CN113884822A - Classification and identification method for partial discharge ultrahigh frequency signals - Google Patents
Classification and identification method for partial discharge ultrahigh frequency signals Download PDFInfo
- Publication number
- CN113884822A CN113884822A CN202110917715.3A CN202110917715A CN113884822A CN 113884822 A CN113884822 A CN 113884822A CN 202110917715 A CN202110917715 A CN 202110917715A CN 113884822 A CN113884822 A CN 113884822A
- Authority
- CN
- China
- Prior art keywords
- classification
- pduhf
- data
- partial discharge
- deformer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 238000012545 processing Methods 0.000 claims abstract description 18
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 11
- 230000007547 defect Effects 0.000 claims abstract description 10
- 238000010606 normalization Methods 0.000 claims abstract description 10
- 230000009467 reduction Effects 0.000 claims abstract description 5
- 239000011159 matrix material Substances 0.000 claims description 33
- 230000006870 function Effects 0.000 claims description 24
- 238000013528 artificial neural network Methods 0.000 claims description 23
- 239000013598 vector Substances 0.000 claims description 18
- 238000009413 insulation Methods 0.000 claims description 9
- 238000013507 mapping Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 5
- PXUQTDZNOHRWLI-OXUVVOBNSA-O malvidin 3-O-beta-D-glucoside Chemical compound COC1=C(O)C(OC)=CC(C=2C(=CC=3C(O)=CC(O)=CC=3[O+]=2)O[C@H]2[C@@H]([C@@H](O)[C@H](O)[C@@H](CO)O2)O)=C1 PXUQTDZNOHRWLI-OXUVVOBNSA-O 0.000 claims description 3
- 239000002245 particle Substances 0.000 claims description 3
- 239000000725 suspension Substances 0.000 claims description 3
- 238000009833 condensation Methods 0.000 claims description 2
- 238000012847 principal component analysis method Methods 0.000 claims description 2
- 230000007246 mechanism Effects 0.000 abstract description 4
- 210000002569 neuron Anatomy 0.000 description 6
- 238000001514 detection method Methods 0.000 description 4
- 230000004913 activation Effects 0.000 description 3
- 238000000513 principal component analysis Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
- G01R31/1254—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials of gas-insulated power appliances or vacuum gaps
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing Relating To Insulation (AREA)
Abstract
The invention relates to a classification identification method of partial discharge ultrahigh frequency signals, which is technically characterized by comprising the following steps: acquiring PDUHF signals caused by faults of different GIS defect types as sample data; normalizing the acquired sample data; performing signal decomposition on each sample data after normalization processing to obtain high-frequency and low-frequency data and performing dimensionality reduction processing; dividing sample data into training data and test data; training an attention encoder model using training data; and carrying out classification recognition on the test data by using the trained attention encoder model. The invention adopts the empirical mode decomposition and the attention mechanism to extract the frequency domain and the time domain characteristics of the partial discharge ultrahigh frequency signals in the gas insulated switchgear, effectively improves the classification and identification accuracy of the partial discharge ultrahigh frequency signals of different fault types of the GIS, and can be widely used for classifying and identifying PDUHF signals caused by different types of faults in the gas insulated switchgear.
Description
Technical Field
The invention belongs to the technical field of high-voltage electrical equipment, relates to the detection of the insulation performance of a gas insulated switchgear, and particularly relates to a classification and identification method of partial discharge ultrahigh frequency signals.
Background
The safety and reliability of Gas Insulated Switchgear (GIS) play a key role in the stable operation of the power grid. However, insulation deterioration due to various reasons is often an important cause of failure of the GIS, and therefore, in order to ensure safety and stability of the power grid, insulation performance detection of the GIS is required.
