Fault diagnosis method, device and equipment for partial discharge of generator stator winding, readable storage medium and program product

Abstract

The invention relates to a fault diagnosis method, device and equipment for partial discharge of a generator stator winding, a readable storage medium and a program product. The method comprises the following steps: acquiring a partial discharge signal of a stator winding; respectively extracting time domain features and frequency domain features of the partial discharge signals to obtain the time domain features and the frequency domain features; inputting the time domain feature into an independent full connection layer to obtain a time domain feature vector; inputting the frequency domain features into the independent full connection layer to obtain frequency domain feature vectors; based on a preset attention mechanism, calculating a first contribution degree of the time domain feature vector to the classification result and a second contribution degree of the frequency domain feature vector to the classification result; performing weighted fusion based on the first contribution degree and the second contribution degree to obtain weighted fusion features; determining a fault type and a fault probability corresponding to the partial discharge signal according to the weighted fusion feature and based on a bidirectional long-short-term memory network; therefore, the comprehensiveness and accuracy of fault monitoring are improved.

Description

Technical Field

[0001] The present application relates to the technical field of partial discharge of generator stator windings, and in particular to a fault diagnosis method, apparatus, computer equipment, computer-readable storage medium, and computer program product for partial discharge of generator stator windings. Background Art

[0002] With the continuous development of power systems, the safe and stable operation of generators is crucial to ensuring power supply. The insulation condition of a generator's stator windings is one of the key factors affecting its reliability, and partial discharge is a key indicator for evaluating the insulation condition of stator windings. Traditional partial discharge detection methods rely on a single feature analysis, making it difficult to fully and accurately reflect the complex characteristics of discharge signals. Summary of the Invention

[0003] Based on this, it is necessary to provide a fault diagnosis method, device, computer equipment, computer-readable storage medium and computer program product for local discharge of generator stator windings that can accurately reflect discharge signals to diagnose faults in order to solve the above technical problems.

[0004] In a first aspect, the present application provides a method for diagnosing partial discharge in a generator stator winding, comprising:

[0005] Obtaining partial discharge signals of stator windings;

[0006] Performing time domain feature extraction on the partial discharge signal to obtain time domain features;

[0007] Perform frequency domain feature extraction on the partial discharge signal to obtain frequency domain features;

[0008] Input the time domain features into the independent fully connected layer to obtain the time domain feature vector;

[0009] Input the frequency domain features into the independent fully connected layer to obtain the frequency domain feature vector;

[0010] Based on the preset attention mechanism, the first contribution of the time domain feature vector to the classification result and the second contribution of the frequency domain feature vector to the classification result are calculated;

[0011] Perform weighted fusion based on the first contribution and the second contribution to obtain a weighted fusion feature;

[0012] According to the weighted fusion features and based on the bidirectional long short-term memory network, the fault type and fault probability corresponding to the partial discharge signal are determined.

[0013] In one embodiment, the step of obtaining a partial discharge signal of a stator winding includes:

[0014] Collect current signals based on current sensors;

[0015] Collect coupled discharge signals based on capacitive sensors;

[0016] A partial discharge signal is determined based on the current signal and the coupled discharge signal.

[0017] In one embodiment, the above-mentioned method for diagnosing partial discharge of a generator stator winding further includes:

[0018] Generate a time-frequency spectrum diagram based on time domain features and frequency domain features;

[0019] Inputting the time-frequency spectrum graph into a preset convolutional model to obtain at least one feature graph;

[0020] According to the fault type and each characteristic diagram, determine the characteristics Figure 1 1. The corresponding feature map gradient;

[0021] Perform global average pooling on the gradient of each feature map to obtain the weight of each feature map;

[0022] Based on each feature map and each weight, a weighted sum is performed to obtain the class activation map;

[0023] Based on the class activation map, a visualization graph is determined and displayed; wherein the visualization graph carries information about the fault location.

[0024] In one embodiment, before the step of extracting time domain features from the partial discharge signal to obtain the time domain features, the fault diagnosis method for partial discharge of the generator stator winding further includes:

[0025] The wavelet threshold denoising algorithm is used to filter out the white noise of the partial discharge signal;

[0026] Normalize the denoised partial discharge signal;

[0027] Segmenting the normalized partial discharge signal according to a preset time window length to obtain multiple sample sequences;

[0028] Multiple sample sequences are spliced ​​together to obtain partial discharge signals.

