Generator partial discharge on-line monitoring method and device based on neutral point coupling method

Abstract

This invention relates to a method and apparatus for online monitoring of partial discharge in generators based on neutral point coupling. The method includes installing a capacitively coupled sensor at the generator neutral point, triggering a data acquisition unit based on the rising edge of a square wave generated by a signal synchronization unit to acquire partial discharge data; preprocessing the partial discharge data to obtain a preprocessed signal; extracting pulse signals from the preprocessed signal and calculating the phase of each pulse signal; removing interference pulses based on the phase to obtain partial discharge pulses; classifying the partial discharge pulses; determining the discharge phase of each type of partial discharge pulse after classification and generating corresponding phase distribution maps; comparing the phase distribution maps with pre-stored typical discharge mode maps, and issuing an early warning based on the comparison results. This method can achieve adaptive detection, analysis, and early warning of partial discharge signals.

Description

Technical Field

[0001] This invention relates to the field of generator partial discharge monitoring, specifically to a method and device for online monitoring of generator partial discharge based on neutral point coupling. Background Technology

[0002] Partial discharge is a significant symptom and cause of insulation degradation in generator stator windings. Online monitoring of partial discharge is a crucial technology for assessing generator insulation condition, preventing sudden insulation faults, and enabling predictive maintenance. Partial discharge monitoring detects transient pulse signals generated by internal insulation defects, acquiring early characteristic information of insulation degradation. Combined with pattern recognition technology, it can achieve fault type diagnosis and early warning, which is of great significance for ensuring the safe and stable operation of large generator sets.

[0003] Existing online monitoring technologies for generator partial discharge are mainly based on electrical measurement methods. Depending on the sensor installation location, they can be divided into high-voltage side coupling and neutral point coupling methods. The high-voltage side coupling method installs the sensor at the high-voltage end of the generator outlet, offering advantages such as high signal coupling efficiency and a good signal-to-noise ratio. However, this method requires the sensor and auxiliary equipment to have extremely high insulation withstand voltage ratings, the installation process is complex and poses safety hazards, and due to the coupling characteristics of partial discharge, the high-voltage side coupling method also suffers from interphase coupling.

[0004] The neutral point coupling method involves installing the sensor at the generator's neutral point. Due to the low potential of the generator neutral point relative to ground (typically zero-sequence voltage), this method offers significant advantages such as safe installation, ease of implementation, and the ability to capture all discharge pulses from the three-phase windings simultaneously, making it a mainstream choice in engineering applications. However, the discharge signal at the neutral point suffers severe attenuation and distortion after propagation through the windings, and is highly susceptible to various electromagnetic interferences in the field (such as carrier communication, thyristor rectification, white noise, etc.), resulting in a low signal-to-noise ratio and difficulty in pulse signal extraction. Furthermore, the discharge signals from the three-phase windings superimpose at the neutral point, exhibiting interphase coupling, making accurate identification of the original discharge pulse's phase from the mixed signal (i.e., "discharge phase identification") a technical challenge.

[0005] Currently, monitoring devices based on neutral point coupling still face some challenges in engineering applications: First, under strong noise backgrounds, traditional thresholding or fixed filtering methods are difficult to extract real partial discharge pulses stably and accurately, which can easily lead to missed or false detections; Second, there is a lack of effective algorithms to automatically and accurately identify key features (such as peak phase) in the phase distribution pattern of discharge pulses, and to reliably distinguish the discharge of the three phases A, B, and C, which limits the accurate location of fault sources and deeper diagnostic analysis.

[0006] Therefore, there is an urgent need for an online monitoring method for generator partial discharge that is applicable to neutral point coupling scenarios, has strong anti-interference capabilities, and can achieve high-precision discharge pulse extraction and automatic discharge phase identification, so as to improve the reliability, accuracy and intelligence level of the monitoring device. Summary of the Invention

[0007] This invention provides a method and device for online monitoring of generator partial discharge based on neutral point coupling. The purpose is to solve the problems of inaccurate extraction of partial discharge pulses and difficulty in identifying discharge phases caused by strong noise interference and superposition of three-phase signals in the existing online monitoring technology for generator partial discharge based on neutral point coupling.

[0008] To achieve the above objectives, the first aspect of the present invention provides a method for online monitoring of partial discharge in a generator based on neutral point coupling, comprising the following steps:

[0009] A capacitive coupling sensor is installed at the neutral point of the generator. The data acquisition unit is triggered by the rising edge of the square wave generated by the signal synchronization unit to collect partial discharge data.

[0010] The partial discharge data is preprocessed to obtain a preprocessed signal;

[0011] Pulse signals are extracted from the preprocessed signal, and the phase of each pulse signal is calculated;

[0012] Based on the phase-eliminating interference pulse, a partial discharge pulse is obtained;

[0013] The partial discharge pulses are classified.

[0014] After determining the discharge phase of each type of partial discharge pulse, the corresponding phase distribution spectrum is generated;

[0015] The phase distribution spectrum is compared with the pre-stored typical discharge mode spectrum, and an early warning is issued based on the comparison results.