At present, one of the common technologies for detecting the insulation performance of the GIS is to diagnose different fault types by classifying and identifying partial discharge signals of the GIS, so as to evaluate the insulation performance of the GIS. In the partial discharge detection, an ultrahigh frequency signal detection technique is a commonly used technique. With the development, popularization and application of related technologies, a large number of Partial Discharge ultrahigh Frequency (PDUHF) signals are generated in operation and maintenance of a transformer substation. However, due to the complexity of the field environment of the substation, the detected PDUHF signals inevitably contain noise signals, the difficulty of classifying and identifying the PDUHF signals is increased by the noise signals, and how to accurately classify and identify the PDUHF signals containing noise is of great significance for reliably evaluating the insulation performance of the GIS and the safe and stable operation of the power grid.
In order to accurately classify and identify noisy PDUHF signals, researchers at home and abroad have proposed some methods in succession, for example, extracting and counting relevant features by combining different methods such as a neural network and a support vector machine after transforming original PDUHF signals by using methods such as wavelet transform and fourier transform, and the like, so as to be used for classifying and identifying PDUHF signals. However, these methods either only focus on the frequency domain characteristics of the PDUHF signal or only focus on the time domain characteristics thereof, and ignore the dual effectiveness of the PDUHF signal in both frequency and time domain.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a classification identification method of partial discharge ultrahigh frequency signals, which considers the frequency domain and time domain characteristics of PDUHF signals and can effectively improve the identification accuracy of the PDUHF signals.
The invention solves the technical problems in the prior art by adopting the following technical scheme:
a classification identification method of partial discharge ultrahigh frequency signals comprises the following steps:
step 1, acquiring PDUHF signals caused by faults of different GIS defect types as sample data;
step 2, carrying out normalization processing on the acquired sample data;
step 3, performing signal decomposition on each sample data after normalization processing to obtain high-frequency and low-frequency data and performing dimensionality reduction processing;
step 4, dividing the sample data into training data and test data;
step 5, training an attention encoder model by using training data;
and 6, classifying and identifying the test data by using the trained attention encoder model.
Moreover, the GIS defect types comprise four types of point discharge, free particle discharge, suspension discharge and insulation discharge.
Further, the normalization processing in step 2 is to compress the discharge voltage of each sample data to be in the range of 0 to 1.
Moreover, the specific implementation method of step 3 is as follows: performing empirical mode decomposition on each sample data to obtain high-frequency signals C corresponding to N eigenmode functions1,i,C2,i,...,CN,iLow frequency signal C corresponding to 1 residual functionr,iA signal matrix [ C ] represented by N eigenmode functions by using a principal component analysis method1,i,C2,i,...,CN,i]Reducing dimension, extracting the first k eigenvectors with the largest eigenvalue to form an eigenvector matrix [ c1,i,c2,i,...,ck,i]And splicing signal vectors corresponding to the residual functions to form a characteristic matrix: xi＝[c1,i,c2,i,...,ck,i,Cr,i]TA feature matrix representing the ith sample data.
Moreover, the training data are the collected PDUHF signals with different defect types, and the test data are the field-detected PDUHF signals to be classified and identified.
Moreover, the attention coding model comprises a deformer coding layer and a classification identification layer connected with the deformer coding layer; the deformer coding layer comprises a multi-head attention module and an Add & Norm module thereof, a full-connection feedforward neural network module and an Add & Norm module thereof, the classification and identification layer comprises a full-connection feedforward neural network module and a Softmax function connected with the full-connection feedforward neural network module, the input end of the deformer coding layer receives a PDUHF signal characteristic matrix, and the classification and identification layer outputs a classification and identification result.
