequency phase interval, and statistically analyzing discharge parameters within the phase window to generate a phase-resolved partial discharge map. The encapsulation output unit is used to encapsulate and output the event waveform frame and the phase-resolved partial discharge spectrum in a hierarchical manner.

[0021] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method and system for processing signals for online monitoring of partial discharge in converter transformers, which has at least the following beneficial effects: Simultaneously, it preserves the transient waveforms and phase statistics of partial discharge: through the parallel division of labor between the two links, the fast link preserves the high-fidelity partial discharge event waveform frames, while the slow link continuously generates PRPD / PRPS spectra, thus resolving the contradiction that a single link cannot simultaneously capture waveform details and long-term statistics.

[0022] Effectively suppress false alarm storms caused by commutation interference: By using power frequency phase synchronization and commutation dead zone gating, false triggers in the high-incidence phase interval of commutation interference are suppressed at the trigger logic level, reducing buffer fullness and bus congestion caused by false alarms.

[0023] To improve the robustness of partial discharge identification in complex noise environments, a chain-like decision mechanism combining threshold initial judgment and waveform morphology re-judgment is adopted to better distinguish between real partial discharge pulses and pseudo pulses.

[0024] Reduce edge processing and communication resource pressure: Statistical processing is placed on slow links, high-fidelity waveform storage is limited to actual triggered events, and hierarchical output is used to replace the full upload of raw data, effectively reducing storage space, bus transmission burden and communication bandwidth pressure.

[0025] Suitable for engineering-based online deployment: The dual-link division of labor, phase synchronization, dead-zone gating and resource optimization structure is suitable for implementation in FPGA or SoC edge platforms, meeting the requirements of real-time performance, stability and long-term operation for on-site online monitoring of converter stations. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 This is a flowchart illustrating an online monitoring signal processing method for partial discharge of a converter transformer according to an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0030] In one exemplary embodiment, such as Figure 1 As shown, a method for processing online monitoring signals of partial discharge in a converter transformer is provided. This method is executed by a computer device, specifically by an edge computing terminal device and a server. In this embodiment, it includes the following steps: Step 101, Signal Acquisition: Acquire the electromagnetic induction signal output by the converter transformer partial discharge monitoring sensor; Specifically, a flexible high-frequency current transformer (HFCT) or other equivalent high-frequency sensor is preferred to collect the electromagnetic induction signal generated by partial discharge in the converter transformer.

[0031] Step 102, Signal Conditioning: The electromagnetic induction signal of the converter transformer is sent to the analog front-end unit for analog signal conditioning.

[0032] Step 103, sampling and digitization: The conditioned analog signal is sampled and digitized to obtain the original sampled digital stream.

[0033] The original sampled digital stream includes: real partial discharge pulses, background noise, narrowband communication interference, and commutated broadband group pulse interference.

[0034] Step 104, Power frequency phase synchronization: Establish a power frequency phase reference, and generate continuous power frequency phase stamps corresponding point-by-point to the original sampled digital stream based on the power frequency synchronization signal.

[0035] Step 104 specifically includes: Zero-crossing detection is performed on the secondary side signal of the PT; phase-locked tracking is performed in the digital domain; a continuous power frequency phase stamp aligned with the sampling clock is generated; and a fixed phase shift is compensated to obtain a calibration phase stamp. The detailed scheme is as follows: The AC voltage signal on the secondary side of a voltage transformer (PT) first undergoes isolation conditioning, amplitude limiting protection, and waveform shaping before being input to a zero-crossing comparator circuit. The zero-crossing comparator circuit outputs a digital synchronization pulse signal corresponding to the zero-crossing moment of the power frequency voltage. Subsequently, a programmable logic device (FPGA) performs clock domain synchronization and edge detection on the digital synchronization pulse signal to extract a continuous sequence of zero-crossing moments. Due to voltage transformer propagation delay, comparator propagation delay, harmonic distortion, and ambient noise, jitter exists at the zero-crossing moments. Linear interpolation based solely on adjacent zero-crossing points is insufficient to obtain a stable and reliable power frequency phase. Therefore, this application preferably employs a fully digital phase-locked loop (DPLL) to track the zero-crossing moment sequence.

