A method and system for extracting time-domain features of partial discharge signals of a generator stator winding

Abstract

The application discloses a kind of generator stator winding partial discharge signal time domain feature extraction method and system, it is related to generator state monitoring technical field, the method includes the following steps: at least two different frequency bands in the partial discharge high-frequency signal after pre-processing respectively setting sliding window, and the standard deviation of white noise in sliding window is calculated, according to the dynamic fluctuation threshold of frequency band of white noise standard deviation, and preset slope threshold and fluctuation number threshold;After filtering thyristor pulse and slot discharge in the partial discharge high-frequency signal after pre-processing, according to the dynamic fluctuation threshold of corresponding frequency band, slope threshold and fluctuation number threshold, from the partial discharge high-frequency signal after filtering, effective feature point is extracted, and the feature quantity of the partial discharge high-frequency signal is corrected based on effective feature point, complete time domain feature extraction.The present application can solve the problem that the feature extraction is not accurate due to the poor adaptability of fixed threshold and the mismatch of single window in the prior art.

Description

Technical Field

[0001] This invention relates to the field of generator condition monitoring technology, specifically to a method and system for extracting time-domain features of partial discharge signals from generator stator windings. Background Technology

[0002] The generator stator winding is a high-risk area for insulation faults, and the partial discharge pulse signals (50kHz~20MHz) it generates are crucial for assessing insulation condition. Currently, existing technologies for processing stator winding partial discharge signals primarily rely on fixed detection thresholds and preset single-window durations. Fixed thresholds have poor noise fluctuation immunity, and the stator winding operating environment is complex, with white noise amplitude varying drastically with operating conditions: during startup, the excitation system's start-up causes noise to surge from 5mV to 20mV; during load adjustments, noise fluctuations can reach ±50%. Existing fixed thresholds (such as 10mV measured under shutdown conditions) misinterpret valid signals as noise when noise increases and vice versa when noise decreases, severely impacting monitoring reliability. A single window duration processes signals across the entire frequency band, neglecting the propagation characteristics of stator winding signals. High-frequency signals (3MHz~20MHz) propagate in a fast-wave mode, attenuating 5~10 times faster than low-frequency signals along the stator bars, requiring a short window to capture transient characteristics. Low-frequency signals (50kHz~3MHz) propagate in a slow-wave mode, less affected by winding inductance and capacitance coupling, resulting in stable signals but prone to persistent interference, necessitating a long window to reduce statistical errors. The single window leads to threshold lag for high-frequency signals and incomplete noise filtering for low-frequency signals. Furthermore, while analysis methods relying on complex time-frequency transformations can provide richer feature information, they are computationally intensive and time-consuming, making it difficult to complete within the high sampling rates (e.g., 30MS / s) and short analysis cycles (e.g., 100ms) required for generator online monitoring, presenting a contradiction between insufficient real-time performance and difficulty in achieving computational accuracy. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for extracting time-domain features of partial discharge signals from generator stator windings, in order to solve the problem of inaccurate feature extraction caused by poor adaptability of fixed thresholds and mismatch of single windows in the prior art.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] In a first aspect, a method for extracting time-domain features of partial discharge signals from generator stator windings includes the following steps:

[0006] Acquire high-frequency partial discharge signals from the generator stator windings and perform preprocessing;

[0007] Sliding windows are set for at least two different frequency bands in the preprocessed partial discharge high-frequency signal, and the white noise standard deviation within the sliding window is calculated. The dynamic fluctuation threshold of the frequency band is calculated based on the white noise standard deviation, and a slope threshold and a fluctuation number threshold are preset.

[0008] After filtering out the thyristor pulses and slot discharges in the preprocessed partial discharge high-frequency signal, effective feature points are extracted from the filtered partial discharge high-frequency signal according to the dynamic fluctuation threshold, slope threshold, and fluctuation number threshold of the corresponding frequency band. Based on the effective feature points, the feature quantities of the partial discharge high-frequency signal are corrected to complete the time-domain feature extraction.

[0009] In some implementations, the step of setting sliding windows for at least two different frequency bands in the preprocessed partial discharge high-frequency signal and calculating the standard deviation of white noise within the sliding windows specifically includes:

[0010] Set a corresponding first sliding window duration and a first sliding step size for the high-frequency band, and set a corresponding second sliding window duration and a second sliding step size for the low-frequency band;

[0011] The standard deviation of white noise in the high-frequency band window is calculated based on the duration of the first sliding window and the first sliding step size. The standard deviation of white noise in the low-frequency band window is calculated based on the duration of the second sliding window and the second sliding step size.