Moreover, the specific implementation method of step 5 includes the following steps:
the PDUHF signal characteristic matrix X is formediIntroducing into multi-head attention module and making Y ═ XiBy the formula
Is calculated to obtainj representations of attention, wherein
Attention h
Spliced matrix and corresponding
Result of matrix multiplication OM＝[H1,H2,...,Hh]WoAs the output of a multi-head attention module;
add using a multi-headed attention module&Norm module pair OMAnd XiThe addition results were normalized by Z-score to give OAN；
Subjecting O to condensationANInputting the data into a full-connection feedforward neural network module for mapping and coding to obtain OFN；
All fourFNAdd to fully connected feedforward neural network module&In Norm module, OFNAnd OANAdding the two layers and carrying out Z-score standardization to obtain an output O of a deformer coding layeren；
Fifthly, if the number of the deformer encoding layers is smaller than that of the deformer encoding layers, carrying out OenIn place of XiPerforming first step to step fifthly in a circulating mode as input of the encoding layer of the next deformer, otherwise, performing step sixteenth;
sixthly, outputting O of a deformer coding layerenAs the input of a classification identification layer, firstly, a fully connected feedforward neural network module carries out mapping processing to obtain a vector O with the output length of 4, wherein 4 elements in the vector correspond to 4 PDUHF signal types;
calculating a softmax function of the vector O to obtain a length-4 vector softmax (O), wherein each element respectively represents the input PDUHF signal XiInputting the probabilities of 4 different types of signals, and selecting the element p with the maximum probability value in softmax (O)iThe corresponding category is XiThe classification recognition result of (2);
and calculating the error between the recognition result and the actual class of the signal by using a cross entropy loss function, and learning model parameters by using a back propagation algorithm until the model converges to obtain a trained attention encoder model.
Furthermore, the number of layers of the deformer encoding layer is 6.
Moreover, the specific implementation method of step 6 is as follows: and inputting the PDUHF signal of the test data into the trained attention encoder model, and executing the steps from the step I to the step II to obtain a classification and identification result of the test data.
The invention has the advantages and positive effects that:
the invention has reasonable design, extracts the frequency domain and time domain characteristics of the partial discharge ultrahigh frequency signals in the gas insulated switchgear by adopting empirical mode decomposition and an attention mechanism, and further learns the characteristic representation capable of effectively classifying different types of partial discharge ultrahigh frequency signals from the frequency domain and time domain characteristics by utilizing an attention encoder model, thereby effectively improving the classification and identification accuracy of the partial discharge ultrahigh frequency signals of different fault types of the GIS, and being widely used for classifying and identifying PDUHF signals caused by different types of faults in the gas insulated switchgear.
Drawings
FIG. 1 is a process flow diagram of the present invention;
FIG. 2 is a diagram of an attention coding model structure according to the present invention.
Detailed Description
The embodiments of the present invention will be described in detail with reference to the accompanying drawings.
The invention provides a classification and identification method of partial discharge ultrahigh frequency signals based on empirical mode decomposition and attention mechanism, as shown in fig. 1, comprising the following steps:
step 1 (S1): and acquiring PDUHF signals caused by faults of different GIS defect types as sample data.
In the present embodiment, the PDUHF signal is classified according to the defect type: point discharge, free particle discharge, suspension discharge and insulation discharge.
Step 2 (S2): and carrying out normalization processing on the acquired sample data.
In the step, four types of sample data (original PDUHF signals) are preprocessed, and the discharge voltage of each sample data is normalized and scaled to be within a range of 0-1, so that the convergence speed in the model training process is increased, and the operation efficiency is improved.
Step 3 (S3): and performing signal decomposition on each sample data after normalization processing to obtain high-frequency and low-frequency data and performing dimensionality reduction processing.
In this step, Empirical Mode Decomposition (EMD) is performed on each sample data (normalized PDUHF signal) to obtain a high-frequency signal C corresponding to N eigenmode functions (IMF)1,i,C2,i,...,CN,iLow frequency signal C corresponding to 1 residual functionr,i. In order to reduce the operation complexity, a signal matrix [ C ] expressed by N IMF functions by using a Principal Component Analysis (PCA) method1,i,C2,i,...,CN,i]Reducing dimension, extracting the first k eigenvectors with the largest eigenvalue to form an eigenvector matrix [ c1,i,c2,i,...,ck,i]The signal vectors corresponding to the residual function on the splice together form a feature matrix, i.e. Xi＝[c1,i,c2,i,...,ck,i,Cr,i]TA feature matrix representing the ith sample data.
Step 4 (S4): the sample data is divided into training data and test data.
