CN115935244B - Single-phase rectifier fault diagnosis method based on data driving - Google Patents
Single-phase rectifier fault diagnosis method based on data driving Download PDFInfo
- Publication number
- CN115935244B CN115935244B CN202310222390.6A CN202310222390A CN115935244B CN 115935244 B CN115935244 B CN 115935244B CN 202310222390 A CN202310222390 A CN 202310222390A CN 115935244 B CN115935244 B CN 115935244B
- Authority
- CN
- China
- Prior art keywords
- data
- fault
- model
- sub
- fault 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 61
- 238000003745 diagnosis Methods 0.000 title claims abstract description 49
- 102100032202 Cornulin Human genes 0.000 claims abstract description 21
- 101000920981 Homo sapiens Cornulin Proteins 0.000 claims abstract description 21
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 11
- 238000005259 measurement Methods 0.000 claims abstract description 4
- 238000011176 pooling Methods 0.000 claims description 21
- 238000000605 extraction Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 14
- 239000011159 matrix material Substances 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 238000012800 visualization Methods 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000012360 testing method Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000007246 mechanism Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000012549 training Methods 0.000 claims description 4
- 230000003190 augmentative effect Effects 0.000 claims description 3
- 238000001228 spectrum Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000012935 Averaging Methods 0.000 claims description 2
- 239000003990 capacitor Substances 0.000 claims description 2
- 238000000489 vacuum metal deposition Methods 0.000 claims description 2
- 230000005764 inhibitory process Effects 0.000 claims 1
- 238000013527 convolutional neural network Methods 0.000 description 14
- 230000000694 effects Effects 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 210000004027 cell Anatomy 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000003672 processing method Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000001419 dependent effect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 210000000352 storage cell Anatomy 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
- Y04S10/52—Outage or fault management, e.g. fault detection or location
Landscapes
- Testing Electric Properties And Detecting Electric Faults (AREA)
Abstract
The invention discloses a fault diagnosis method of a single-phase rectifier based on data driving, which comprises the following steps: s1, constructing a model of a single-phase PWM rectifier, and obtaining network current measurement and direct-current side voltage fault data of a normal state of the single-phase PWM rectifier, an IGBT open circuit, an anti-parallel diode open circuit, a series resonance circuit inductance open circuit and a capacitance open circuit; s2, performing VMD decomposition on the obtained normal state data and fault data to obtain an intrinsic mode component IMF serving as an input feature vector of a subsequent fault diagnosis network; s3, constructing a single-phase rectifier fault diagnosis sub-model based on CRNN; and S4, integrating the features extracted by the two sub-models, and finally converting the output data into probability values of corresponding categories through softmax. The invention solves the problems that the fault signals of the IGBT and the anti-parallel diode are insensitive to direct-current side voltage and the fault signals of the elements of the series resonance circuit are insensitive to network side current, and the fault elements are accurately positioned.
Description
Technical Field
The invention relates to the technical field of single-phase rectifier fault detection, in particular to a single-phase rectifier fault diagnosis method based on data driving.
Background
The AC-DC-AC traction transmission system is widely applied to the field of high-speed railway traction systems. The power electronic transformer plays a role in power conversion in the traction drive system, and mainly comprises a high-frequency transformer and a power electronic converter. The power electronic transformer has a complex structure, a large number of parts and frequent hidden faults. The faults of the main devices of the power electronic transformer often cause abnormal fluctuation of voltage and current, and endanger the operation safety of high-speed trains. The existing power electronic transformer fault diagnosis method is mostly model-based, and the basic idea is to calculate fault residual errors by using a mathematical model and perform fault diagnosis through residual error analysis. However, in practical engineering applications, the nonlinearity and discreteness of the switching devices limit the accuracy of the analytical model. Therefore, in the fault diagnosis of a complex nonlinear system, it is urgent to study a fault diagnosis method with little dependence on the model. In this process, data-based fault diagnosis methods are more prone to leverage the rich device state information contained in the system detection data, exploring the inherent links between these data and fault patterns.