[0012] In some implementations, calculating the dynamic fluctuation threshold of the frequency band based on the white noise standard deviation specifically includes:

[0013] Based on the standard deviation of white noise within the high-frequency band window, the dynamic fluctuation threshold of the high-frequency band is calculated using the following formula: Thres1_H = 1.5×2σ1, where Thres1_H is the dynamic fluctuation threshold of the high-frequency band, and σ1 is the standard deviation of white noise within the high-frequency band window;

[0014] Based on the standard deviation of white noise within the low-frequency window, the dynamic fluctuation threshold of the low-frequency band is calculated using the following formula: Thres1_L = 1.2×3σ2, where Thres1_L is the dynamic fluctuation threshold of the low-frequency band, and σ2 is the standard deviation of white noise within the low-frequency window.

[0015] In some implementations, effective feature points are extracted from the filtered partial discharge high-frequency signal based on the dynamic fluctuation threshold, slope threshold, and fluctuation number threshold for the corresponding frequency band, specifically including:

[0016] Calculate the amplitude change, rise slope, and number of oscillations of the filtered partial discharge high-frequency signal;

[0017] Among the high-frequency partial discharge signals, signal points that simultaneously satisfy the following conditions are selected as valid feature points: the amplitude change is greater than the dynamic fluctuation threshold, the rising slope is greater than the slope threshold, and the number of oscillations is greater than the fluctuation number threshold.

[0018] In some implementations, the characteristic quantities include: pulse amplitude, number of peaks and troughs, duration, and equivalent oscillation velocity.

[0019] In some implementations, the high-frequency signal of partial discharge from the generator stator winding is acquired and preprocessed, specifically including:

[0020] Remove the DC offset of the partial discharge high-frequency signal and eliminate single-point abrupt changes in the partial discharge high-frequency signal.

[0021] Secondly, a generator stator winding partial discharge signal extraction system includes:

[0022] The signal acquisition module is used to acquire and preprocess the high-frequency partial discharge signal of the generator stator winding.

[0023] The dynamic fluctuation threshold calculation module is used to set sliding windows for at least two different frequency bands in the preprocessed partial discharge high-frequency signal, calculate the white noise standard deviation within the sliding window, calculate the dynamic fluctuation threshold of the frequency band based on the white noise standard deviation, and preset the slope threshold and fluctuation number threshold.

[0024] The feature extraction and correction module is used to filter out the thyristor pulses and slot discharges in the preprocessed partial discharge high-frequency signal, extract effective feature points from the filtered partial discharge high-frequency signal according to the dynamic fluctuation threshold, slope threshold and fluctuation number threshold of the corresponding frequency band, and correct the feature quantity of the partial discharge high-frequency signal based on the effective feature points to complete the time domain feature extraction.

[0025] Thirdly, an electronic device includes a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein the processor executes the computer program to implement the steps of the method for extracting time-domain features of partial discharge signals from generator stator windings.

[0026] Fourthly, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for extracting time-domain features of partial discharge signals from generator stator windings.

[0027] Fifthly, a computer program product comprising a computer program that, when executed by a processor, implements the steps of the method for extracting time-domain features of partial discharge signals from generator stator windings.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] This invention sets sliding windows for at least two different frequency bands in the preprocessed partial discharge high-frequency signal, allowing for analysis of high-frequency and low-frequency signals using windows of different time lengths. This adapts to their rapidly changing and slowly changing propagation characteristics, avoiding the problems of high-frequency signal truncation or insufficient low-frequency noise suppression that occur when using a single window. By calculating the white noise standard deviation within the sliding window, and then calculating the dynamic fluctuation threshold of the frequency band based on the white noise standard deviation, the threshold is dynamically adjusted according to the actual noise level within the window, overcoming the high misjudgment rate of fixed thresholds when noise fluctuates. After filtering out thyristor pulses and slot discharges from the preprocessed partial discharge high-frequency signal, effective feature points are extracted from the filtered partial discharge high-frequency signal based on the dynamic fluctuation threshold, slope threshold, and oscillation number threshold for the corresponding frequency band. This allows for joint judgment in the time domain using dynamic amplitude thresholds, fixed slope thresholds, and oscillation number thresholds, thereby accurately identifying partial discharge pulses in strong interference environments. Finally, by correcting the characteristic quantities of the partial discharge high-frequency signal based on the effective feature points, the output characteristic quantities can reflect the distortion generated by the signal propagation in the stator winding, thereby improving the accuracy of feature evaluation.