[0016] Furthermore, the method for preprocessing the partial discharge data to obtain a preprocessed signal includes:

[0017] A high-pass filter is used to filter the partial discharge data to reduce white noise;

[0018] Data within one power frequency cycle is extracted from the partial discharge data after white noise reduction;

[0019] The data is windowed using a Blackman-trapezoidal convolution window of length N to obtain a windowed signal. The Blackman-trapezoidal convolution window is obtained by convolving a Blackman window and a trapezoidal window, both of length N / 2.

[0020] Perform a Fast Fourier Transform on the windowed signal to obtain its spectrum, calculate the amplitude of the spectrum, and calculate a threshold based on the spectrum amplitude.

[0021] The amplitude values ​​are iterated over. For a spectrum with an amplitude value greater than or equal to the threshold, the average amplitude value of the first five spectra is taken as the corrected amplitude value of the spectrum. For a spectrum with an amplitude value less than the threshold, its original amplitude value is retained.

[0022] The reconstructed partial discharge data are obtained by performing an inverse Fourier transform on the corrected spectrum.

[0023] By iterating through all partial discharge data after white noise reduction, the discrete spectrum interference is reduced.

[0024] Furthermore, the method for extracting pulse signals from the preprocessed signal includes:

[0025] Calculate the kurtosis of the preprocessed signal;

[0026] Extract signal data with a kurtosis value greater than 3 as pulse signals;

[0027] The formula for calculating the kurtosis is as follows:

[0028]

[0029] in, For raucousness; This represents the total number of data points within a signal segment involved in the calculation. Indicates the first... The value of each data point; This represents the average value of the preprocessed signal; The standard deviation of the preprocessed signal is represented by .

[0030] Furthermore, methods for calculating the phase of each pulse signal include:

[0031] The power frequency period corresponding to the pulse signal, from 0° to 360°, is divided into 360 phase windows, each with a width of 1°.

[0032] For the Each pulse signal is used to determine the location of the data point where the absolute value of its amplitude is the maximum.

[0033] Based on the location of the data points, the phase of the pulse signal is calculated using the following formula:

[0034]

[0035] in, Let P be the phase of the pulse signal, and P be the location of the data point where the absolute value of the amplitude is maximum. The sampling frequency of the data acquisition unit. The power frequency is indicated by "round".

[0036] Furthermore, the method based on the phase-eliminating interference pulse includes: traversing the phase from 0° to 60° at 1° intervals. If in phase , , , , , If pulses are present in both phases, the Pearson correlation coefficient is calculated based on the pulse waveform characteristics of the aforementioned phases. If the Pearson correlation coefficient between any two pulses is greater than 0.6, they are considered interference pulses and are discarded. The value ranges from 0 to 60; wherein the pulse waveform feature is a time-domain waveform or a frequency-domain feature.

[0037] Furthermore, the method for classifying the partial discharge pulses includes:

[0038] Principal component analysis was performed to reduce the dimensionality of the characteristic spectrum of the partial discharge pulse.

[0039] Unsupervised clustering algorithms are applied to cluster the dimensionality-reduced feature data to complete the classification. The unsupervised clustering algorithms include, but are not limited to, density-based noise-based spatial clustering algorithms or density peak clustering algorithms.

[0040] Furthermore, methods for determining the discharge phase of each type of partial discharge pulse after classification and generating corresponding phase distribution maps include:

[0041] For a set of partial discharge pulses to be processed, the kernel density estimation function of its phase in the range of 0° to 360° is calculated using the von Mises distribution kernel function based on the phase data of each pulse.

[0042] Multiple initial phase points are selected within the range of 0° to 360°. Each initial phase point is iteratively updated along the gradient direction of the kernel density estimation function until the phase change is less than a set threshold or the maximum number of iterations is reached, thereby obtaining at least one peak phase candidate point.

[0043] The candidate peak phase points are deduplicated to obtain the accurate peak phase;

[0044] Based on the precise peak phase and the preset phase determination conditions, the discharge phase of this type of partial discharge pulse is determined;

[0045] Based on the determined discharge phase, the phases of various partial discharge pulses are corrected and aligned, and their phase distribution within the power frequency cycle is statistically analyzed to generate a phase distribution map.

[0046] Furthermore, the formula for calculating the kernel density estimation function is as follows:

[0047]

[0048] in, Here is the kernel density estimation function; This represents the total number of pulses of this type. The kernel function is the von Mises distribution kernel function. The phase point to be estimated; For the first The phase of each pulse;

[0049] The von Mises distribution kernel function The calculation formula is:

[0050]

[0051] in, For concentration parameters; It is a zero-order modified Bessel function;

[0052] The formula for calculating the gradient direction iterative update is as follows:

[0053]

[0054] in, For the first Phase value of the next iteration For the first Phase value of the next iteration For learning rate, Kernel density estimation function At point The gradient at that point.

[0055] Furthermore, the phase determination conditions include:

[0056] If condition A is met, then this type of partial discharge pulse is determined to be phase A discharge, where condition A is:

[0057]

[0058] If condition B is met, then this type of partial discharge pulse is determined to be a phase B discharge, where condition B is:

[0059]

[0060] If condition C is met, then this type of partial discharge pulse is determined to be a phase C discharge, where condition C is:

[0061]

[0062] in, This indicates the precise peak phase corresponding to a positive discharge; This indicates the precise peak phase corresponding to the negative discharge.

