CN115062538A - Converter fault diagnosis method and system based on attention mechanism and convolution NN - Google Patents
Converter fault diagnosis method and system based on attention mechanism and convolution NN Download PDFInfo
- Publication number
- CN115062538A CN115062538A CN202210639082.9A CN202210639082A CN115062538A CN 115062538 A CN115062538 A CN 115062538A CN 202210639082 A CN202210639082 A CN 202210639082A CN 115062538 A CN115062538 A CN 115062538A
- Authority
- CN
- China
- Prior art keywords
- fault
- feature
- attention mechanism
- branch
- diagnosis
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- 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
- G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
- G01R31/54—Testing for continuity
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Tests Of Electronic Circuits (AREA)
Abstract
The invention discloses a converter fault diagnosis method and system based on attention mechanism and convolution NN, comprising the following steps: establishing a simulation model of the double-active-bridge converter, and selecting a fault diagnosis signal; collecting fault diagnosis signal samples of different power switching devices; a feature extraction module is constructed by utilizing convolution operation, a de-noising module and a feature fusion module are constructed by utilizing an attention mechanism, fault features of three diagnosis signals in an input diagnosis sample are extracted and fused, and a multi-branch convolution neural network model is constructed; and inputting the fault diagnosis signal sample into a multi-branch convolutional neural network for training, and diagnosing and positioning the open-circuit fault of the double-active-bridge converter according to the trained network. The invention utilizes the attention mechanism to construct a fault diagnosis model, removes noise in diagnosis signals, integrates fault characteristics of multiple diagnosis signals, effectively improves the diagnosis accuracy, and utilizes the convolutional neural network to diagnose, thereby improving the intellectualization of diagnosis.
Description
Technical Field
The invention relates to the field of power electronic converter fault diagnosis, in particular to a converter fault diagnosis method and system based on an attention mechanism and convolution NN.
Background
The dual active bridge converter has the characteristics of high power density, bidirectional energy flow, soft switching and electric energy isolation, and the reliability of the dual active bridge converter is very important in electric energy conversion and power system transmission systems. The fault of the double-active-bridge converter is mainly the fault of a power switching device, the fault of the power switching device is divided into a short-circuit fault and an open-circuit fault, and the short-circuit fault is relatively easy to monitor, so that the fault diagnosis of the double-active-bridge converter at present is mainly the open-circuit fault of the power switching device. The method for diagnosing the open-circuit fault of the power switching device is mainly divided into a model-based method and a data-driven method. Model-based approaches require the accurate mathematical model of the converter to be built, but the crosstalk and contribution between power switches to date makes it difficult to build an accurate mathematical model. The data-driven method is not used for establishing a mathematical model, but is used for mining the corresponding relation between monitoring data and fault types through machine learning.
At present, a deep learning algorithm is widely applied to fault diagnosis, however, the current algorithm is mostly used for processing a single input signal sample, the dual-active-bridge converter belongs to the problem of multi-signal fault diagnosis, and the above deep learning model is not suitable for processing the problem of fault diagnosis of the dual-active-bridge converter. And the key of the multi-signal fault diagnosis problem is to fully utilize and learn useful characteristics of a plurality of diagnostic signals, so that the research on a characteristic fusion technology is necessary. Fault features contained in the diagnosis signals have certain redundant features and can interfere fault diagnosis classification, and a common feature fusion algorithm gives the same importance weight to a plurality of feature graphs and ignores the importance degree of different features. Therefore, how to effectively fuse the extracted features according to the importance of different features is very important for improving the fault diagnosis performance. In addition, the presence of a large amount of noise in the monitored diagnostic signal also affects the accuracy of the fault diagnosis.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a converter fault diagnosis method and system based on attention mechanism and convolution NN, which can process a plurality of diagnostic signals, eliminate noise characteristics in the signals, effectively fuse the fault characteristics of the plurality of diagnostic signals after denoising, and improve the classification accuracy of the network. And the accuracy of fault diagnosis of the double-active-bridge converter is improved by utilizing a deep learning algorithm.
The technical scheme adopted by the invention for solving the technical problems is as follows:
the invention provides a converter fault diagnosis method based on an attention mechanism and convolution NN, which comprises the following steps:
and 6, inputting the data to be diagnosed into the trained multi-branch convolutional neural network fault diagnosis model and outputting a final classification result, thereby realizing fault diagnosis of the double-active-bridge converter.
Further, the method for constructing the original data sample set in step 1 of the present invention comprises:
the original data sample set is: DATA i ={A i,1 ,A i,2 ,...,A i,j ,...,A i,N }i∈[1,3]Where "3" indicates three diagnostic signals in the raw data set, A i,j For the monitoring value corresponding to the jth point in the ith diagnostic signal monitoring data, j belongs to [1, N ∈]N is the total number of monitoring data points; the combination of the three diagnostic signal monitoring data corresponds to a power switch tube fault state of the double-active-bridge converter.
Further, the method in step 2 of the present invention specifically comprises:
when a multi-branch feature extraction module is constructed by utilizing convolution operation, the fault features of the diagnosis signals are directly extracted by utilizing one-dimensional convolution operation; meanwhile, batch standardization, ReLU activation function and maximum pooling operation are carried out after each one-dimensional convolution operation, so that overfitting, gradient disappearance or gradient explosion of the model are prevented; the calculation formula of the convolution operation is as follows:wherein x represents the operation of convolution,andrespectively of the ith channel in the ith and l-1 th layers of the profile X,is the convolution kernel between the ith channel and the jth channel of the ith layer of the feature map X;is an offset, M j Representing an input feature set; the formula for the ReLU activation function is:wherein x represents each feature point of the input feature map; the maximum pooling operation is calculated asWherein S represents the size of the sliding window for pooling operation, y j Representing the characteristics of the ith channel in the output characteristic diagram.
Further, the method in step 3 of the present invention specifically includes:
constructing a denoising module by combining an attention mechanism and a soft threshold function, and extracting noise features in the feature map by a feature extraction module, wherein the attention mechanism can acquire an extracted original feature map M 0 Is equal to [ tau ] channel threshold 1 ,τ 2 ,...,τ C ]Inputting the extracted characteristic diagram and the channel threshold value into a soft threshold value function for processing, and being capable of adaptively removing the noise characteristic in the extracted characteristic diagram to obtain a reconstructed characteristic diagram M 1 (ii) a Wherein, the soft threshold value calculation formula is as follows:
wherein, the first and the second end of the pipe are connected with each other,representation feature map M 0 And (4) adding a denoising module to each convolution branch according to the characteristics of the ith channel to remove noise in the original diagnostic signal data to obtain a denoising characteristic map.
Further, the method in step 4 of the present invention specifically comprises:
constructing a feature fusion module through an attention mechanism, learning and extracting the importance of different channel features of the feature graph, and fusing a plurality of feature graphs according to the channel feature importance of the feature graph; m is a group of 1 ,M 2 And M 3 For three fused feature maps, M is obtained by the attention mechanism 1 ,M 2 And M 3 Corresponding channel feature importance sets ρ, ω andrespectively multiplying the channel feature importance set and the corresponding de-noising feature map set, and then adding to obtain a fused feature map, wherein the calculation formula is as follows:
where ρ is i ,ω i Andrespectively represent ρ, ω andthe importance of the ith channel characteristic in (ii),andcharacteristic diagram M of three branch channels respectively 1 ,M 2 And M 3 The characteristics of the ith channel.
Further, the method in step 5 of the present invention specifically includes:
the constructed multi-branch convolutional neural network fault diagnosis is directly used for fault diagnosis of a plurality of diagnostic signals, the plurality of diagnostic signals are respectively input into different convolutional branches, a feature extraction module of each convolutional branch extracts fault features of each diagnostic signal, a denoising module in each convolutional branch removes noise information in a corresponding fault feature map, a feature fusion module fuses the denoised feature maps of the plurality of convolutional branches, and the obtained fusion feature maps are input into a classifier for classification after being subjected to full connection and activation function processing.