[0030] Furthermore, by setting corresponding first sliding window duration and first sliding step size for the high-frequency band, and corresponding second sliding window duration and second sliding step size for the low-frequency band, independent window parameters are configured for the high and low frequency bands respectively. This allows the window length and step size to accurately match the time-domain characteristics of the signals in each frequency band. Using a shorter first window duration and step size is beneficial for capturing the transient details of high-frequency signals, while using a longer second window duration and step size is beneficial for performing more thorough statistics on low-frequency signals, thereby improving the stability of noise estimation and the timeliness of signal detection within their respective frequency bands.

[0031] Furthermore, among the high-frequency partial discharge signals, signal points that simultaneously satisfy the following conditions are selected as valid feature points: the amplitude change is greater than the dynamic fluctuation threshold, the rising slope is greater than the slope threshold, and the number of oscillations is greater than the fluctuation number threshold. These signal points must simultaneously cross the dynamic amplitude threshold, the fixed slope threshold, and the number of oscillations threshold, thereby effectively eliminating interference that only meets some of the characteristics and significantly improving the authenticity and representativeness of the extracted feature points. Attached Figure Description

[0032] Figure 1 A flowchart illustrating a method for extracting time-domain features of partial discharge signals from generator stator windings, provided in an embodiment of the present invention;

[0033] Figure 2 This is an overall technical flowchart of a method for extracting time-domain features of partial discharge signals from generator stator windings, provided by an embodiment of the present invention.

[0034] Figure 3 This is a structural diagram of a generator stator winding partial discharge signal time-domain feature extraction system provided in an embodiment of the present invention;

[0035] Figure 4 The logic diagram for calculating the dynamic fluctuation threshold provided in the embodiments of the present invention is shown. Detailed Implementation

[0036] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described in detail below with reference to the accompanying drawings. The content is for explanation rather than limitation of the present invention.

[0037] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification of this invention are intended to cover a non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, systems, products or devices.

[0038] The following explains the terminology used in the embodiments:

[0039] PDDC-24: The specific model number of the coupling capacitor;

[0040] MS / s: Mega Samples per second, refers to the sampling rate;

[0041] mV: millivolt;

[0042] μs: microsecond;

[0043] ns: nanosecond;

[0044] MHz: Megahertz;

[0045] kHz: kilohertz;

[0046] MW: Megawatt;

[0047] The present invention provides a method for extracting partial discharge signals of a generator stator winding based on a sliding window dynamic fluctuation threshold, which includes the following steps:

[0048] (1) Collect high-frequency partial discharge signals of the stator winding, synchronously record the power frequency phase, and perform baseline correction and deburring preprocessing; the sampling rate of the high-frequency partial discharge signals is 30MS / s, the single acquisition duration is 100ms, and the preprocessing includes removing DC offset and eliminating single-point mutations, the DC offset ≤ 0.5mV, and the mutation threshold < 2mV;

[0049] (2) Set sliding windows in different frequency bands: for the high-frequency band (3MHz - 20MHz), use a 20ms sliding window and a 4ms sliding step, and for the low-frequency band (50KHz - 3MHz), use a 50ms sliding window and a 10ms sliding step; the sliding windows in the high-frequency band cover [0 - 20ms), [4 - 24ms)..., and the sliding windows in the low-frequency band cover [0 - 50ms), [10 - 60ms)..., and the window overlap rate is 80% for both.

[0050] (3) Calculate the dynamic fluctuation threshold based on the standard deviation of white noise within the window: the dynamic fluctuation threshold in the high-frequency band Thres1_H = 1.5×2σ1, the dynamic fluctuation threshold in the low-frequency band Thres1_L = 1.2×3σ2, the slope threshold Thres2 = 6mV / μs, and the fluctuation number threshold Thres3 = 4; σ1 is the standard deviation of white noise within the high-frequency band window, σ2 is the standard deviation of white noise within the low-frequency band window, and the white noise is determined by extracting signal segments with an amplitude < 5mV and passing the normality test.