[0029] In one embodiment, the above-mentioned method for diagnosing partial discharge of a generator stator winding further includes:

[0030] Based on the time domain features, frequency domain features and fault types, a feature-fault type mapping table is determined.

[0031] In one embodiment, the above-mentioned method for diagnosing partial discharge of a generator stator winding further includes:

[0032] The severity of the fault is determined based on the fault probability and at least one preset probability threshold.

[0033] In a second aspect, the present application further provides a fault diagnosis device for partial discharge of a generator stator winding, comprising:

[0034] An acquisition module, used for acquiring a partial discharge signal of a stator winding;

[0035] A time domain feature extraction module, configured to extract time domain features from the partial discharge signal to obtain time domain features;

[0036] A frequency domain feature extraction module, configured to extract frequency domain features from the partial discharge signal to obtain frequency domain features;

[0037] A first input module is used to input the time domain features into an independent fully connected layer to obtain a time domain feature vector;

[0038] A second input module is used to input the frequency domain features into the independent fully connected layer to obtain a frequency domain feature vector;

[0039] A contribution determination module, configured to calculate a first contribution of the time-domain feature vector to the classification result and a second contribution of the frequency-domain feature vector to the classification result based on a preset attention mechanism;

[0040] a weighted fusion module, configured to perform weighted fusion based on the first contribution and the second contribution to obtain a weighted fusion feature;

[0041] A fault determination module is used to determine the fault type and fault probability corresponding to the partial discharge signal according to the weighted fusion features and based on a bidirectional long short-term memory network.

[0042] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements any step in the above-mentioned fault diagnosis method for partial discharge of the generator stator winding.

[0043] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step of the above-mentioned method for diagnosing partial discharge of a generator stator winding.

[0044] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements any step in the above-mentioned fault diagnosis method for partial discharge of a generator stator winding.

[0045] The above-mentioned generator stator winding partial discharge fault diagnosis method, apparatus, computer device, computer-readable storage medium, and computer program product obtain a stator winding partial discharge signal and extract time domain features based on the partial discharge signal to obtain time domain features. The obtained time domain features are input into an independent fully connected layer to obtain a time domain feature vector corresponding to the time domain features. Correspondingly, the obtained frequency domain features are input into the above-mentioned independent fully connected layer to obtain a frequency domain feature vector corresponding to the frequency domain features. Based on a preset attention mechanism, a first contribution of the time domain feature vector to the classification result and a second contribution of the frequency domain feature vector to the classification result are calculated. A weighted fusion is then performed based on the obtained first and second contributions to obtain a corresponding weighted fusion feature. Based on the weighted fusion feature and a bidirectional long short-term memory network, the fault type and fault probability corresponding to the partial discharge signal can be determined. The generator stator winding partial discharge fault diagnosis method can comprehensively determine the fault type and fault probability based on the time domain features and frequency domain features of the partial discharge signal, with high comprehensiveness and accuracy. Furthermore, the fault diagnosis method for partial discharge of the generator stator winding can realize automatic monitoring, reduce the input of human resources, and realize intelligent fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments of the present application or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 FIG2 is an application environment diagram of a fault diagnosis method for partial discharge of a generator stator winding according to an embodiment;

[0048] Figure 2 1 is a flow chart of a method for diagnosing partial discharge in a generator stator winding fault in one embodiment;

[0049] Figure 3 1 is a flow chart of a fault location step in a fault diagnosis method for partial discharge of a generator stator winding in one embodiment;

[0050] Figure 4 1 is a structural block diagram of a fault diagnosis device for partial discharge of a generator stator winding according to one embodiment;

[0051] Figure 5 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0053] The fault diagnosis method for partial discharge of the generator stator winding provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store parameters of an independent fully connected layer and parameters of a bidirectional long short-term memory network required by server 104. The data storage system can be integrated with server 104, or placed on a cloud or other network server. Server 104 obtains partial discharge signals from the stator winding and extracts time-domain features based on the partial discharge signals to obtain time-domain features. It also extracts frequency-domain features based on the partial discharge signals to obtain frequency-domain features. The obtained time-domain features are input into an independent fully connected layer to obtain a time-domain feature vector corresponding to the time-domain features. Correspondingly, the obtained frequency-domain features are input into the independent fully connected layer to obtain a frequency-domain feature vector corresponding to the frequency-domain features. Based on a preset attention mechanism, a first contribution of the time-domain feature vector to the classification result and a second contribution of the frequency-domain feature vector to the classification result can be calculated. A weighted fusion is then performed based on the obtained first and second contributions to obtain a corresponding weighted fusion feature. Based on the weighted fusion feature and the bidirectional long short-term memory network, the fault type and fault probability corresponding to the partial discharge signal can be determined. After server 104 transmits the determined fault type and probability to terminal 102, staff can view a visualization of the fault type and probability on terminal 102. Terminal 102 may include, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, smart car devices, and projectors. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. Head-mounted devices may include virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, and the like. Server 104 may be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services.

