An adaptive partial discharge pC calibration method based on a single-chip microcomputer

Abstract

The application provides a single-chip microcomputer-based adaptive partial discharge pC calibration method, comprising: for the plurality of continuous phase windows, judging the signal state of the initial pulse peak value sequence in each phase window based on the change of the amplitude data in the initial pulse peak value sequence between adjacent power frequency periods; for the discharge clear window, obtaining a discharge characteristic peak value in the discharge clear window, performing amplitude concentration trend processing on the discharge characteristic peak value, and extracting a clear discharge reference value reflecting the true discharge intensity; calculating the link transfer gain corresponding to the current power frequency period according to the stable interference reference value and the clear discharge reference value, performing amplitude conversion on the amplitude data in the initial pulse peak value sequence according to the link transfer gain, and obtaining a partial discharge calibration value.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an adaptive partial discharge pC calibration method based on a microcontroller. Background Technology

[0002] Partial discharge detection is indispensable for assessing the insulation condition of high-voltage power equipment. Microcontroller-based PC calibration is a crucial step in converting measured electrical signals into actual discharge quantities, and its accuracy directly determines the reliability of subsequent fault diagnosis and equipment lifespan prediction. In converter stations, frequency converter drives, and other similar applications, the microcontroller needs to continuously perform this calibration on-site to ensure long-term reliability of monitoring results. Currently, the common practice is to filter out all non-discharge components as noise and calibrate the link gain only based on the amplitude of the extracted discharge pulses. This approach assumes that the discharge signal, at any power frequency phase, reflects the true transmission characteristics of the link better than interference. The problem is that on-site interference is not random noise; it has a clear coupling relationship with the power frequency, and this assumption is not always valid. Processes such as converter valve triggering and thyristor commutation generate strong pulses that are strictly phase-locked with the power frequency, appearing in fixed phase intervals within a few phase ranges of the power frequency cycle with extremely high repeatability. This phase-locked characteristic presents the first challenge: interference does not uniformly spread throughout the entire cycle but rather concentrates and repeatedly appears within a predictable phase window. This leads to a second, more challenging situation: when these fixed windows happen to cover the phase where the actual discharge occurs, the weak discharge characteristics are completely suppressed by strong interference. The amplitude of the discharge pulse extracted from the residual components fluctuates wildly between adjacent power frequency cycles, with the fluctuation exceeding the stability of the interference itself, resulting in a reversal phenomenon where the more it is removed, the more chaotic it becomes. To give a concrete example, the peak value of the discharge pulse read by the microcontroller in a certain phase window may be a few millivolts in one cycle and jump to tens of millivolts in the next cycle. The calculated pC value fluctuates back and forth, while the peak value of the interference pulses discarded as noise in the same window remains stable at the same level. In this case, insisting on using the discharge characteristics as a reference makes the calibration of this phase segment completely ineffective, and this is not the case throughout the entire cycle; the discharge signal in other phase windows is still clearly distinguishable. Therefore, in field environments where power frequency phase-locked interference periodically covers the discharge phase and the signal identifiability fluctuates significantly with the phase, how to enable a microcontroller with limited computing power to determine which is the more stable and more realistic reference for the link characteristics in different segments of the power frequency cycle, and avoid continuing to rely on distorted discharge characteristics in the window where the signal is submerged, has become a key issue in ensuring the consistent reliability of partial discharge pC calibration across the entire phase range. Summary of the Invention

[0003] This invention provides an adaptive partial discharge pC calibration method based on a microcontroller, mainly comprising:

[0004] The original electrical signal generated during the operation of high-voltage power equipment is acquired, the initial pulse peak sequence within the power frequency cycle is extracted from the original electrical signal, and multiple continuous phase windows are divided within the power frequency cycle according to the time information corresponding to the initial pulse peak sequence.

[0005] For the multiple consecutive phase windows, based on the changes in amplitude data in the initial pulse peak sequence between adjacent power frequency cycles, the signal state of the initial pulse peak sequence within each phase window is determined;

[0006] If the signal state is in the first state, the current phase window is determined to be a clear discharge window; if the signal state is in the second state, the current phase window is determined to be an interference coverage window.

[0007] For the interference coverage window, the peak values ​​of interference pulses for multiple consecutive cycles are obtained, and the peak values ​​of interference pulses are classified to remove discrete abnormal extreme values ​​and extract stable interference reference values.

[0008] For the clear discharge window, the discharge characteristic peak value is obtained, and the amplitude central trend processing is performed on the discharge characteristic peak value to extract the clear discharge reference value.

[0009] Based on the stable interference reference value and the clear discharge reference value, the link transfer gain corresponding to the current power frequency cycle is jointly determined when both exist. When only one of them exists, the link transfer gain is determined according to the existing reference value. The amplitude data in the initial pulse peak sequence is converted according to the link transfer gain to obtain the partial discharge calibration value.

[0010] Furthermore, the step of extracting the initial pulse peak sequence within the power frequency cycle from the original electrical signal includes:

[0011] The original electrical signal is obtained by a high-frequency current sensor installed on the grounding wire of the high-voltage power equipment;

[0012] The original electrical signal is sent to the acquisition port of the microcontroller at a preset sampling frequency;

[0013] Peak detection is performed on sampling points whose amplitude exceeds a preset noise floor threshold within the power frequency cycle to extract the initial pulse peak sequence. The preset noise floor threshold is the upper limit of the amplitude of the static noise of the link. At the same time, the timestamp corresponding to each peak point in the initial pulse peak sequence is recorded.

[0014] Furthermore, the step of dividing the power frequency period into multiple consecutive phase windows based on the time information corresponding to the initial pulse peak sequence includes:

[0015] Based on the relative time difference between the timestamp corresponding to each peak point in the initial pulse peak sequence and the zero-crossing point of the power frequency reference signal, each peak point is mapped to the corresponding phase angle within the power frequency cycle, and a continuous phase window is divided on the power frequency cycle at uniform intervals, with the zero-crossing point as the starting position.

[0016] Each peak point in the initial pulse peak sequence is assigned to its corresponding phase window to obtain a pulse peak distribution with phase window labels.