[0051] (4) Specifically filter thyristor pulses and slot discharges. The time window of thyristor pulses is 3.3ms ± 0.2ms, with an attenuation of 0.6 times. Slot discharges are signals with a phase of -30° to 30° and a slope < Thres2. Based on the dynamic fluctuation threshold, slope threshold, and fluctuation number threshold, extract effective feature points from the filtered high-frequency partial discharge signals; the judgment conditions for effective feature points are: in the high-frequency band, the amplitude change amount |ΔV| > Thres1_H and |rising slope| > Thres2, in the low-frequency band, the amplitude change amount |ΔV| > Thres1_L and |rising slope| > Thres2, where ΔV is the local amplitude change amount, and slope = ΔV / 33ns.

[0052] (5) Calculate modified time-domain characteristic quantities: pulse amplitude (with position coefficient), peak-valley number, duration (fast wave compensation), equivalent oscillation speed (with type coefficient). The correction coefficient for pulse amplitude is: for signals collected at the neutral point × 1.2, and for signals collected at the outgoing terminal × 0.9; the formula for the equivalent oscillation speed is ×k, is the equivalent oscillation speed, For the number of peaks and troughs, For duration, k is the type coefficient, with the inter-turn discharge type coefficient being 1.0 and the slot discharge type coefficient being 0.8.

[0053] Based on the above, such as Figure 1 and Figure 2 As shown, this embodiment provides a method for extracting time-domain features of partial discharge signals from generator stator windings, including the following steps:

[0054] Step 1: Acquire high-frequency partial discharge signals from the generator stator windings and perform preprocessing;

[0055] Specifically, the first step is to deploy sensors by installing PDDC-24 coupling capacitors at the generator neutral point or stator winding output terminals. These capacitors have a bandwidth of 50kHz to 20MHz and acquire high-frequency partial discharge signals. Sampling parameters include a sampling rate of 30ms / s, a single acquisition duration of 100ms, and simultaneous recording of the stator winding's power frequency phase information for subsequent phase correlation analysis. Preprocessing includes baseline correction (removing DC offset) and pulse de-glitching (eliminating single-point abrupt changes caused by sampling noise) on the raw signal.

[0056] Step 2: Set sliding windows for at least two different frequency bands in the preprocessed partial discharge high-frequency signal, calculate the white noise standard deviation within the sliding window, calculate the dynamic fluctuation threshold of the frequency band based on the white noise standard deviation, and preset the slope threshold and fluctuation number threshold.

[0057] Specifically, the design and verification of the frequency band sliding window are as follows: Verification of the window duration for the high-frequency band (3MHz~20MHz, fast wave mode): Experiments show that after a high-frequency signal propagates for 10m in the stator winding, its amplitude attenuates by >80%, and its effective duration is <20ms. A 20ms window can cover 98% of the complete high-frequency pulse waveform; if the window is <15ms, the pulse truncation rate is >30%; if the window is >25ms, the noise interference increases by 20%. Therefore, the window duration is set to 20ms. Verification of the sliding step size: An excessively large step size will lead to missed signal detection; when the step size is >8ms, the missed detection rate is >15%; an excessively small step size will increase the computational load; when the step size is <2ms, the computation time increases by a factor of 3. Therefore, the step size is set to 4ms, overlapping with the previous window by 16ms, with an overlap rate of 80%.

[0058] Window duration verification for the low-frequency band (50kHz~3MHz, slow wave mode): Low-frequency signals propagate and attenuate slowly, with a duration >50ms, but the noise fluctuation period is approximately 50ms. A 50ms window balances the accuracy of noise statistics, i.e., standard deviation error <5%; if the window is <40ms, the noise statistical error >15%; if the window is >60ms, the threshold response lag >20ms. Therefore, the window duration is set to 50ms. Sliding step size verification: When the step size is >15ms, the low-frequency pulse false negative rate is >10%; when the step size is <5ms, the computational load increases by 2 times. Therefore, the step size is set to 10ms, overlapping the previous window by 40ms, with an overlap rate of 80%.

[0059] Window sequence generation: Starting from the signal start time (t=0ms), the high-frequency band windows are [0~20ms), [4~24ms), [8~28ms)...; the low-frequency band windows are [0~50ms), [10~60ms), [20~70ms)..., to achieve full-time signal coverage.