[0054] In an exemplary embodiment, Figure 2 As shown, a fault diagnosis method for partial discharge of a generator stator winding is provided. Figure 1The server 104 in FIG. 1 is used as an example to illustrate the invention, including:

[0055] S202: Acquire a partial discharge signal of the stator winding.

[0056] The partial discharge signal from the stator winding is a key indicator for assessing the insulation condition of a generator. Its characteristics and detection methods are directly related to the safe operation of the generator. The partial discharge signal from the stator winding appears as a high-frequency pulse with an ultra-narrow pulse width. Typically, the rise time of this high-frequency pulse is less than 1ns, and the frequency range is mainly concentrated in the 40MHz-350MHz range.

[0057] Partial discharge signals originate from air gaps, delaminations, or poor conductor contact in the stator winding insulation. When the operating voltage exceeds the air gap breakdown field strength, the air in the air gap ionizes and generates discharge pulses.

[0058] S204: extracting time domain features from the partial discharge signal to obtain time domain features.

[0059] Time-domain feature extraction focuses on the temporal variations of partial discharge signals, such as waveform amplitude, rise time, and pulse width. Time-domain features naturally suppress periodic narrowband interference (such as power frequency signals). Specifically, in time-domain feature extraction, setting appropriate thresholds and pulse widths for signal screening effectively eliminates noise at non-target frequencies, thereby retaining high-frequency partial discharge signals.

[0060] Among them, time domain features include statistical features, waveform features and fractal dimension; statistical features include mean, variance, kurtosis, etc.; waveform features include rising edge slope and pulse width; fractal dimension includes quantized signal complexity.

[0061] S206 , extracting frequency domain features from the partial discharge signal to obtain frequency domain features.

[0062] Frequency domain feature extraction converts partial discharge signals into the frequency domain and analyzes their frequency components, such as the spectrum and energy distribution of partial discharge signals. Frequency domain analysis is more robust to non-stationary interference (such as random pulse noise) and is suitable for complex electromagnetic environments.

[0063] Among them, the frequency domain features include speech-like features and energy distribution. For example, the speech-like features in the partial discharge signal can be extracted based on Mel-scale Frequency Cepstral Coefficients (MFCC); and the energy distribution of multiple frequency bands can be obtained based on wavelet packet decomposition.

[0064] S208: Input the time domain features into an independent fully connected layer to obtain a time domain feature vector.

[0065] S210, input the frequency domain features into an independent fully connected layer to obtain a frequency domain feature vector.

[0066] An independent fully-connected layer is an independent fully-connected layer. In a fully-connected layer, each node is connected to all nodes in the previous layer, integrating the extracted features (time domain features or frequency domain features). An independent fully-connected layer exists relatively independently within a neural network and does not share neural network parameters (such as weights and biases) with other layers. Through an independent fully-connected layer, the input time domain features are mapped to a new feature space, resulting in a time domain feature vector. Similarly, through an independent fully-connected layer, the input frequency domain features are mapped to a new feature space, resulting in a frequency domain feature vector. Each dimension represents an abstract representation of the time domain information or the frequency domain information. Therefore, an independent fully-connected layer can independently process features from different modalities, avoiding mutual interference between features while allowing the model to optimize parameters separately. Therefore, inputting time domain and frequency domain features into an independent fully-connected layer separately avoids mutual interference between the time domain and frequency domain features, allowing for independent optimization.

[0067] S212: Based on a preset attention mechanism, calculate a first contribution of the time domain feature vector to the classification result and a second contribution of the frequency domain feature vector to the classification result.

[0068] The preset attention mechanism can be a self-attention mechanism, a channel attention mechanism, a multi-head attention mechanism, a hybrid attention mechanism, etc. The selection of the attention mechanism can be determined based on the weights between the dimensions within the time domain feature vector and the weights between the dimensions within the frequency domain feature vector. The dimensions within the time domain feature vector include: statistical feature dimensions including maximum value, minimum value, peak-to-peak value, mean, variance, standard deviation, mean square value, and root mean square value; dimensionless feature dimensions including kurtosis, skewness, form factor, crest factor, impulse factor, and margin factor; entropy feature dimensions including power spectrum entropy, singular spectrum entropy, energy entropy, approximate entropy, sample entropy, permutation entropy, and fuzzy entropy; time-correlated feature dimensions including frame energy and frame zero-crossing rate; and graph metric feature dimensions including time domain function integration graph metrics and time domain function separation graph metrics. The dimensions within the frequency domain feature vector include: frequency component dimension, frequency band energy dimension, statistical feature dimensions including heart rate, mean square frequency, root mean square frequency, frequency variance, and frequency standard deviation; frequency ratio dimension; and feature point dimension.

[0069] The attention mechanism quantifies the contribution of each modality through dynamic weight assignment. Specifically, time-domain feature vectors, including waveform amplitude, rise time, and pulse width, and frequency-domain feature vectors, including spectrum and energy distribution, are considered independent modalities. The attention layer in the attention mechanism generates weights by calculating the correlation between modalities. It then determines the first and second contributions based on a contribution calculation method. The contribution calculation method can be based on an attention score function (Softmax). The weights are determined by the importance of the time-domain and frequency-domain features (such as their relevance to the fault type). The specific weight determination can refer to the query-key matching in the self-attention mechanism.

[0070] Specifically, the time-domain feature vector and the frequency-domain feature vector are mapped to a unified dimensional space to achieve dimensional alignment between the two. Based on a preset attention mechanism, such as a cross-attention mechanism, attention weights are calculated to obtain the corresponding attention weights for the time-domain feature vector and the frequency-domain feature vector. The attention weights are then summed over the sequence dimension to obtain the global weights of the time-domain feature vector and the frequency-domain feature vector. The higher the global weight, the greater the contribution of the modality (the time-domain feature vector or the frequency-domain feature vector) to the classification result. Based on game theory, the average marginal contribution of the time-domain feature vector and the frequency-domain feature vector across all feature combinations is calculated to obtain the independent contributions of the time-domain feature vector and the frequency-domain feature vector, respectively. The independent contribution of the time-domain feature vector is the first contribution, and the independent contribution of the frequency-domain feature vector is the second contribution.

[0071] S214: Perform weighted fusion based on the first contribution and the second contribution to obtain a weighted fusion feature.

[0072] The rise time of the first contribution reflects the transient characteristics of the partial discharge signal, while the energy distribution of the second contribution reveals the degree of insulation material aging. By weightedly fusing the different dimensions reflected by the characteristics of these two modalities, we can systematically integrate the complementary information in the partial discharge signal.

[0073] S216: Determine the fault type and fault probability corresponding to the partial discharge signal based on the weighted fusion features and a bidirectional long short-term memory network. Fault types include corona discharge, insulation crack, and insulation breakdown.

[0074] The weighted fused features are input into a bidirectional long short-term memory network (LSTM). The BLSTM captures the temporal dependencies within the weighted fused features, thereby determining the fault type and probability corresponding to the partial discharge signal. Compared to a unidirectional LSTM, the BLSTM can simultaneously process both forward and backward temporal information, better capturing the long-range dependencies of the weighted fused features and determining the temporal evolution of the partial discharge signal. Forward temporal information refers to the weighted fused features from the past to the present, in chronological order. Backward temporal information refers to the weighted fused features from the future to the present, in chronological order.

[0075] Specifically, the bidirectional long short-term memory network includes a forward LSTM (Long Short-Term Memory) and a backward LSTM. The forward LSTM processes the weighted fusion features in chronological order to capture the historical dependencies of the weighted fusion features, and the backward LSTM processes the weighted fusion features in reverse time to capture the future contextual information of the weighted fusion features. The historical dependencies of the weighted fusion features and the future contextual information of the weighted fusion features are then concatenated at each time step. After sequence data modeling is performed based on the concatenation results, a time-frequency diagram can be obtained. By analyzing the time-frequency diagram, the fault type and fault probability corresponding to the partial discharge signal can be obtained.

[0076] Specifically, the time-frequency graph is directly correlated with the fault type and probability. High-frequency components in the time-frequency graph can correspond to insulation breakdown. Compared to low-frequency components, the greater the number of high-frequency components, the greater the fault probability. Therefore, the greater the number of high-frequency components, the higher the fault probability, and the more accurately the fault type is insulation breakdown. A rising edge slope greater than a preset slope value can correspond to the presence of corona discharge. Considering the global rising edge slope of the partial discharge signal, the greater the number of rising edges with a rising edge slope greater than the preset slope value, the greater the fault probability. Therefore, in this case, the higher the fault probability, the more accurately the fault type is corona discharge. The preset slope value can be 10 kV / ms.

[0077] The above-mentioned fault diagnosis method for partial discharge of a generator stator winding obtains a partial discharge signal of the stator winding and extracts time domain features based on the partial discharge signal to obtain time domain features. The above-mentioned frequency domain features are extracted based on the partial discharge signal to obtain frequency domain features. The obtained time domain features are input into an independent fully connected layer to obtain a time domain feature vector corresponding to the time domain features. Correspondingly, the obtained frequency domain features are input into the above-mentioned independent fully connected layer to obtain a frequency domain feature vector corresponding to the frequency domain features. Based on a preset attention mechanism, a first contribution of the time domain feature vector to the classification result and a second contribution of the frequency domain feature vector to the classification result are calculated. A weighted fusion is then performed based on the obtained first and second contributions to obtain a corresponding weighted fusion feature. Based on the weighted fusion feature and a bidirectional long short-term memory network, the fault type and fault probability corresponding to the partial discharge signal can be determined. The above-mentioned fault diagnosis method for partial discharge of a generator stator winding can comprehensively determine the fault type and fault probability based on the time domain features and frequency domain features of the partial discharge signal, with high comprehensiveness and accuracy. Furthermore, the fault diagnosis method for partial discharge of the generator stator winding can realize automatic monitoring, reduce the input of human resources, and realize intelligent fault diagnosis.

[0078] In an exemplary embodiment, the step of obtaining a partial discharge signal of a stator winding includes:

[0079] The current signal is collected based on a current sensor, wherein the current sensor may be a high-frequency current sensor that can detect the high-frequency component of the partial discharge signal and reduce the distortion of the partial discharge signal.

[0080] The coupled discharge signal is collected using a capacitive sensor. The capacitive sensor can be an 80pF epoxy mica sensor. Mica is an insulating material with high-temperature and high-voltage resistance. Epoxy mica is a composite material combining mica and epoxy resin. This epoxy mica sensor has improved mechanical strength, insulation properties, and excellent resistance to electromagnetic interference, enabling it to operate in harsh environments.

[0081] A partial discharge signal is determined based on the current signal and the coupled discharge signal.

[0082] Based on the time difference between the reception of the current signal and the coupled discharge signal, the partial discharge signal of the stator winding can be determined.

[0083] In an exemplary embodiment, Figure 3 As shown, the above-mentioned fault diagnosis method for partial discharge of the generator stator winding also includes:

[0084] S302: Generate a time-spectrum diagram based on the time-domain features and the frequency-domain features, wherein the time-spectrum diagram includes a time-frequency matrix (time-frequency-energy distribution).

[0085] After fusing the time-domain waveform from the time-domain features with the frequency-domain energy distribution from the frequency-domain features, a corresponding time-spectrogram can be determined. The time-spectrogram can be generated using a signal processing method, such as a short-time Fourier transform (STFT). The generated time-spectrogram can be in the form of a two-dimensional image, such as a 256×256 pixel image.

[0086] S304: Input the time-frequency spectrum graph into a preset convolution model to obtain at least one feature graph.

[0087] The time-frequency spectrum is input into the pre-trained convolutional model, and the feature map is obtained from the penultimate convolutional layer. Generally, the feature map of this layer can retain the rich spatial information of the image.

[0088] The preset convolutional model can be based on a preset model and pre-trained according to a public dataset (such as ImageNet), and then fine-tuned using a dataset of partial discharge signals from the generator. The preset model must have the ability to preserve feature space, such as Residual Neural Network (ResNet) and Visual Geometry Group (VGG).

[0089] S306, determine the fault type and the characteristic diagrams Figure 1 The corresponding feature map gradient.

[0090] For each fault type, the gradient of the fault type score with respect to the feature map is calculated. These gradients reflect the degree of influence of each position in the feature map on the fault type score.

[0091] S308: Perform global average pooling on the gradients of each feature map to obtain the weight of each feature map.

[0092] The weight of each feature map can reflect the importance of each feature map to the fault type.

[0093] S310 , performing weighted summation based on each feature map and each weight to obtain a class activation map.

[0094] By multiplying each feature map with the corresponding weight and then summing all weighted feature maps, we can get the class activation map.

[0095] S312: Determine and display a visualization map based on the class activation map; wherein the visualization map carries information about the fault location.

[0096] Normalize the class activation map and apply the ReLU (rectified linear unit) function to remove negative values ​​to produce a visualization. This visualization can be a heat map, with red representing high activation areas.

[0097] Analyzing the visualization can pinpoint fault locations in the stator winding. Specifically, if the visualization shows a concentration of high-frequency anomalies within a certain time period, it can be determined that insulation defects exist in a certain section of the stator winding.

[0098] Furthermore, the high-frequency harmonics or pulse clusters in the visualization can be analyzed to locate the critical fault frequency points.

[0099] In an exemplary embodiment, the above-mentioned method for diagnosing partial discharge of a generator stator winding further includes:

[0100] The time domain features and frequency domain features are verified based on Grad-CAM to obtain verification results.

[0101] According to the verification results, the reliability of the time domain features and frequency domain features can be determined to exclude false activation areas.

[0102] In an exemplary embodiment, before the step of extracting time domain features from the partial discharge signal to obtain time domain features, the method further includes:

[0103] A wavelet threshold denoising algorithm is used to filter out white noise from the partial discharge signal. This denoising process can improve the signal-to-noise ratio of the partial discharge signal, preventing subsequent feature extraction and classification from being misled by noise.

[0104] Normalize the denoised partial discharge signal. Normalizing the partial discharge signal ensures consistency between model training and testing, thereby improving generalization. The amplitude of the partial discharge signal can be normalized to the range [0, 1].

[0105] The normalized partial discharge signal is segmented according to a preset time window length to obtain multiple sample sequences. Framing can process non-stationary signals in the partial discharge signal and extract dynamic features, facilitating the identification of different fault types. Furthermore, framing reduces computational effort and improves the efficiency of the generator stator winding partial discharge fault diagnosis method. The preset time window length can be 5ms.

[0106] Multiple sample sequences are spliced ​​together to obtain partial discharge signals.

[0107] The signal processing mentioned above can effectively suppress noise interference, thereby improving the diagnostic reliability of the fault diagnosis method for partial discharge of the generator stator winding under load conditions.

[0108] In an exemplary embodiment, the above-mentioned method for diagnosing partial discharge of a generator stator winding further includes:

[0109] Based on the time domain features, frequency domain features and fault types, a feature-fault type mapping table is determined.

[0110] When the time domain features and / or frequency domain features are determined in the next analysis, there is no need for complex calculation and analysis. Instead, the corresponding fault type can be determined by simply querying the mapping relationship in the feature-fault type mapping table. The operation is simple and consumes little computing power resources.

[0111] In an exemplary embodiment, the above-mentioned method for diagnosing partial discharge of a generator stator winding further includes:

[0112] The severity of the fault is determined based on the fault probability and at least one preset probability threshold.

[0113] Multiple probability thresholds can be set, and different fault levels can be divided according to different probability thresholds. Taking the setting of three probability thresholds as an example, the three probability thresholds are the first probability threshold, the second probability threshold, and the third probability threshold, and the first probability threshold is less than the second probability threshold, and the second probability threshold is less than the third probability threshold. When the probability of failure is less than or equal to the first probability threshold, it can be determined that there is a potential hidden danger; when the probability of failure is greater than the first probability threshold and the probability of failure is less than or equal to the second probability threshold, it can be determined that there is a general defect; when the probability of failure is greater than the second probability threshold and the probability of failure is less than or equal to the third probability threshold, it can be determined that there is a serious defect; when the probability of failure is greater than the third probability threshold, it can be determined that there is a serious fault. Among them, the first probability threshold can be 20%, the second probability threshold can be 40%, and the third probability threshold can be 90%.

[0114] This fault diagnosis method for partial discharge in generator stator windings achieves a fault identification accuracy of 98.7% on the IEEE CIGRE standard dataset, a 12% improvement over traditional methods. Through adaptive feature selection, it maintains an accuracy of over 92% when the signal-to-noise ratio is below 10d. The physical mapping of fault features in the feature-fault type mapping table enables maintenance personnel to quickly locate the root cause of the fault. The single-sample diagnosis time of this fault diagnosis method for partial discharge in generator stator windings can be less than 20ms, meeting the requirements of online monitoring and thus enabling real-time monitoring.

[0115] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0116] Based on the same inventive concept, embodiments of the present application further provide a generator stator winding partial discharge fault diagnosis device for implementing the aforementioned generator stator winding partial discharge fault diagnosis method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more of the following embodiments of the generator stator winding partial discharge fault diagnosis device can be found in the above-described limitations of the generator stator winding partial discharge fault diagnosis method and are not further elaborated here.

[0117] In an exemplary embodiment, Figure 4 As shown, a fault diagnosis device 400 for partial discharge of a generator stator winding is provided, comprising: an acquisition module 402, a time domain feature extraction module 404, a frequency domain feature extraction module 406, a first input module 408, a second input module 410, a contribution determination module 412, a weighted fusion module 414, and a fault determination module 416, wherein:

[0118] The acquisition module 402 is used to acquire a partial discharge signal of the stator winding.

[0119] The time domain feature extraction module 404 is used to extract the time domain features of the partial discharge signal to obtain the time domain features.

[0120] The frequency domain feature extraction module 406 is used to extract frequency domain features from the partial discharge signal to obtain frequency domain features.

[0121] The first input module 408 is used to input the time domain features into the independent fully connected layer to obtain a time domain feature vector.

[0122] The second input module 410 is used to input the frequency domain features into an independent fully connected layer to obtain a frequency domain feature vector.

[0123] The contribution determination module 412 is used to calculate the first contribution of the time domain feature vector to the classification result and the second contribution of the frequency domain feature vector to the classification result based on a preset attention mechanism.

[0124] The weighted fusion module 414 is configured to perform weighted fusion based on the first contribution and the second contribution to obtain a weighted fusion feature.

[0125] The fault determination module 416 is used to determine the fault type and fault probability corresponding to the partial discharge signal according to the weighted fusion features and based on the bidirectional long short-term memory network.

[0126] In an exemplary embodiment, the acquisition module 402 includes: a first acquisition module, a second acquisition module, and a partial discharge signal determination module.

[0127] The first acquisition module is used to acquire current signals based on the current sensor.

[0128] The second acquisition module is used to acquire the coupled discharge signal based on the capacitive sensor.

[0129] The partial discharge signal determination module is used to determine the partial discharge signal according to the current signal and the coupled discharge signal.

[0130] In an exemplary embodiment, the above-mentioned generator stator winding partial discharge fault diagnosis device 400 further includes: a time-frequency spectrum generation module, a feature map determination module, a feature map gradient determination module, a weight determination module, a class activation map determination module and a visualization map determination module.

[0131] The time-spectrum graph generation module is used to generate a time-spectrum graph based on time domain features and frequency domain features.

[0132] The feature map determination module is used to input the time-frequency spectrum map into a preset convolution model to obtain at least one feature map.

[0133] The feature map gradient determination module is used to determine the feature map gradient according to the fault type and each feature map. Figure 1 The corresponding feature map gradient.

[0134] The weight determination module is used to perform global average pooling on the gradients of each feature map to obtain the weight of each feature map.

[0135] The class activation map determination module is used to perform weighted summation based on each feature map and each weight to obtain a class activation map.

[0136] The visualization graph determination module is used to determine and display a visualization graph based on the class activation map; wherein the visualization graph carries information about the fault location.

[0137] In an exemplary embodiment, the generator stator winding partial discharge fault diagnosis device 400 further includes: a denoising module, a normalization module, a framing module, and a splicing module.

[0138] The denoising module is used to filter out the white noise of the partial discharge signal by using a wavelet threshold denoising algorithm.

[0139] The normalization module is used to normalize the denoised partial discharge signal.

[0140] The framing module is used to segment the normalized partial discharge signal according to a preset time window length to obtain multiple sample sequences.

[0141] The splicing module is used to splice multiple sample sequences to obtain partial discharge signals.

[0142] In an exemplary embodiment, the generator stator winding partial discharge fault diagnosis device 400 further includes a mapping module.

[0143] The mapping module is used to determine a feature-fault type mapping table based on time domain features, frequency domain features and fault types.

[0144] In an exemplary embodiment, the generator stator winding partial discharge fault diagnosis device 400 further includes a fault degree determination module.