[0063] To achieve the above objectives, a second aspect of the present invention provides an online monitoring device for partial discharge of a generator based on the neutral point coupling method, comprising:

[0064] A capacitive coupling sensor is installed at the neutral point of the generator to couple partial discharge signals.

[0065] The signal synchronization unit is used to acquire the phase of the generator A-phase outlet voltage and generate a square wave synchronization signal. The rising edge of the square wave signal is synchronized with the zero phase point of the power frequency voltage.

[0066] The data acquisition unit, connected to the capacitively coupled sensor and the signal synchronization unit, is used to acquire partial discharge data triggered by the rising edge of the square wave synchronization signal.

[0067] A data preprocessing unit, connected to the data acquisition unit, is used to preprocess the acquired partial discharge data to reduce white noise and discrete spectrum interference and obtain a preprocessed signal.

[0068] A pulse extraction and processing unit, connected to the data preprocessing unit, is used to extract pulse signals from the preprocessed signal, calculate the phase of each pulse signal, and remove interference pulses based on the phase to obtain partial discharge pulses;

[0069] A pulse classification unit, connected to the pulse extraction and processing unit, is used to classify the partial discharge pulses;

[0070] The discharge phase analysis and spectrum generation unit is connected to the pulse classification unit and is used to determine the discharge phase of each type of partial discharge pulse after classification and generate the corresponding phase distribution spectrum.

[0071] Storage unit, used to store pre-stored typical discharge mode maps;

[0072] The early warning unit, connected to the discharge phase analysis and spectrum generation unit and the storage unit, is used to compare the generated phase distribution spectrum with the pre-stored typical discharge mode spectrum and issue an early warning based on the comparison result.

[0073] The beneficial effects of this invention are:

[0074] Compared with existing technologies, the present invention provides a generator partial discharge online monitoring method and device based on neutral point coupling. In the pulse extraction stage, high-pass filtering and discrete spectrum interference reduction based on spectrum correction are applied to the acquired data to effectively suppress broadband and narrowband noise. Furthermore, an adaptive pulse detection method based on kurtosis is employed to accurately separate the partial discharge pulse from the preprocessed signal, and periodic interference pulses of the same phase are eliminated based on phase information, thus significantly improving the accuracy and anti-interference capability of pulse extraction. Secondly, in the discharge phase identification stage, a peak phase localization method based on kernel density estimation and gradient ascent is used to estimate the probability density of the classified pulse phase distribution and find the density peak, accurately capturing the main phase characteristics after the three-phase signals are superimposed. Subsequently, based on preset peak phase logic judgment conditions for different discharge polarities, the phase to which the discharge pulse belongs can be uniquely determined, effectively solving the problem of discharge phase localization caused by inter-phase signal coupling and superposition. Ultimately, by generating high-quality phase distribution maps that can be compared with typical patterns in IEC standards, accurate diagnosis and early warning of insulation defect types were achieved, forming a solution for reliably extracting discharge signals from strong noise and accurately identifying discharge phases. Attached Figure Description

[0075] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0076] Figure 1 This is a flowchart of an online monitoring method for partial discharge of a generator based on the neutral point coupling method disclosed in an embodiment of the present invention.

[0077] Figure 2 This is a typical PRPD mode diagram of internal discharge disclosed in an embodiment of the present invention, wherein... Figure 2 (a) in the diagram is a typical phase-analyzed partial discharge pattern of internal cavity discharge. Figure 2 (b) in the diagram is a typical phase-analyzed partial discharge pattern of internal layered discharge in insulation. Figure 2 (c) in the diagram is a typical phase-analyzed partial discharge pattern of layered discharge between the conductor and the insulation.

[0078] Figure 3 This is a typical PRPD mode diagram of a slot discharge disclosed in an embodiment of the present invention.

[0079] Figure 4 This is a typical PRPD mode diagram of end discharge disclosed in an embodiment of the present invention, wherein... Figure 4 (a) in the figure shows a typical phase-analyzed partial discharge (PRPD) pattern of end corona discharge. Figure 4 (b) in the diagram shows a typical phase-analyzed partial discharge (PRPD) pattern with identical end discharges. Figure 4 (c) in the figure is a typical phase-analyzed partial discharge (PRPD) pattern of end surface creepage. Detailed Implementation

[0080] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0081] According to embodiments of the present invention, it should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer device such as a set of computer-executable instructions, and although a logical order is shown in the following manufacturing method, in some cases the steps shown or described may be performed in a different order than that shown here.

[0082] like Figure 1 As shown, this invention provides an online monitoring method for partial discharge of a generator based on the neutral point coupling method, comprising the following steps:

[0083] Step S100: Install the capacitive coupling sensor at the neutral point of the generator, and trigger the data acquisition unit based on the rising edge of the square wave generated by the signal synchronization unit to acquire partial discharge data.

[0084] Step S200: Perform data preprocessing on the partial discharge data to obtain a preprocessed signal;

[0085] Step S300: Extract pulse signals from the preprocessed signal and calculate the phase of each pulse signal;

[0086] Step S400: Based on the phase-removed interference pulse, a partial discharge pulse is obtained;

[0087] Step S500: Classify the partial discharge pulses;

[0088] Step S600: Determine the discharge phase of each type of partial discharge pulse after classification and generate the corresponding phase distribution spectrum;

[0089] Step S700: Compare the phase distribution map with the pre-stored typical discharge mode map, and issue an early warning based on the comparison result.

[0090] Understandably, this method aims to achieve real-time monitoring and early warning of generator insulation status. By safely installing sensors at the neutral point and synchronously triggering data acquisition, the method effectively acquires raw signals containing three-phase discharge information. After preprocessing to suppress noise and interference, partial discharge pulses are accurately extracted using kurtosis detection and phase screening. The pulses are then classified through cluster analysis, and kernel density estimation and gradient ascent algorithms are used to accurately identify the discharge phase and its corresponding phase. Finally, through intelligent comparison of phase distribution maps and typical patterns, discharge type diagnosis and early warning are completed. This method effectively solves the problems of difficult signal extraction, large noise interference, and insufficient accuracy in distinguishing three-phase discharges in neutral point monitoring, significantly improving the reliability, automation level, and fault identification capability of partial discharge monitoring.

[0091] In this embodiment, step S100 involves the hardware deployment of the monitoring device and the synchronous triggering of the acquisition process. First, a capacitive coupling sensor is installed at the neutral point grounding lead of the generator to be monitored. The sensor consists of a high-frequency high-voltage capacitor and a Rogowski coil. The high-frequency high-voltage capacitor is responsible for coupling the transient partial discharge pulse signal generated by the stator winding insulation defect and conducted to the neutral point, while the Rogowski coil is used to sense and measure the pulse current signal. The two work together to extract the discharge signal.

[0092] Installing at this low potential point avoids the insulation risks associated with high-voltage side installation and allows for the collection of discharge signals from the three-phase windings. Subsequently, a signal synchronization unit is activated, consisting of an isolation transformer and conditioning equipment. The isolation transformer is connected to the secondary side (e.g., phase A) of the generator outlet voltage transformer (PT) to obtain a phase voltage signal synchronized with the power grid frequency. This voltage signal is then isolated and stepped down (e.g., to 10V) before being converted into a square wave signal of the same frequency by the subsequent conditioning equipment, ensuring that each rising edge of the square wave signal precisely corresponds to the 0° phase point (i.e., zero-crossing point) of the original sinusoidal phase voltage. This square wave signal is output as a global synchronization clock to the data acquisition unit.

[0093] It should be noted that the signal synchronization unit can also be connected to the voltage transformer at the output of phase B or phase C of the generator, using the voltage phase of the corresponding phase as the synchronization reference. If phase B voltage triggering is used, then in the phase determination conditions, conditions A, B, and C correspond to the discharge of phases B, C, and A, respectively; if phase C voltage triggering is used, then conditions A, B, and C correspond to the discharge of phases C, A, and B, respectively. In practical applications, the synchronization phase can be flexibly selected according to the convenience of on-site wiring or monitoring requirements, but it is necessary to ensure that the phase determination conditions are consistent with the selected synchronization phase to guarantee the accuracy of the discharge phase identification results.

[0094] When the data acquisition unit receives the rising edge of the square wave, it immediately starts a data acquisition window and synchronously samples the analog signal from the capacitively coupled sensor at a pre-set high sampling frequency (not less than 100MHz) to obtain partial discharge data.

[0095] In this embodiment, step S200 preprocesses the acquired raw partial discharge data to suppress or eliminate two main types of interference: broadband white noise and narrowband discrete spectrum interference, thereby improving the signal-to-noise ratio of the signal.

[0096] The specific methods for preprocessing are as follows:

[0097] First, the data preprocessing unit processes the raw collected data. A high-pass digital filter is applied for filtering. This filter can effectively remove low-frequency background noise and some power frequency fundamental components from the signal, while retaining high-frequency components containing partial discharge pulse information, thus achieving preliminary suppression of broadband white noise and obtaining an intermediate signal.

[0098] Discrete spectral interference is typically generated by carrier communication, switching of power electronic equipment, etc., and manifests in the frequency domain as discrete spectral lines with amplitudes significantly higher than the background. This invention employs a refined method based on spectrum correction to reduce it, the specific steps of which are as follows:

[0099] Step S201: Take data from the partial discharge data after white noise reduction within one power frequency cycle. ,in Indicates less than integers, , This indicates the power frequency, i.e., 50Hz. The sampling frequency of the data acquisition unit. Indicates rounding down;

[0100] Step S202, using a length of Multiplying the Blackman-trapezoidal convolution window w(n) by x(n) yields the windowed signal. :

[0101]

[0102] in, It consists of lengths respectively It is obtained by convolving the Blackman window and the trapezoidal window, and the calculation formula is:

[0103]

[0104] in, , These are the time-domain expressions for the Blackman window and the trapezoidal window, respectively. " indicates convolution operation;

[0105] Step S203: Windowing signal After performing a Fast Fourier Transform, the spectrum is obtained. The spectral amplitude is calculated, and a threshold is calculated based on the spectral amplitude. :

[0106]

[0107] in, represents the standard deviation of the spectral amplitude, and ln represents the logarithm with the natural constant e as the base;

[0108] Step S204: Traverse the spectrum amplitude, and for amplitudes greater than or equal to the threshold... The spectrum is calculated by taking the average of the first five frequencies as the amplitude of the spectrum. For frequencies with amplitudes less than a threshold, ... The original spectrum is retained, thus obtaining the corrected spectrum;

[0109] Step S205: Based on the corrected spectrum, perform an inverse Fourier transform to obtain the reconstructed partial discharge data, i.e., the preprocessed signal. Iterate through all discharge data to complete discrete spectrum interference reduction.