[0060] Specifically, the dynamic fluctuation threshold is calculated as follows: Figure 4 As shown, for the signal within each window, the threshold is calculated in real time based on the statistical characteristics of white noise: White noise feature extraction. For the signal within the window, white noise is separated through the following steps: Extract all segments with amplitude <5mV in the signal and preliminarily determine them as noise; Perform a normality test on the noise segments, with skewness <0.1 and kurtosis <3.2, conforming to a normal distribution; Calculate the noise standard deviation σ, where the standard deviation of white noise in the high-frequency window is σ1, and the standard deviation of white noise in the low-frequency window is σ2. Fast wave noise is extracted from the high-frequency window, and slow wave noise is extracted from the low-frequency window.

[0061] For the dynamic fluctuation threshold Thres1, where the high-frequency dynamic fluctuation threshold Thres1_H = 1.5×2σ1, the principle is: 2σ1 covers 95% of the white noise, multiplied by a coefficient of 1.5. Experimental verification shows that the high-frequency pulse detection rate can be >98%, the noise false detection rate is <2%, and it is suitable for the characteristics of fast wave amplitude.

[0062] For the low-frequency dynamic fluctuation threshold Thres1_L = 1.2×3σ2, the principle is that 3σ2 covers 99.7% of white noise. Multiplying it by a coefficient of 1.2, experimental verification shows that the low-frequency pulse detection rate can be >95%, the noise false detection rate can be <5%, and the anti-interference ability can be enhanced.

[0063] For the slope threshold Thres2, experimental measurements show that the rising slope of the partial discharge pulses in the stator winding ≥ 8 mV / μs, with an average of 10 mV / μs for turn-to-turn discharges and 8 mV / μs for end-winding discharges, while the slope of interference pulses < 6 mV / μs, with an average of 3 mV / μs for slot discharges and 4 mV / μs for the thyristor attenuation section. Therefore, Thres2 is set to 6 mV / μs to distinguish effective signals from interference.

[0064] For the fluctuation count threshold Thres3, experimental statistics show that partial discharge pulses will generate ≥ 4 peak-valley oscillations due to winding reflections, with an average of 6 oscillations for turn-to-turn discharges and 4 oscillations for end-winding discharges, while the number of oscillations of interference pulses < 3, with an average of 2 oscillations for thyristor pulses and no obvious oscillations for narrowband interference. Therefore, Thres3 is set to 4 to eliminate short-period interference.

[0065] Step 3: After filtering the thyristor pulses and slot discharges in the preprocessed high-frequency partial discharge signals, extract effective feature points from the filtered high-frequency partial discharge signals according to the dynamic fluctuation threshold, slope threshold, and fluctuation count threshold in the corresponding frequency band, and correct the characteristic quantities of the high-frequency partial discharge signals based on the effective feature points to complete the time-domain feature extraction.

[0066] Specifically, targeted filtering for specific interference. For the filtering of thyristor pulses, based on its fixed period of 3.3 ms, mark through a time window: identify the occurrence times t1, t2, t3 of the first 3 pulses, calculate the period T = (t3 - t1) / 2, if |T - 3.3 ms| < 0.2 ms, it is determined as a thyristor pulse; for the signals at subsequent times t = t3 + n×T ± 0.2 ms (n = 1, and 2,...), perform amplitude attenuation with an attenuation coefficient of 0.6. Experimental verification shows that the interference amplitude can be reduced to less than 1 / 3 of the partial discharge signal amplitude, without affecting feature extraction. For the filtering of slot discharge interference, combine phase and slope features: extract the signals in the phase range of -30° to 30°, which is the concentrated area of slot discharges; calculate the rising slope of the signals in this range. If the slope < Thres2 (6 mV / μs), it is determined as a slot discharge and directly removed. Experimental verification shows that the removal rate > 94% and no effective signals are misdeleted.

[0067] Effective feature point extraction involves calculating the local change ΔV = V(k) - V(k-1) and the cumulative change ΣV = V(k) - V(start) point by point in the filtered signal. Extreme points satisfying the following conditions are retained: High frequency band: |ΔV| > Thres1_H and |rising slope| > Thres2; Low frequency band: |ΔV| > Thres1_L and |rising slope| > Thres2; where the slope = ΔV / Δt (Δt = 1 / 30MS / s ≈ 33ns, Δt is the sampling interval). That is, signal points in the partial discharge high-frequency signal that simultaneously satisfy the following conditions—amplitude change greater than the dynamic fluctuation threshold, rising slope greater than the slope threshold, and oscillation number greater than the fluctuation number threshold—are selected as effective feature points.