As an important component of a traction drive system, the performance of a single-phase Pulse Width Modulation (PWM) rectifier will directly affect the performance of the drive system, with power semiconductors being the most fragile. Thus, open and short circuit faults of Insulated Gate Bipolar Transistors (IGBTs) have become common faults of grid-side rectifiers. In general, when a short-circuit fault occurs, voltage and current rise sharply in a short time, and then are converted from the short-circuit fault to an open-circuit fault. Therefore, the open circuit fault analysis and diagnosis of the IGBT is the core of the fault diagnosis of the rectifier. In addition, abnormal voltage stress caused by the anti-parallel diode open circuit fault is far higher than the blocking voltage of the IGBT, and the IGBT can be caused to have overvoltage fault in a short time. For the intermediate dc circuit, the diagnosis method of the series resonant circuit element failure has been studied relatively little.
The model-based method is the most widely applied method in IGBT open circuit fault diagnosis, the diagnosis performance of the method is highly dependent on model precision, but the model precision is difficult to be ensured, and particularly, the complexity of a traction system and the influence of current direction change on the work of the anti-parallel diode are considered. The signal-based and machine-learning-based fault diagnosis method is different from the model-based fault diagnosis method in that accurate modeling is not required, but faults are diagnosed by analyzing amplitude-frequency characteristics of voltage and current. Therefore, the effectiveness of the data-driven approach is greatly affected by the feature extraction performance. The signal processing-based method basically uses only extraction with shallow features including amplitude information, frequency information, energy information, and the like. Through analysis of the characteristic information, the corresponding relation between the fault signal and the fault mode can be established. However, it is often difficult to establish such a relationship directly through the characteristics of the data itself. Currently, three key problems that must be solved in diagnosing single-phase rectifier faults using a data-driven method are: (1) The method has good generalization and robustness, and can perform fault diagnosis on IGBT, diode and series resonant circuit element on different single-phase rectifiers to replace manual diagnosis so as to realize intelligent diagnosis. (2) The method has the characteristic of short time, and because the critical components are abnormal, parameters such as module current and the like can be changed greatly, so that the whole module works abnormally in a short time. (3) The algorithm model must have high precision and high stability, the single-phase rectifier has a complex system structure, various components and the fault device can be accurately diagnosed, so that the stable operation of the train is ensured.
Disclosure of Invention
The invention provides a single-phase rectifier fault diagnosis method based on data driving, aiming at the problems existing in the prior data driving method for single-phase rectifier fault diagnosis. Comprehensive fault diagnosis of the single-phase rectifier is achieved by combining a Variational Modal Decomposition (VMD) method and a dual-model convolutional neural network (CRNN).
The invention provides a fault diagnosis method of a single-phase rectifier based on data driving, which comprises the following steps:
s1, building a model of a single-phase PWM rectifier through a dSPACE hardware circuit test platform, and obtaining network current measurement and direct-current side voltage fault data of the single-phase PWM rectifier in a normal state, an IGBT open circuit, an anti-parallel diode open circuit, a series resonance circuit inductance open circuit and a capacitor open circuit through the platform.
S2, performing VMD decomposition on the obtained normal state data and fault data to obtain an intrinsic mode component IMF, and taking the intrinsic mode component IMF as an input feature vector of a subsequent fault diagnosis network. The method specifically comprises the following substeps:
s21, searching K IMF components with specific sparsity through a constraint variation model, so that the estimated bandwidth sum of the components is minimum, and limiting constraint conditions to be the sum of the components and equal to the original signal. Where a particular sparsity refers to the percentage of zero elements in the matrix or dataset relative to the total number of elements in the dataset.
S22, obtaining each IMF component u through Hilbert transformation for obtaining K IMFs with limited bandwidths k (t) single side inter-spectral and then estimating the center frequency ω of each IMF k And its index signal
Multiplying, modulating the spectrum of the mode to the corresponding baseband, and then calculating the analytic signal gradient square norm L 2 And constructing a variation mode.