[0110] After processing in step S200, the final output is the preprocessed signal. White noise that differs significantly from the partial discharge pulse shape and discrete spectrum interference with fixed frequency characteristics are effectively suppressed in this signal, significantly highlighting the true partial discharge pulse components.

[0111] In this embodiment, step S300 is to obtain the preprocessed signal with noise suppressed. In this process, the pulse signals generated by partial discharge are identified and located, and the precise phase of each pulse relative to the power frequency voltage is calculated.

[0112] This step is divided into two parts: pulse signal extraction and pulse phase calculation.

[0113] For pulse signal extraction, partial discharge pulses are transient and non-Gaussian, with kurtosis values ​​much higher than stationary background noise. The specific method for extracting pulse signals from the preprocessed signal includes:

[0114] Calculate the kurtosis of the preprocessed signal;

[0115] Extract signal data with a kurtosis value greater than 3 as pulse signals;

[0116] The formula for calculating the kurtosis is as follows:

[0117]

[0118] in, For raucousness; This represents the total number of data points within a signal segment involved in the calculation. Indicates the first... The value of each data point; This represents the average value of the preprocessed signal; The standard deviation of the preprocessed signal is represented by .

[0119] The "3" mentioned above is an empirical kurtosis threshold. When the calculated kurtosis value... If the kurtosis exceeds this threshold, it is determined that the data segment is likely to contain a partial discharge pulse. All data points within this data segment (or pulse data after further precise localization) are marked and extracted as candidate pulse signals.

[0120] For each extracted candidate pulse signal, its specific phase within the power frequency cycle needs to be determined. This phase is a key parameter for subsequent interference removal, discharge phase identification, and PRPD pattern generation. Therefore, the methods for calculating the phase of each pulse signal specifically include:

[0121] Based on the extracted pulse data, its power frequency phase from 0 to 360° is divided into 360 phase windows, each with a width of 1°; for the... For each pulse, the phase corresponding to the position with the maximum absolute value of the pulse amplitude is taken as the pulse phase. All pulse phases are recorded. The formula for calculating the pulse phase is:

[0122]

[0123] in, Let P be the phase of the pulse signal, and P be the location of the data point where the absolute value of the amplitude is maximum. The sampling frequency of the data acquisition unit. The power frequency is indicated by "round".

[0124] Understandably, the output of step S300 is a set of pulse data and its corresponding phase values. This process fully utilizes the kurtosis characteristics of partial discharge pulses for adaptive extraction and combines them with a high-precision synchronous clock for phase calibration.

[0125] In this embodiment, as shown in step S400, the method for eliminating interference pulses based on the phase includes: traversing the phases from 0° to 60° at 1° intervals. If in phase , , , , , If pulses are present in both phases, the Pearson correlation coefficient is calculated based on the pulse waveform characteristics of the aforementioned phases. If the Pearson correlation coefficient between any two pulses is greater than 0.6, they are considered interference pulses and are discarded. The value ranges from 0 to 60; wherein the pulse waveform feature is a time-domain waveform or a frequency-domain feature.

[0126] In this embodiment, step S500 automatically groups (classifies) the purified partial discharge pulse set obtained in step S400, classifying pulses with similar discharge characteristics into one category. Because a generator may have multiple types of insulation defects or exhibit multiple discharge modes at the same defect, distinguishing them helps to more accurately locate the fault source and assess the insulation condition.

[0127] The implementation of this step mainly includes the following three stages: feature calculation and graph construction, feature dimensionality reduction, and unsupervised clustering.

[0128] For feature quantity calculation and spectrum construction: The pulse classification unit first extracts a set of quantization parameters, i.e., features, for each partial discharge pulse that can characterize its waveform, spectrum, or statistical properties. Features may include, but are not limited to, equivalent time width and equivalent bandwidth.

[0129] Each pulse is represented by a multidimensional feature vector constructed using all its characteristic quantities. Collecting all the feature vectors of all pulses together creates a feature map that comprehensively reflects the characteristics of all pulses.

[0130] For feature dimensionality reduction, since the initially extracted features may be numerous and correlated, direct clustering is inefficient and susceptible to noise interference. Therefore, principal component analysis is used to reduce the dimensionality of the feature map.

[0131] For unsupervised clustering, on the dimensionality-reduced feature data, unsupervised clustering algorithms, including but not limited to density-based noise spatial clustering (DBSCAN) and density peak clustering (DPC), are applied to classify partial discharge pulses.

[0132] In this embodiment, step S600 performs discharge phase identification for each type of partial discharge pulse classified in step S500 and generates a corresponding phase distribution (PRPD) map. The classification of the partial discharge pulses employs a discharge phase identification method based on kernel density estimation and gradient ascent, including the following steps:

[0133] Step S601: Based on this type of partial discharge pulse Calculate the density estimate for each pulse phase:

[0134]

[0135] in, Here is the kernel density estimation function; The total number of pulses of this type, taking a positive integer value; The phase point to be estimated; For the first The phase of each pulse, its value ; The calculation uses the ring distance; The kernel function for the von Mises distribution is calculated using the following formula:

[0136]

[0137] in, This is a concentration parameter, taking positive real numbers. It can be selected through cross-validation, and the initial value can be... Make an estimate. Indicates the phase standard deviation; It is a zero-order modified Bessel function;

[0138] Step S602: Estimating the kernel density function Uniform selection within the range of 0° to 360° There are 12 starting points, for example, one every 30°, and each starting point... The new phase value is iteratively updated along the gradient direction, as shown in the following equation:

[0139] in, For the first Phase value of the next iteration For the first Phase value of the next iteration The learning rate should be a small positive number (e.g., 0.1). A value that is too large will cause oscillations, while a value that is too small will result in slow convergence. Kernel density estimation function At point The gradient (i.e., derivative) at each point, after each iteration. Map back Interval.

[0140] Step S603, when When the value is less than a set threshold (e.g., 1°) or the maximum number of iterations is reached, the iteration stops, and the phase at this point is a candidate point for the peak phase.

[0141] Step S604: Remove duplicates from all candidate peak phase points, treating points within a set tolerance (e.g., within 1°) as the same point, thereby obtaining the accurate peak phase. ;

[0142] Step S605: The peak phase can be further divided according to the local discharge polarity (positive discharge or negative discharge). and ;

[0143] If condition A is met, then this type of partial discharge pulse is determined to be phase A discharge, where condition A is:

[0144]

[0145] If condition B is met, then this type of partial discharge pulse is determined to be phase B discharge, and the phase of the partial discharge pulse is corrected by adding 120° to the specified value. The interval is defined as follows:

[0146]

[0147] If condition C is met, then this type of partial discharge pulse is determined to be a phase C discharge, and the phase of the partial discharge pulse is corrected by adding 240° to the specified value. The interval is defined by condition C:

[0148]

[0149] in, This indicates the precise peak phase corresponding to a positive discharge; This indicates the precise peak phase corresponding to the negative discharge.

[0150] In step S700, the early warning unit compares and analyzes the generated phase distribution maps (PRPD maps) of various types with clearly defined discharge phases with the typical discharge mode maps of different insulation defects pre-stored in the storage unit. The typical mode maps are constructed based on international standards (such as IEC 60034) and historical fault case databases, covering the phase-amplitude distribution characteristics of typical faults. During the comparison process, the similarity between the generated map and the typical mode in key features such as phase concentration interval, amplitude distribution shape, and symmetry is calculated, and a judgment is made based on a preset matching threshold. If the PRPD map of a certain type of discharge successfully matches a typical fault mode, the device determines that the current generator has the corresponding insulation defect type and automatically triggers an early warning. The early warning result is published in real time through the data display unit and can be stored in the data storage unit or uploaded to the remote monitoring platform, thereby realizing online diagnosis and early risk warning of the generator insulation status and providing direct basis for operation and maintenance decisions.

[0151] To more intuitively understand the process of comparing the spectrum with typical discharge modes described in step S700, this embodiment illustrates several typical PRPD modes in conjunction with the accompanying drawings.

[0152] Figure 2This is a typical PRPD mode diagram of internal discharge disclosed in an embodiment of the present invention. Wherein, Figure 2 (a) in the figure is a typical phase-analyzed partial discharge (PRPD) pattern of internal cavity discharge; Figure 2 (b) in the diagram is a typical PRPD pattern of internal delamination discharge in insulation; Figure 2 (c) in the diagram is a typical PRPD pattern of delamination discharge between the conductor and the insulation.

[0153] Figure 3 This is a typical PRPD mode diagram of a slot discharge disclosed in an embodiment of the present invention.

[0154] Figure 4 This is a typical PRPD mode diagram of end discharge disclosed in an embodiment of the present invention. Wherein, Figure 4 (a) in the image shows a typical PRPD pattern of end corona discharge; Figure 4 (b) in the diagram is a typical PRPD pattern with the same end discharge; Figure 4 (c) in the diagram shows a typical PRPD pattern of end surface creepage. These three types of end discharges have different characteristics in phase distribution and amplitude statistics due to their different locations and mechanisms, and can be used to accurately distinguish fault types.

[0155] In the early warning judgment process of step S700, the device compares the generated PRPD map with... Figure 2 , Figure 3 , Figure 4 The typical patterns shown are compared for features. By analyzing the similarity between the generated spectrum and these standard patterns in dimensions such as phase distribution concentration, amplitude statistical regularity, and graphic symmetry, the automatic identification and classification of discharge types can be effectively completed, thus providing a visual and quantitative basis for the accurate assessment of insulation status.

[0156] The complete process involved in this method is implemented by a set of collaborative online monitoring devices for generator partial discharge based on the neutral point coupling method. This device includes:

[0157] A capacitive coupling sensor installed at the generator neutral point is used for safe and non-invasive coupling of partial discharge signals.

[0158] The signal synchronization unit is used to obtain the voltage phase from the generator A-phase outlet PT and generate a synchronization square wave; the data acquisition unit is used to acquire data at high speed under the triggering of the synchronization square wave.

[0159] The data preprocessing unit is responsible for reducing white noise and discrete spectrum interference;

[0160] The pulse extraction and processing unit performs kurtosis-based pulse extraction and phase calculation, and removes in-phase interference.

[0161] The pulse classification unit automatically classifies pulses through principal component analysis and unsupervised clustering.

[0162] The discharge phase analysis and spectrum generation unit uses an algorithm based on kernel density estimation and gradient ascent to accurately identify the discharge phase and generate PRPD spectra.

[0163] Storage unit, pre-stores various typical discharge mode diagrams;

[0164] And an early warning unit, responsible for map comparison and early warning issuance.

[0165] The aforementioned units operate under the coordination of the data analysis and device control unit based on the ARM+FPGA architecture, and perform human-computer interaction through the data display unit.

[0166] The application of this device has brought significant technical benefits: First, the sensor is installed at the neutral point, requiring no operation on the high-voltage side and completely unaffecting the safe and stable operation of the generator; second, the pulse extraction and preprocessing methods employed are computationally efficient and highly resistant to interference, effectively addressing complex electromagnetic noise on-site and suitable for high-speed continuous data acquisition in engineering projects; third, the proposed discharge phase identification method based on kernel density estimation and gradient ascent enables discharge phase analysis of a single discharge source, effectively solving the discharge phase location problem caused by the superposition of three-phase signals and interphase cross-coupling; fourth, the entire device is highly integrated, operates stably and reliably, and can automatically complete the entire process from signal acquisition, processing, analysis to early warning, realizing online intelligent monitoring and early warning of partial discharge, providing strong technical support for condition-based maintenance and insulation fault prevention of generators.

[0167] According to another aspect of the embodiments of this application, an electronic device is also provided, including a processor and a memory, wherein the processor is configured to implement the steps of the method when executing a computer program stored in the memory.

[0168] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0169] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0170] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0171] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0172] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for online monitoring of partial discharge in generators based on neutral point coupling, characterized in that, Includes the following steps: A capacitive coupling sensor is installed at the neutral point of the generator. The data acquisition unit is triggered by the rising edge of the square wave generated by the signal synchronization unit to collect partial discharge data. The partial discharge data is preprocessed to obtain a preprocessed signal; Pulse signals are extracted from the preprocessed signal, and the phase of each pulse signal is calculated; Based on the phase-eliminating interference pulse, a partial discharge pulse is obtained; The partial discharge pulses are classified. After determining the discharge phase of each type of partial discharge pulse, the corresponding phase distribution spectrum is generated. The phase distribution spectrum is compared with the pre-stored typical discharge mode spectrum, and an early warning is issued based on the comparison results; Methods for determining the discharge phase of each type of partial discharge pulse after classification and generating the corresponding phase distribution spectrum include: For a set of partial discharge pulses to be processed, the kernel density estimation function of its phase in the range of 0° to 360° is calculated using the von Mises distribution kernel function based on the phase data of each pulse. Multiple initial phase points are selected within the range of 0° to 360°. Each initial phase point is iteratively updated along the gradient direction of the kernel density estimation function until the phase change is less than a set threshold or the maximum number of iterations is reached, thereby obtaining at least one peak phase candidate point. The candidate peak phase points are deduplicated to obtain the accurate peak phase; Based on the precise peak phase and the preset phase determination conditions, the discharge phase of this type of partial discharge pulse is determined; Based on the determined discharge phase, the phases of various partial discharge pulses are corrected and aligned, and their phase distribution within the power frequency cycle is statistically analyzed to generate a phase distribution map.

2. The online monitoring method for partial discharge of generators based on neutral point coupling as described in claim 1, characterized in that, The method for preprocessing the partial discharge data to obtain a preprocessed signal includes: A high-pass filter is used to filter the partial discharge data to reduce white noise; Data within one power frequency cycle is extracted from the partial discharge data after white noise reduction; The data is windowed using a Blackman-trapezoidal convolution window of length N to obtain a windowed signal. The Blackman-trapezoidal convolution window is obtained by convolving a Blackman window and a trapezoidal window, both of length N / 2. Perform a Fast Fourier Transform on the windowed signal to obtain its spectrum, calculate the amplitude of the spectrum, and calculate a threshold based on the spectrum amplitude. The amplitude values ​​are iterated over. For a spectrum with an amplitude value greater than or equal to the threshold, the average amplitude value of the first five spectra is taken as the corrected amplitude value of the spectrum. For a spectrum with an amplitude value less than the threshold, its original amplitude value is retained. The reconstructed partial discharge data are obtained by performing an inverse Fourier transform on the corrected spectrum. By iterating through all partial discharge data after white noise reduction, the discrete spectrum interference is reduced.

3. The online monitoring method for partial discharge of generators based on neutral point coupling as described in claim 1, characterized in that, The method for extracting pulse signals from the preprocessed signal includes: Calculate the kurtosis of the preprocessed signal; Extract signal data with a kurtosis value greater than 3 as pulse signals; The formula for calculating the kurtosis is as follows: in, For raucousness; This represents the total number of data points within a signal segment involved in the calculation. Indicates the first... The value of each data point; This represents the average value of the preprocessed signal; The standard deviation of the preprocessed signal is represented by .

4. The online monitoring method for partial discharge of generators based on neutral point coupling as described in claim 1, characterized in that, Methods for calculating the phase of each pulse signal include: The power frequency period corresponding to the pulse signal, from 0° to 360°, is divided into 360 phase windows, each with a width of 1°. For the Each pulse signal is used to determine the location of the data point where the absolute value of its amplitude is the maximum. Based on the location of the data points, the phase of the pulse signal is calculated using the following formula: in, Let P be the phase of the pulse signal, and P be the location of the data point where the absolute value of the amplitude is maximum. The sampling frequency of the data acquisition unit. The power frequency is indicated by "round".

5. The online monitoring method for partial discharge of generators based on neutral point coupling as described in claim 1, characterized in that, The method for eliminating interference pulses based on the phase includes: traversing the phases within the range of 0° to 60° at 1° intervals. If in phase , , , , , If pulses are present in both phases, the Pearson correlation coefficient is calculated based on the pulse waveform characteristics of the aforementioned phases. If the Pearson correlation coefficient between any two pulses is greater than 0.6, they are considered interference pulses and are discarded. The value ranges from 0 to 60; wherein the pulse waveform feature is a time-domain waveform or a frequency-domain feature.

6. The online monitoring method for partial discharge of generators based on neutral point coupling as described in claim 1, characterized in that, The method for classifying the partial discharge pulses includes: Principal component analysis was performed to reduce the dimensionality of the characteristic spectrum of the partial discharge pulse. Unsupervised clustering algorithms are applied to cluster the dimensionality-reduced feature data to complete the classification. The unsupervised clustering algorithms include, but are not limited to, density-based noise-based spatial clustering algorithms or density peak clustering algorithms.

7. The online monitoring method for partial discharge of generators based on neutral point coupling as described in claim 1, characterized in that, The formula for calculating the kernel density estimation function is as follows: in, Here is the kernel density estimation function; This represents the total number of pulses of this type. The kernel function is the von Mises distribution kernel function. The phase point to be estimated; For the first The phase of each pulse; The von Mises distribution kernel function The calculation formula is: in, For concentration parameters; It is a zero-order modified Bessel function; The formula for calculating the gradient direction iterative update is as follows: in, For the first Phase value of the next iteration For the first Phase value of the next iteration For learning rate, Kernel density estimation function At point The gradient at that point.

8. The online monitoring method for partial discharge of a generator based on neutral point coupling as described in claim 7, characterized in that, The phase determination conditions include: If condition A is met, then this type of partial discharge pulse is determined to be phase A discharge, where condition A is: If condition B is met, then this type of partial discharge pulse is determined to be a phase B discharge, where condition B is: If condition C is met, then this type of partial discharge pulse is determined to be a phase C discharge, where condition C is: in, This indicates the precise peak phase corresponding to a positive discharge; This indicates the precise peak phase corresponding to a negative discharge.

9. A generator partial discharge online monitoring device based on neutral point coupling method, characterized in that, include: A capacitive coupling sensor is installed at the neutral point of the generator to couple partial discharge signals. The signal synchronization unit is used to acquire the phase of the generator A-phase outlet voltage and generate a square wave synchronization signal. The rising edge of the square wave signal is synchronized with the zero phase point of the power frequency voltage. The data acquisition unit, connected to the capacitively coupled sensor and the signal synchronization unit, is used to acquire partial discharge data triggered by the rising edge of the square wave synchronization signal. A data preprocessing unit, connected to the data acquisition unit, is used to preprocess the acquired partial discharge data to reduce white noise and discrete spectrum interference and obtain a preprocessed signal. A pulse extraction and processing unit, connected to the data preprocessing unit, is used to extract pulse signals from the preprocessed signal, calculate the phase of each pulse signal, and remove interference pulses based on the phase to obtain partial discharge pulses; A pulse classification unit, connected to the pulse extraction and processing unit, is used to classify the partial discharge pulses; The discharge phase analysis and spectrum generation unit is connected to the pulse classification unit and is used to determine the discharge phase of each type of partial discharge pulse after classification and generate the corresponding phase distribution spectrum. Storage unit, used to store pre-stored typical discharge mode maps; The early warning unit, connected to the discharge phase analysis and spectrum generation unit and the storage unit, is used to compare the generated phase distribution spectrum with the pre-stored typical discharge mode spectrum and issue an early warning based on the comparison result. Methods for determining the discharge phase of each type of partial discharge pulse after classification and generating the corresponding phase distribution spectrum include: For a set of partial discharge pulses to be processed, the kernel density estimation function of its phase in the range of 0° to 360° is calculated using the von Mises distribution kernel function based on the phase data of each pulse. Multiple initial phase points are selected within the range of 0° to 360°. Each initial phase point is iteratively updated along the gradient direction of the kernel density estimation function until the phase change is less than a set threshold or the maximum number of iterations is reached, thereby obtaining at least one peak phase candidate point. The candidate peak phase points are deduplicated to obtain the accurate peak phase; Based on the precise peak phase and the preset phase determination conditions, the discharge phase of this type of partial discharge pulse is determined; Based on the determined discharge phase, the phases of various partial discharge pulses are corrected and aligned, and their phase distribution within the power frequency cycle is statistically analyzed to generate a phase distribution map.

Images