[0068] The modified characteristic quantity calculation adapted to the stator winding is based on the propagation attenuation characteristics of the stator winding signal, and the characteristic quantity is corrected: pulse amplitude (A) e ): Half the peak-to-peak value, multiplied by the propagation distance correction factor (e.g., multiply by 1.2 for neutral point acquisition signals, multiply by 0.9 for outgoing terminal acquisition signals, and compensate for winding attenuation); number of peaks and valleys (n) pv ): The total number of peaks and troughs within the effective fluctuation range; Duration (t_d): The time difference between the start and end points of the fluctuation, combined with propagation mode correction, with a 0.5μs increment added to the fast wave signal to compensate for the truncation error caused by high-frequency attenuation; Equivalent oscillation velocity (f e The formula is: ×k, where k is a pulse type coefficient unique to the stator winding, with k=1.0 for inter-turn discharge and k=0.8 for slot discharge, obtained based on stator winding simulation experiments.

[0069] Based on the above, the stator winding of a 300MW hydro-generator was used as the monitoring object. The sensor was installed at the neutral point, with a PDDC-24 coupling capacitor at the neutral point. The sampling rate was 30MS / s, with 100ms of data collected per cycle, totaling 3 million points. The power frequency phase was recorded synchronously from 0° to 360°. The experimental signals included: effective signals: the main frequency of the inter-turn partial discharge pulse was 1.2MHz, and the main frequency of the end discharge pulse was 0.8MHz; interference signals: the period of the thyristor pulse was 3.3ms, and the frequency was 0.4MHz; the phase of the slot discharge interference was -20° to 20°, with a slope of 3mV / μs; and the amplitude range of white noise was 0 to 15mV.

[0070] The steps are detailed below:

[0071] Step 1: Signal Acquisition

[0072] Collect high-frequency signals of the stator winding through a neutral-point coupling capacitor, with a sampling rate of 30 MS / s and a single continuous acquisition duration of ≥10 s (to ensure complete coverage of the noise fluctuation period).

[0073] Step 2: Sliding window setting

[0074] (1) Window parameters: Use a rectangular sliding window with a window duration of 50 ms, determined according to the noise change period of the stator winding. Measured results show that the noise fluctuation within 50 ms is ≤10%. If the window is too short, the statistics will be inaccurate; if it is too long, the response will be lagged. (2) Sliding method: The window starts from the starting time of the acquired signal (t = 0 ms) and slides to the right every 10 ms, that is, it moves 10 ms each time and overlaps with the previous window by 40 ms to ensure data continuity, forming a continuously covered window sequence. For example, the first window: 0 - 50 ms, the second window: 10 - 60 ms, the third window: 20 - 70 ms... and so on.

[0075] Step 3: Calculation of dynamic fluctuation threshold

[0076] Process the signals within each sliding window as follows: (1) Calculate the white noise characteristics within the window: Statistically analyze the amplitude distribution of the white noise within the window and calculate the standard deviation σ, which reflects the degree of noise dispersion. (2) Dynamic fluctuation threshold Thres1: Thres1 = 1.2 × 3σ. 3σ covers 99.7% of the noise, and 1.2 times reserves a 20% margin to cope with sudden noise. (3) Slope threshold Thres2: Based on the minimum rising slope of the partial discharge pulses within the window, the measured effective pulse slope of the stator winding is ≥8 mV / μs, so Thres2 = 5 mV / μs. (4) Fluctuation count threshold Thres3: Statistically analyze the average oscillation count of the partial discharge pulses within the window. The measured value is ≥4 times, so Thres3 = 4, which is different from the general fixed value of 3.

[0077] Step 4: Interference filtering

[0078] (1) Thyristor pulse: The signal within the time window of 3.3 ms ± 0.2 ms is attenuated by 0.6 times, and the amplitude drops from 600 mV to 360 mV. (2) Slot discharge interference: Within the phase range of -20° to 20°, if the slope = 3 mV / μs < Thres2, it is directly eliminated. (3) Extraction of effective feature points: If the local change amount of a turn-to-turn discharge pulse = 20 mV (> Thres1_L = 18 mV) and the slope = 10 mV / μs (> Thres2), it is determined as an effective feature point.