[0017] Furthermore, determining the signal state of the initial pulse peak sequence within each phase window based on the changes in amplitude data in the initial pulse peak sequence between adjacent power frequency cycles includes:

[0018] Within each phase window, the amplitude data belonging to that phase window are extracted sequentially according to the power frequency cycle from the initial pulse peak sequence;

[0019] The amplitude data of adjacent power frequency cycles falling into the same phase window are paired one by one according to the cycle number;

[0020] Calculate the absolute value of the difference between the paired amplitudes, and summarize multiple consecutive power frequency cycles to obtain the period difference sequence of the current phase window.

[0021] Furthermore, the step of summarizing multiple consecutive power frequency cycles to obtain the period difference sequence of the current phase window includes:

[0022] The mean of the absolute values ​​of the differences in the periodic difference sequence is used as the fluctuation amplitude of the current phase window, and the standard deviation of the absolute values ​​of the differences is used as the stability of the current phase window.

[0023] The normalized fluctuation amplitude and the normalized stability are obtained by dividing the fluctuation amplitude and the stability by the mean value of the amplitude data in the current phase window, respectively. The two are then added together to obtain the dispersion index of the current phase window.

[0024] If the dispersion index is greater than the preset judgment threshold, the status label of the phase window is set to the first state; if the dispersion index is not greater than the preset judgment threshold, the status label of the phase window is set to the second state.

[0025] A window-level signal status label set covering the entire power frequency cycle is formed based on the status labels corresponding to all phase windows.

[0026] Furthermore, the step of determining the current phase window as a clear discharge window if the signal state is in the first state, and determining the current phase window as an interference coverage window if the signal state is in the second state, includes:

[0027] For each phase window in the window-level signal state label set, read the signal state corresponding to the phase window;

[0028] The phase window in the first state is marked as the discharge clear window, and the phase window in the second state is marked as the interference coverage window. The category labels of all phase windows are arranged sequentially on the power frequency cycle according to the phase angle, so as to obtain the phase window attribute distribution that distinguishes the interference coverage window and the discharge clear window.

[0029] Furthermore, for the interference coverage window, acquiring the peak values ​​of interference pulses for multiple consecutive periods, classifying the peak values ​​of the interference pulses to remove discrete outliers, and extracting stable interference reference values ​​includes:

[0030] In the phase window attribute distribution, locate each interference coverage window, extract the peak points that fall into the interference coverage window within a consecutive number of power frequency cycles from the initial pulse peak sequence according to the period number, and collect the amplitudes of the extracted peak points into an interference pulse peak set.

[0031] The K-means clustering algorithm is used to cluster the amplitude samples in the set of interference pulse peaks into a preset number of clusters. The cluster with the largest number of samples is taken as the main cluster. Samples in the cluster whose number of samples is less than a preset percentage threshold and samples whose distance from the centroid of the main cluster exceeds a preset distance threshold are treated as discrete outliers and are removed.

[0032] The arithmetic mean of all amplitude samples in the remaining main cluster sample set is used as the stable interference reference value for the current interference coverage window.

[0033] Furthermore, for the clear discharge window, obtaining the discharge characteristic peak value, performing amplitude central tendency processing on the discharge characteristic peak value, and extracting the clear discharge reference value includes:

[0034] In the phase window attribute distribution, each clear discharge window is located, and the peak points that fall into the clear discharge window within multiple consecutive power frequency cycles are extracted from the initial pulse peak sequence and aggregated into a discharge characteristic peak sequence.

[0035] After sorting the discharge characteristic peak sequence by amplitude, samples at the beginning and end of the line, each accounting for a preset proportion, are removed to obtain the pruned discharge characteristic samples. The median of the pruned discharge characteristic samples is then taken as the clear discharge reference value for the current clear discharge window.

[0036] Furthermore, the step of jointly determining the link transfer gain corresponding to the current power frequency cycle based on the stable interference reference value and the clear discharge reference value when both exist, and determining the link transfer gain based on the existing reference value when only one exists, includes:

[0037] When both exist, the ratio of the stable interference reference value to the pre-calibrated interference reference amplitude is used as the interference path gain component, and the ratio of the clear discharge reference value to the pre-calibrated discharge reference amplitude is used as the discharge path gain component.

[0038] The link transmission gain is obtained by weighting and summing the interference path gain component and the discharge path gain component according to a preset weighted fusion weight. When only one of them exists, the path gain component corresponding to the existing reference value is taken as the link transmission gain.

[0039] When both are invalid, the most recent effective link transmission gain is used as the link transmission gain for the current power frequency cycle. The preset weighted fusion weight is determined according to the proportion of the number of windows in the phase window attribute distribution between the interference coverage window and the discharge clarity window.

[0040] Furthermore, the step of performing amplitude conversion on the amplitude data in the initial pulse peak sequence based on the link transfer gain to obtain the partial discharge calibration value includes:

[0041] The amplitude conversion result is obtained by dividing each amplitude data in the initial pulse peak sequence by the link transfer gain.

[0042] The amplitude conversion result is scaled and mapped according to a preset picocoulomb scale, which is a linear correspondence between voltage amplitude and picocoulomb value.

[0043] The sampling point segment in the original electrical signal that continuously exceeds the preset noise floor threshold is taken as the pulse boundary, and the peak value of the calibration value within the same pulse boundary is taken as the partial discharge calibration value corresponding to the pulse.

[0044] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0045] This invention discloses an adaptive partial discharge (PD) calibration method based on a microcontroller. Addressing the problem of distorted PD amplitude calibration caused by interference within the power frequency cycle affecting the original electrical signal during high-voltage power equipment operation, the method acquires the original electrical signal and extracts the initial pulse peak sequence within the power frequency cycle. It then divides continuous phase windows based on time information and determines the signal state of each window based on amplitude changes between adjacent power frequency cycles, thereby distinguishing between interference-covered windows where signal coverage occurs and clear discharge windows where coverage is not. For interference-covered windows, discrete abnormal extreme values ​​are eliminated by classifying and processing the peak values ​​of interference pulses over multiple cycles, and stable interference reference values ​​are extracted. For clear discharge windows, clear discharge reference values ​​reflecting the true discharge intensity are extracted through amplitude central tendency processing. This invention jointly determines the link transmission gain when both reference values ​​exist; when only one exists, the gain is determined based on the existing reference value. Based on this, amplitude data is converted to obtain the PD calibration value. This achieves accurate PD calibration with phase adaptation under complex interference environments, significantly improving the anti-interference capability and measurement accuracy of PD detection. Attached Figure Description

[0046] Figure 1 This is a flowchart of an adaptive partial discharge pC calibration method based on a microcontroller according to the present invention.