In this step, in order to verify that the trained classification model is valid, the sample data needs to be divided into two parts, namely training data and test data. The training data is used for training the model, so that the model learns the specific abstract characteristics of different types of PDUHF signals in the iterative training process and is used for classification and identification. The test data does not participate in the training of the model, and the trained model does not contain any information of the test data, so that the classification and identification effects of the model in the test data can be used for evaluating whether the trained model is effective or not. In the specific implementation, the training data refers to different types of collected PDUHF signals, and the test data refers to field-detected PDUHF signals to be classified and identified.
Step 5 (S5): the attention encoder model is trained using training data.
In this step, the feature matrix of the training sample is transmitted to the attention coding model as an input for training the model. As shown in fig. 2, the attention coding model includes: a deformer encoding layer and a classification identification layer connected thereto, in this embodiment, the deformer encoding layer is a 6-layer deformer encoding layer. The input end of the deformer coding layer receives a PDUHF signal characteristic matrix
And the output end of the deformer coding layer outputs a coding result to the input end of the deformer coding layer and the input end of the classification and identification layer, and the output end of the classification and identification layer outputs a classification and identification result. Wherein the deformer encoding layer comprises a multi-headed attention module and its Add&Norm module, full-connection feedforward neural network module and Add thereof&Norm module, PDUHF signal feature matrix
(where d + k +1 represents signals of different frequency domain characteristics and l is the signal length) to a multi-head attention module and its Add&Among Norm modules, the multi-head attention module and Add thereof&Output of Norm module to full-connection feedforward neural network module and Add thereof&Norm module. The classification identification layer comprises a fully-connected feedforward neural network module and a Softmax function connected with the fully-connected feedforward neural network module.
The specific implementation method of the step comprises the following steps:
in step 501(S501), a PDUHF signal feature matrix X is formediIntroducing into multi-head attention module and making Y ═ XiBy the formula
Calculating to obtain j attention expression
Attention h
Spliced matrix and corresponding
Result of matrix multiplication OM＝[H1,H2,...,Hh]WoAs the output of a multi-head attention module.
Step 502(S502) Add Using Multi-head attention Module&Norm module pair OMAnd XiThe addition results were normalized by Z-score to give OAN。
In step 503(S503), O is addedANInputting the data into a full-connection feedforward neural network module for mapping and coding to obtain OFN。
In step 504(S504), O is addedFNAdd to fully connected feedforward neural network module&In Norm module, OFNAnd OANAdding the two layers and carrying out Z-score standardization to obtain an output O of a deformer coding layeren。
Step 505(S505) of comparing O if the deformer encoding layer number is less than 6enIn place of XiS501, S502, S503, S504, S505 are performed as input of the next deformer encoding layer, otherwise, S506 is performed.
Step 506(S506) of encoding the output O of the 6-layer deformer encoding layerenAs the input of the classification identification layer, firstly, the fully connected feedforward neural network module carries out mapping processing to obtain a vector O with the output length of 4, and 4 elements in the vector correspond to 4 PDUHF signal types.
In step 507(S507), a softmax function is performed on the vector O to obtain a vector softmax (O) having a length of 4, each element representing the inputted PDUHF signal XiInputting the probabilities of 4 different types of signals, and selecting the element p with the maximum probability value in softmax (O)iThe corresponding category is XiThe classification recognition result of (2).
And step 508, calculating the error between the recognition result and the actual class of the signal by using a cross entropy loss function, and learning model parameters by using a back propagation algorithm until the model converges to obtain a trained attention encoder model.
Step 6 (S6): and (3) carrying out classification identification on the test data (PDUHF signals to be identified) by using the trained attention encoder model.
In this step, test data PDUHF signal X is appliedtInputting the data into a trained attention encoder model, and obtaining test data X according to the steps of S501, S502, S503, S504, S505, S506 and S507tThe classification recognition result of (2).
A specific example is given below to illustrate the process of the present invention:
s1: 100000 data are randomly selected from the collected PDUHF signals to form a sample data set, and the PDUHF signals in the sample data set and the corresponding categories thereof are known, namely: the PDUHF signal belongs to and only belongs to one of the four types. Each PDUHF signal is expressed as
Where l is the signal length.