S23, introducing a penalty factor alpha and Lagrange multiplier lambda to convert the constrained variation problem into an unconstrained variation problem so as to solve the variation problem and obtain an augmented Lagrange expression.
S24, updating iteration solution saddle points by adopting an alternate direction multiplier algorithm to obtain an optimal solution so as to decompose an original signal into K IMF components.
S3, constructing a single-phase rectifier fault diagnosis sub-model based on the CRNN, wherein the single-phase rectifier fault diagnosis sub-model comprises a CRNN current sub-model and a CRNN voltage sub-model.
Step S3 comprises the following sub-steps:
s31, building a feature primary extraction module based on 1D-CNN, wherein a convolutional layer of the CNN consists of two parts: the first part is a convolution layer C1 and a convolution layer C2, and convolution operation is carried out to extract the structural characteristics of the adjacent domains; the second part is the pooling layers S1 and S2, and downsampling operation is performed to eliminate redundant information of the feature map.
In this step, the respective signals in the normal state and the failure are transferred as inputs to the CNN layer through the one-dimensional convolution filter; obtaining a characteristic diagram through convolution operation of an l-th convolution layer
Wherein,,
and->
Respectively representing the weight and offset value, M, of the j-th layer convolution filter j Is the number of input feature maps;
the pooling process after the convolution process, the maximum pooling operation formula is as follows:
wherein,,
and->
Representing the weights and bias values of the max pooling layer, dowm () represents the max pooling function.
S32, constructing a feature secondary extraction module based on SRU, wherein the SRU network operates in a matrix multiplication mode, and each gating structure needs to process input x through an activation function t Finally, reset the gate and cell internal state and input x t Obtain output h t The method comprises the steps of carrying out a first treatment on the surface of the The calculation formula is as follows:
f t ＝α f (W f *x t )+b f
r t ＝α r (W r *x t )+b r
h t ＝r t ⊙tanh(c t )+(1-r t )⊙x t
in the formula, is
Linear representation, f t Is forgetful door r t To reset the gate c t Representing an intrinsic state; alpha f And alpha r Respectively forget gate f And reset gate r S activation function of (2); w, W f And W is r Respectively linear representation +.>
Forgetting gate f And reset gate r Weights of (2); b f And b r Respectively indicate forgetting gate f And reset gate r Deviation of (2); the corresponding element is multiplied by; tanh () represents a hyperbolic tangent activation function in a hidden state.
S33, introducing an attention mechanism before the full connection layer. The present invention uses SEnet (Squeeze-and-Excitation Network), which considers the relationship between feature channels, adding attention mechanisms to the feature channels. Firstly, carrying out global pooling on each feature map through squeeze operation, and averaging the feature maps into a real value; then carrying out an expitaton operation, wherein in the process, the dimension of the C channels is reduced and then the C channels are expanded back; finally, the output of the association is regarded as the importance of each channel after feature selection, and the important features are improved and the unimportant features are restrained by multiplying the importance of each channel by the previous features in a weighted manner.
S34, respectively taking IMFs obtained by decomposing the network side current and the direct current side voltage through VMDs as inputs of the built network model to form a double-model framework, namely a CRNN current sub-model and a CRNN voltage sub-model, and training and calculating in the current sub-model and the voltage sub-model.
S4, integrating the extracted features of the two sub-models through a flat layer, converting the extracted feature information into a tag space through a full connection layer to finish data classification, and outputting the data classification as follows:
wherein D is i 、b i Is a learning parameter of the full-connection layer,
input data and output data respectively;
the output data is converted into probability values of corresponding categories through softmax, and the expression is as follows:
wherein y is i The i-th parameter value (i=1, 2,) for vector y.
The confusion matrix and the T-SNE are visualized as visualization tools for analyzing the identification result of the test sample, wherein the visualization tools are used for visualizing the algorithm performance in a matrix form, and the visualization tools are used for presenting the visualization effect of sample distribution through feature mapping.
Compared with the prior art, the invention has the following advantages:
1. according to the invention, the data is acquired by constructing the model of the single-phase PWM rectifier through the dSPACE hardware circuit test platform, so that the problems that fault data are difficult to acquire and the data quantity is small are solved, and the comprehensiveness and reliability of fault diagnosis are improved on the premise of ensuring the diagnosis safety.
2. According to the invention, the difference between the original similar fault signals is increased by the VMD signal processing method, the stable fault characteristics are obtained from massive data, and the problem that when the IGBT or the diode has an open-circuit fault, the IGBT and the diode on the same bridge arm are difficult to identify due to the similar fault characteristics is solved. The method can specify the number of modes to be decomposed, avoids the condition of useless components, reduces the steps of subsequent feature processing, and solves the problems of end effect and mode aliasing which are easy to occur in the conventional signal processing method by a mirror image expansion method.
3. The invention designs a CRNN network, which solves the problems that when the 1D-CNN network extracts the structural characteristics of signal adjacent domains, the mining of the time sequence characteristics contained in the structural characteristics is insufficient, the recursive structure of the RNN network is beneficial to the extraction of the time sequence characteristics of a model, but the operation speed of the model is sacrificed, and the two pain points of gradient elimination and over-fitting phenomenon are easy to generate. The CRNN network enables the fault recognition model to have the advantages of high CNN training speed and high RNN recognition accuracy.
4. The invention designs a dual-model architecture, which respectively establishes voltage and current sub-models, and respectively takes network side current and direct current side voltage IMF components after VMD decomposition as the inputs of the sub-models. The problem that the fault signals of the IGBT and the anti-parallel diode are insensitive to the voltage of the direct current side and the fault signals of the series resonant circuit element are insensitive to the current of the network side is solved, the current of the network side and the voltage of the direct current side are fully utilized, and the purpose of accurately positioning the fault element is achieved.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Drawings
FIG. 1 is a flow chart of a single-phase rectifier fault diagnosis method based on data driving.
Fig. 2 is a single phase PWM rectifier topology.
Fig. 3 is a VMD exploded flow diagram.
FIG. 4 is a schematic diagram of a dual-model CRNN framework.
Fig. 5 is a confusion matrix result diagram.
Detailed Description
The preferred embodiments of the present invention will be described below with reference to the accompanying drawings, it being understood that the preferred embodiments described herein are for illustration and explanation of the present invention only, and are not intended to limit the present invention.
As shown in fig. 1 to 4, the data-driven-based fault diagnosis method for the single-phase rectifier of the present invention comprises the following steps:
step S1: and constructing a model of the single-phase PWM rectifier through a dSPACE hardware circuit test platform, and obtaining network test current and direct-current side voltage fault data of the single-phase PWM rectifier in a normal state, with an IGBT open circuit and an anti-parallel diode open circuit, and with a series resonance circuit with an inductance open circuit and a capacitance open circuit.
dSPACE can control virtual objects by using a real controller to realize limit simulation test on a real environment. Therefore, the data based on dSPACE is used as actual measurement data, and has a good verification effect on the effectiveness of the method. The system is composed of a Digital Signal Processor (DSP) as an executor of a control algorithm, wherein a control chip is TMS320F28335, a dSPACE simulator and an upper computer as a control interface. The PWM pulse signal is input to dsace by the DS2103 board. The network side voltage UN, the rectifying circuit input current iN and the main circuit direct current side voltage Udc are output to the AD sampling module of the DSP by the DS2002 board and displayed by an oscilloscope. As shown in fig. 2, the topology circuit of the single-phase PWM rectifier adopts a DQ decoupling control strategy, so that the network side voltage and the network side current of the rectifier are consistent in phase, can both be kept to operate in a unit power factor state, and the network side current has less harmonic content and excellent steady-state performance.
Step S2: and performing VMD decomposition on the obtained normal data and fault data to obtain an intrinsic mode component (IMF), and taking the intrinsic mode component (IMF) as an input feature vector of a subsequent fault diagnosis network. The sampling frequency is set to 100000HZ, that is, 100000 data can be collected in 1s, wherein there is a lot of useless information and characteristics, so that data preprocessing is needed. An obvious feature of VMD is that the number K of modal components needs to be set before the signal is decomposed by VMD. For some scenarios where the number of implicit modes of the signal is not predicted, the determination of the K value is very important for the VMD. Therefore, considering that the decomposed IMFs have independent center frequencies, the original data is subjected to fast Fourier transform to obtain a spectrogram of the data, and the components with larger amplitudes are generally considered to have larger energy, which means to contain more features, soIt is reasonable to choose such components as features. After determining the value of K from the spectrogram, VMD decomposition can be performed. A flowchart of VMD decomposition is shown in fig. 3. In the figure, { u k }＝{u 1 ,u 2 ,u 3 ,…,u K K IMFs, { omega } represents k }＝{ω 1 ,ω 2 ,ω 3 ,…,ω K -the center frequency of each component; lambda is Lagrangian multiplier, alpha is penalty factor, tau is fidelity coefficient, lambda is Fourier transform, and n is iteration number. In the final satisfaction, ε is the discrimination accuracy, here 10e-6 is taken. K IMF components with specific sparsity are sought through constraint variation models, so that the estimated bandwidth sum of the components is minimum, and constraint conditions are the sum of the components and equal to the original signal. To obtain K IMFs with limited bandwidth, hilbert transformation is performed to obtain each IMF component u k (t) single side inter-spectral and then estimating the center frequency ω of each IMF k And its index signal
Multiplying the spectrum of the mode to the corresponding baseband, and then calculating the square norm L of the analytic signal gradient 2 And constructing a variation mode. And introducing a penalty factor alpha and Lagrange multiplier lambda to convert the constrained variation problem into an unconstrained variation problem so as to solve the variation problem and obtain an augmented Lagrange expression. And updating the iteration solving saddle by adopting an alternate direction multiplier algorithm to obtain an optimal solution so as to decompose the original signal into K IMF components. The network side current and the direct current side voltage are subjected to frequency spectrum analysis to obtain an optimal value of K which is 3, and then VMD decomposition is carried out to obtain the corresponding IMF.
Step S3: and constructing a single-phase rectifier fault diagnosis sub-model based on the CRNN, wherein the single-phase rectifier fault diagnosis sub-model comprises a CRNN current sub-model and a CRNN voltage sub-model.
A downsampling layer is introduced first, and excessive network parameters are easy to generate due to the fact that the sample length is large, so that overfitting is caused, the network parameter scale is greatly reduced, the robustness of an algorithm is enhanced, and the network operation time is greatly shortened. CNNs have the property of sparse weights, which can be significantly smaller than the input by usingConvolution filters detect small and meaningful features. This means that CNN reduces the number of parameters that need to be stored, significantly improving the efficiency of feature extraction. The convolutional layer of CNN is generally composed of two parts: the first part carries out convolution operation to extract characteristics; (2) The second part carries out pooling operation and adjusts the output of the convolution layer. In CRNN, the convolution layer function is regarded as a feature extractor. The signals in normal state and failure are passed as inputs to the CNN layer through a one-dimensional convolution filter. Through the l-th convolution layer (l ε l) c ) Is convolved to obtain a feature map
Wherein the method comprises the steps of
And->
Respectively representing the weight and offset value, M, of the j-th layer convolution filter j Is the number of input feature maps.
The pooling process plays a secondary extraction role after the convolution process. Max pooling is used to reduce the dimensionality of data and preserve useful information:
wherein the method comprises the steps of
And->
Representing the weights and bias values of the max pooling layer, dowm () represents the max pooling function.