[0047] Figure 2 This is a schematic diagram of an adaptive partial discharge pC calibration method based on a microcontroller according to the present invention.

[0048] Figure 3 This is another schematic diagram of an adaptive partial discharge pC calibration method based on a microcontroller according to the present invention. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0050] like Figures 1-3 This embodiment of an adaptive partial discharge pC calibration method based on a microcontroller may specifically include:

[0051] S101. Obtain the original electrical signal generated during the operation of the high-voltage power equipment, extract the initial pulse peak sequence within the power frequency cycle from the original electrical signal, and divide the power frequency cycle into multiple continuous phase windows according to the time information corresponding to the initial pulse peak sequence.

[0052] The original electrical signal is acquired by a high-frequency current sensor installed on the grounding wire of high-voltage power equipment. This original electrical signal is then sent to the acquisition port of a microcontroller at a preset sampling frequency. Peak detection is performed on sampling points whose amplitude exceeds a preset noise floor threshold within the power frequency cycle to extract the initial pulse peak sequence. Simultaneously, the timestamp corresponding to each peak point in the initial pulse peak sequence is recorded. The preset noise floor threshold is the upper limit of the amplitude of the link static noise. Based on the relative time difference between the timestamp and the zero-crossing point of the power frequency reference signal, each peak point is mapped to the corresponding phase angle within the power frequency cycle. Starting from the zero-crossing point, continuous phase windows are divided at uniform intervals along the power frequency cycle. Each peak point in the initial pulse peak sequence is assigned to its corresponding phase window, resulting in a pulse peak distribution labeled with a phase window.

[0053] In one embodiment, in high-voltage power equipment including converter valves of converter stations, high-voltage windings of frequency converter drive devices, etc., a Rogowski coil-type high-frequency current sensor is installed at its grounding wire or the grounding lead of the casing. The passband range of the high-frequency current sensor is 3MHz to 30MHz, covering the high-frequency band where the main energy of partial discharge pulses is located. The induced signal is sent to the analog front end on the microcontroller side through a shielded coaxial cable.

[0054] Specifically, after the analog front-end performs impedance matching and amplitude limiting protection on the original electrical signal, the high-speed acquisition port built into the microcontroller discretizes the waveform at a preset sampling frequency. The preset sampling frequency ranges from 20MHz to 100MHz, and 50MHz can be selected as an example. This range is sufficient to cover the main spectral energy of partial discharge pulses from 3MHz to 30MHz, while also taking into account the sampling and processing capabilities of the microcontroller to meet the requirement of capturing the leading edge of nanosecond-level discharge pulses. The preset noise floor threshold is determined by acquiring a static waveform without discharge or commutation pulses under equipment shutdown or no-load conditions. The absolute value of this waveform is taken and then slid across the time domain point by point according to a sliding window. The sampling point with the largest absolute value in each window is taken as the envelope value of that window, forming an upper envelope curve. The length of the sliding window is set to 1% of the power frequency period, for example, 0.2ms at a 50Hz power frequency. The maximum value of the upper envelope curve is then multiplied by 1.2 as the upper limit of the static noise amplitude of the link.

[0055] For example, the microcontroller compares the current amplitude with the preset noise floor threshold at each sampling point, regards a segment of sampling points that continuously exceeds the threshold as a candidate pulse, takes the sampling point with the largest absolute value within the range of the segment of sampling points as the peak point, includes the amplitude of the peak point in the initial pulse peak sequence, and multiplies the sampling sequence number of the peak point by the sampling interval to obtain the corresponding timestamp.

[0056] It is understandable that several candidate pulses will be detected within the power frequency cycle, thus forming an initial pulse peak sequence with timestamps.

[0057] Preferably, the power frequency reference signal is taken from the secondary side voltage transformer or synchronous sampling channel of the equipment. The microcontroller uses the zero-crossing moment when the reference signal changes from negative to positive as the starting position of the power frequency cycle. The timestamp of each peak point is subtracted from the most recent zero-crossing moment, divided by the power frequency cycle length, and multiplied by 360 degrees to obtain the phase angle of the peak point within the power frequency cycle. The 0-degree to 360-degree range is divided into several consecutive phase windows at uniform intervals. The number of phase windows ranges from 12 to 72, and the phase span of each phase window is from 5 to 30 degrees. 36 phase windows can be selected, with each window corresponding to a 10-degree phase span. This range takes into account both phase resolution and statistical sufficiency of pulse samples within each window. Too many windows may lead to sparse samples within a single window, while too few windows may make it impossible to accurately locate the phase interval covered by interference. Each peak point is assigned to the phase window into which its phase angle falls, forming a pulse peak distribution labeled with a phase window.

[0058] S102. For the multiple consecutive phase windows, based on the change of amplitude data in the initial pulse peak sequence between adjacent power frequency cycles, determine the signal state of the initial pulse peak sequence in each phase window.

[0059] For the multiple consecutive phase windows, amplitude data belonging to that phase window are extracted from the initial pulse peak sequence in the order of the power frequency cycle within each phase window. Amplitude data falling into the same phase window in adjacent power frequency cycles are paired one by one according to the cycle number to obtain adjacent cycle amplitude pairs under the current phase window. The absolute value of the difference between the adjacent cycle amplitude pairs is calculated, and multiple consecutive power frequency cycles are summarized to obtain the cycle difference sequence of the current phase window. According to the cycle difference sequence, the mean of the absolute values ​​of the differences is taken as the fluctuation amplitude of the current phase window, and the standard deviation of the absolute values ​​of the differences is taken as the stability of the current phase window. The fluctuation amplitude and the stability are divided by the mean of the amplitude data in the current phase window to obtain the normalized fluctuation amplitude and the normalized stability, respectively. The two are added together to obtain the dispersion index of the current phase window. If the dispersion index is greater than a preset judgment threshold, the state label of the phase window is set to the first state; if the dispersion index is not greater than the preset judgment threshold, the state label of the phase window is set to the second state. Based on the status labels corresponding to all phase windows, the initial pulse peak sequence in each phase window is marked as the signal state indicated by the status label, forming a window-level signal state label set covering the entire power frequency cycle, and the signal state of the initial pulse peak sequence in each phase window is determined.