S2: to pair
According to the formula
Carrying out normalization processing to obtain Si。
S3: decomposing each sample data S by EMD algorithmiObtaining high-frequency signals C corresponding to N (N may vary due to different signal types) IMFs1,i,C2,i,...,CN,iLow frequency signal C corresponding to 1 residual functionr,i. Using PCA to signal matrix [ C ]1,i,C2,i,...,CN,i]Dimension reduction is performed, and the first k (k is 3 in this example) eigenvectors with the largest eigenvalue are extracted to form an eigenvector matrix [ c1,i,c2,i,c3,i]The signal vectors corresponding to the residual function on the splice, together forming a feature matrix, i.e. Xi＝[c1,i,c2,i,c3,i,Cr,i]TA feature matrix representing the ith sample data, wherein
S4: 80000 data are randomly selected from the sample data set as training data, and the remaining 20000 data are used as test data.
S5: the feature matrix X of each training sampleiInputting the information into an attention coding model, and training the model.
S501：XiThe multi-head attention module is transmitted into the attention coding layer of the layer 1, and Y is made to be XiBy the formula
Calculating to obtain j attention expression
In this example we have a total of 6 attention representations: h1,H2,...,H6. The matrix after splicing the 6 attention points and the corresponding
Result of matrix multiplication OM＝[H1,H2,...,Hh]WoAs an output of multi-headed attention.
S502：Add&Norm module pair according to formula OAN＝((OM+Xi) -mu)/sigma is obtained after Z-score standardization
S503：OANInputting into a fully-connected feedforward neural network module with a 3-layer structure, wherein the number of neurons in the 1 st layer of the network is 4, the number of neurons in the 2 nd layer is 10, the number of neurons in the 3 rd layer is 4, and activatingThe function is a ReLU activation function. Mapping the fully-connected feedforward neural network module to obtain OFN。
S504：OFNInput to another Add&Norm module according to formula Oen＝((OFN+OAN) Z-score normalization to give Oen。
S505: if the number of deformer encoding layers is less than 6, then OenIn place of XiS501, S502, S503, S504, S505 are performed as input of the next deformer encoding layer, otherwise, S506 is performed.
S506: encoding the output O of the 6-layer deformer layerenThe method is characterized in that the method is used as the input of a classification identification layer, firstly, the method is input into a full-connection feedforward neural network module, the network structure is that the number of neurons in a layer 1 is 4, the number of neurons in a layer 2 is 20, the number of neurons in a layer 3 is 4, an activation function is a ReLU activation function, and a vector O with the output length of 4 is obtained through mapping processing, wherein 4 elements in O correspond to 4 PDUHF signal types.
S507: performing softmax function calculation on the O to obtain a vector softmax (O) with the length of 4, wherein each element respectively represents the input PDUHF signal XiInputting the probabilities of 4 different types of signals, and selecting the element p with the maximum probability value in softmax (O)iThe corresponding category is XiThe classification recognition result of (2).
And S508, calculating the error between the recognition result and the actual class of the signal by using a cross entropy loss function, setting the batch processing size to be 64, setting the dropout rate to be 10%, and optimizing the model parameters by using an SGD (generalized regression) optimization algorithm with the learning rate of 0.001 until the model converges to obtain the trained attention encoder model.
S6 testing data PDUHF signal XtInputting the data into a trained attention encoder model, executing the budgets of the steps 501, S502, S503, S504, S505, S506 and S507 to obtain test data XtThe classification recognition result of (2).
Through the steps, the function of classifying and identifying the partial discharge ultrahigh frequency signals based on empirical mode decomposition and attention mechanism can be realized, and through actual measurement, the accuracy of classification and identification can be effectively improved.
It should be emphasized that the embodiments described herein are illustrative rather than restrictive, and thus the present invention is not limited to the embodiments described in the detailed description, but also includes other embodiments that can be derived from the technical solutions of the present invention by those skilled in the art.