[0145] The fault severity determination module is used to determine the severity of the fault based on the fault probability and at least one preset probability threshold.

[0146] Each module in the aforementioned generator stator winding partial discharge fault diagnosis device can be implemented in whole or in part via software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0147] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 5As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store weights. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a fault diagnosis method for partial discharge of a generator stator winding is implemented.

[0148] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0149] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory and the processor executes the steps of any one of the above-mentioned methods for diagnosing partial discharge in a generator stator winding.

[0150] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the computer program implements the steps of any of the above-mentioned methods for diagnosing partial discharge of a generator stator winding.

[0151] In one embodiment, a computer program product is provided, comprising a computer program, which implements the steps of any one of the above-mentioned methods for diagnosing partial discharge in a generator stator winding when the computer program is executed by a processor.

[0152] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), quantum computing-based data processing logic devices, artificial intelligence (AI) processors, and the like.

[0153] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0154] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A fault diagnosis method for partial discharge of a generator stator winding, characterized in that: The method comprises: Obtaining partial discharge signals of stator windings; performing time domain feature extraction on the partial discharge signal to obtain a time domain feature; Performing frequency domain feature extraction on the partial discharge signal to obtain frequency domain features; Inputting the time domain features into an independent fully connected layer to obtain a time domain feature vector; Inputting the frequency domain features into the independent fully connected layer to obtain a frequency domain feature vector; Calculating, based on a preset attention mechanism, a first contribution of the time-domain feature vector to the classification result and a second contribution of the frequency-domain feature vector to the classification result; Performing weighted fusion based on the first contribution and the second contribution to obtain a weighted fusion feature; According to the weighted fusion features and based on a bidirectional long short-term memory network, the fault type and fault probability corresponding to the partial discharge signal are determined.

2. The method according to claim 1, characterized in that The obtaining of the partial discharge signal of the stator winding includes: Collect current signals based on current sensors; Collect coupled discharge signals based on capacitive sensors; The partial discharge signal is determined according to the current signal and the coupled discharge signal.

3. The method according to claim 1, characterized in that Also includes: generating a time-frequency spectrum diagram according to the time-domain features and the frequency-domain features; Inputting the time-frequency spectrum into a preset convolutional model to obtain at least one feature map; Determining, according to the fault type and each of the characteristic maps, a characteristic map gradient corresponding to each of the characteristic maps; Performing global average pooling on the gradients of each feature map to obtain the weight of each feature map; Performing weighted summation based on each of the feature maps and each of the weights to obtain a class activation map; A visualization graph is determined and displayed based on the class activation map; wherein the visualization graph carries information about the fault location.

4. The method according to claim 1, wherein Before the step of extracting time domain features from the partial discharge signal to obtain time domain features, the method further includes: Using a wavelet threshold denoising algorithm to filter out white noise from the partial discharge signal; Normalize the denoised partial discharge signal; Segmenting the normalized partial discharge signal according to a preset time window length to obtain multiple sample sequences; The multiple sample sequences are spliced ​​together to obtain the partial discharge signal.

5. The method according to claim 1, wherein Also includes: A feature-fault type mapping table is determined based on the time domain feature, the frequency domain feature, and the fault type.

6. The method according to claim 1, characterized in that The method further comprises: The severity of the fault is determined based on the fault probability and at least one preset probability threshold.

7. A fault diagnosis device for partial discharge of a generator stator winding, characterized in that: The device comprises: An acquisition module, used for acquiring a partial discharge signal of a stator winding; A time domain feature extraction module, configured to extract time domain features from the partial discharge signal to obtain time domain features; A frequency domain feature extraction module, configured to extract frequency domain features from the partial discharge signal to obtain frequency domain features; A first input module is used to input the time domain features into an independent fully connected layer to obtain a time domain feature vector; A second input module is used to input the frequency domain features into the independent fully connected layer to obtain a frequency domain feature vector; A contribution determination module, configured to calculate a first contribution of the time-domain feature vector to the classification result and a second contribution of the frequency-domain feature vector to the classification result based on a preset attention mechanism; a weighted fusion module, configured to perform weighted fusion based on the first contribution and the second contribution to obtain a weighted fusion feature; A fault determination module is used to determine the fault type and fault probability corresponding to the partial discharge signal according to the weighted fusion features and based on a bidirectional long short-term memory network.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images