[0060] In one embodiment, a significant characteristic of power frequency phase-locked interference is that its pulse position repeatedly appears relative to the power frequency zero-crossing point, thus the amplitude falling within the same phase window exhibits high consistency between adjacent power frequency cycles. In contrast, the generation of a real partial discharge pulse is significantly random, and the amplitude falling within the same phase window fluctuates considerably between adjacent power frequency cycles. Based on this physical difference, by quantifying the fluctuation of amplitude data between adjacent power frequency cycles, an index reflecting the stability of the initial pulse peak sequence within each phase window is obtained.

[0061] Specifically, the microcontroller maintains a sliding buffer of length N. This buffer stores amplitude data for each phase window within the most recent N power frequency cycles, numbered according to the power frequency cycle. The value of N ranges from 8 to 32 cycles. This length range ensures that the number of samples used for statistics is sufficient to reflect amplitude fluctuation characteristics while avoiding lag in response to changes in link gain due to an excessively long buffer. A length of 16 cycles can be selected. For the k-th phase window, amplitude data from the i-th and (i+1)-th cycles falling within that phase window are sequentially retrieved from the buffer. If multiple peak points are detected within the same phase window, the peak with the largest amplitude within that window in that cycle is selected for pairing, resulting in adjacent cycle amplitude pairs. The absolute value of the difference between the amplitudes of all adjacent cycle amplitude pairs is taken and arranged in cycle number order to obtain the cycle difference sequence for the k-th phase window. The length of this cycle difference sequence is N-1.

[0062] It is understood that the period difference sequence essentially reflects the activity level of the pulse amplitude variation with the period at the same phase position, and its value distribution characteristics directly correspond to the stability differences of the pulse source within the phase window.

[0063] Specifically, the periodic difference sequence is characterized in two dimensions. The fluctuation amplitude is taken as the arithmetic mean of all elements in the periodic difference sequence, reflecting the overall level of the amplitude difference of the phase window; the stability is taken as the sample standard deviation of the periodic difference sequence, reflecting the dispersion of the amplitude difference itself.

[0064] For example, let the fluctuation amplitude be denoted as A, the stability as S, and the arithmetic mean of all amplitude data within the current phase window as M. Then, the normalized fluctuation amplitude is A / M, the normalized stability is S / M, and the dispersion index D = A / M + S / M. By normalizing by dividing by M, the dimensional differences caused by the different pulse fundamental amplitudes within different phase windows are eliminated, making the dispersion index comparable between windows.

[0065] Preferably, the preset judgment threshold is determined as follows: Static waveforms are collected under equipment shutdown or no-load conditions, divided into the same sliding buffer length and the same phase window. The static dispersion D0 is obtained according to the calculation process of the dispersion index. D0 is multiplied by an amplification factor to obtain the preset judgment threshold. The amplification factor ranges from 3 to 6, used to raise the judgment threshold during dynamic operation above the static noise fluctuation level, avoiding misjudgments caused by static noise. For example, when the static dispersion D0 is 0.05, an amplification factor of 4 is used, resulting in a preset judgment threshold of 0.20. If the dispersion index is greater than the preset judgment threshold, the status label of the phase window is set to the first state; if the dispersion index is not greater than the preset judgment threshold, the status label of the phase window is set to the second state.

[0066] It should be noted that the first state corresponds to a situation where the amplitude jumps sharply between adjacent periods, reflecting a lack of cross-period consistency in the pulses detected within that window; the second state corresponds to a situation where the amplitude remains stable between adjacent periods, reflecting that the pulses detected within that window possess cross-period consistency. The state labels corresponding to each phase window within the range of 0 degrees to 360 degrees are arranged in phase angle order, marking the signal state of the initial pulse peak sequence within each phase window, forming a window-level signal state label set covering the entire power frequency cycle. This window-level signal state label set characterizes the overall pattern of signal properties changing with phase over the power frequency cycle, providing a criterion for subsequently selecting different calibration references in different phase segments.

[0067] S103. If the signal state is the first state, it is determined that the initial pulse peak sequence has not been flooded with signal within the current phase window, and the current phase window is determined to be a clear discharge window. If the signal state is the second state, it is determined that the initial pulse peak sequence has been flooded with signal within the current phase window, and the current phase window is determined to be an interference coverage window.

[0068] For each phase window in the set of window-level signal state labels, the signal state corresponding to the phase window is read. If the signal state is the first state, the current phase window is marked as a clear discharge window. If the signal state is the second state, the current phase window is marked as an interference coverage window. The category labels of all phase windows are arranged sequentially on the power frequency cycle according to the phase angle order to obtain the phase window attribute distribution that distinguishes between interference coverage windows and clear discharge windows.

[0069] In one embodiment, the first state and the second state correspond to two types of physical backgrounds, respectively. The first state reflects poor consistency of pulse amplitude between adjacent power frequency cycles within the current phase window, which is consistent with the characteristics of large fluctuations in real partial discharge signals and belongs to a clear discharge window; the second state reflects stable pulse amplitude between adjacent power frequency cycles within the current phase window, which is consistent with the characteristics of repeated occurrence of power frequency phase-locked interference and belongs to an interference coverage window.

[0070] Specifically, the microcontroller traverses each phase window in the window-level signal state label set in phase angle order. If the current phase window's state label is the first state, then the current phase window's category label is set to "clear discharge window"; if the current phase window's state label is the second state, then the current phase window's category label is set to "interference coverage window".