[0079] Step 5: Signal extraction and judgment

[0080] For real-time signals, each point is checked to see if it meets the following criteria: amplitude change > Thres1; rising slope > Thres2; number of oscillations > Thres3. If all criteria are met, it is determined to be a valid partial discharge signal; otherwise, it is noise.

[0081] like Figure 3 As shown, this embodiment provides a time-domain feature extraction system for partial discharge signals of generator stator windings, including:

[0082] The signal acquisition module is used to acquire and preprocess the high-frequency partial discharge signal of the generator stator winding.

[0083] The dynamic fluctuation threshold calculation module is used to set sliding windows for at least two different frequency bands in the preprocessed partial discharge high-frequency signal, calculate the white noise standard deviation within the sliding window, calculate the dynamic fluctuation threshold of the frequency band based on the white noise standard deviation, and preset the slope threshold and fluctuation number threshold.

[0084] The feature extraction and correction module is used to filter out the thyristor pulses and slot discharges in the preprocessed partial discharge high-frequency signal, extract effective feature points from the filtered partial discharge high-frequency signal according to the dynamic fluctuation threshold, slope threshold and fluctuation number threshold of the corresponding frequency band, and correct the feature quantity of the partial discharge high-frequency signal based on the effective feature points to complete the time domain feature extraction.

[0085] This embodiment addresses the misjudgment problem caused by fixed thresholds during noise fluctuations by dynamically updating the threshold within a 50ms sliding window, achieving noise self-adaptation. For the unique interference of SCR pulses (periodic filtering) and slot discharges (phase filtering), the rejection efficiency is improved to 92%, far exceeding the general method (75%). Without complex time-frequency transformations, 3 million points can be analyzed within 100ms, meeting the real-time requirements of online stator winding monitoring. The extracted features are directly correlated with stator winding insulation defect types (such as inter-turn discharge and end discharge), and can be directly used for subsequent defect identification, aligning with insulation assessment needs.

[0086] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0087] This embodiment also provides a computer device, which includes a processor and a memory. The memory is used to store a computer program (in this embodiment, the computer program includes a calculation component and an iterative component, capable of model calculation and model updating). The computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to realize the corresponding method flow or corresponding function. The processor of this embodiment can be used for the operation of a method for extracting time-domain features of partial discharge signals of generator stator windings.

[0088] This embodiment also provides a storage medium, specifically a computer-readable storage medium (Memory). A computer-readable storage medium is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the time-domain feature extraction method for partial discharge signals of generator stator windings in the above embodiment.

[0089] This embodiment also provides a computer program product, which includes a computer program that, when executed by a processor, implements the corresponding steps of the time-domain feature extraction method for partial discharge signals of generator stator windings in the above embodiment.

[0090] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0091] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for extracting time-domain features of partial discharge signals from generator stator windings, characterized in that, Includes the following steps: Acquire high-frequency partial discharge signals from the generator stator windings and perform preprocessing; Sliding windows are set for at least two different frequency bands in the preprocessed partial discharge high-frequency signal, and the white noise standard deviation within the sliding window is calculated. The dynamic fluctuation threshold of the frequency band is calculated based on the white noise standard deviation, and a slope threshold and a fluctuation number threshold are preset. After filtering out the thyristor pulses and slot discharges in the preprocessed partial discharge high-frequency signal, effective feature points are extracted from the filtered partial discharge high-frequency signal according to the dynamic fluctuation threshold, slope threshold and fluctuation number threshold of the corresponding frequency band. Based on the effective feature points, the feature quantity of the partial discharge high-frequency signal is corrected to complete the time domain feature extraction. For at least two different frequency bands in the preprocessed partial discharge high-frequency signal, a sliding window is set, and the standard deviation of white noise within the sliding window is calculated, specifically including: Set a corresponding first sliding window duration and a first sliding step size for the high-frequency band, and set a corresponding second sliding window duration and a second sliding step size for the low-frequency band; The standard deviation of white noise in the high-frequency band window is calculated based on the first sliding window duration and the first sliding step size. The standard deviation of white noise in the low-frequency band window is calculated based on the second sliding window duration and the second sliding step size. The dynamic fluctuation threshold of the frequency band is calculated based on the white noise standard deviation, specifically including: Based on the standard deviation of white noise within the high-frequency band window, the dynamic fluctuation threshold of the high-frequency band is calculated using the following formula: Thres1_H = 1.5×2σ1, where Thres1_H is the dynamic fluctuation threshold of the high-frequency band, and σ1 is the standard deviation of white noise within the high-frequency band window; Based on the standard deviation of white noise within the low-frequency window, the dynamic fluctuation threshold of the low-frequency band is calculated using the following formula: Thres1_L = 1.2×3σ2, where Thres1_L is the dynamic fluctuation threshold of the low-frequency band, and σ2 is the standard deviation of white noise within the low-frequency window.