Claims (9)
1. A classification identification method of partial discharge ultrahigh frequency signals is characterized in that: the method comprises the following steps:
step 1, acquiring PDUHF signals caused by faults of different GIS defect types as sample data;
step 2, carrying out normalization processing on the acquired sample data;
step 3, performing signal decomposition on each sample data after normalization processing to obtain high-frequency and low-frequency data and performing dimensionality reduction processing;
step 4, dividing the sample data into training data and test data;
step 5, training an attention encoder model by using training data;
and 6, classifying and identifying the test data by using the trained attention encoder model.
2. The method for classifying and identifying partial discharge ultrahigh frequency signals according to claim 1, wherein: the GIS defect types comprise point discharge, free particle discharge, suspension discharge and insulation discharge.
3. The method for classifying and identifying partial discharge ultrahigh frequency signals according to claim 1, wherein: the normalization process in step 2 is to compress the discharge voltage of each sample data to be in the range of 0 to 1.
4. The method for classifying and identifying partial discharge ultrahigh frequency signals according to claim 1, wherein: the tool of the step 3The implementation method comprises the following steps: performing empirical mode decomposition on each sample data to obtain high-frequency signals C corresponding to N eigenmode functions1,i,C2,i,...,CN,iLow frequency signal C corresponding to 1 residual functionr,iA signal matrix [ C ] represented by N eigenmode functions by using a principal component analysis method1,i,C2,i,...,CN,i]Reducing dimension, extracting the first k eigenvectors with the largest eigenvalue to form an eigenvector matrix [ c1,i,c2,i,...,ck,i]And splicing signal vectors corresponding to the residual functions to form a characteristic matrix: xi＝[c1,i,c2,i,...,ck,i,Cr,i]TA feature matrix representing the ith sample data.
5. The method for classifying and identifying partial discharge ultrahigh frequency signals according to claim 1, wherein: the training data are the collected PDUHF signals with different defect types, and the test data are the field-detected PDUHF signals to be classified and identified.
6. The method for classifying and identifying partial discharge ultrahigh frequency signals according to claim 1, wherein: the attention coding model comprises a deformer coding layer and a classification identification layer connected with the deformer coding layer; the deformer coding layer comprises a multi-head attention module and an Add & Norm module thereof, a full-connection feedforward neural network module and an Add & Norm module thereof, the classification and identification layer comprises a full-connection feedforward neural network module and a Softmax function connected with the full-connection feedforward neural network module, the input end of the deformer coding layer receives a PDUHF signal characteristic matrix, and the classification and identification layer outputs a classification and identification result.
7. The method for classifying and identifying partial discharge ultrahigh frequency signals according to claim 6, wherein: the specific implementation method of the step 5 comprises the following steps:
the PDUHF signal characteristic matrix X is formediIntroducing into multi-head attention module and making Y ═ XiBy the formula
Calculating to obtain j attention expression
Attention h
Spliced matrix and corresponding
Result of matrix multiplication OM＝[H1,H2,...,Hh]WoAs the output of a multi-head attention module;
add using a multi-headed attention module&Norm module pair OMAnd XiThe addition results were normalized by Z-score to give OAN；
Subjecting O to condensationANInputting the data into a full-connection feedforward neural network module for mapping and coding to obtain OFN；
All fourFNAdd to fully connected feedforward neural network module&In Norm module, OFNAnd OANAdding the two layers and carrying out Z-score standardization to obtain an output O of a deformer coding layeren；
Fifthly, if the number of the deformer encoding layers is smaller than that of the deformer encoding layers, carrying out OenIn place of XiPerforming first step to step fifthly in a circulating mode as input of the encoding layer of the next deformer, otherwise, performing step sixteenth;
sixthly, outputting O of a deformer coding layerenAs the input of a classification identification layer, firstly, a fully connected feedforward neural network module carries out mapping processing to obtain a vector O with the output length of 4, wherein 4 elements in the vector correspond to 4 PDUHF signal types;
calculating a softmax function of the vector O to obtain a vector softmax (O) with the length of 4, wherein each element is respectivelyPDUHF signal X representing an inputiInputting the probabilities of 4 different types of signals, and selecting the element p with the maximum probability value in softmax (O)iThe corresponding category is XiThe classification recognition result of (2);
and calculating the error between the recognition result and the actual class of the signal by using a cross entropy loss function, and learning model parameters by using a back propagation algorithm until the model converges to obtain a trained attention encoder model.