[0071] For example, phase windows detected within the phase interval at the moment of converter valve triggering are mostly identified as interference coverage windows, while phase windows detected in phase intervals far from the commutation moment are mostly identified as clear discharge windows. Further, the interference coverage window markers and clear discharge window markers corresponding to all phase windows are arranged in order of phase angle from 0 degrees to 360 degrees to obtain the phase window attribute distribution. This phase window attribute distribution clarifies the window category to which each phase interval belongs in the power frequency cycle, forming the boundary division results for the two types of phase segments: the interference coverage window set and the clear discharge window set.

[0072] S104. For the interference coverage window, obtain the peak values ​​of interference pulses for multiple consecutive cycles within the interference coverage window, classify the peak values ​​of interference pulses to remove discrete abnormal extreme values, and extract stable interference reference values.

[0073] For the interference coverage window, each interference coverage window is located in the phase window attribute distribution. The peak points that fall into the interference coverage window within a consecutive number of power frequency cycles are extracted from the initial pulse peak sequence according to the period number. The amplitudes of the extracted peak points are collected in periodic order to obtain the interference pulse peak set of the current interference coverage window. Based on the set of interference pulse peaks, the amplitude samples in the set of interference pulse peaks are clustered into a preset number of clusters using the K-means clustering algorithm. The preset number can be 2 to 4. This range ensures that the main interference cluster can be distinguished from possible discrete anomalous clusters, while avoiding excessive computational complexity due to too many clusters. For example, 3 clusters can be used. The cluster with the largest number of samples is taken as the main cluster. Samples within the cluster whose sample number is less than a preset percentage threshold and samples whose distance from the centroid of the main cluster exceeds a preset distance threshold are considered discrete outliers and are removed. The preset percentage threshold is 10%, that is, clusters with a sample number percentage of less than 10% are considered outlier clusters composed of outliers and are removed. The preset distance threshold is 3 times the standard deviation of the samples within the main cluster. For example, when the standard deviation of the samples within the main cluster is 2mV, the distance threshold is 6mV, thus obtaining the remaining main cluster sample set. Based on the main cluster sample set, the arithmetic mean of all amplitude samples in the main cluster sample set is used as the stable interference reference value of the current interference coverage window. All interference coverage windows in the phase window attribute distribution are traversed to obtain the stable interference reference value corresponding to the interference coverage window.

[0074] In one embodiment, the physical sources of power frequency phase-locked loop (PLL) interference include the thyristor trigger pulse of the converter valve and the commutation pulse of the silicon controlled rectifier (SCR). The phase position of these pulses within the power frequency cycle is determined by the trigger angle, ensuring strict phase-locking with the power frequency. The amplitude of the PLL interference under steady-state operation is jointly determined by the trigger circuit parameters and link transmission characteristics, and theoretically should remain at a relatively stable level over multiple consecutive power frequency cycles. A small number of extreme amplitudes deviating from this stable level are typically introduced by transient disturbances in the field, sensor mechanical jitter, or instantaneous changes in sampling, and need to be eliminated before extracting the interference reference value.

[0075] Specifically, the microcontroller traverses the phase window attribute distribution in phase angle order to identify all phase windows labeled as interference coverage windows. For each interference coverage window, peak points falling within the interference coverage window within multiple consecutive power frequency cycles are extracted from the initial pulse peak sequence in period number order. The number of power frequency cycles ranges from 16 to 64 cycles. This range ensures that the sample size for statistics is sufficient to exclude occasional disturbances while avoiding deviations caused by link drift due to excessively long spans. For example, 32 cycles can be selected. The amplitudes of the extracted peak points are aggregated in period order to obtain the interference pulse peak set of the current interference coverage window.

[0076] It is understandable that the samples in the set of interference pulse peaks include both the majority of samples representing stable interference and a small number of abnormal samples that deviate from the main distribution, which need to be distinguished through classification.

[0077] Specifically, for the one-dimensional amplitude samples in the set of interference pulse peaks, the K-means clustering algorithm is used for classification. The number of clusters K in the K-means clustering algorithm ranges from 2 to 4. The initial centroids are selected from the set of interference pulse peaks at equal intervals according to the amplitude quantiles. When K is 2, the amplitudes corresponding to the 25% and 75% quantiles are selected as the initial centroids. When K is 3, the amplitudes corresponding to the 20%, 50%, and 80% quantiles are selected as the initial centroids. When K is 4, the amplitudes corresponding to the 15%, 40%, 65%, and 90% quantiles are selected as the initial centroids. For each amplitude sample in the set of interference pulse peaks, calculate the absolute difference between it and each centroid, and classify it into the cluster with the smallest absolute difference. After classification, update the centroid position according to the arithmetic mean of the amplitude samples in each cluster. Repeat the iterative process of classification and centroid update until the centroid displacement of two adjacent iterations is less than a preset convergence threshold or the number of iterations reaches a preset upper limit. The preset convergence threshold is 1% of the standard deviation of the samples in the main cluster. For example, when the standard deviation of the samples in the main cluster is 2mV, the convergence threshold is 0.02mV. The preset upper limit ranges from 20 to 50 times, and can be exemplarily taken as 30 times to ensure that the clustering can converge in most cases, while avoiding infinite iteration that consumes too many processing resources.

[0078] Preferably, after iterative convergence, the cluster with the largest number of samples is selected as the master cluster. Samples within a cluster whose sample size is less than a preset threshold (10%) of the total number of interference pulse peak values ​​are discarded as discrete outliers. Simultaneously, samples whose absolute amplitude difference from the centroid of the master cluster exceeds a preset distance threshold (3 times the standard deviation of the samples within the master cluster) are also discarded. The remaining samples constitute the master cluster sample set. Further, the arithmetic mean of all amplitude samples in the master cluster sample set is calculated, and this arithmetic mean is used as the stable interference benchmark value for the current interference coverage window. The stable interference benchmark value reflects the stable amplitude center of the power frequency phase-locked loop interference after crossing multiple power frequency cycles within the current interference coverage window, and can serve as a quantitative description of the link's interference path transmission characteristics.

[0079] It should be noted that the microcontroller traverses each interference coverage window in the phase window attribute distribution according to the above process to obtain the set of stable interference reference values ​​corresponding to the interference coverage window, covering all phase segments that are determined to be interference coverage in the power frequency cycle.