[0036] The all-digital phase-locked loop (PLL) comprises a phase detection module, a loop filter module, and a numerically controlled oscillator (CNC) module. The phase detection module compares the deviation between the actually detected zero-crossing moment and the locally predicted zero-crossing moment to generate a phase error signal. The loop filter module smooths and dynamically corrects the phase error signal. The CNC oscillator module continuously outputs the accumulated power frequency phase value within each high-speed sampling clock cycle based on the corrected phase step. Therefore, between any two zero-crossing points, the system can continuously generate a continuously changing power frequency phase value for each sampling point, rather than obtaining discrete synchronization information only at the zero-crossing moment.

[0037] For the Each high-speed sampling point corresponds to a power frequency phase. It can be represented as: ; in, Indicates the first The power frequency phase corresponding to each high-speed sampling point This indicates the phase step size within the current sampling period; The phase step is dynamically updated based on the correction result output by the all-digital phase-locked loop, and the generated phase value can be mapped to... or Within the phase range.

[0038] Furthermore, considering that voltage transformers, conditioning circuits, zero-crossing comparator circuits, and digital processing links may introduce fixed delays, this application performs fixed phase shift compensation on the aforementioned continuous power frequency phase values ​​to obtain calibrated power frequency phase stamps corresponding to each point of the partial discharge sampling data. The fixed phase shift compensation can be determined through on-site calibration, offline testing, or comparison with a standard power frequency reference.

[0039] The resulting calibrated power frequency phase stamp can be correlated point-by-point with the original high-speed sampling data, or with each partial discharge triggering event, and can be used for at least the following purposes: Dead-zone shielding or gating suppression is performed on samples or events located in the preset high-incidence phase interval of commutation interference. Partial discharge pulses are mapped onto the corresponding power frequency phase window to generate PRPD, PRPS, or other phase-resolved statistical results.

[0040] This step enables each sampling point or event in the original sampled digital stream to be associated with the corresponding power frequency phase interval.

[0041] Step 105, Signal preprocessing: The original sampled digital stream is preprocessed to obtain a preprocessed digital stream.

[0042] Preprocessing preferably includes the following sub-steps: Step 1051: Bandpass Filtering The original sampled digital stream is bandpass filtered to suppress low-frequency drift and low-frequency noise.

[0043] Step 1052: Notch Suppression Notch filtering is applied to the bandpass filtered signal to suppress narrowband communication interference frequencies.

[0044] Step 106, Dual-link parallel digital signal processing: The preprocessed digital stream is sent to the fast link and the slow link for dual-link parallel signal processing; wherein, the fast link is used to perform full-rate transient partial discharge waveform capture, including initial judgment based on threshold, re-judgment based on waveform morphology, and commutation dead-zone gating based on the continuous power frequency phase stamp, and the event waveform frame is captured when the trigger condition is met; the slow link is used to perform multi-rate statistical feature analysis, including downsampling the data stream, mapping the effective pulses to the power frequency phase interval, and statistically analyzing the discharge parameters within the phase window to generate a phase-resolved partial discharge spectrum.

[0045] The "fast link" refers to the use of a high-speed ADC (e.g., a sampling rate of 250 MSPS, more than 8 times the highest frequency of the partial discharge signal (30MHz)) for signal sampling and digitization, followed by subsequent processing. Therefore, the obtained raw sampled digital stream can retain the waveform variations of the partial discharge signal with high fidelity. A partial discharge event refers to the high-speed acquisition triggered by the fast link. The acquired digital (signal) stream may contain a partial discharge signal (valid partial discharge event) or may not (invalid partial discharge event, where the signal amplitude increases due to noise or sudden interference, exceeding the trigger threshold).

[0046] The term "multi-rate" in slow links refers to the highest sampling rate (data rate, specifically 250 Mbps) of fast links. Multi-rate is achieved through decimation: different decimation factors are used to obtain different data rates.

[0047] The reason for decimation is to perform long-term statistical feature analysis (such as multi-cycle statistical analysis using power frequency cycles (20ms) as the unit). If the analysis is performed directly at the highest data rate of 250M, on the one hand, such a high data rate will bring huge data processing pressure to the back-end computing unit; on the other hand, since the sampling rate is much higher than the highest frequency of the partial discharge signal, the sampled digital stream is "oversampled". Decimation can reduce the data rate, thereby reducing the computational processing pressure, but at the same time, it will not lose signal feature information (according to the Nyquist sampling theorem: if the sampling frequency is greater than or equal to twice the highest frequency component in the original signal, the original continuous signal can be reconstructed from the discrete sampling points without loss).