The invention provides a converter fault diagnosis system based on an attention mechanism and convolution NN, which comprises:
the data acquisition module is used for establishing a simulation model of the double-active-bridge converter, selecting leakage inductance current of each power switch tube in a fault state, acquiring an original data sample set by taking the midpoint voltage of any bridge arm of the primary bridge and the midpoint voltage of any bridge arm of the secondary bridge as diagnosis and diagnosis signals, and performing classified encoding on the fault state of the double-active-bridge converter according to the position of a power switch device with a fault;
the fault feature extraction module is used for constructing a multi-branch feature extraction module by utilizing convolution operation, inputting a plurality of diagnostic signals in a sample into each branch respectively, and extracting fault features of the diagnostic signals to obtain a fault feature map;
the denoising module is constructed by combining an attention mechanism with a soft threshold function, and the diagnostic signal features extracted by the multi-branch feature extraction module are input into the denoising module to eliminate the noise features in the diagnostic signal, so that a denoising feature map is obtained;
the feature fusion module is constructed by utilizing an attention mechanism, and the multi-branch denoising feature maps are fused according to the channel importance of each feature map obtained by the attention mechanism to obtain a fusion feature map;
the network training module is used for dividing an original data sample set into a training set and a verification set, the training set trains the multi-branch convolutional neural network fault diagnosis model based on the attention mechanism, and the verification set verifies the trained fault diagnosis model;
and the diagnosis module is used for directly inputting newly acquired double-active-bridge converter fault diagnosis data into a trained network for diagnosis and fault positioning in the later diagnosis process.
The invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of any of the methods described above.
In general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:
the method comprises the steps of collecting three signals of leakage inductance current, midpoint voltage of any bridge arm of a primary bridge and midpoint voltage of any bridge arm of a secondary bridge of the double-active-bridge converter under the open-circuit faults and normal modes of different power switching devices as an original data set, and performing classified coding on the fault states of the double-active-bridge converter according to the positions of the power switching devices with faults; constructing a feature extraction module by using convolution operation, constructing a denoising module by using an attention mechanism and combining a soft threshold function, eliminating noise features, and constructing a feature fusion module by using the attention mechanism; and constructing a multi-branch convolutional neural network by utilizing the feature extraction module, the denoising module and the feature fusion module. Training the multi-branch convolutional neural network based on the attention mechanism by adopting an original data set, repeating the steps aiming at the double-active-bridge converter to be tested to obtain the original data set of the double-active-bridge converter to be tested, and obtaining a fault diagnosis result by utilizing the trained multi-branch convolutional neural network. The multi-branch convolution neural network constructed by the invention utilizes the module constructed by the attention mechanism to remove noise in the diagnosis signal and fuse the fault characteristics of the multi-diagnosis signal, thereby greatly improving the diagnosis accuracy of the double-active-bridge converter.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a schematic flow chart of a method provided by an embodiment of the present invention;
FIG. 2 is a simulated topology diagram of a dual active bridge converter according to an embodiment of the present invention;
FIG. 3 is a multi-branch convolutional neural network structure based on attention mechanism according to an embodiment of the present invention;
FIG. 4 is a visualization of a feature after fusing features provided by embodiments of the present invention;
FIG. 5 is a block diagram of a multi-branch convolutional neural network and classification using a softmax classifier provided by an embodiment of the present invention;
FIG. 6 is a graph illustrating accuracy variations in a training and testing process for a dual active bridge converter according to an embodiment of the present invention;
fig. 7 is a graph illustrating a variation of a loss function in a training and testing process of a dual active bridge converter according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
Example one
Fig. 1 is a schematic flow chart of a method provided by an embodiment of the present invention, where the method shown in fig. 1 includes the following steps:
(1) establishing a simulation model of the double-active-bridge converter, selecting leakage inductance current, midpoint voltage of any bridge arm of a primary bridge and midpoint voltage of any bridge arm of a secondary bridge as fault diagnosis signals, classifying and coding fault states of the double-active-bridge converter according to the position of a power switch device with a fault, and collecting diagnosis signals of the converter under different open-circuit faults as an original data sample set;
(2) constructing a multi-branch feature extraction module by utilizing convolution operation, respectively inputting a plurality of diagnostic signals in a sample into each branch, and extracting fault features of the diagnostic signals to obtain a fault feature map;
(3) constructing a denoising module by combining an attention mechanism with a soft threshold function, inputting the diagnostic signal characteristics extracted by the multi-branch characteristic extraction module into the denoising module to eliminate the noise characteristics in the diagnostic signal, and obtaining a denoising characteristic map;
(4) and constructing a feature fusion module by using an attention mechanism, and fusing the multi-branch denoising feature map according to the channel importance of each feature map obtained by the attention mechanism to obtain a fusion feature map.