[0080] S105. For the clear discharge window, obtain the discharge characteristic peak value within the clear discharge window, perform amplitude concentration trend processing on the discharge characteristic peak value, and extract the clear discharge reference value that reflects the true discharge intensity.

[0081] For each clear discharge window, locate the clear discharge window in the phase window attribute distribution. Extract peak points that fall within the clear discharge window within multiple consecutive power frequency cycles from the initial pulse peak sequence, numbered periodically. Collect the amplitudes of these extracted peak points to form the discharge characteristic peak sequence for the current clear discharge window. Based on this discharge characteristic peak sequence, sort the sequence by amplitude from smallest to largest. Remove samples from the beginning and end of the sorted sequence, each representing a predetermined proportion, to obtain pruned discharge characteristic samples. The predetermined proportion ranges from 5% to 20%, and can be set to 10%. This proportion is sufficient to remove extreme outliers from the sample distribution while retaining a sufficient sample size to ensure statistical representativeness. Take the median of the pruned discharge characteristic samples as the amplitude central tendency result. Use this amplitude central tendency result as the clear discharge benchmark value for the current clear discharge window. Traverse all clear discharge windows in the phase window attribute distribution to obtain a clear discharge benchmark value that reflects the true discharge intensity.

[0082] In one embodiment, the generation of a real partial discharge pulse is random, and its single amplitude is affected by microscopic disturbances during the discharge channel formation process, resulting in instantaneous peaks or troughs. Directly using the amplitude of a certain power frequency cycle as a reference for discharge intensity can easily introduce large periodic fluctuations. Based on this physical background, amplitude concentration trend processing is performed on the discharge characteristic peaks falling within the clear discharge window in multiple consecutive power frequency cycles to suppress the bias of extreme samples at the beginning and end of the cycle on the reference value, thereby obtaining a stable reference that reflects the true discharge intensity.

[0083] Specifically, the microcontroller sequentially locates each phase window labeled as a clear discharge window in the phase window attribute distribution. It extracts peak points that fall into the current clear discharge window within multiple consecutive power frequency cycles from the initial pulse peak sequence in period number order. The number of consecutive power frequency cycles ranges from 16 to 64 cycles, usually 32 cycles. This range is consistent with the statistical cycle of the interference coverage window, ensuring that the benchmark values ​​of the two types of windows are comparable, while the sample size is sufficient to suppress random fluctuations of a single pulse. The amplitudes of the extracted peak points are collected in period order to form the discharge characteristic peak sequence of the current clear discharge window.

[0084] Preferably, the discharge characteristic peak sequence is sorted in ascending order of amplitude. Samples representing a predetermined proportion are removed from both the beginning and end of the sorted sequence to obtain pruned discharge characteristic samples. The predetermined proportion ranges from 5% to 20%, and can be set to 10%, meaning the 10% of samples with the smallest amplitude are removed from the beginning, and the 10% of samples with the largest amplitude are removed from the end. Then, the median of the pruned discharge characteristic samples is taken as the amplitude central tendency result, and this amplitude central tendency result is used as the clear discharge benchmark value for the current clear discharge window.

[0085] Furthermore, the microcontroller traverses each clear discharge window in the phase window attribute distribution according to the above process to obtain the clear discharge reference value set corresponding to the clear discharge window, covering all phase segments that are determined to be clear discharges in the power frequency cycle.

[0086] S106. Based on the stable interference reference value and the clear discharge reference value, when both exist, determine the link transfer gain corresponding to the current power frequency cycle based on the stable interference reference value and the clear discharge reference value. When only one exists, determine the link transfer gain based on the existing reference value. Perform amplitude conversion on the amplitude data in the initial pulse peak sequence based on the link transfer gain to obtain the partial discharge calibration value.

[0087] For the current power frequency cycle, the availability of the stable interference reference value and the clear discharge reference value is determined. If the reference value is greater than zero and the number of samples extracted within the corresponding window is not less than a preset sample number threshold, the reference value is determined to be valid. If both the stable interference reference value and the clear discharge reference value are valid, it is determined to be a dual-reference available state. If only one of them is valid, it is determined to be a single-reference available state, and the reference availability identifier corresponding to the current power frequency cycle is obtained. The link transmission gain of the current power frequency cycle is determined according to the reference availability identifier. In the dual-reference available state, the ratio of the stable interference reference value to the pre-calibrated interference reference amplitude is used as the interference path gain component, and the ratio of the clear discharge reference value to the pre-calibrated discharge reference amplitude is used as the discharge path gain component. The two are weighted and summed according to a preset weighted fusion weight to obtain the link transmission gain. The preset weighted fusion weight ranges from 0 to 1, and the sum of the weights of the two components is 1. In the single-reference available state, the path gain component corresponding to the valid one is directly used as the link transmission gain. Based on the link transfer gain, each amplitude data in the initial pulse peak sequence is divided by the link transfer gain to obtain the amplitude conversion result corresponding to the amplitude data. The amplitude conversion result is then scaled according to a preset picocoulomb scale, which is a linear correspondence between voltage amplitude and picocoulomb value established during the calibration phase, to obtain the partial discharge calibration value corresponding to each amplitude data in the initial pulse peak sequence.

[0088] In one embodiment, the link transfer gain reflects the conversion relationship between the amplitude measured at the sensor acquisition end and the actual discharge quantity. It should remain consistent within the same power frequency cycle; that is, regardless of whether the interference path or the discharge path is used as a reference, the obtained gain value should be at a similar level. Based on this physical consistency, by simultaneously using the stable interference reference value extracted by the interference coverage window and the clear discharge reference value extracted by the clear discharge window as references within the same power frequency cycle, it is possible to avoid continuing to rely on distorted discharge characteristics in phase segments where the signal is submerged, while not completely discarding the link information reflected by the discharge characteristics.

[0089] Specifically, the availability of the stable interference reference value and the clear discharge reference value within the current power frequency cycle is determined. If the reference value is greater than zero and the number of samples extracted within the corresponding window is not less than a preset sample number threshold, the reference value is deemed valid. The preset sample number threshold is set to 8 samples; that is, the statistical results of at least 8 samples are considered representative. Statistical results below this threshold are excessively affected by a few extreme values ​​and are discarded. When both the stable interference reference value and the clear discharge reference value are valid, it is determined to be a dual-reference availability state; when only one is valid, it is determined to be a single-reference availability state. The state determination result is recorded as the reference availability identifier for the current power frequency cycle.

