Generator stator insulation partial discharge diagnosis method based on graph neural network

Abstract

The invention provides a generator stator insulation partial discharge diagnosis method based on a graph neural network, and relates to the technical field of fault diagnosis, and the method comprises the steps: testing a generator to obtain a fault sample, and carrying out the preprocessing of the fault sample; calculating a partition map characteristic quantity corresponding to a fault characteristic region, and constructing a map model characteristic matrix according to the partition map characteristic quantity and data in the PRPD map; establishing a graph structure model by taking each fault feature region as a node, establishing directed connection among the nodes, and constructing an edge relation matrix; training the graph neural network; and inputting the real-time graph model feature matrix into the trained graph neural network to obtain a corresponding generator insulation fault type. By processing various data in a fault sample, fusion of multi-source partial discharge data features of the same generator is realized, multi-feature insulation state analysis and diagnosis are carried out, and the reliability of a diagnosis result is greatly improved.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a method for diagnosing partial discharge of generator stator insulation based on graph neural network. Background Art

[0002] Diagnosing insulation status through partial discharge detection data analysis is a widely used insulation status analysis and diagnosis technology. The sources of partial discharge data include online monitoring, regular live detection, and offline testing. Online monitoring can obtain continuous long-term detection data, which can be used for long-term insulation status trend analysis, but the noise interference is large; live detection is carried out under special operating conditions of the generator. Live detection can be combined with operating condition data for analysis, and the correlation between insulation faults, operating conditions and partial discharge can be used to improve the reliability of diagnosis; offline testing is usually carried out in a state where the generator is isolated from the system, which can fully eliminate the interference and noise of the external system and obtain a partial discharge signal with a high signal-to-noise ratio; in addition, in order to increase the monitoring range of partial discharge of winding insulation, different detection frequency bands will be selected for partial discharge detection to explore the insulation status of more stator bars. The reliability of the current diagnosis results is low. Summary of the invention

[0003] The present application aims to solve one of the technical problems in the related art at least to some extent.

[0004] To this end, the first objective of this application is to propose a generator stator insulation partial discharge diagnosis method based on graph neural network.

[0005] The second object of the present application is to provide a device.

[0006] The third objective of the present application is to provide an electronic device.

[0007] A fourth objective of the present application is to provide a computer-readable storage medium.

[0008] A fifth object of the present application is to provide a computer program product.

[0009] To achieve the above-mentioned purpose, the first embodiment of the present application proposes a generator stator insulation partial discharge diagnosis method based on a graph neural network, comprising:

[0010] Testing the generator to obtain a fault sample, and preprocessing the fault sample to divide a phase-resolved partial discharge (PRPD) spectrum therein into a plurality of fault characteristic regions;

[0011] Calculate the partition map feature quantity corresponding to each of the fault characteristic areas, and construct a graph model feature matrix according to the partition map feature quantity and the data in the PRPD map;

[0012] A graph structure model is established with each of the fault characteristic areas as a node, a directed connection is established between the nodes, and an edge relationship matrix is ​​constructed;

[0013] Training a graph neural network according to the edge relationship matrix, the graph model feature matrix and the corresponding labels;

[0014] The real-time operation data collected in real time in the generator is partitioned and the real-time partition map feature quantity is calculated, and the real-time map model feature matrix is ​​constructed according to the real-time partition map feature quantity and the data in the PRPD map;

[0015] The real-time graph model feature matrix is ​​input into the trained graph neural network to obtain the corresponding generator insulation fault type.

[0016] Optionally, the fault sample includes at least one of the following: partial discharge pulse record raw data Data_d, PRPD spectrum Data_p, test voltage V, detection range R, detection frequency band F, fault type Y;

[0017] The preprocessing of the fault sample to divide the PRPD spectrum therein into a plurality of fault characteristic areas includes:

[0018] By analyzing the records of the partial discharge pulses, the discharge characteristics at different phases can be extracted, thereby forming the PRPD spectrum;

[0019] A multi-source partial discharge fault sample library is constructed according to the fault samples, and structured processing is performed on the data in the multi-source partial discharge fault sample library.

[0020] Optionally, the performing structured processing on the data in the multi-source partial discharge fault sample library includes:

[0021] Convert the detection range R, detection frequency band F and fault type Y from character strings into corresponding numerical values;

[0022] Extracting pulse recording data corresponding to the spectrum from the data table of the partial discharge pulse recording raw data to generate table data;

[0023] The list data is partitioned according to coordinates, and the corresponding PRPD map is divided into mutually related fault characteristic areas.

[0024] Optionally, the calculating of the partition map feature quantity corresponding to each of the fault feature areas includes:

[0025] Calculate the partition spectrum feature quantity of each fault characteristic area according to the PRPD spectrum, wherein the partition spectrum feature quantity includes: discharge pulse duty cycle K in the area, number of discharge pulses NQN in the area, and discharge pulse amplitude QM in the area;

[0026] The list X is constructed according to the discharge pulse duty cycle K within the region, the number of discharge pulses NQN within the region, and the discharge pulse amplitude QM within the region.

[0027] Optionally, constructing a graph model feature matrix according to the partition graph feature quantity and the data in the PRPD graph includes:

[0028] Constructing a node feature vector according to the partition map feature quantity of each fault feature area and the data in the PRPD map;

[0029] The graph model feature matrix is ​​constructed according to the node feature vectors corresponding to all nodes.

[0030] Optionally, the establishing of a graph structure model using each of the fault characteristic regions as a node, establishing directed connections between the nodes and constructing an edge relationship matrix includes:

[0031] According to the correlation between the positive and negative polarities of each region in the PRPD spectrum, the logic of the mutual correlation between the front and rear phase regions of the same polarity establishes a directed connection between each node;

[0032] The edge relationship matrix is ​​constructed according to the connection relationship between two nodes.

[0033] Optionally, partitioning the list data by coordinates to divide the corresponding PRPD map into interrelated fault feature areas includes:

[0034] The size and position of the fault characteristic region are adjusted according to the coordinate parameters, or the number of the fault characteristic regions is adjusted according to the coordinate parameters.

[0035] To achieve the above-mentioned purpose, the second embodiment of the present application proposes a generator stator insulation partial discharge diagnosis device based on a graph neural network, comprising:

[0036] A preprocessing module, used for testing the generator to obtain fault samples, and preprocessing the fault samples to divide the phase-resolved partial discharge (PRPD) spectrum therein into a plurality of fault characteristic regions;

[0037] A partitioning module, used to calculate the partition map feature quantity corresponding to each of the fault characteristic areas, and to construct a graph model feature matrix according to the partition map feature quantity and the data in the PRPD map;

[0038] A model building module, used to establish a graph structure model with each of the fault feature areas as a node, establish directed connections between the nodes and construct an edge relationship matrix;

[0039] A training module, used for training a graph neural network according to the edge relationship matrix, the graph model feature matrix and the corresponding labels;

[0040] A feature extraction module is used to partition the real-time operation data collected in real time in the generator and calculate the real-time partition map feature quantity, and construct a real-time map model feature matrix according to the real-time partition map feature quantity and the data in the PRPD map;

[0041] The diagnostic module is used to input the real-time graph model feature matrix into the trained graph neural network to obtain the corresponding generator insulation fault type.

[0042] To achieve the above-mentioned purpose, the third aspect of the present application provides an electronic device, including: a processor, and a memory communicatively connected to the processor;

[0043] The memory stores computer-executable instructions;

[0044] The processor executes the computer-executable instructions stored in the memory to implement the method as described in any one of the first aspects.

[0045] To achieve the above-mentioned purpose, the fourth aspect embodiment of the present application proposes a computer-readable storage medium, in which computer-readable storage medium is stored computer execution instructions, and when the computer execution instructions are executed by a processor, they are used to implement the method as described in any one of the first aspects.

[0046] To achieve the above-mentioned purpose, the fifth aspect of the present application proposes a computer program product, which implements any method in the first aspect when executed by a processor.

[0047] The graph neural network-based generator stator insulation partial discharge diagnosis method, device, electronic device and storage medium provided in the present application realize the fusion of multi-source partial discharge data features for the same generator by processing multiple data in the fault samples, and carry out multi-feature insulation status analysis and diagnosis, thereby greatly improving the reliability of the diagnosis results.

[0048] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0050] Figure 1 A schematic diagram of a flow chart of a generator stator insulation partial discharge diagnosis method based on a graph neural network provided in an embodiment of the present application;

[0051] Figure 2 It is a schematic diagram of the characteristics of the partition map;

[0052] Figure 3 It is a structural diagram of the graph structure model;

[0053] Figure 4 This is a schematic diagram of the training results;

[0054] Figure 5 A schematic diagram of the structure of a generator stator insulation partial discharge diagnostic device based on a graph neural network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0056] Diagnosing insulation status through partial discharge detection data analysis is a widely used insulation status analysis and diagnosis technology. The sources of partial discharge data include online monitoring, regular live detection, and offline testing. Online monitoring can obtain continuous long-term detection data, which can be used for long-term insulation status trend analysis, but the noise interference is large; live detection is carried out under special operating conditions of the generator. Live detection can be combined with operating condition data for analysis, and the correlation between insulation faults, operating conditions and partial discharge can be used to improve diagnostic reliability; offline testing is usually carried out in a state where the generator is isolated from the system, which can fully eliminate the interference and noise of the external system and obtain a partial discharge signal with a high signal-to-noise ratio; in addition, in order to increase the monitoring range of partial discharge of winding insulation, different detection frequency bands will be selected for partial discharge detection to explore the insulation status of more stator bars.