RNN adaptationTasks such as natural semantic analysis and time sequence modeling. The RNN module of the CRNN uses SRU units that have faster computational power than LSTM. And (3) building a feature secondary extraction module based on the SRU, wherein the SRU realizes parallelization acceleration in two aspects of structure and operation optimization. The SRU pre-processes the input data relatively independently at each recursive feature extraction, thereby performing the recursive feature extraction by relatively lightweight parallel operations. SRU simplifies history information h t Expanding the operation parallelism similar to 1D-CNN, adopting the updating mechanism of the reserved information storage cell to construct a reset gate control cell gate r Dynamically adjusting the recursion step to eliminate the output gate o The recursive step length produces a gradient vanishing phenomenon over a long period. In SRU, gating cell states
No longer depends on the historical iteration state h t-1 Introducing reset gate state r t Improving the time sequence feature extraction mode. Therefore, the SRU realizes parallelization acceleration from two aspects of structure and operation optimization. The advantages of SRU are embodied as: SRUs with the same number of layers extract features faster and less information loss than LSTM.
The attention mechanism is introduced before the full connection layer, important information can be focused with high weight, irrelevant information can be ignored with low weight, and the weight can be continuously adjusted, so that important information can be selected under different conditions, and therefore, the method has better expandability and robustness. The present invention uses SEnet (Squeeze-and-Excitation Network), which considers the relationship between feature channels, adding attention mechanisms to the feature channels. First, global pooling is performed on each feature map through a squeeze operation, and the feature maps are averaged into a real value. The real number has to some extent a global receptive field. This operation can enable features near the data input to also have global receptive fields. The exact operation follows, and as the result of the squeeze operation, the network outputs a feature map of 1xC size, and the weight w is used to learn the direct correlation of the C channels. In the process, the dimension of the C channels is reduced and then the C channels are expanded. The method has the advantages of reducing the network calculation amount on one hand and increasing the non-linearity capability of the network on the other hand. Finally, the output of the association is regarded as the importance of each channel after feature selection, and the importance is multiplied to the previous feature in a multiplication weighting mode, so that the function of improving the important feature and suppressing the unimportant feature is realized.
The feature primary extraction module of the CRNN framework consists of two alternately stacked 1D-CNN convolution and pooling layers, and the feature secondary extraction module consists of 4 SRU circulation layers. The 1D-CNN-based feature primary extraction module comprises a convolution layer and a pooling layer, and is used for receiving a fault signal reconfirmer of the single-phase rectifier t Extracting structural feature maps in (t=1, 2, …, T):
F i ＝(F i.1 ,F i,2 ,…F i,j ),j＝1,2,…,m k
SRU-based feature primary extraction module receives structural feature map F i Extracting time sequence characteristics h t (t＝1,2,…T)。
IMFs of net side current and direct side voltage decomposed by VMD are used as inputs, respectively, and training and calculation are performed in current and voltage sub-models.
Step S4: and integrating and outputting the features extracted by the two sub-models.
The DCRNN is shown in fig. 4, where features extracted from two sub-models are integrated by a flat layer, and extracted feature information is converted into a tag space by a full connection layer to complete data classification, and its output may be expressed as:
wherein D is i B i Is the learning parameter of the full connection layer. Then, the output data is converted into a probability value of the corresponding category through softmax, and the expression is as follows:
wherein y is i The i-th parameter value (i=1, 2,) for vector y.
The confusion matrix and the T-SNE are visualized as visualization tools for analyzing the identification result of the test sample, wherein the visualization tools are used for visualizing the algorithm performance in a matrix form, and the visualization tools are used for presenting the visualization effect of sample distribution through feature mapping.
The visual result of the confusion matrix shows that the DCRNN is very sensitive to the fault characteristics of the single-phase rectifier, and can effectively distinguish whether the operation state of the single-phase rectifier is faulty or not and distinguish which specific part of the single-phase rectifier is faulty.
T-SNE is a random optimization dimension reduction algorithm, and is mainly used for dimension reduction visualization of nonlinear high-dimension data, and feature distribution is visualized in a two-dimensional or three-dimensional low-dimension space. The core of the T-SNE algorithm is to minimize the original data distribution and K-L divergence and search for a proper low-dimensional map. Complexity (prep) is used as a key adjustable parameter of the T-SNE to characterize the estimation range of the potential adjacent points of the sample point. The setting interval of the perp value is between [5 and 10 ]]. For fault identification of single-phase rectifier, the high-dimensional feature extracted by DCRNN is set as X bogie ＝{x 1 ,x 2 ,…,x n T-SNE based dimension reduction to obtain low-dimensional feature distribution Y bogie ＝{y 1 ,y 2 ,…,y n }. T-SNE-based single-phase rectifier fault identification visualization, in a low-dimensional feature space, sample boundaries between different fault categories are obvious. The DCRNN has excellent feature extraction capability from the aspect of feature distribution, so that the DCRNN has better recognition precision and speed in the fault recognition of the single-phase rectifier.