(5) And inputting the obtained characteristic diagram into a classifier, and constructing a multi-branch convolutional neural network fault diagnosis model. The collected raw data set is divided into a training set and a test set. And training the multi-branch convolutional neural network fault diagnosis model by using a training set, and testing the trained fault diagnosis model by using a test set.
(6) And inputting the data to be diagnosed into the trained network and outputting a final classification result to realize fault diagnosis of the double-active-bridge converter.
In this embodiment, the step (1) may be implemented as follows:
the circuit simulation topological diagram of the double-active-bridge converter is shown in fig. 2, and is classified according to the open-circuit fault of a single power tube in actual operation, wherein the double-active-bridge converter comprises 9 fault states including a normal state and an S state 1 Open circuit fault, S 2 Open circuit fault, S 3 Open circuit fault, S 4 Open circuit fault, Q 1 Open circuit fault, Q 2 Open circuit fault, Q 3 Open circuit fault sum Q 4 Open circuit failure. The fault category, i.e. the code, is shown in table 1. And selecting a primary bridge S with a leakage inductance current 1 /S 2 Bridge arm middle point voltage and secondary bridge Q 1 /Q 2 The midpoint voltage of the bridge arm is a fault diagnosis signal, 1000 samples are collected under each fault state respectively, 9000 data samples can be obtained, each sample comprises three diagnosis signal data, and each diagnosis signal data has 1000 points;
table 1: fault status and coding
In this embodiment, the step (2) can be implemented as follows:
and setting three convolution characteristic extraction branches according to the number of the diagnostic signals, setting two one-dimensional convolution layers in each branch to extract the characteristics of the diagnostic signals, and setting the sizes of convolution kernels to be 64 and 3 respectively. And simultaneously performing batch normalization, a ReLU activation function and a maximum pooling operation after each convolution operation, wherein the calculation formula of the convolution operation is as follows:wherein x represents the operation of convolution,andrespectively of the ith channel in the ith and l-1 th layers of the profile X,is the convolution kernel between the ith and jth channels of the ith layer of the feature map X.Is an offset, M j Representing an input feature set. The formula for the ReLU activation function is:where x represents each feature point of the input feature map. The maximum pooling operation is calculated asWherein S represents the size of the sliding window for pooling operation, y j Representing the characteristics of the ith channel in the output characteristic diagram. The size of the feature map extracted by each branch after feature extraction is 32 × 250 × 1, wherein "32" indicates that the number of channels of the feature map is 32, "250" indicates that the width of the feature map is 250, and "1" indicates that the height of the feature map is 1.
In this embodiment, the step (3) may be implemented as follows:
in order to effectively eliminate noise features irrelevant to fault classification in a diagnosis signal, a denoising module is constructed by combining an attention mechanism and a soft threshold function to eliminate the noise features in a feature map of the fault diagnosis signal extracted by a feature extraction module, wherein an attention mechanism structural diagram is shown in fig. 3, C, W and H are respectively the number, width and length of channels of the feature map, and α is the importance of the obtained channels. Wherein the attention deviceThe extracted original feature map M can be obtained by performing a global average pooling, two convolutions, a batch normalization, a ReLU activation and a Sigmoid operation on the input feature map 0 The channel importance of (2) is multiplied by the input feature map to obtain the denoising threshold value tau [ tau ] of each channel 1 ,τ 2 ,...,τ C ]And C ═ 16. Inputting the extracted feature map and the channel threshold into a soft threshold function for processing, and adaptively removing the noise features in the extracted feature map to obtain a reconstructed feature map M 1 . Wherein, the soft threshold value calculation formula is as follows:
wherein the content of the first and second substances,representation feature map M 0 The ith channel. And respectively using a denoising module for each branch to obtain a denoising characteristic map. Each branch feature map size after denoising was 32 × 250 × 1.
In the embodiment of the present invention, other models may be used for the attention mechanism, and the embodiment of the present invention is not limited uniquely.
In this embodiment, the step (4) may be implemented as follows:
by constructing the adaptive feature fusion module through the attention mechanism, a plurality of feature maps can be fused according to the channel feature importance of the feature maps, wherein the attention mechanism structure diagram is shown in fig. 3, C, W and H are the channel number, width and length of the feature maps respectively, and α is the acquired channel importance. With three characteristic maps M 1 ,M 2 And M 3 For example, the attention mechanism may obtain M 1 ,M 2 And M 3 Corresponding channel feature importance sets ρ, ω andchannel feature importance set and corresponding denoising feature atlasThe feature maps after fusion are obtained by adding the products after multiplication respectively, and the calculation formula is as follows:
where ρ is i ,ω i Andrespectively represent ρ, ω andthe importance of the ith channel characteristic in (ii),andcharacteristic diagram M of three branch channels respectively 1 ,M 2 And M 3 The ith channel is characterized by C-16. The size of the fused feature map after feature fusion was 32 x 250 x 1. The visualization result of the features after the features are fused is shown in fig. 4, and better separability is shown among the features.
In the embodiment of the present invention, other models may be used for the attention mechanism, and the embodiment of the present invention is not limited uniquely.
In this embodiment, in step (5), the advantages of learning the fault features and the corresponding fault categories by deep learning are utilized, and fault feature features, adaptive denoising, adaptive feature fusion and intelligent fault state identification of the multi-branch convolutional neural network based on the attention mechanism are completed. The whole learning process of the multi-branch convolutional neural network mainly comprises the following steps: initializing parameters of a network; the input data is transmitted forward through each operation layer to obtain an output value; solving an error between an output value of the network and a target value; and when the error is larger than the expected value, the error is transmitted back to the network, and the errors of the operation layers are sequentially obtained. When the error is equal to or less than the desired value, the training is ended.
Specifically, the step (5) may be implemented by:
(5.1) fault classification of a dual active bridge converter comprises, step 1: dividing an original data sample set into a training sample set and a test sample set according to the proportion of 7: 3; step 2: a multi-branch convolutional neural network is constructed by utilizing a feature extraction module, a denoising module and a feature fusion module, and is classified by using a softmax classifier, and the structure is shown in FIG. 5. Initializing network parameters, training and testing each data sample by using the constructed network as shown in table 2, and step 3: calculating the classification accuracy and giving a fault diagnosis result;
table 2: diagnostic model hyper-parameter settings
(5.2) accuracy and loss function changes of the training and testing process of the double-active-bridge converter are shown in fig. 6 and 7, and the training effect of the double-active-bridge converter fault diagnosis method is good. The results of the fault classification in the case of noise are shown in table 2, where the noise content in the diagnostic signal is represented by the signal-to-noise ratio, with a higher signal-to-noise ratio indicating a lower noise content. Compared with the fault classification results of the mainstream classification algorithm one-dimensional convolutional neural network 1-D CNN and the long-and-short time memory network LSTM, the method for diagnosing the double-active-bridge converter fault is proved to be advanced.
TABLE 2 comparison of the results of the classification of the double-active-bridge converter faults
Signal to noise ratio/dB | 1-D CNN/% | LSTM/% | Method as mentioned/%) |
-2 | 77.23±2.12 | 79.31±2.58 | 97.22±0.54 |
0 | 85.19±1.69 | 86.20±1.56 | 98.34±0.43 |
2 | 92.54±0.89 | 90.64±1.11 | 99.17±0.65 |
4 | 95.76±0.63 | 94.28±0.93 | 99.36±0.34 |
6 | 96.82±0.69 | 95.67±0.48 | 99.67±0.23 |
8 | 97.37±0.67 | 96.98±0.53 | 99.81±0.13 |
The invention relates to a converter fault diagnosis method based on an attention mechanism and convolution NN, which comprises the steps of constructing a feature extraction module by utilizing convolution operation, constructing a denoising module by utilizing the attention mechanism and a soft threshold function, constructing a feature fusion module by utilizing the attention mechanism, extracting fault features of a plurality of diagnosis signals, eliminating fault samples in the diagnosis signals, fusing the denoised important features of the diagnosis signals into a feature map, and fully extracting the important features of the fault diagnosis signals related to fault classification. The method can improve the feature extraction capability of the network on multiple fault diagnosis signals and the anti-noise capability of the network, and improves the fault diagnosis accuracy of the double-active-bridge converter.
Example two
In another embodiment of the present invention, there is also provided an attention mechanism and convolution NN based transformer fault diagnosis system, including:
the data acquisition module is used for establishing a simulation model of the double-active-bridge converter, selecting leakage inductance current of each power switch tube in a fault state, acquiring an original data set by taking the midpoint voltage of any bridge arm of the primary bridge and the midpoint voltage of any bridge arm of the secondary bridge as diagnosis and diagnosis signals, and classifying and coding the fault state of the double-active-bridge converter according to the position of a power switch device with a fault;
the fault feature extraction module is used for constructing a multi-branch feature extraction module by utilizing convolution operation, inputting a plurality of diagnostic signals in a sample into each branch respectively, and extracting fault features of the diagnostic signals to obtain a fault feature map;
the denoising module is constructed by combining an attention mechanism with a soft threshold function, and the diagnostic signal features extracted by the multi-branch feature extraction module are input into the denoising module to eliminate the noise features in the diagnostic signal, so that a denoising feature map is obtained;
and the feature fusion module is constructed by utilizing an attention mechanism, and the multi-branch denoising feature map is fused according to the channel importance of each feature map obtained by the attention mechanism to obtain a fusion feature map.
The network training module is used for dividing the original data set into a training set and a verification set, the training set trains the multi-branch convolutional neural network fault diagnosis model based on the attention mechanism, and the verification set verifies the trained fault diagnosis model;
and the diagnosis module is used for directly inputting newly acquired double-active-bridge converter fault diagnosis data into a trained network for diagnosis and fault positioning in the later diagnosis process.
The specific implementation of each module may refer to the description of the above method embodiment, and this embodiment will not be repeated.
It should be noted that, according to the implementation requirement, each step/component described in the present application can be divided into more steps/components, and two or more steps/components or partial operations of the steps/components can be combined into new steps/components to achieve the purpose of the present invention.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (8)
1. A method for diagnosing transducer faults based on attention mechanism and convolution NN, the method comprising the steps of:
step 1, establishing a simulation model of a double-active-bridge converter, selecting leakage inductance current, midpoint voltage of any bridge arm of a primary bridge and midpoint voltage of any bridge arm of a secondary bridge as fault diagnosis signals, classifying and coding fault states of the double-active-bridge converter according to the position of a power switch device with a fault, and collecting diagnosis signals of the converter under different open-circuit faults as an original data sample set;
step 2, constructing a multi-branch feature extraction module by utilizing convolution operation, respectively inputting a plurality of diagnostic signals in an original data sample set into each branch, and extracting fault features of the diagnostic signals to obtain a fault feature map;
step 3, constructing a denoising module by combining an attention mechanism with a soft threshold function, inputting the diagnostic signal features extracted by the multi-branch feature extraction module into the denoising module to eliminate the noise features in the diagnostic signal, and obtaining a denoising feature map;
step 4, constructing a feature fusion module by using an attention mechanism, and fusing multi-branch denoising feature maps according to the channel importance of each feature map obtained by the attention mechanism to obtain a fusion feature map;
step 5, inputting the obtained fusion characteristic graph into a classifier, and constructing a multi-branch convolutional neural network fault diagnosis model; dividing an acquired original data sample set into a training set and a testing set, training a multi-branch convolutional neural network fault diagnosis model by using the training set, and testing the trained fault diagnosis model by using the testing set;
and 6, inputting the data to be diagnosed into the trained multi-branch convolutional neural network fault diagnosis model and outputting a final classification result, thereby realizing fault diagnosis of the double-active-bridge converter.
2. The method for diagnosing transducer failure based on attention mechanism and convolution NN according to claim 1, wherein the method for constructing the original data sample set in step 1 is as follows:
the original data sample set is: DATA i ={A i,1 ,A i,2 ,...,A i,j ,...,A i,N }i∈[1,3]Where "3" indicates three diagnostic signals in the raw data set, A i,j For the monitoring value corresponding to the jth point in the ith diagnostic signal monitoring data, j belongs to [1, N ∈]N is the total number of monitoring data points; the combination of the three diagnostic signal monitoring data corresponds to a power switch tube fault state of the double-active-bridge converter.
3. The method for diagnosing converter faults based on an attention mechanism and convolution NN as claimed in claim 2, wherein the method in step 2 is specifically:
when a multi-branch feature extraction module is constructed by using convolution operation, one-dimensional convolution operation is used for direct extractionTaking fault characteristics of the diagnosis signal; meanwhile, batch standardization, ReLU activation function and maximum pooling operation are carried out after each one-dimensional convolution operation, so that overfitting, gradient disappearance or gradient explosion of the model are prevented; the calculation formula of the convolution operation is as follows:wherein x represents the operation of convolution,andrespectively of the ith channel in the ith and l-1 th layers of the profile X,is the convolution kernel between the ith channel and the jth channel of the ith layer of the feature map X;is an offset, M j Representing an input feature set; the formula for the ReLU activation function is:wherein x represents each feature point of the input feature map; the maximum pooling operation is calculated asWherein S represents the size of the sliding window for pooling operation, y j Representing the characteristics of the ith channel in the output characteristic diagram.
4. The method for diagnosing converter faults based on an attention mechanism and convolution NN as claimed in claim 3, wherein the method in step 3 is specifically:
an attention mechanism and a soft threshold function are combined to construct a denoising module, and a feature extraction module extracts noise in a feature mapAcoustic features, in which the attention mechanism enables the acquisition of an extracted raw feature map M 0 Is equal to [ tau ] channel threshold 1 ,τ 2 ,…,τ C ]Inputting the extracted characteristic diagram and the channel threshold value into a soft threshold value function for processing, and being capable of adaptively removing the noise characteristic in the extracted characteristic diagram to obtain a reconstructed characteristic diagram M 1 (ii) a Wherein, the soft threshold value calculation formula is as follows:
5. The method for diagnosing transducer faults based on attention mechanism and convolution NN as claimed in claim 4, wherein the method in step 4 is specifically:
constructing a feature fusion module through an attention mechanism, learning and extracting the importance of different channel features of the feature graph, and fusing a plurality of feature graphs according to the channel feature importance of the feature graph; m 1 ,M 2 And M 3 For three fused feature maps, M is obtained by the attention mechanism 1 ,M 2 And M 3 Corresponding channel feature importance sets ρ, ω andrespectively multiplying the channel feature importance set and the corresponding de-noising feature map set, and then adding to obtain a fused feature map, wherein the calculation formula is as follows:
6. The method for diagnosing transducer faults based on attention mechanism and convolution NN as claimed in claim 5, wherein the method in step 5 is specifically:
the constructed multi-branch convolutional neural network fault diagnosis is directly used for fault diagnosis of a plurality of diagnostic signals, the plurality of diagnostic signals are respectively input into different convolutional branches, a feature extraction module of each convolutional branch extracts fault features of each diagnostic signal, a denoising module in each convolutional branch removes noise information in a corresponding fault feature map, a feature fusion module fuses the denoised feature maps of the plurality of convolutional branches, and the obtained fusion feature maps are input into a classifier for classification after being subjected to full connection and activation function processing.
7. An attention-based and convolution NN based transducer fault diagnosis system, comprising:
the data acquisition module is used for establishing a simulation model of the double-active-bridge converter, selecting leakage inductance current of each power switch tube in a fault state, acquiring an original data sample set by taking the midpoint voltage of any bridge arm of the primary bridge and the midpoint voltage of any bridge arm of the secondary bridge as diagnosis and diagnosis signals, and classifying and coding the fault state of the double-active-bridge converter according to the position of a power switch device with a fault;
the fault feature extraction module is used for constructing a multi-branch feature extraction module by utilizing convolution operation, inputting a plurality of diagnostic signals in a sample into each branch respectively, and extracting fault features of the diagnostic signals to obtain a fault feature map;
the denoising module is constructed by combining an attention mechanism with a soft threshold function, and the diagnostic signal features extracted by the multi-branch feature extraction module are input into the denoising module to eliminate the noise features in the diagnostic signal, so that a denoising feature map is obtained;
the feature fusion module is constructed by utilizing an attention mechanism, and the multi-branch denoising feature maps are fused according to the channel importance of each feature map obtained by the attention mechanism to obtain a fusion feature map;
the network training module is used for dividing an original data sample set into a training set and a verification set, the training set trains the multi-branch convolutional neural network fault diagnosis model based on the attention mechanism, and the verification set verifies the trained fault diagnosis model;
and the diagnosis module is used for directly inputting newly acquired double-active-bridge converter fault diagnosis data into a trained network for diagnosis and fault positioning in the later diagnosis process.
8. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210639082.9A CN115062538A (en) | 2022-06-07 | 2022-06-07 | Converter fault diagnosis method and system based on attention mechanism and convolution NN |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210639082.9A CN115062538A (en) | 2022-06-07 | 2022-06-07 | Converter fault diagnosis method and system based on attention mechanism and convolution NN |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115062538A true CN115062538A (en) | 2022-09-16 |
Family
ID=83199475
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210639082.9A Pending CN115062538A (en) | 2022-06-07 | 2022-06-07 | Converter fault diagnosis method and system based on attention mechanism and convolution NN |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115062538A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116610916A (en) * | 2023-05-18 | 2023-08-18 | 兰州理工大学 | Multi-signal self-adaptive fusion cascade H-bridge inverter fault diagnosis method |
CN116798630A (en) * | 2023-07-05 | 2023-09-22 | 广州视景医疗软件有限公司 | Myopia physiotherapy compliance prediction method, device and medium based on machine learning |
-
2022
- 2022-06-07 CN CN202210639082.9A patent/CN115062538A/en active Pending
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116610916A (en) * | 2023-05-18 | 2023-08-18 | 兰州理工大学 | Multi-signal self-adaptive fusion cascade H-bridge inverter fault diagnosis method |
CN116610916B (en) * | 2023-05-18 | 2023-11-21 | 兰州理工大学 | Multi-signal self-adaptive fusion cascade H-bridge inverter fault diagnosis method |
CN116798630A (en) * | 2023-07-05 | 2023-09-22 | 广州视景医疗软件有限公司 | Myopia physiotherapy compliance prediction method, device and medium based on machine learning |
CN116798630B (en) * | 2023-07-05 | 2024-03-08 | 广州视景医疗软件有限公司 | Myopia physiotherapy compliance prediction method, device and medium based on machine learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112710969B (en) | Open-circuit fault diagnosis method for switching tube of single-phase half-bridge five-level inverter | |
CN109635928B (en) | Voltage sag reason identification method based on deep learning model fusion | |
CN115062538A (en) | Converter fault diagnosis method and system based on attention mechanism and convolution NN | |
CN115081316A (en) | DC/DC converter fault diagnosis method and system based on improved sparrow search algorithm | |
CN112685961B (en) | Method and system for predicting remaining service life of analog circuit | |
CN107894564A (en) | A kind of analog-circuit fault diagnosis method based on intersection wavelet character | |
CN102636742B (en) | Large-scale analogue circuit fault diagnosis method based on wavelet neural network | |
CN113702895B (en) | Online quantitative evaluation method for error state of voltage transformer | |
CN103267941B (en) | Method for testing fault modes of integrated switching current circuit | |
CN103995237A (en) | Satellite power supply system online fault diagnosis method | |
CN110596506A (en) | Converter fault diagnosis method based on time convolution network | |
CN102279358A (en) | MCSKPCA based neural network fault diagnosis method for analog circuits | |
CN112180204A (en) | Power grid line fault diagnosis method based on electric quantity information | |
CN113869145A (en) | Circuit fault diagnosis method and system for light-weight gradient elevator and sparrow search | |
CN114358085A (en) | Line fault diagnosis method and device based on heterogeneous model fusion | |
CN117289013A (en) | Data processing method and system for pulse current test | |
CN115019129A (en) | Double-active-bridge converter fault diagnosis method and system based on time sequence imaging and image fusion | |
CN114355173B (en) | Analog filter circuit fault diagnosis method based on multi-input residual error network | |
CN116578922A (en) | Valve cooling system fault diagnosis method and device based on multichannel convolutional neural network | |
CN116127856A (en) | Permanent magnet synchronous motor fault diagnosis method based on deep neural network | |
CN110908365A (en) | Unmanned aerial vehicle sensor fault diagnosis method and system and readable storage medium | |
CN116401572A (en) | Power transmission line fault diagnosis method and system based on CNN-LSTM | |
CN113311364B (en) | Permanent magnet synchronous motor inverter open-circuit fault diagnosis method based on multi-core SVM | |
CN115456013A (en) | Wind turbine generator converter power module fault diagnosis method based on operation data | |
Guangquan et al. | Analog circuit incipient fault diagnosis using deep neural Network |
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 |