[0057] Partial discharge data from different sources for the same generator can reflect the insulation status of the winding to a certain extent. They are correlated, but they also have large differences due to different detection parameters. If the multi-source partial discharge data features for the same generator are integrated and multi-feature insulation status analysis and diagnosis are carried out, the reliability of the diagnosis results will be greatly improved. However, how to integrate the multi-source partial discharge data features is a difficult problem in the industry.

[0058] The purpose of the present invention is to solve the shortcomings of the existing generator insulation partial discharge analysis and diagnosis technology, that is, the reliability of the diagnosis model is not high and the generalization ability is weak; and to provide an insulation fault diagnosis method for generators by fusing multi-source partial discharge data. Specifically, it includes the establishment of a multi-source partial discharge database, the extraction of detection parameter features, the extraction of graph feature parameters, the construction of a fault diagnosis graph model, the training of a graph neural network diagnosis model, and the application of a diagnosis model.

[0059] The embodiment of the present application provides a generator stator insulation partial discharge diagnosis method based on a graph neural network. Figure 1 The present invention provides a flow chart of a generator stator insulation partial discharge diagnosis method based on a graph neural network. Figure 1 As shown, the method comprises the following steps:

[0060] Step 101, testing the generator to obtain a fault sample, and preprocessing the fault sample to divide the phase-resolved partial discharge PRPD spectrum therein into a plurality of fault characteristic regions;

[0061] Step 102, calculating the partition map feature quantity corresponding to each of the fault characteristic areas, and constructing a graph model feature matrix according to the partition map feature quantity and the data in the PRPD map;

[0062] Step 103, establishing a graph structure model with each of the fault characteristic regions as a node, establishing directed connections between the nodes and constructing an edge relationship matrix;

[0063] Step 104, training a graph neural network according to the edge relationship matrix, the graph model feature matrix and the corresponding labels;

[0064] Step 105, partitioning the real-time operation data collected in real time in the generator and calculating the real-time partition map feature quantity, and constructing a real-time map model feature matrix according to the real-time partition map feature quantity and the data in the PRPD map;

[0065] Step 106: input the real-time graph model feature matrix into the trained graph neural network to obtain the corresponding generator insulation fault type.

[0066] In this embodiment, the fault sample acquisition: using the generator sample, confirm the existence of insulation fault in the generator stator winding insulation through appearance inspection, ultraviolet imaging, etc., such as corona corrosion at the stator winding slot, corona corrosion between phases at the stator winding end, and internal electrical and thermal aging of the insulation. Carry out live or online partial discharge test, offline ultra-high frequency partial discharge test, and offline low-frequency partial discharge test for the same faulty generator, obtain the partial discharge data of the insulation fault of the same generator through different methods, and synchronously record the test parameters voltage, detection range, detection frequency band, generator temperature, humidity, load, and fault type label, and establish a multi-source partial discharge fault sample library Data.

[0067] Optionally, the fault sample includes at least one of the following: partial discharge pulse record raw data Data_d, PRPD spectrum Data_p, test voltage V, detection range R, detection frequency band F, fault type Y;

[0068] The preprocessing of the fault sample to divide the PRPD spectrum therein into a plurality of fault characteristic areas includes:

[0069] By analyzing the records of the partial discharge pulses, the discharge characteristics at different phases can be extracted, thereby forming the PRPD spectrum;

[0070] A multi-source partial discharge fault sample library is constructed according to the fault samples, and structured processing is performed on the data in the multi-source partial discharge fault sample library.

[0071] In a possible embodiment, the fault sample library is as shown in Table 1:

[0072]

[0073] Table 1

[0074] Optionally, the performing structured processing on the data in the multi-source partial discharge fault sample library includes:

[0075] Convert the detection range R, detection frequency band F and fault type Y from character strings into corresponding numerical values;

[0076] Extracting pulse recording data corresponding to the spectrum from the data table of the partial discharge pulse recording raw data to generate table data;

[0077] The list data is partitioned according to coordinates, and the corresponding PRPD map is divided into mutually related fault characteristic areas.

[0078] In this embodiment, the detection range is replaced by the key value in the dictionary Range_dir={'2-34[mV]':1,'5-85[mV]':2,'10-170[mV]':3,'20-340[mV]':4,'50-850[mV]':5,'100-1700[mV]':6,'200-3400[mV]':7,'500-8500[mV]':8,'1-17[V]':9};

[0079] The detection frequency band F is replaced according to the key value in the dictionary Class_dir={'Directional(BT,TGAB)':1,'Differential(HT,PDAB)':2,'Directional BUS(TGA)':3,'Low Frequency(50kHz)':4};

[0080] The fault type Y is replaced by the key value in the dictionary y_dir = {"notch discharge":0,"end discharge":1,"notch plus end discharge":2}.

[0081] In a possible embodiment, the replaced fault samples are shown in Table 2:

[0082]

[0083] Table 2

[0084] The original data of the spectrum is structured and the pulse recording data corresponding to the spectrum is extracted from the xlsx data table to generate 32*100*3 3D list data:

[0085] Data = [

[0086] [[x 00 ,y 00 ,z 00 ],[x 01 ,y 00 ,z 01 ],[x 02 ,y 00 ,z 02 ]......[x 98 ,y 00 ,z 02 ],[x 99 ,y 00 ,z 99 ]]

[0087] [[x 00 ,y 15 ,z 1500 ],[x 01 ,y15 ,z 1501 ],[x 02 ,y 15 ,z 1502 ]......[x 98 ,y 15 ,z 198 ],[x 99 ,y 15 ,z 1599 ]]

[0088] [[x 00 ,y 31 ,z 3100 ],[x 01 ,y 31 ,z 3101 ],[x 02 ,y 31 ,z 3102 ]......[x 98 ,y 31 ,z 3198 ],[x 99 ,y 31 ,z 3199 ]] ]

[0090] Further, based on the list data, the data is partitioned by coordinates, and the PRPD diagram is divided into 20 interrelated fault feature areas, namely A1-A14, B1-B4, and C1-C2. The start and end coordinates of the corresponding areas are recorded in the dictionary regions.

[0091] For example, regions=

[0092] {'A1':[[0,0],[7,15]],'A2':[[8,0],[15,15]],'A3'[[16,0],[24,15]],'A4':[[25,0],[49,15]],'A5':[[50,0],[74,15]],'A6' :[[75,0],[95,15]],'A7':[[96,0],[99,15]],'A8':[[0,16],[24,31]],'A9':[[25,16],[45,31]],'A10':[[46,16],[49,31]],'A 11':[[50,16],[57,31]],'A12':[[58,16],[65,31]],'A13':[[66,16],[74,31]],'A14':[[75,16],[99,31]],'B1':[[0,0],[99,7 ]],'B2':[[0,8],[99,15]],'B3':[[0,16],[99,23]],'B4':[[0,24],[99,31]],'C1':[[0,0],[99,15]],'C2':[[0,16],[99,31]]}

[0093] Optionally, the step 102 of calculating the partition map feature quantity corresponding to each of the fault feature areas includes:

[0094] Calculate the partition spectrum feature quantity of each fault characteristic area according to the PRPD spectrum, wherein the partition spectrum feature quantity includes: discharge pulse duty cycle K in the area, number of discharge pulses NQN in the area, and discharge pulse amplitude QM in the area;

[0095] The list X is constructed according to the discharge pulse duty cycle K within the region, the number of discharge pulses NQN within the region, and the discharge pulse amplitude QM within the region.

[0096] In this embodiment, the discharge pulse duty cycle K in the region refers to the ratio of the time period in which the discharge pulse actually occurs to the total time in a certain time period. This ratio reflects the proportion of the discharge activity in the entire observation time. The calculation formula is:

[0097] K = cumulative time of discharge pulse appearance / total observation time

[0098] MQN: This indicator directly reflects the number of local discharge pulses within a certain period of time or under certain conditions. It is obtained by counting, that is, how many discharge pulses are detected within the set observation window. The calculation formula is:

[0099]

[0100] The discharge pulse amplitude QM in the region describes the energy of each discharge pulse. It can be an average value or a maximum value, depending on specific needs. If you want to calculate the average amplitude, the formula is:

[0101]

[0102] If the maximum amplitude is of interest, simply select the maximum value among all the measured values. It should be noted that the calculation of the above indicators depends on the experimental conditions and data collection methods. In practical applications, these indicators often need to be adjusted and optimized in combination with specific test environments and equipment. In addition, due to the complexity and randomness of partial discharge, the calculation of these indicators may also be affected by noise and other non-discharge factors. Therefore, they should be treated with caution and considered comprehensively in combination with multiple means during analysis.

[0103] Figure 2 is a schematic diagram of the partition map feature, X = [K, NQN, QM], and we get Figure 2 The partition map feature diagram shown includes 20 fault feature areas, namely A1-A14, B1-B4, and C1-C2.

[0104] Optionally, the method further comprises: marking data samples, labeling samples according to actual problems existing in the generator, such as recording the existence of corona problems and end discharge problems confirmed by maintenance inspection in the Y_dir dictionary; for example:

[0105] Y_dir = {Corona problem: 0, End discharge problem: 1, Corona and end: 2, ...}

[0106] Optionally, the step 102 constructs a graph model feature matrix according to the partition graph feature quantity and the data in the PRPD graph, including:

[0107] Constructing a node feature vector according to the partition map feature quantity of each fault feature area and the data in the PRPD map;

[0108] The graph model feature matrix is ​​constructed according to the node feature vectors corresponding to all nodes.

[0109] In this embodiment, the node feature vector is constructed using a one-dimensional array x i express:

[0110] x i =[Ki, NQNi, QMi, Vi, Mi, Fi, Ti, Hi, Zi]

[0111] x iis the feature vector of the i-th partition, i.e., node, i = 0, 1, 2, 3, .... n-1, n represents the number of nodes in the graph structure, i.e., the number of graph partitions, x i In the figure, the duty cycle of the pulse count in the i-th area in the Ki spectrum, the sum of the number of pulse counts in the i-th area in the NQNi spectrum, and the maximum amplitude of the pulse count in the i-th area in the QMi spectrum; Vi obtains the test voltage at the time of the pulse count, Mi obtains the test range at the time of the pulse count, Fi obtains the detection frequency band at the time of the pulse count; Ti obtains the winding temperature at the time of the pulse count, Hi obtains the winding humidity at the time of the pulse count, and Zi obtains the stator vibration at the time of the pulse count.

[0112] X id =[x 0 ,x 1 ,x 2 ,.....x n-3 ,x n-2 ,x n-1 ]

[0113] n is the number of nodes or partitions, 1, 2, 3...n represents the node sequence number; x 0 ,x 1 ,x 2 ,.....x n-3 ,x n-2 ,x n-1 represents the n feature vectors of nodes 1 to n in turn; X id It is the graphical model feature matrix of the sample No. id.

[0114] In a possible embodiment, taking 105 samples as an example

[0115] Construct a one-dimensional array X of node feature vectors i

[0116] X A1 =[0.578,619,74.4,10500,3,2]

[0117] X A2 =[0.539,639,85.0,10500,3,2]

[0118] X A3 =[0.340,435,47.8,10500,3,2]

[0119] X A7 =[0.393,298,74.4,10500,3,2]

[0120] X A10 =[0.438,404,74.4,10500,3,2]

[0121] X A11 =[0.587,723,79.7,10500,3,2]

[0122] X A12 =[0.461,675,47.8,10500,3,2]

[0123] X A13 =[0.292,414,37.2,10500,3,2]

[0124] Construct a single sample graph feature vector two-dimensional array X id

[0125] X 105 =[X A1, X A2, X A3, X A7, X A10, X A11, X A12 ,X A13 ] [

[0127] [0.578, 619, 74.4, 10500, 3, 2],

[0128] [0.539, 639, 85.0, 10500, 3, 2],

[0129] [0.340, 435, 47.8, 10500, 3, 2],

[0130] [0.393, 298, 74.4, 10500, 3, 2],

[0131] [0.438,404,74.4,10500,3,2],

[0132] [0.587,723,79.7,10500,3,2],

[0133] [0.461, 675, 47.8, 10500, 3, 2],

[0134] [0.292,414,37.2,10500,3,2] ]

[0136] Construct other samples in the sample library and summarize the sample features into feature library X, where the length of X is the number of samples in the fault library.

[0137] Get the Y field in the sample data as the fault label list;

[0138] Y=[0,0,0,0,0,0,0,0,0,0,......1,1,1,1,1,1,1]

[0139] Optionally, the step 103 establishes a graph structure model using each of the fault feature regions as a node, establishes directed connections between the nodes and constructs an edge relationship matrix, including:

[0140] According to the correlation between the positive and negative polarities of each region in the PRPD spectrum, the logic of the mutual correlation between the front and rear phase regions of the same polarity establishes a directed connection between each node;

[0141] The edge relationship matrix is ​​constructed according to the connection relationship between two nodes.

[0142] In this embodiment, a graph structure is established to partition 20 regions from A1 to C2, and each region is used as a node to establish a graph structure model. According to the correlation between the positive and negative polarities of each region in the graph, and the mutual correlation logic between the front and rear phase regions of the same polarity, a directed connection is established between the nodes.

[0143] In addition to this node, a two-way connection is established between it and other nodes.