8. The method for classification and identification of partial discharge uhf signals according to claim 6 or 7, wherein: the number of layers of the deformer encoding layer is 6.
9. The method for classifying and identifying partial discharge ultrahigh frequency signals according to claim 7, wherein: the specific implementation method of the step 6 comprises the following steps: and inputting the PDUHF signal of the test data into the trained attention encoder model, and executing the steps from the step I to the step II to obtain a classification and identification result of the test data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110917715.3A CN113884822A (en) | 2021-08-11 | 2021-08-11 | Classification and identification method for partial discharge ultrahigh frequency signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110917715.3A CN113884822A (en) | 2021-08-11 | 2021-08-11 | Classification and identification method for partial discharge ultrahigh frequency signals |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113884822A true CN113884822A (en) | 2022-01-04 |
Family
ID=79010944
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110917715.3A Pending CN113884822A (en) | 2021-08-11 | 2021-08-11 | Classification and identification method for partial discharge ultrahigh frequency signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113884822A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116304979A (en) * | 2023-03-02 | 2023-06-23 | 兰州交通大学 | Attention mechanism-based multi-feature fusion partial discharge type identification method |
-
2021
- 2021-08-11 CN CN202110917715.3A patent/CN113884822A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116304979A (en) * | 2023-03-02 | 2023-06-23 | 兰州交通大学 | Attention mechanism-based multi-feature fusion partial discharge type identification method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11409933B2 (en) | Method for diagnosing analog circuit fault based on cross wavelet features | |
| CN106443316B (en) | Multi-information detection method and device for deformation state of power transformer winding | |
| CN110287983B (en) | Single-classifier anomaly detection method based on maximum correlation entropy deep neural network | |
| CN110726898B (en) | Power distribution network fault type identification method | |
| CN110070102B (en) | Method for establishing sequence-to-sequence model for identifying power quality disturbance type | |
| CN112926728B (en) | Small sample turn-to-turn short circuit fault diagnosis method for permanent magnet synchronous motor | |
| CN113673346A (en) | Motor vibration data processing and state recognition method based on multi-scale SE-Resnet | |
| CN116578940A (en) | Bearing fault new type identification and diagnosis method based on mixed depth self-coding | |
| CN112596016A (en) | Transformer fault diagnosis method based on integration of multiple one-dimensional convolutional neural networks | |
| CN113884822A (en) | Classification and identification method for partial discharge ultrahigh frequency signals | |
| CN116089857A (en) | Transformer fault identification method based on CEEMDAN-DBN | |
| CN112305379A (en) | Mode identification method and system for GIS insulation defect | |
| CN114861778A (en) | Method for rapidly classifying rolling bearing states under different loads by improving width transfer learning | |
| CN113112039A (en) | Active power distribution system initial fault identification method based on time-frequency memory recurrent neural network | |
| CN111275109A (en) | Power equipment state data characteristic optimization method and system based on self-encoder | |
| CN112285632A (en) | Electromagnetic current transformer fault diagnosis method based on VMD and sample entropy | |
| CN115310837A (en) | Complex electromechanical system fault detection method based on causal graph attention neural network | |
| CN114139577A (en) | Fault arc identification method and device based on one-dimensional convolution self-coding network | |
| CN110631827B (en) | Gearbox fault diagnosis method based on improved collaborative representation classification | |
| CN113361631A (en) | Insulator aging spectrum classification method based on transfer learning | |
| Tian et al. | Fault Diagnosis for Rolling Bearings Based on Recurrence Plot and Convolutional Neural Network | |
| CN117892117B (en) | Fault positioning method and system for power transmission line of power distribution network | |
| Wang et al. | An Automatic Identification Framework for Complex Power Quality Disturbances Based on Ensemble CNN | |
| CN112240964B (en) | Method for identifying fault type of power distribution network | |
| Prabhu et al. | Deep Learning Based Power Transformer Monitoring Using Partial Discharge Patterns. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |