Automatic pattern recognition method and system for partial discharge defects of generator stator insulation

Abstract

The present application relates to the technical field of power equipment fault diagnosis, and particularly relates to a method and system for automatically identifying partial discharge defects of a generator stator insulation; the method comprises the following steps: dividing the generator stator core into a plurality of monitoring areas according to the axial direction and the circumferential direction, synchronously collecting multiple types of signals in each area, and obtaining an original data set with area identification; preprocessing the original data set, extracting four types of features in the time domain, the frequency domain, the phase and the space, and obtaining a feature set; calculating the discharge probability of each monitoring area according to the feature set, determining the defect type according to the probability threshold, and tracing the discharge area; the system comprises the following modules: a partial discharge signal collection module, a multi-dimensional feature set construction module, and a partial discharge defect type identification and area tracing module; through the above-mentioned mode, the effect of cooperatively realizing defect type identification and area tracing is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power equipment fault diagnosis, and particularly relates to an automatic pattern recognition method and system for generator stator insulation partial discharge defects. BACKGROUND

[0002] As the core equipment of the power system, the stator insulation performance of the generator directly determines the operation life and safety stability of the equipment. In the long-term operation process, under the combined action of electricity, heat, mechanical stress and environmental factors, the stator insulation is prone to aging, damage and other problems, which causes partial discharge phenomenon. The type and characteristics of the partial discharge directly reflect the nature and severity of the insulation defects.

[0003] There are two core problems in the traditional diagnosis method: one is that the signal collection is mostly overall monitoring, without regional layout, and the operation and maintenance need to be checked in a large range, which is high in cost and low in efficiency; the other is that the feature extraction is limited to a single dimension of time domain, frequency domain or phase, ignoring the signal space correlation, and it is difficult to distinguish similar discharge types in different regions, and the recognition accuracy is affected.

[0004] In view of the above, it is very necessary to provide an automatic pattern recognition method and system for generator stator insulation partial discharge defects, which cooperatively realizes defect type recognition and region tracing. SUMMARY

[0005] The purpose of the present application is to provide an automatic pattern recognition method and system for generator stator insulation partial discharge defects, which cooperatively realizes defect type recognition and region tracing.

[0006] To achieve the above purpose, an automatic pattern recognition method for generator stator insulation partial discharge defects is adopted, which comprises the following steps:

[0007] The generator stator core is divided into multiple monitoring regions according to the axial direction and the circumferential direction, and multiple types of signals in each region are synchronously collected to obtain an original data set with region identification;

[0008] The original data set is preprocessed, and four types of features of time domain, frequency domain, phase and space are extracted to obtain a feature set;

[0009] The discharge probability of each monitoring region is calculated according to the feature set, the defect type is determined according to the probability threshold, and the discharge region is traced.

[0010] In the step of dividing the generator stator core into multiple monitoring regions according to the axial direction and the circumferential direction, synchronously collecting multiple types of signals in each region, and obtaining an original data set with region identification:

[0011] According to the structure parameters of the generator stator core, the monitoring region division rule is determined;

[0012] three types of sensors are arranged in each divided monitoring area;

[0013] An electric grid voltage phase signal is acquired, the three types of sensors in the monitoring area are connected to a multi-channel data acquisition card, and a sampling request is triggered to start synchronously in each channel;

[0014] A corresponding area identification code, a collection time stamp, and a sensor type label are added to each collected signal data frame, the signal data with labels are integrated and stored to form an original data set with area identification.

[0015] In the step of determining the monitoring area division rule according to the structural parameters of the generator stator core:

[0016] The stator core is divided into several axial areas along the axial direction at intervals, and divided into several circumferential areas along the circumferential direction according to the number of magnetic pole pairs, and each monitoring area is assigned a unique spatial identification code.

[0017] In the step of arranging three types of sensors in each divided monitoring area:

[0018] The high-frequency current sensor is connected in series at the stator winding ground line of the corresponding area, for collecting the discharge current signal;

[0019] The ultrahigh-frequency sensor is fixed at the core slot position through an insulating support, for collecting the ultrahigh-frequency electromagnetic radiation signal;

[0020] The ultrasonic sensor is attached to the outer wall of the core through a coupling agent, for collecting the ultrasonic vibration signal.

[0021] In the step of preprocessing the original data set and extracting four types of features in time domain, frequency domain, phase, and space to obtain a feature set:

[0022] The signals in the original data set are denoised, the abnormal values in the denoised signal data are removed, and the data is converted into standardized data;

[0023] Four types of features in time domain, frequency domain, phase, and space are extracted from the standardized data.

[0024] In the step of extracting four types of features in time domain, frequency domain, phase, and space from the standardized data:

[0025] In terms of time domain features, the discharge pulse is identified and the pulse peak value, rise time, fall time, pulse width, and signal amplitude variance are calculated;

[0026] In terms of frequency domain features, the signal is subjected to Fourier transform, and the signal main frequency, energy proportion of different frequency bands, spectral entropy, and harmonic component amplitude are extracted;

[0027] In the aspect of phase characteristics, a correlation graph is drawn in combination with a phase synchronization signal, and a phase distribution histogram, a phase barycenter, a phase dispersion degree and a phase correlation coefficient are calculated.

[0028] In the aspect of spatial characteristics, a signal amplitude difference, a signal arrival time difference and a signal correlation coefficient of the same type of sensors in adjacent regions are calculated.

[0029] In the aspect of spatial characteristics, a signal amplitude difference, a signal arrival time difference and a signal correlation coefficient of the same type of sensors in adjacent regions are calculated.

[0030] The extracted features are processed, principal components are retained, and a local topological structure of optimized features is embedded to form a feature set.

[0031] In the aspect of spatial characteristics, a signal amplitude difference, a signal arrival time difference and a signal correlation coefficient of the same type of sensors in adjacent regions are calculated.

[0032] According to the feature set, probability values of four types of defect types, i.e., corona discharge, surface discharge, internal discharge and suspended potential discharge, are calculated to determine the defect type;

[0033] According to the feature set, probability values of four types of defect types, i.e., corona discharge, surface discharge, internal discharge and suspended potential discharge, are calculated to determine the defect type;

[0034] In the aspect of spatial characteristics, a signal amplitude difference, a signal arrival time difference and a signal correlation coefficient of the same type of sensors in adjacent regions are calculated.

[0035] A first discharge threshold and a second discharge threshold are set, and if the discharge probability of a single region is greater than or equal to the first discharge threshold, the region is determined as a discharge region;

[0036] If the discharge probabilities of multiple regions are between the first threshold and the second threshold, theoretical signal attenuation values of the regions are calculated and compared with actual measurement values, and the region with the minimum error is determined as a discharge region.

[0037] The application also provides an automatic pattern recognition system for partial discharge defects of a generator stator insulation, which comprises a partial discharge signal acquisition module, a multi-dimensional feature set construction module and a partial discharge defect type recognition and region tracing module.

[0038] The partial discharge signal acquisition module is used for dividing a generator stator core into multiple monitoring regions in an axial direction and a circumferential direction, synchronously acquiring multiple types of signals in each region and obtaining an original data set with region identification.

[0039] The multi-dimensional feature set construction module is used for pre-processing the original data set and extracting four types of features, i.e., time domain, frequency domain, phase and space, to obtain a feature set.

[0040] The partial discharge defect type identification and area tracing module is used for calculating discharge probability of each monitoring area according to the feature set, determining the defect type according to a probability threshold, and tracing the discharge area.

[0041] The automatic pattern recognition method and system of the generator stator insulation partial discharge defect of the application adopts the partial discharge signal acquisition module, the multi-dimensional feature set construction module and the partial discharge defect type identification and area tracing module to perform the following steps: the generator stator core is divided into a plurality of monitoring areas according to the axial and circumferential directions, a plurality of types of signals in each area are synchronously acquired to obtain an original data set with area identification; the original data set is preprocessed, and four types of features in time domain, frequency domain, phase and space are extracted to obtain a feature set; discharge probability of each monitoring area is calculated according to the feature set, the defect type is determined according to a probability threshold, and the discharge area is traced; through the above-mentioned manner, the effect of cooperatively realizing defect type identification and area tracing is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0043] Figure 1 is a step flow chart of the automatic pattern recognition method of the generator stator insulation partial discharge defect of the application.

[0044] Figure 2 is a step flow chart of S100 of the application.

[0045] Figure 3 is a step flow chart of S200 of the application.

[0046] Figure 4 is a step flow chart of S300 of the application.

[0047] Figure 5 is a structure principle diagram of the automatic pattern recognition system of the generator stator insulation partial discharge defect of the application.

[0048] Figure 6 is a structure principle diagram of the electronic device of the application.

[0049] 401-partial discharge signal acquisition module, 402-multi-dimensional feature set construction module, 403-partial discharge defect type identification and area tracing module. DETAILED DESCRIPTION

[0050] The exemplary embodiments will be described in detail herein with reference to several drawings. Descriptions of well-known functions and constructions can be omitted to help emphasize aspects of the exemplary embodiments. The following description is presented to enable any person skilled in the art to make and use the application. Descriptions of specific embodiments are included for purposes of disclosure, but are not intended to limit the scope of the application. Details are included for the purpose of disclosure.

[0051] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0052] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is to be understood that the term "and / or" as used herein encompasses all possible combinations of particular items listed apart from disjunctively worded limitations of various claims, so that if a particular claim is divided into disjunctively worded subclaims, references in that particular claim to "and / or" are taken to mean that the particular items so referred to can be taken individually or in any combination of one or more of the particular items.

[0053] Please refer to Figures 1-4 The application provides an automatic pattern recognition method for partial discharge defects of a generator stator insulation, comprising the following steps:

[0054] S100: dividing the generator stator core into a plurality of monitoring areas in the axial and circumferential directions, synchronously collecting multi-type signals of each area, and obtaining an original data set with area identification.

[0055] In the embodiment, the generator stator core is divided into a plurality of monitoring areas in the axial and circumferential directions, multi-type signals of each area are synchronously collected, and an original data set with area identification is obtained. The specific process is as follows:

[0056] S101: according to the structural parameters of the generator stator core, dividing the core into a plurality of axial areas in the axial direction according to the interval, dividing the core into a plurality of circumferential areas in the circumferential direction according to the number of magnetic pole pairs, and assigning a unique spatial identification code to each monitoring area;

[0057] S102: arranging three types of sensors in each divided monitoring area; connecting a high-frequency current sensor in series at the grounding line of the stator winding of the corresponding area to collect a discharge current signal; fixing an ultrahigh-frequency sensor at the core slot position through an insulating support to collect an ultrahigh-frequency electromagnetic radiation signal; and adhering an ultrasonic sensor to the outer wall of the core through a coupling agent to collect an ultrasonic vibration signal;

[0058] S103: Obtain a power grid voltage phase signal, connect the three types of sensors in the monitoring area to a multi-channel data acquisition card, and trigger a synchronous start sampling request for each channel;

[0059] S104: Add a corresponding area identification code, a collection time stamp, and a sensor type label to each collected signal data frame, integrate the labeled signal data, and store it to form an original data set with area identification.

[0060] In the above process, according to the actual structural parameters of the generator stator core, including the core length, inner diameter, and the number of magnetic pole pairs, the monitoring area division rule is determined, and the core is divided into several axial areas at an interval of 200-300 mm along the axial direction, and divided into several circumferential areas according to the number of magnetic pole pairs along the circumferential direction, ensuring that the area of each monitoring area does not exceed 0.5 square decimeters (i.e., 50 square millimeters), and a unique spatial identification code is assigned to each area. In each divided monitoring area, three types of sensors are arranged, the high-frequency current sensor is connected in series at the stator winding ground line of the corresponding area, which is used to collect the discharge current signal in the frequency band of 1-100 MHz, which covers the main current signal frequency of the generator stator insulation partial discharge (including corona discharge, surface discharge, internal discharge, and suspended potential discharge), the ultrahigh frequency sensor is fixed on the core slot position through an insulating support, which is used to collect the ultrahigh frequency electromagnetic radiation signal, and the ultrasonic sensor is attached to the outer wall of the core through a coupling agent, which is used to collect the ultrasonic vibration signal, and the distance between the ultrahigh frequency sensor and the ultrasonic sensor in the same area is not more than 50 mm.

[0061] Connect the phase synchronization module to the power grid to obtain the power grid voltage phase signal, connect the three types of sensors in each area to the multi-channel data acquisition card, set the sampling frequency of the data acquisition card to not less than 100 MHz and the sampling bit number to not less than 16 bits, and trigger a synchronous start sampling for each channel through the phase synchronization signal; in the signal collection process, add a corresponding area identification code, a collection time stamp, and a sensor type label to each collected signal data frame, integrate and store all the labeled signal data to form an original data set with area identification.

[0062] S200: Preprocess the original data set and extract four types of features in time domain, frequency domain, phase, and space to obtain a feature set.

[0063] In this embodiment, the original data set is preprocessed, and four types of features in time domain, frequency domain, phase, and space are extracted to obtain a feature set. The specific process is as follows:

[0064] S201: Denoise the signals in the original data set, remove the outliers in the denoised signal data, and convert the data to standardized data;

[0065] S202: Extract four types of features from standardized data: time domain, frequency domain, phase, and spatial features. In the time domain, identify discharge pulses and calculate pulse peak value, rise time, fall time, pulse width, and signal amplitude variance. In the frequency domain, perform Fourier transform on the signal to extract the signal's dominant frequency, energy proportion of different frequency bands, spectral entropy, and harmonic component amplitude. In the phase domain, combine phase synchronization signals to draw relevant spectra and calculate the phase distribution histogram, phase centroid, phase dispersion, and phase correlation coefficient. In the spatial domain, calculate the signal amplitude difference, signal arrival time difference, and signal correlation coefficient between adjacent sensors of the same type.

[0066] S203: Process the extracted features, retain the principal components and embed the local topological structure of the optimized features to form a feature set.

[0067] In the above process, wavelet threshold denoising method is used to denoise the signal in the original dataset. An improved db6 wavelet basis is selected to perform multi-level decomposition of the signal. The threshold coefficient is dynamically adjusted according to the signal-to-noise ratio. When the signal-to-noise ratio is less than 20dB, the threshold coefficient is set between 0.8 and 1.0. When the signal-to-noise ratio is not less than 20dB, the threshold coefficient is set between 0.5 and 0.7. The denoised signal is obtained by threshold processing and inverse wavelet transform reconstruction.

[0068] Outlier removal is performed on the denoised signal data. First, outliers that are significantly outside the normal distribution range of the data are screened and removed based on the 3σ criterion. Then, the local outlier factor of the remaining data is calculated. Data with a local outlier factor greater than 1.8 are identified as potential outliers and removed. Subsequently, the Z-score standardization method is used to convert the processed data into standardized data with a mean of 0 and a standard deviation of 1.

[0069] Four types of features are extracted from standardized data. In terms of time domain features, discharge pulses are identified and pulse peak value, rise time, fall time, pulse width, and signal amplitude variance are calculated. In terms of frequency domain features, Fourier transform is performed on the signal to extract the signal's dominant frequency, energy proportion of different frequency bands, spectral entropy, and harmonic component amplitude. In terms of phase features, a correlation spectrum is plotted in conjunction with the phase synchronization signal, and the phase distribution histogram, phase centroid, phase dispersion, and phase correlation coefficient are calculated. In terms of spatial features, the signal amplitude difference, signal arrival time difference, and signal correlation coefficient of similar sensors in adjacent areas are calculated.

[0070] A hybrid dimensionality reduction algorithm combining principal component analysis and local linear embedding is used to process the extracted high-dimensional features. First, principal component analysis is used to retain principal components with a cumulative contribution rate of not less than 95%. Then, local linear embedding is used to optimize the local topology of the features to reduce the loss of feature information, and finally form a feature set with appropriate dimensionality.

[0071] S300: Calculate the discharge probability of each monitoring area based on the feature set, determine the defect type based on the probability threshold, and trace the discharge area.

[0072] In this embodiment, the discharge probability of each monitoring area is calculated based on the feature set, the defect type is determined based on the probability threshold, and the discharge area is traced. The specific process is as follows:

[0073] S301: Calculate the probability values ​​of four defect types—corona discharge, surface discharge, internal discharge, and floating potential discharge—based on the feature set to determine the defect type;

[0074] S302: Obtain the discharge probability value of each monitoring area and trace the discharge area.

[0075] Furthermore, in the steps of obtaining the discharge probability value of each monitoring area and tracing the discharge area:

[0076] A first discharge threshold and a second discharge threshold are set. If the discharge probability of a single region is greater than or equal to the first discharge threshold, then the region is determined to be a discharge region.

[0077] If the discharge probability of multiple regions is between the first threshold and the second threshold, then the theoretical signal attenuation value of each region is calculated and compared with the actual measured value. The region with the smallest error is determined as the discharge region.

[0078] In the above process, a CNN-BiLSTM-Transformer hybrid model based on the attention mechanism is constructed. The model includes a CNN layer, a BiLSTM layer, a Transformer layer and a dual-branch output layer. The CNN layer is used to extract the local correlation of features, the BiLSTM layer is used to capture the temporal dependency of features, the Transformer layer is used to enhance the spatial interaction of features, and the dual-branch output layer is used to output the probability of defect type and the probability of regional discharge, respectively.

[0079] The feature set was divided into training and testing sets in a 7:3 ratio. Data augmentation was performed on the training set, including adding small Gaussian noise and stretching / compressing the time axis. Then, the model training parameters were set, with the initial learning rate set between 0.001 and 0.005, the batch size set between 32 and 64, and the number of training epochs set between 60 and 120. The model was trained using the cross-entropy loss function and the AdamW optimization algorithm. An early stopping strategy was enabled during training; training was stopped and the optimal model parameters were saved when the test set loss did not decrease for 8 consecutive epochs.

[0080] Input the feature set corresponding to the signal to be identified into the trained model. The defect type output branch of the model will output the probability values ​​of four defect types: corona discharge, surface discharge, internal discharge, and floating potential discharge. Select the defect type with the highest probability and not less than 0.75 as the final determined defect type. If the maximum probability is less than 0.75, a secondary detection is triggered.

[0081] The model outputs the discharge probability value of each monitoring area through the regional discharge probability output branch. A first discharge threshold and a second discharge threshold are set. For example, if the first discharge threshold is 0.8 and the second discharge threshold is 0.6, if the discharge probability of a single area is not less than 0.8, then the area is directly determined to be a discharge area. If the discharge probability of multiple areas is between 0.6 and 0.8, then the theoretical signal attenuation value of each area is calculated by combining the signal attenuation model based on the stator core material and structure and compared with the actual measurement value. The area with the smallest error is determined to be a discharge area.

[0082] Corresponding to the aforementioned embodiments of the automatic pattern recognition method for partial discharge defects in generator stator insulation, this application also provides embodiments of an automatic pattern recognition system for partial discharge defects in generator stator insulation.

[0083] Figure 5 This is a block diagram illustrating an automatic pattern recognition system for partial discharge defects in generator stator insulation according to an exemplary embodiment. (Refer to...) Figure 5 The system may include: a partial discharge signal acquisition module 401, a multi-dimensional feature set construction module 402, and a partial discharge defect type identification and area tracing module 403; wherein:

[0084] The partial discharge signal acquisition module 401 is used to divide the generator stator core into multiple monitoring areas according to the axial and circumferential directions, and simultaneously acquire multiple types of signals in each area to obtain a raw dataset with area identification.

[0085] The multi-dimensional feature set construction module 402 is used to preprocess the original dataset and extract four types of features: time domain, frequency domain, phase, and space, to obtain a feature set.

[0086] The partial discharge defect type identification and area tracing module 403 is used to calculate the discharge probability of each monitoring area based on the feature set, determine the defect type based on the probability threshold, and trace the discharge area.

[0087] In this embodiment, the partial discharge signal acquisition module 401 divides the generator stator core into multiple monitoring areas along the axial and circumferential directions, and synchronously acquires multiple types of signals from each area to obtain an original dataset with area identifiers. The multi-dimensional feature set construction module 402 preprocesses the original dataset and extracts four types of features: time domain, frequency domain, phase, and space, to obtain a feature set. The partial discharge defect type identification and area tracing module 403 calculates the discharge probability of each monitoring area based on the feature set, determines the defect type based on the probability threshold, and traces the discharge area. Through the above methods, the effect of collaboratively identifying defect types and tracing areas is achieved.

[0088] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0089] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0090] Accordingly, this application also provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; and, when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the automatic pattern recognition method for partial discharge defects in generator stator insulation as described above. Figure 6 The diagram shown is a hardware structure diagram of any device with data processing capabilities, used in an automatic pattern recognition system for partial discharge defects in generator stator insulation according to an embodiment of the present invention. (Except for...) Figure 6 In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0091] Accordingly, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the automatic pattern recognition method for partial discharge defects in generator stator insulation as described above. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0092] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0093] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. An automatic pattern recognition method for partial discharge defects in generator stator insulation, characterized in that, Includes the following steps: The generator stator core is divided into multiple monitoring areas according to the axial and circumferential directions. Multiple types of signals from each area are collected simultaneously to obtain the original dataset with area identification. The original dataset is preprocessed, and four types of features—time domain, frequency domain, phase, and spatial—are extracted to obtain a feature set. The discharge probability of each monitoring area is calculated based on the feature set, the defect type is determined based on the probability threshold, and the discharge area is traced. Specifically: Based on the feature set, the probability values ​​of four defect types—corona discharge, surface discharge, internal discharge, and floating potential discharge—are calculated to determine the defect type. Obtain the discharge probability value of each monitoring area and trace the discharge area; set a first discharge threshold and a second discharge threshold. If the discharge probability of a single area is greater than or equal to the first discharge threshold, then the area is determined to be a discharge area; if the discharge probability of multiple areas is between the first threshold and the second threshold, then calculate the theoretical signal attenuation value of each area and compare it with the actual measurement value, and determine the area with the smallest error as the discharge area. In the steps of dividing the generator stator core into multiple monitoring areas according to the axial and circumferential directions, synchronously collecting multiple types of signals from each area, and obtaining a raw dataset with area identification: Based on the structural parameters of the generator stator core, the monitoring area division rules are determined; the core is divided into several axial areas according to the spacing along the axial direction, and into several circumferential areas according to the average number of magnetic pole pairs along the circumference, and each monitoring area is assigned a unique spatial identification code. Three types of sensors are deployed in each designated monitoring area: a high-frequency current sensor is connected in series with the stator winding grounding wire of the corresponding area to collect discharge current signals; an ultra-high frequency sensor is fixed to the slot of the iron core with an insulating bracket to collect ultra-high frequency electromagnetic radiation signals; and an ultrasonic sensor is attached to the outer wall of the iron core with a coupling agent to collect ultrasonic vibration signals. Acquire the grid voltage phase signal, connect the three types of sensors in the monitoring area to the multi-channel data acquisition card, and trigger the synchronous start sampling request of each channel; Each acquired signal data frame is assigned a corresponding area identifier, acquisition timestamp, and sensor type label. The tagged signal data is then integrated and stored to form a raw dataset with area identifiers.

2. The automatic pattern recognition method for partial discharge defects in generator stator insulation as described in claim 1, characterized in that, In the steps of preprocessing the original dataset and extracting four types of features—time domain, frequency domain, phase, and spatial—to obtain the feature set: The signals in the original dataset are denoised to remove outliers and the data is then converted into standardized data. Four types of features are extracted from standardized data: time domain, frequency domain, phase, and spatial features.

3. The automatic pattern recognition method for partial discharge defects in generator stator insulation as described in claim 2, characterized in that, In the steps of extracting four types of features—time domain, frequency domain, phase, and spatial—from standardized data: In terms of time domain characteristics, the discharge pulse is identified and the pulse peak value, rise time, fall time, pulse width, and signal amplitude variance are calculated. In terms of frequency domain characteristics, Fourier transform is performed on the signal to extract the signal's main frequency, energy proportion of different frequency bands, spectral entropy, and harmonic component amplitude. Regarding phase characteristics, relevant spectra are plotted in conjunction with phase synchronization signals, and phase distribution histograms, phase centroids, phase dispersion, and phase correlation coefficients are calculated. In terms of spatial characteristics, the signal amplitude difference, signal arrival time difference, and signal correlation coefficient of similar sensors in adjacent areas are calculated.

4. The automatic pattern recognition method for partial discharge defects in generator stator insulation as described in claim 3, characterized in that, After the steps of extracting four types of features—time domain, frequency domain, phase, and spatial—from standardized data: The extracted features are processed to retain principal components and embed the local topological structure of optimized features to form a feature set.

5. An automatic pattern recognition system for partial discharge defects in generator stator insulation, applied to the automatic pattern recognition method for partial discharge defects in generator stator insulation as described in claim 1, characterized in that, It includes a partial discharge signal acquisition module, a multi-dimensional feature set construction module, and a partial discharge defect type identification and area tracing module; among which: The partial discharge signal acquisition module is used to divide the generator stator core into multiple monitoring areas according to the axial and circumferential directions, and simultaneously acquire multiple types of signals in each area to obtain a raw dataset with area identification. The multi-dimensional feature set construction module is used to preprocess the original dataset and extract four types of features: time domain, frequency domain, phase, and space, to obtain a feature set. The partial discharge defect type identification and area tracing module is used to calculate the discharge probability of each monitoring area based on the feature set, determine the defect type based on the probability threshold, and trace the discharge area.

Images