[0144] Construct the graph model edge relationship matrix E: The connection relationship between each node is represented by a two-dimensional matrix E, and the elements E in the matrix ij Indicates whether there is a connection relationship between nodes i and j, that is, a directed connection edge from node i to j. If yes, then E ij =1, if it does not exist, then E ij =0, n represents the number of nodes in the graph structure i = 0, 1, 2, 3....n-1, j = 0, 1, 2, 3....n-1, the element E in the matrix ij The value 0 or 1 is determined according to the relationship between two nodes in the specific physical model; according to the partition n = 20, the E matrix is ​​as follows:

[0145]

[0146] In a possible embodiment, the structural diagram of the graph structure model is as follows: Figure 3 As shown in the figure, due to the limitation of the type and quantity of fault samples, graph modeling is performed for corona problem: 0 and end discharge problem: 1. When modeling, only the phase partitions related to the two types of faults are selected as nodes, namely A1, A2, A3, A7, A10, A11, A12, A13, a total of 8 regions. Build a graph model. The number of nodes is 8, 0, 1, 2, 3, 4, 5, 6, 7, corresponding to A1, A2, A3, A7, A10, A11, A12, A13, a total of 8 regions.

[0147] The model edge relationship matrix is ​​as follows:

[0148]

[0149] Optional, training of graph neural network: taking the graph model edge relationship two-dimensional matrix E and the graph feature matrix X corresponding to each fault sample data set and the corresponding fault label Y as input, the graph neural network is trained and verified to obtain the multi-source partial discharge data fusion analysis and diagnosis model of the hydro-generator stator.

[0150] In this embodiment, a graph neural network algorithm is used to train the model, and the network structure parameters are as follows:

[0151] Graph convolution layer 1, input feature dimension 6, output feature dimension 128;

[0152] Graph downsampling, input feature dimension is 128, the number of retained nodes is 80% of the number of input nodes, and the output feature dimension remains unchanged;

[0153] Graph convolutional layer 2, input feature dimension 128, output feature dimension 128;

[0154] Graph downsampling, input feature dimension is 128, the number of retained nodes is 80% of the number of input nodes, and the output feature dimension remains unchanged;

[0155] Graph convolution layer 3, input feature dimension 128, output feature dimension 128;

[0156] Graph downsampling, input feature dimension is 128, the number of retained nodes is 80% of the number of input nodes, and the output feature dimension remains unchanged;

[0157] Fully connected layer 1, input feature dimension 128, output feature dimension 128;

[0158] Fully connected layer 2, input feature dimension 128, output feature dimension 64;

[0159] Fully connected layer 3, input feature dimension 64, output feature dimension 1;

[0160] Batch normalization layer 1, normalizes the 128-dimensional features;

[0161] Batch normalization layer 2, normalizing the 64-dimensional features;

[0162] Activation function 1: ReLU;

[0163] Activation function 2: ReLU;

[0164] Model training includes:

[0165] Optimizer: torch.optim.Adam optimizes the parameters of the model;

[0166] Learning rate: lr is set to 0.001, which is the step size taken by the optimizer each time it updates the parameters;

[0167] Loss function: torch.nn.BCELoss binary cross entropy loss function, torch.nn.CrossEntropyLoss loss function for multi-classification;

[0168] Training steps: forward propagation, loss calculation, back propagation and parameter update;

[0169] Training parameters: Epoch = 400, batch_size = 200;

[0170] The training results are shown in the figure below: Figure 4 As shown, after 400 cycles, the loss dropped below 0.1;

[0171] Training result evaluation score: roc_auc_score: 0.8.

[0172] Carry out one of the following tests on live partial discharge detection, offline high-frequency partial discharge, and offline low-frequency partial discharge on a new generator with unknown winding insulation fault, or directly obtain sample data through online monitoring, and generate a fault feature matrix two-dimensional array X after partition calculation. id Input the trained diagnostic model and the diagnostic output result is the generator insulation fault type.

[0173] Optionally, partitioning the list data by coordinates to divide the corresponding PRPD map into interrelated fault feature areas includes:

[0174] The size and position of the fault characteristic region are adjusted according to the coordinate parameters, or the number of the fault characteristic regions is adjusted according to the coordinate parameters.

[0175] In this embodiment, the size and position of the A1 to C1 partitions can be adjusted through the coordinate parameters of the corresponding key values ​​in the dictionary regions, and the number of partitions can also be reduced or increased to accommodate the extraction of more key features.

[0176] The eigenvector is x i =[Ki, NQNi, QMi, Vi, Mi, Fi, Ti, Hi, Zi] can also be adjusted according to the sample data to increase or decrease features. The more features, the better the diagnostic effect.

[0177] As time goes by, the types of fault samples in the sample database increase, and the fault labels also increase. Retraining the graph neural network on this basis will improve the diagnostic ability of the graph neural network.