[0048] Multi-rate statistical characterization involves performing signal characteristic analysis (such as time / frequency domain analysis of the signal, and phase domain analysis combined with power frequency synchronization signal) on multiple digital streams obtained after extraction.

[0049] Specifically, the fast link processing steps are as follows: Fast link execution full-rate transient partial discharge waveform capture; The fast link performs acquisition event triggering and partial discharge waveform capture at high sampling rates, preferably including the following sub-steps: Step 1061a: Initial Threshold Judgment An adaptive threshold algorithm is used for initial screening. In a preferred embodiment, an improved Otsu threshold algorithm is used.

[0050] Let the candidate threshold be T. Based on this threshold, samples are divided into background and impulse classes. Then the objective function is... It can be represented as: ; in, Background class probability; For pulse class probabilities; The mean of the background class; The mean of the pulse class; Skewness; Weighting functions for skewnesses.

[0051] When the signal amplitude or energy exceeds the threshold determined by the objective function, the initial threshold judgment is considered to have passed.

[0052] Step 1062a: Waveform morphology verification For suspected pulses that pass the initial threshold judgment, kurtosis morphology verification is further performed to distinguish between real partial discharge pulses and pseudo pulses.

[0053] Preferably, the formula for calculating the kurtosis value is: ;in, The length of the sliding window; For the first in the window Each sample value; The mean of the window; For window kurtosis.

[0054] when If the kurtosis exceeds the preset threshold, the pulse is determined to have a spike-like non-Gaussian characteristic, and the pulse passes the re-evaluation.

[0055] Step 1063a: Commutation dead zone gating check Based on the calibration phase stamp obtained in step 104, determine whether the current sampling point or the current suspected event falls within the preset high-incidence phase range of commutation interference: If the partial discharge event is located in the commutation dead zone (i.e., the preset high-incidence phase interval of commutation interference), but the initial threshold judgment and waveform morphology re-judgment are both passed, then a special mark is made: the partial discharge event is located in the commutation dead zone.

[0056] If it is not located in the commutation dead zone, and both the initial threshold judgment and the waveform morphology re-judgment pass, then proceed to step 1064a.

[0057] Step 1064a: Circular buffer freezing and partial discharge event frame stitching When the triggering conditions described in steps 1061a to 1063a are met, the circular buffer is frozen, and the pre-trigger window data, the data near the trigger point, and the subsequent trigger window data are extracted.

[0058] The above data are spliced ​​together to form a partial discharge event waveform frame.

[0059] Step 1065a: DMA burst write event storage area The event waveform frame is written to the event storage area via DMA for subsequent uploading and diagnostic use.

[0060] Slow link execution multi-rate statistical characteristic analysis.

[0061] The slow link performs decimation downsampling and phase statistics processing on the preprocessed digital stream, preferably including the following sub-steps: Step 1061b: Multi-stage extraction processing Multi-stage decimation processing is performed using a cascaded integrator-comb filter (CIC) and a half-band filter (HBF) to reduce data rate and computational overhead.

[0062] In this application, the fast link uses a high-speed ADC to sample and digitize the partial discharge signal at a sampling rate of 250 MSPS. This sampling rate is more than twice that of the highest frequency component of the partial discharge signal (approximately 30 MHz), representing a typical oversampling scenario. The slow link is used to perform long-term statistical feature analysis (e.g., multi-cycle statistics in units of 20 ms at the power frequency). If processed directly at the raw data rate of 250 MSPS, it would place a huge burden on the back-end computing unit's data throughput and computational demands. Therefore, it is necessary to gradually reduce the data rate through multi-stage decimation processing, while adhering to the Nyquist sampling theorem to ensure that no effective signal feature information is lost during the decimation process.

[0063] The multi-stage decimation process employs a multi-stage cascaded architecture combining a cascaded integrator-comb filter (CIC) and a half-band filter (HBF), as detailed below: Level 1: CIC Filtering and High-Magnification Primary Decimation A CIC filter consists of an N-stage integrator and an N-stage comb, with a downsampling factor R used for decimation. The system transfer function of the CIC filter is: ;in, To extract factors, The cascade order is denoted as .

[0064] The CIC filter operates as follows: the integrator operates in the high-speed sampling clock domain, accumulating the input signal point by point; after decimation factor... After downsampling (i.e., each (One input sample retains one output sample). The comb section operates in the slowed-down clock domain, performing differential operations on the downsampled signal. The CIC filter requires no multipliers, only adders and delay registers, resulting in extremely low hardware resource consumption. It is particularly suitable for implementing high-rate primary decimation in FPGAs.

[0065] In a preferred embodiment of this application, the decimation factor of the CIC filter It can be set to 8~64 (e.g.) ), cascade order It can be set to 3 to 5 orders (e.g.) After CIC primary extraction, the original data rate of 250 MSPS can be reduced to approximately 15.625 MSPS. (For example) or lower, thereby significantly reducing the computational speed requirements of subsequent filters.

[0066] Because the amplitude-frequency response of the CIC filter is non-flat within the passband. Due to the function's attenuation characteristics (i.e., droop exists within the passband) and limited stopband attenuation, CIC decimation usually requires subsequent compensation and fine filtering.

[0067] Level 2 and subsequent levels: HBF filtering and fine extraction The digital stream decimated by CIC enters a single-stage or multi-stage half-band filter (HBF) for further fine decimation. An HBF is a special type of symmetrical FIR filter whose cutoff frequency is exactly half the Nyquist frequency of the input signal, and about half of the filter coefficients are zero. Therefore, the number of multipliers required is only half that of a regular FIR filter of the same order, making it ideal for implementing efficient 2x decimation in FPGAs.

[0068] Each HBF stage performs a 2x extraction (i.e., the extraction factor is 2), and this can be achieved by cascading multiple HBF stages. Total extraction times ( (This refers to the number of HBF cascades). Before decimation, the HBF performs low-pass filtering on the signal to suppress high-frequency components that will cause aliasing due to decimation, ensuring that the decimated signal meets the Nyquist criterion and does not produce spectral aliasing. In a preferred embodiment, 2 to 3 HBF stages are cascaded after CIC, with each stage performing a 2x decimation.

[0069] The overall effect of multi-level extraction is as follows: After cascading processing of CIC and multi-level HBF, the total extraction factor is... for: in, For CIC extraction factors, For HBF cascade series.

[0070] by , For example, the total sampling factor is The original 250 MSPS data stream was reduced to approximately 7.8125 MSPS after multiple stages of extraction, reaching a data rate suitable for subsequent statistical feature analysis. The extraction factor of CIC and the number of HBF cascades can be flexibly adjusted according to actual monitoring needs and FPGA resources.

[0071] This application adopts a multi-stage decimation architecture cascaded with CIC and HBF, which has the following technical advantages: CIC undertakes high-rate primary decimation without the need for multipliers, resulting in extremely low hardware resource consumption; HBF undertakes fine decimation and passband compensation, with half of the coefficients being zero, and the multiplication operation is only half that of ordinary FIR; the multi-stage cascaded step-by-step speed reduction allows for controllable filter order and computational complexity at each stage, and the overall architecture is suitable for pipelined implementation on FPGA or SoC edge platforms, meeting the real-time requirements of online monitoring.

[0072] Step 1062b: Power frequency phase mapping Based on continuous power frequency phase stamps, the extracted signal samples or identified valid pulses are mapped to... Within the power frequency phase range.

[0073] Mapping the extracted digital signal samples or identified valid partial discharge pulses to the power frequency phase interval mainly relies on the continuous power frequency phase stamps generated in step 104. These continuous power frequency phase stamps correspond point-to-point with the original high-speed sampled digital stream and can be transmitted to the extracted slow-link data stream through sampling point indexes or timestamp synchronization relationships. The specific mapping process is as follows: 1. Determine the sampling time corresponding to the sample or pulse: For each extracted digital signal sample, its corresponding sampling point index in the original high-speed sampling digital stream is deduced based on its sequence number in the extracted data stream and the total extraction rate.

[0074] For a valid partial discharge pulse identified from a signal sample, the peak point, energy center point, or trigger point of the pulse can be selected as the representative sampling point of the pulse.

[0075] 2. Read the corresponding continuous power frequency phase stamp: Based on the above representative sampling point index, the corresponding power frequency phase value is read from the continuous power frequency phase stamp sequence generated by the power frequency synchronization unit.

[0076] 3. Phase normalization processing: If the accumulated phase value exceeds one power frequency cycle, a modulo operation is performed to bring it within the range of one standard power frequency cycle.

[0077] ; in, For the first The cumulative value of continuous power frequency phase corresponding to each sampling point This is the normalized power frequency phase.

[0078] 4. Map to the corresponding phase window: Divide one power frequency cycle into There are 1 phase window, and the width of each phase window is: Then phase value The corresponding phase window number can be represented as: ; in, The phase window is numbered, and its value range is [value range missing]. .

[0079] 5. Write the corresponding phase window: Once the phase window number is determined, the statistical information such as amplitude, energy, count, and occurrence period number corresponding to the sample or the effective partial discharge pulse is updated to the corresponding phase window for subsequent formation of the phase-resolved statistical matrix.

[0080] Through the above processing, each extracted signal sample or effective partial discharge pulse can be accurately associated with the power frequency phase interval corresponding to its occurrence time, thereby realizing the correspondence between the partial discharge pulse and the power frequency phase, and providing basic data for generating PRPD and PRPS spectra.

[0081] Preferably, one power frequency cycle is divided into multiple phase windows, for example: 64 phase windows; 128 phase windows; 256 phase windows; 512 phase windows.

[0082] The main purpose of dividing a power frequency cycle into multiple phase windows is to perform phase-resolved statistics on partial discharge pulses, enabling the correlation between partial discharge activity and the power frequency phase to be quantified and graphically represented. Specifically, dividing the cycle into multiple phase windows has the following functions: 1. Achieving phase-resolved statistics: Partial discharge is usually related to the electric field strength at the insulation defect, and the electric field strength changes periodically with the phase of the power frequency voltage. Therefore, by dividing a power frequency cycle into multiple phase windows, the occurrence frequency, amplitude, and energy distribution of partial discharge pulses in different phase intervals can be statistically analyzed, thereby reflecting the phase characteristics of partial discharge.

[0083] 2. Improve the ability to identify partial discharge types: Different types of insulation defects typically have different phase distribution characteristics. For example, internal discharge, surface discharge, and floating potential discharge may have different phase, amplitude distributions, and repetition rates within the power frequency cycle. Phase window statistics can provide a characteristic basis for subsequent defect type identification.

[0084] 3. Suppressing the influence of random noise: Random noise typically lacks stable power frequency phase correlation, while real partial discharges often exhibit repetitive characteristics within specific phase intervals. Through multi-cycle phase window accumulation statistics, the influence of random noise can be weakened, enhancing the identifiability of stable discharge modes.

[0085] 4. Supports PRPD / PRPS map generation: Both PRPD and PRPS maps require power frequency phase as the basic coordinate. After phase window division, discharge pulses in different cycles can be accumulated, arranged, and statistically analyzed according to their phase positions to form two-dimensional or three-dimensional phase-resolved maps.

[0086] 5. Balancing resolution and computational load: The number of phase windows can be configured according to engineering requirements, such as 64, 128, 256, or 512 phase windows. A higher number of phase windows results in higher phase resolution; a lower number of phase windows results in lower computational and storage requirements. Therefore, the choice can be made flexibly based on the device's computing power, communication bandwidth, and diagnostic accuracy requirements.

[0087] Therefore, dividing the data into multiple phase windows is not simply a data grouping process, but a key step in establishing a statistical correlation between partial discharge events and power frequency phases, which can significantly improve the interpretability, anti-interference ability, and long-term trend analysis capability of partial discharge patterns.

[0088] Step 1063b: Phase window statistics Statistically analyze at least one of the following parameters within each phase window: Pulse counting; Peak amplitude; Average energy; Pulse density; Cyclical trend value.

[0089] Step 1064b: Generate PRPD / PRPS maps Partial discharge statistical maps are generated based on phase window statistical results, including but not limited to: Phase-resolved partial discharge (PRPD) pattern. Phase-Resolved Pulse Sequence (PRPS) map.

[0090] The generation of partial discharge statistical maps (PRPD and PRPS maps) is based on the phase window statistical results of the slow link output. The core technology involves mapping the phase domain of the partial discharge pulse based on continuous power frequency phase stamps, performing multi-parameter cumulative statistics within the phase window, and forming a map data matrix that can be displayed and used for diagnostic analysis at the back end. The specific process is as follows.

[0091] 1. Power frequency phase window division: Divide a complete power frequency cycle (0°~360°) into equal parts. There are 1 phase window, and the phase width of each phase window is 1 / 2. . The value can be configured according to the statistical resolution requirements, with preferred values ​​including 64, 128, 256 or 512.

[0092] No. The phase intervals corresponding to each phase window are: ; 2. Partial discharge pulse identification and phase mapping: The slow link, based on the extracted digital stream, identifies and records information (pulse amplitude, energy, and peak time) of the partial discharge pulse according to the decision threshold (amplitude threshold, energy threshold, or adaptive threshold) and in coordination with the fast link. Then, it maps the partial discharge pulse to the power frequency phase interval. For details, please refer to "II. Mapping Partial Discharge Pulses to Power Frequency Phase Interval" above.

[0093] 3. Cumulative statistics of multiple parameters within the phase window: Within each phase window, multidimensional parameter accumulation statistics are performed on all partial discharge pulses falling within that window. The statistically analyzed parameters include at least one or more of the following: (1) Pulse counting Record the first entry within the statistical period The total number of valid pulses detected cumulatively in each phase window reflects the frequency of partial discharge events within that phase interval.

[0094] (2) Peak amplitude Record number The maximum amplitude value among all effective pulses within a phase window, i.e. ,in, For the first The set of pulses within a phase window. This parameter reflects the maximum intensity of the discharge event within that phase interval.

[0095] (3) Average energy : Calculate the first The average value of all effective pulse energies within each phase window, i.e. ,in, For the first The energy of each pulse. This parameter reflects the average intensity level of the discharge event within this phase interval.

[0096] (4) Pulse density : Calculate the first unit of time or unit power frequency cycle. The pulse generation density in each phase window, i.e. ,in, This is the statistical duration. This parameter reflects the intensity of discharge activity along the phase distribution.

[0097] (5) Periodic trend value: The statistical parameters of the same phase window within multiple consecutive power frequency cycles are tracked to record their changes over time, which are used for long-term evolution analysis of insulation status.

[0098] 4. Data Matrix Construction and Map Generation After multiple power frequency cycles (statistical cycles) Contains After accumulating statistics for each power frequency cycle, the statistical results of each phase window are organized into a spectral data matrix. 1) PRPD map generation PRPD (Phase-Resolved Partial Discharge) plots are plotted using the power frequency phase window number as the horizontal axis, the discharge amplitude or discharge quantity as the vertical axis, and the pulse count or pulse density within each (phase, amplitude) cell as the third-dimensional statistic (color depth or z-axis height can be used), forming a two-dimensional scatter plot or a three-dimensional histogram.

[0099] Specifically, the partial discharge pulses can be written into a two-dimensional statistical matrix according to the "phase window number" and "amplitude interval number": in, Indicates the phase window number. Indicates the amplitude range number, This indicates the number of pulses that fall within the phase window and amplitude range.

[0100] Backend based on matrix A PRPD (Partial Discharge Phase Detection) map is generated, where color intensity or dot density represents the number of pulse repetitions. This map reflects the concentration, amplitude distribution, and repetition rate of partial discharges at the power frequency phase.

[0101] 2) PRPS map generation PRPS (Phase-Resolved Pulse Sequence) map. When generating a PRPS map, in addition to the power frequency phase number and discharge amplitude, the power frequency cycle number or time series information is also incorporated. Specifically, each effective partial discharge pulse is written into a three-dimensional data structure according to its power frequency cycle, phase window, and amplitude, which can be represented as: ; in, Indicates the power frequency cycle number. Indicates the phase window number. This indicates the pulse amplitude, maximum amplitude, or equivalent statistical amplitude within the phase window of that period.

[0102] This results in a pulse sequence map based on "cycle number - power frequency phase - discharge amplitude". The PRPS map can reflect the evolution of partial discharge over time or power frequency cycle, and is suitable for observing the persistence, intermittency and development trend of discharge activity.

[0103] 5. Image encapsulation and output: After generating the PRPD or PRPS spectral data, the system encapsulates information such as the spectral matrix, number of phase windows, amplitude scale, number of statistical periods, timestamp, and device number, and periodically outputs it to the backend application platform. The backend platform can then use this data to perform spectral plotting, trend analysis, insulation status assessment, and alarm determination.

[0104] Using the above-mentioned technical means, continuously acquired partial discharge signals can be converted into statistical spectra with power frequency phase characteristics, thereby enabling long-term monitoring and trend diagnosis of partial discharge activity in converter transformers.

[0105] Step 107, Encapsulation and Output: Encapsulate and output the event waveform frame and the phase-resolved partial discharge spectrum in a hierarchical manner.

[0106] The event waveform frames output by the fast link and the statistical features output by the slow link are encapsulated and output separately.

[0107] Preferably: Fast link outputs key partial discharge event waveform frames; Slow link outputs periodic statistical matrices or spectral parameters; The backend application platform receives the data and performs visualization, trend analysis, and alarm processing.

[0108] Based on the same inventive concept, this application also provides a converter transformer partial discharge online monitoring signal processing system for implementing the converter transformer partial discharge online monitoring signal processing method described above. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more converter transformer partial discharge online monitoring signal processing system embodiments provided below can be found in the limitations of the converter transformer partial discharge online monitoring signal processing method described above, and will not be repeated here.

[0109] In one exemplary embodiment, a converter transformer partial discharge online monitoring signal processing system is provided, comprising: The signal sensing unit is used to collect electromagnetic induction signals generated by partial discharge in the converter transformer. An analog front-end unit, connected to the signal sensing unit, is used to perform analog signal conditioning on the output signal of the signal sensing unit; A sampling digitization unit, connected to the analog front-end unit, is used to sample and digitize the analog signal output by the analog front-end unit to obtain the original sampled digital stream; A power frequency synchronization unit is used to generate continuous power frequency phase stamps that correspond point-by-point to the original sampled digital stream based on the power frequency synchronization signal. A signal preprocessing unit, connected to the high-speed sampling unit, is used to preprocess the original sampled digital stream. The edge heterogeneous processing unit is used to perform dual-link parallel digital signal processing (fast link and slow link) on the preprocessed digital stream. The fast link is used to perform full-rate transient partial discharge waveform capture, including initial judgment based on a threshold, re-judgment based on waveform morphology, and commutation dead-zone gating based on the continuous power frequency phase stamp, capturing event waveform frames when trigger conditions are met. The slow link is used to perform multi-rate statistical feature analysis, including downsampling the data stream, mapping effective pulses to the power frequency phase interval, and statistically analyzing discharge parameters within the phase window to generate a phase-resolved partial discharge map. The encapsulation output unit is used to encapsulate and output the event waveform frame and the phase-resolved partial discharge spectrum in a hierarchical manner.

[0110] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0111] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for processing online monitoring signals of partial discharge in a converter transformer, characterized in that, The online monitoring signal processing method for partial discharge of the converter transformer includes: Collect electromagnetic induction signals output by the partial discharge monitoring sensor of the converter transformer; The electromagnetic induction signal is sent to the analog front-end unit for analog signal conditioning; The conditioned analog signal is sampled and digitized to obtain the original sampled digital stream; Establish a power frequency phase reference, and generate continuous power frequency phase stamps that correspond point-by-point to the original sampled digital stream based on the power frequency synchronization signal; The original sampled digital stream is preprocessed to obtain a preprocessed digital stream; The preprocessed digital stream is fed into a fast link and a slow link for dual-link parallel signal processing. The fast link is used to perform full-rate transient partial discharge waveform capture, including initial judgment based on thresholds, re-judgment based on waveform morphology, and commutation dead-zone gating based on the continuous power frequency phase stamp. When the trigger condition is met, the event waveform frame is captured. The slow link is used to perform multi-rate statistical feature analysis, including downsampling the data stream, mapping effective pulses to the power frequency phase interval, and statistically analyzing discharge parameters within the phase window to generate a phase-resolved partial discharge spectrum. The event waveform frame and the phase-resolved partial discharge map are encapsulated and output in a hierarchical manner.

2. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, The establishment of the power frequency phase reference, based on the power frequency synchronization signal, generates continuous power frequency phase stamps corresponding point-by-point to the original sampled digital stream, specifically including: The AC voltage signal on the secondary side of the voltage transformer is acquired and zero-crossing detection is performed to obtain the digital synchronization pulse signal; Edge detection is performed on the digital synchronization pulse signal to extract the zero-crossing time sequence; A fully digital phase-locked loop is used to track the zero-crossing time sequence, generate a phase error signal, and correct the phase step. Based on the corrected phase step, the power frequency phase accumulation value is continuously output in each high-speed sampling clock cycle; The continuous power frequency phase stamp is obtained by performing fixed phase shift compensation on the accumulated power frequency phase value.

3. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, The original digital sample stream undergoes signal preprocessing, specifically including: The original digital sample stream is bandpass filtered using a bandpass filter to suppress low-frequency drift and low-frequency noise; Notch filters are used to notch-process the bandpass filtered signal in order to suppress known narrowband communication interference frequencies.

4. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, The initial threshold-based judgment includes: using an adaptive threshold algorithm to make a threshold decision on the preprocessed digital stream signal; when the amplitude or energy of the digital stream signal exceeds the adaptive threshold, the initial threshold judgment is deemed to have passed. The adaptive threshold algorithm is an improved Otsu threshold algorithm, including: Let the candidate threshold be T, and divide the samples into background class and impulse class according to the candidate threshold; Calculate the probability of the background class, the probability of the impulse class, the mean of the background class, the mean of the impulse class, and the skewness; The objective function is determined based on the background class probability, impulse class probability, background class mean, impulse class mean, and skewness. Determine the optimal threshold based on the objective function; When the signal amplitude or energy exceeds the optimal threshold, the threshold is initially determined to be passed.

5. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, The re-judgment based on waveform morphology specifically includes: Perform kurtosis morphological verification on suspected pulses that pass the initial threshold judgment, and calculate the kurtosis of the signal within the sliding window; When the kurtosis is greater than the preset kurtosis threshold, the suspected pulse is determined to have a spike non-Gaussian characteristic, and the re-evaluation is passed.

6. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, Commutation dead-time gating based on the continuous power frequency phase stamp specifically includes: Based on the continuous power frequency phase stamp, determine whether the current sampling point or the current suspected event falls into the preset high-incidence phase interval of commutation interference; If it is located in the commutation dead zone, i.e., the preset high-incidence phase interval of commutation interference, but the initial threshold judgment and waveform morphology re-judgment are both passed, then it is specially marked: the partial discharge event is located in the commutation dead zone. If it is not located in the commutation dead zone, and both the initial threshold judgment and waveform morphology re-judgment pass, formal triggering is allowed.

7. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, The step of capturing the event waveform frame when the triggering condition is met specifically includes: Freeze the circular cache when the trigger condition is met; Extract data from the pre-triggered window, data near the trigger point, and subsequent triggered window; The pre-trigger window data, the data near the trigger point, and the subsequent trigger window data are spliced ​​together to form an event waveform frame, which is then written to the event storage area via DMA burst.

8. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, The downsampling process of the data stream specifically includes: A cascaded integrator comb filter (CIC) and a half-band filter (HBF) are used to perform multi-stage decimation processing on the preprocessed digital stream.

9. The online monitoring signal processing method for partial discharge of converter transformers according to claim 1, characterized in that, The generation of the phase-resolved partial discharge pattern specifically includes: Divide one power frequency cycle into multiple power frequency phase windows; Based on the continuous power frequency phase stamp, the extracted signal samples or identified valid pulses are mapped into the corresponding power frequency phase window; Within each phase window, at least one of the following is statistically analyzed: pulse count, peak amplitude, average energy, pulse density, or period trend value, to obtain the phase window statistical results. PRPD or PRPS maps are generated based on the phase window statistics.

10. An online monitoring system for partial discharge of a converter transformer, characterized in that, The converter transformer partial discharge online monitoring system includes: The signal sensing unit is used to collect the electromagnetic induction signal output by the partial discharge monitoring sensor of the converter transformer; An analog front-end unit, connected to the signal sensing unit, is used to send electromagnetic induction signals into the analog front-end unit for analog signal conditioning; A sampling digitization unit, connected to the analog front-end unit, is used to sample and digitize the conditioned analog signal to obtain the original sampled digital stream; A power frequency synchronization unit is used to generate continuous power frequency phase stamps that correspond point-by-point to the original sampled digital stream based on the power frequency synchronization signal. A signal preprocessing unit, connected to the high-speed sampling unit, is used to preprocess the original sampled digital stream. The edge heterogeneous processing unit is used to perform dual-link parallel digital signal processing (fast link and slow link) on the preprocessed digital stream. The fast link is used to perform full-rate transient partial discharge waveform capture, including initial judgment based on a threshold, re-judgment based on waveform morphology, and commutation dead-zone gating based on the continuous power frequency phase stamp, capturing event waveform frames when trigger conditions are met. The slow link is used to perform multi-rate statistical feature analysis, including downsampling the data stream, mapping effective pulses to the power frequency phase interval, and statistically analyzing discharge parameters within the phase window to generate a phase-resolved partial discharge map. The encapsulation output unit is used to encapsulate and output the event waveform frame and the phase-resolved partial discharge spectrum in a hierarchical manner.

Images