[0090] It should be noted that if both are invalid, the link transfer gain of the most recent valid power frequency cycle is read from the non-volatile memory area of ​​the microcontroller as the link transfer gain of the current power frequency cycle. For each power frequency cycle in which a valid link transfer gain is obtained, the microcontroller will overwrite this value in the historical gain buffer area of ​​the non-volatile memory area with a timestamp. The historical gain buffer area only retains the most recent valid gain to avoid gaps in the link transfer gain.

[0091] In one possible implementation, the pre-calibrated interference reference amplitude and the pre-calibrated discharge reference amplitude are obtained during the calibration phase before the equipment leaves the factory. During the calibration phase, a standard discharge pulse of known picocoulomb quantity is injected into the link under test, and a simulated interference pulse of known amplitude is injected at the trigger phase of the converter valve. The steady-state amplitude of the interference path output by the link is measured as the interference reference amplitude, and the steady-state amplitude of the discharge path output by the link is measured as the discharge reference amplitude. Both reference amplitudes are written into the non-volatile memory area of ​​the microcontroller.

[0092] Specifically, under dual-reference availability, the stable interference reference value is denoted as U1, and the pre-calibrated interference reference amplitude is denoted as R1, then the interference path gain component G1 = U1 / R1; the clear discharge reference value is denoted as U2, and the pre-calibrated discharge reference amplitude is denoted as R2, then the discharge path gain component G2 = U2 / R2. The two components are weighted and summed according to a preset weighting fusion weight to obtain the link transmission gain G = w1 × G1 + w2 × G2 for the current power frequency cycle, where w1 is the weight of the interference path gain component, w2 is the weight of the discharge path gain component, and the values ​​of w1 and w2 are both between 0 and 1, and w1 + w2 = 1. The preset weighting fusion weight is determined according to the proportion of the number of windows in the phase window attribute distribution between the interference coverage window and the clear discharge window. The number of interference coverage windows is denoted as N1, and the number of clear discharge windows is denoted as N2, then w1 = N1 / (N1 + N2), w2 = N2 / (N1 + N2).

[0093] Preferably, when the number of interference coverage windows is equal to the number of clear discharge windows, both w1 and w2 are set to 0.5, achieving equal-weighted fusion of the gain components of the two paths. Under single-reference availability, if only the stable interference reference value is valid, the interference path gain component G1 is directly used as the link transfer gain; if only the clear discharge reference value is valid, the discharge path gain component G2 is directly used as the link transfer gain. This degradation mechanism maintains continuous output of the link transfer gain even when the signal discernibility fluctuates significantly with phase. Further, each amplitude data in the initial pulse peak sequence is divided by the link transfer gain to obtain the amplitude conversion result corresponding to that amplitude data. The physical meaning of the amplitude conversion result is the equivalent amplitude of the pulse converted to the link input.

[0094] In one embodiment, the preset picocoulomb scale is a linear correspondence between voltage amplitude and picocoulomb value established during the calibration phase. During the calibration phase, several standard discharge pulses with known picocoulomb values ​​are sequentially injected into the link input. The equivalent amplitude sequence and the corresponding picocoulomb value sequence at the link input are recorded. The least squares method is used to fit the scale slope and scale intercept, and the fitting result is written into the non-volatile memory area of ​​the microcontroller as the preset picocoulomb scale. The amplitude conversion result is scaled according to the preset picocoulomb scale, that is, the amplitude conversion result is multiplied by the scale slope and added to the scale intercept to obtain the partial discharge calibration value corresponding to each amplitude data in the initial pulse peak sequence.

[0095] It should be noted that, through the above-mentioned state-by-state processing of dual-reference available state and single-reference available state, the microcontroller can select the more stable and more realistic reference basis in different phase segments of the power frequency cycle, and obtain a partial discharge calibration value that remains consistent throughout the entire phase range.

[0096] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A microcontroller-based adaptive partial discharge pC calibration method, characterized in that, The method includes: The original electrical signal generated during the operation of high-voltage power equipment is acquired, the initial pulse peak sequence within the power frequency cycle is extracted from the original electrical signal, and multiple continuous phase windows are divided within the power frequency cycle according to the time information corresponding to the initial pulse peak sequence. For the multiple consecutive phase windows, based on the changes in amplitude data in the initial pulse peak sequence between adjacent power frequency cycles, the signal state of the initial pulse peak sequence within each phase window is determined; If the signal state is in the first state, the current phase window is determined to be a clear discharge window; if the signal state is in the second state, the current phase window is determined to be an interference coverage window. For the interference coverage window, the peak values ​​of interference pulses for multiple consecutive cycles are obtained, and the peak values ​​of interference pulses are classified to remove discrete abnormal extreme values ​​and extract stable interference reference values. For the clear discharge window, the discharge characteristic peak value is obtained, and the amplitude central trend processing is performed on the discharge characteristic peak value to extract the clear discharge reference value. Based on the stable interference reference value and the clear discharge reference value, the link transfer gain corresponding to the current power frequency cycle is jointly determined when both exist. When only one of them exists, the link transfer gain is determined according to the existing reference value. The amplitude data in the initial pulse peak sequence is converted according to the link transfer gain to obtain the partial discharge calibration value.

2. The microcontroller-based adaptive partial discharge pC calibration method according to claim 1, characterized in that, Extracting the initial pulse peak sequence within the power frequency cycle from the original electrical signal includes: The original electrical signal is obtained by a high-frequency current sensor installed on the grounding wire of the high-voltage power equipment; The original electrical signal is sent to the acquisition port of the microcontroller at a preset sampling frequency; Peak detection is performed on sampling points whose amplitude exceeds a preset noise floor threshold within the power frequency cycle to extract the initial pulse peak sequence. The preset noise floor threshold is the upper limit of the amplitude of the static noise of the link. At the same time, the timestamp corresponding to each peak point in the initial pulse peak sequence is recorded.

3. The microcontroller-based adaptive partial discharge pC calibration method according to claim 1, characterized in that, The step of dividing the power frequency period into multiple consecutive phase windows based on the time information corresponding to the initial pulse peak sequence includes: Based on the relative time difference between the timestamp corresponding to each peak point in the initial pulse peak sequence and the zero-crossing point of the power frequency reference signal, each peak point is mapped to the corresponding phase angle within the power frequency cycle, and a continuous phase window is divided on the power frequency cycle at uniform intervals, with the zero-crossing point as the starting position. Each peak point in the initial pulse peak sequence is assigned to its corresponding phase window to obtain a pulse peak distribution with phase window labels.

4. The microcontroller-based adaptive partial discharge pC calibration method according to claim 1, characterized in that, The step of determining the signal state of the initial pulse peak sequence within each phase window based on the change of amplitude data in the initial pulse peak sequence between adjacent power frequency cycles includes: Within each phase window, the amplitude data belonging to that phase window are extracted sequentially according to the power frequency cycle from the initial pulse peak sequence; The amplitude data of adjacent power frequency cycles falling into the same phase window are paired one by one according to the cycle number; Calculate the absolute value of the difference between the paired amplitudes, and summarize multiple consecutive power frequency cycles to obtain the period difference sequence of the current phase window.

5. The microcontroller-based adaptive partial discharge pC calibration method according to claim 4, characterized in that, The process of summarizing multiple consecutive power frequency cycles to obtain the period difference sequence for the current phase window includes: The mean of the absolute values ​​of the differences in the periodic difference sequence is used as the fluctuation amplitude of the current phase window, and the standard deviation of the absolute values ​​of the differences is used as the stability of the current phase window. The normalized fluctuation amplitude and the normalized stability are obtained by dividing the fluctuation amplitude and the stability by the mean value of the amplitude data in the current phase window, respectively. The two are then added together to obtain the dispersion index of the current phase window. If the dispersion index is greater than the preset judgment threshold, the status label of the phase window is set to the first state; if the dispersion index is not greater than the preset judgment threshold, the status label of the phase window is set to the second state. A window-level signal status label set covering the entire power frequency cycle is formed based on the status labels corresponding to all phase windows.

6. The microcontroller-based adaptive partial discharge pC calibration method according to claim 5, characterized in that, If the signal state is in the first state, then the current phase window is determined to be a clear discharge window; if the signal state is in the second state, then the current phase window is determined to be an interference coverage window, including: For each phase window in the window-level signal state label set, read the signal state corresponding to the phase window; The phase window in the first state is marked as the discharge clear window, and the phase window in the second state is marked as the interference coverage window. The category labels of all phase windows are arranged sequentially on the power frequency cycle according to the phase angle, so as to obtain the phase window attribute distribution that distinguishes the interference coverage window and the discharge clear window.

7. The microcontroller-based adaptive partial discharge pC calibration method according to claim 6, characterized in that, For the interference coverage window, the process involves acquiring the peak values ​​of interference pulses over multiple consecutive periods, classifying the interference pulse peak values ​​to remove discrete outliers, and extracting stable interference baseline values, including: In the phase window attribute distribution, locate each interference coverage window, extract the peak points that fall into the interference coverage window within a consecutive number of power frequency cycles from the initial pulse peak sequence according to the period number, and collect the amplitudes of the extracted peak points into an interference pulse peak set. The K-means clustering algorithm is used to cluster the amplitude samples in the set of interference pulse peaks into a preset number of clusters. The cluster with the largest number of samples is taken as the main cluster. Samples in the cluster whose number of samples is less than a preset percentage threshold and samples whose distance from the centroid of the main cluster exceeds a preset distance threshold are treated as discrete outliers and are removed. The arithmetic mean of all amplitude samples in the remaining main cluster sample set is used as the stable interference reference value for the current interference coverage window.

8. The microcontroller-based adaptive partial discharge pC calibration method according to claim 6, characterized in that, For the clear discharge window, the process involves acquiring discharge characteristic peak values, performing amplitude central tendency processing on the discharge characteristic peak values, and extracting a clear discharge reference value, including: In the phase window attribute distribution, each clear discharge window is located, and the peak points that fall into the clear discharge window within multiple consecutive power frequency cycles are extracted from the initial pulse peak sequence and aggregated into a discharge characteristic peak sequence. After sorting the discharge characteristic peak sequence by amplitude, samples at the beginning and end of the line, each accounting for a preset proportion, are removed to obtain the pruned discharge characteristic samples. The median of the pruned discharge characteristic samples is then taken as the clear discharge reference value for the current clear discharge window.

9. The microcontroller-based adaptive partial discharge pC calibration method according to claim 1, characterized in that, The process of jointly determining the link transmission gain corresponding to the current power frequency cycle based on the stable interference reference value and the clear discharge reference value, when both exist, and determining the link transmission gain based on the existing reference value when only one exists, includes: When both exist, the ratio of the stable interference reference value to the pre-calibrated interference reference amplitude is used as the interference path gain component, and the ratio of the clear discharge reference value to the pre-calibrated discharge reference amplitude is used as the discharge path gain component. The link transfer gain is obtained by weighting and summing the interference path gain component and the discharge path gain component according to a preset weighted fusion weight. When only one of them exists, the path gain component corresponding to the existing reference value is taken as the link transfer gain. When both are invalid, the most recent effective link transmission gain is used as the link transmission gain for the current power frequency cycle. The preset weighted fusion weight is determined according to the proportion of the number of windows in the phase window attribute distribution between the interference coverage window and the discharge clarity window.

10. The microcontroller-based adaptive partial discharge pC calibration method according to claim 1, characterized in that, The step of performing amplitude conversion on the amplitude data in the initial pulse peak sequence based on the link transfer gain to obtain the partial discharge calibration value includes: The amplitude conversion result is obtained by dividing each amplitude data in the initial pulse peak sequence by the link transfer gain. The amplitude conversion result is scaled and mapped according to a preset picocoulomb scale, which is a linear correspondence between voltage amplitude and picocoulomb value. The sampling point segment in the original electrical signal that continuously exceeds the preset noise floor threshold is taken as the pulse boundary, and the peak value of the calibration value within the same pulse boundary is taken as the partial discharge calibration value corresponding to the pulse.

Images