[0178] In a possible embodiment, the diagnostic analysis application obtains 30 samples of generator slot corona faults from the sample library, inputs the data into the model classification diagnosis, and outputs the following prediction list, where the actual fault is slot corona with a label of 0.

[0179] Classification results: ['0.006','0.005','0.003','0.006','0.439','0.455','0.453','0.450','0.080','0.088','0.087','0.042','0.132','0.095','0.082','0.056','0.050','0.080','0.095','0.123','0.108','0.153','0.166','0.194','0.235','0.213','0.082','0.070','0.081','0.079']

[0180] Actual result: [0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0]

[0181] Total number of categories: 30 Correct number: 24

[0182] The percentage of correct classification is: 80.00%

[0183] At present, there are few mature methods for evaluating the insulation status of large generator stators. The most feasible method is the analysis and evaluation based on partial discharge of insulation. In recent years, the detection system for partial discharge of generator insulation has developed rapidly, with frequency bands ranging from low frequency kHZ to ultra-high frequency GHZ. Sensors include capacitive sensors, high frequency CT, and ultra-high frequency antennas. Anti-interference technologies include polarity identification, directional separation, and timing separation. Test methods include online monitoring, live detection, and offline testing. Due to differences in detection systems and generator structures, there are large differences in the partial discharge data output by the system, which brings great trouble to analysis and diagnosis. Although there are analysis methods such as phase analysis and PRPD spectrum, experienced professionals are still needed for identification and diagnosis, and reliable intelligent analysis and diagnosis have not yet been achieved. Based on the partial discharge data output by the system, the present invention fully considers the factors that have a greater impact on insulation faults, such as test method detection frequency, test voltage, generator operating voltage, insulation temperature, and vibration, and integrates partial discharge data, test parameters, and generator operating parameters from different sources to construct an insulation fault analysis and diagnosis method with multi-dimensional features, thereby improving the reliability of the diagnosis results.

[0184] In order to implement the above embodiments, the present application also proposes a generator stator insulation partial discharge diagnosis device based on graph neural network. Figure 5 The present invention provides a schematic diagram of a generator stator insulation partial discharge diagnosis device based on a graph neural network. Figure 5 As shown, the device comprises:

[0185] A preprocessing module 510 is used to test the generator to obtain fault samples, and preprocess the fault samples to divide the phase-resolved partial discharge PRPD spectrum therein into multiple fault characteristic areas;

[0186] A partitioning module 520 is used to calculate the partition map feature quantity corresponding to each of the fault characteristic areas, and to construct a graph model feature matrix according to the partition map feature quantity and the data in the PRPD map;

[0187] A model building module 530 is used to build a graph structure model using each of the fault feature areas as a node, establish directed connections between the nodes and construct an edge relationship matrix;

[0188] A training module 540 is used to train a graph neural network according to the edge relationship matrix, the graph model feature matrix and the corresponding labels;

[0189] The feature extraction module 550 is used to partition the real-time operation data collected in real time in the generator and calculate the real-time partition map feature quantity, and construct the real-time map model feature matrix according to the real-time partition map feature quantity and the data in the PRPD map;

[0190] The diagnostic module 5560 is used to input the real-time graph model feature matrix into the trained graph neural network to obtain the corresponding generator insulation fault type.

[0191] In order to implement the above embodiments, the present application also proposes an electronic device, comprising: a processor, and a memory communicatively connected to the processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory to implement the method provided by the above embodiments.

[0192] In order to implement the above embodiments, the present application also proposes a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the methods provided by the above embodiments.

[0193] In order to implement the above embodiments, the present application also proposes a computer program product, including a computer program, which implements the methods provided by the above embodiments when executed by a processor.

[0194] The collection, storage, use, processing, transmission, provision and disclosure of user personal information involved in this application are in compliance with relevant laws and regulations and do not violate public order and good morals.

[0195] It should be noted that personal information from users should be collected for legitimate and reasonable purposes and should not be shared or sold outside of these legitimate uses. In addition, such collection / sharing should be carried out after receiving the user's informed consent, including but not limited to notifying the user to read the user agreement / user notice and sign the agreement / authorization including authorization of relevant user information before the user uses the function. In addition, any necessary steps should be taken to protect and safeguard access to such personal information data and ensure that others who have access to personal information data comply with its privacy policy and procedures.

[0196] The present application is expected to provide an implementation scheme for users to selectively block the use or access of personal information data. That is, the present disclosure is expected to provide hardware and / or software to prevent or block access to such personal information data. Once the personal information data is no longer needed, the risk can be minimized by limiting data collection and deleting the data. In addition, when applicable, such personal information is de-identified to protect the privacy of the user.