2. The method for extracting time-domain features of partial discharge signals from generator stator windings according to claim 1, characterized in that, Based on the dynamic fluctuation threshold, slope threshold, and fluctuation number threshold of the corresponding frequency band, effective feature points are extracted from the filtered partial discharge high-frequency signal, specifically including: Calculate the amplitude change, rise slope, and number of oscillations of the filtered partial discharge high-frequency signal; Among the high-frequency partial discharge signals, signal points that simultaneously satisfy the following conditions are selected as valid feature points: the amplitude change is greater than the dynamic fluctuation threshold, the rising slope is greater than the slope threshold, and the number of oscillations is greater than the fluctuation number threshold.

3. The method for extracting time-domain features of partial discharge signals from generator stator windings according to claim 1, characterized in that, The characteristic quantities include: pulse amplitude, number of peaks and troughs, duration, and equivalent oscillation velocity.

4. The method for extracting time-domain features of partial discharge signals from generator stator windings according to claim 1, characterized in that, The process involves acquiring and preprocessing high-frequency partial discharge signals from the generator stator windings, specifically including: Remove the DC offset of the partial discharge high-frequency signal and eliminate single-point abrupt changes in the partial discharge high-frequency signal.

5. A system for extracting time-domain features of partial discharge signals from generator stator windings, characterized in that, include: The signal acquisition module is used to acquire and preprocess the high-frequency partial discharge signal of the generator stator winding. The dynamic fluctuation threshold calculation module is used to set sliding windows for at least two different frequency bands in the preprocessed partial discharge high-frequency signal, calculate the white noise standard deviation within the sliding window, calculate the dynamic fluctuation threshold of the frequency band based on the white noise standard deviation, and preset the slope threshold and fluctuation number threshold. The feature extraction and correction module is used to filter out the thyristor pulses and slot discharges in the preprocessed partial discharge high-frequency signal, extract effective feature points from the filtered partial discharge high-frequency signal according to the dynamic fluctuation threshold, slope threshold and fluctuation number threshold of the corresponding frequency band, and correct the feature quantity of the partial discharge high-frequency signal based on the effective feature points to complete the time domain feature extraction. For at least two different frequency bands in the preprocessed partial discharge high-frequency signal, a sliding window is set, and the standard deviation of white noise within the sliding window is calculated, specifically including: Set a corresponding first sliding window duration and a first sliding step size for the high-frequency band, and set a corresponding second sliding window duration and a second sliding step size for the low-frequency band; The standard deviation of white noise in the high-frequency band window is calculated based on the first sliding window duration and the first sliding step size. The standard deviation of white noise in the low-frequency band window is calculated based on the second sliding window duration and the second sliding step size. The dynamic fluctuation threshold of the frequency band is calculated based on the white noise standard deviation, specifically including: Based on the standard deviation of white noise within the high-frequency band window, the dynamic fluctuation threshold of the high-frequency band is calculated using the following formula: Thres1_H = 1.5×2σ1, where Thres1_H is the dynamic fluctuation threshold of the high-frequency band, and σ1 is the standard deviation of white noise within the high-frequency band window; Based on the standard deviation of white noise within the low-frequency window, the dynamic fluctuation threshold of the low-frequency band is calculated using the following formula: Thres1_L = 1.2×3σ2, where Thres1_L is the dynamic fluctuation threshold of the low-frequency band, and σ2 is the standard deviation of white noise within the low-frequency window.

6. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable in the processor. When the processor executes the computer program, it implements the time-domain feature extraction method for partial discharge signals of generator stator windings as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for extracting time-domain features of partial discharge signals from generator stator windings as described in any one of claims 1 to 4.

8. A computer program product, the computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the method for extracting time-domain features of partial discharge signals of generator stator windings as described in any one of claims 1 to 4.

Images