[0197] In the description of the aforementioned embodiments, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0198] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of this application, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0199] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0200] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute the instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0201] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0202] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0203] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0204] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A generator stator insulation partial discharge diagnosis method based on graph neural network, characterized in that: The following steps are involved: Testing the generator to obtain a fault sample, and preprocessing the fault sample to divide a phase-resolved partial discharge (PRPD) spectrum therein into a plurality of fault characteristic regions; Calculate the partition map feature quantity corresponding to each of the fault characteristic areas, and construct a graph model feature matrix according to the partition map feature quantity and the data in the PRPD map; A graph structure model is established with each of the fault characteristic areas as a node, a directed connection is established between the nodes, and an edge relationship matrix is ​​constructed; Training a graph neural network according to the edge relationship matrix, the graph model feature matrix and the corresponding labels; The real-time operation data collected in real time in the generator is partitioned and the real-time partition map feature quantity is calculated, and the real-time map model feature matrix is ​​constructed according to the real-time partition map feature quantity and the data in the PRPD map; The real-time graph model feature matrix is ​​input into the trained graph neural network to obtain the corresponding generator insulation fault type.

2. The method according to claim 1, characterized in that The fault sample includes at least one of the following: partial discharge pulse record raw data Data_d, PRPD spectrum Data_p, test voltage V, detection range R, detection frequency band F, fault type Y; The preprocessing of the fault sample to divide the PRPD spectrum therein into a plurality of fault characteristic areas includes: By analyzing the records of the partial discharge pulses, the discharge characteristics at different phases can be extracted, thereby forming the PRPD spectrum; A multi-source partial discharge fault sample library is constructed according to the fault samples, and structured processing is performed on the data in the multi-source partial discharge fault sample library.

3. The method according to claim 2, characterized in that The structural processing of the data in the multi-source partial discharge fault sample library includes: Convert the detection range R, detection frequency band F and fault type Y from character strings into corresponding numerical values; Extracting pulse recording data corresponding to the spectrum from the data table of the partial discharge pulse recording raw data to generate table data; The list data is partitioned according to coordinates, and the corresponding PRPD map is divided into mutually related fault characteristic areas.

4. The method according to claim 3, characterized in that The calculating of the partition spectrum feature quantity corresponding to each of the fault characteristic areas includes: Calculate the partition spectrum feature quantity of each fault characteristic area according to the PRPD spectrum, wherein the partition spectrum feature quantity includes: discharge pulse duty cycle K in the area, number of discharge pulses NQN in the area, and discharge pulse amplitude QM in the area; The list X is constructed according to the discharge pulse duty cycle K within the region, the number of discharge pulses NQN within the region, and the discharge pulse amplitude QM within the region.

5. The method according to claim 4, characterized in that The step of constructing a graph model feature matrix according to the partition graph feature quantity and the data in the PRPD graph includes: Constructing a node feature vector according to the partition map feature quantity of each fault feature area and the data in the PRPD map; The graph model feature matrix is ​​constructed according to the node feature vectors corresponding to all nodes.

6. The method according to claim 5, characterized in that The method of establishing a graph structure model with each of the fault characteristic regions as a node, establishing directed connections between the nodes and constructing an edge relationship matrix includes: According to the correlation between the positive and negative polarities of each region in the PRPD spectrum, the logic of the mutual correlation between the front and rear phase regions of the same polarity establishes a directed connection between each node; The edge relationship matrix is ​​constructed according to the connection relationship between two nodes.

7. The method according to claim 3, characterized in that Partitioning the list data by coordinates to divide the corresponding PRPD map into interrelated fault feature areas includes: The size and position of the fault characteristic region are adjusted according to the coordinate parameters, or the number of the fault characteristic regions is adjusted according to the coordinate parameters.

8. A generator stator insulation partial discharge diagnostic device based on graph neural network, characterized in that: include: A preprocessing module, used for testing the generator to obtain fault samples, and preprocessing the fault samples to divide the phase-resolved partial discharge (PRPD) spectrum therein into a plurality of fault characteristic regions; A partitioning module, used to calculate the partition map feature quantity corresponding to each of the fault characteristic areas, and to construct a graph model feature matrix according to the partition map feature quantity and the data in the PRPD map; A model building module, used to establish a graph structure model with each of the fault feature areas as a node, establish directed connections between the nodes and construct an edge relationship matrix; A training module, used for training a graph neural network according to the edge relationship matrix, the graph model feature matrix and the corresponding labels; A feature extraction module is used to partition the real-time operation data collected in real time in the generator and calculate the real-time partition map feature quantity, and construct a real-time map model feature matrix according to the real-time partition map feature quantity and the data in the PRPD map; The diagnostic module is used to input the real-time graph model feature matrix into the trained graph neural network to obtain the corresponding generator insulation fault type.

9. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 7 when executed by a processor.

Images