The whole scheme of the invention has a complete diagnosis flow. The diagnosis performance of the model-based method is highly dependent on the model precision, and accurate modeling is very difficult, so that a data-driven single-phase rectifier fault diagnosis method is adopted.
The mode of data acquisition adopts a dSPACE hardware circuit test platform to build a model of the single-phase PWM rectifier, so that the setting of faults is safer, and the situation that the whole circuit is crashed due to the faults of a certain element is avoided. The detailed data of the network side current and the direct current side voltage can be obtained without installing an additional sensor by using the acquisition mode, the problems that fault data are difficult to obtain and the data quantity is small are solved, and the comprehensiveness and the reliability of fault diagnosis are improved on the premise of ensuring the diagnosis safety.
Because of the topology structure of the single-phase rectifier, the fault states of the IGBTs and the anti-parallel diodes on the same bridge arm are similar, and the two bridge arms are provided with four IGBTs and four anti-parallel diodes. The VMD decomposition method based on the signal processing method increases the difference between the originally similar fault signals, determines the mode number to be decomposed by comparing with a spectrogram, accurately extracts the most obvious characteristic of each fault signal, and overcomes the problems of end-point effect and mode component aliasing existing in the EMD method.
The method combines the advantages of the CNN network and the RNN network, and gives consideration to stability, accuracy, generalization and rapidity. Since the fault characteristics of the IGBT and the anti-parallel diode are mainly embodied on the network side current and the fault characteristics of the series resonant circuit element are mainly embodied on the direct current side voltage, a dual-mode architecture is designed, and the network side current and the direct current side voltage IMF components after VMD decomposition are respectively used as the inputs of the sub-models by respectively establishing voltage sub-models. The grid side current and the direct current side voltage are fully utilized, faults of IGBT, anti-parallel diode and series resonance circuit elements are effectively distinguished, and the purpose of accurately positioning the fault elements is achieved. The visual result of the confusion matrix (fig. 5) shows that DCRNN is very sensitive to fault characteristics of the single-phase rectifier, and the accuracy of distinguishing whether the operation state of the single-phase rectifier has faults is 100%, and the accuracy of identifying each fault of the single-phase rectifier is 95.83%, so that the diagnosis requirement is met.
The present invention is not limited to the above-mentioned embodiments, but is intended to be limited to the following embodiments, and any modifications, equivalents and modifications can be made to the above-mentioned embodiments without departing from the scope of the invention.
Claims (7)
1. A fault diagnosis method of a single-phase rectifier based on data driving is characterized by comprising the following steps:
s1, building a model of a single-phase PWM rectifier through a dSPACE hardware circuit test platform, and obtaining network current measurement and direct-current side voltage fault data of the single-phase PWM rectifier in a normal state, an IGBT open circuit, an anti-parallel diode open circuit, a series resonance circuit inductance open circuit and a capacitor open circuit through the platform;
s2, performing VMD decomposition on the obtained normal state data and fault data to obtain an intrinsic mode component IMF, and taking the intrinsic mode component IMF as an input feature vector of a subsequent fault diagnosis network;
s3, constructing a single-phase rectifier fault diagnosis sub-model based on CRNN, wherein the single-phase rectifier fault diagnosis sub-model comprises a CRNN current sub-model and a CRNN voltage sub-model;
and S4, integrating the extracted features of the two sub-models through a flat layer, converting the extracted feature information into a tag space through a full connection layer to finish data classification, and finally converting output data into probability values of corresponding categories through softmax.
2. The data-driven single-phase rectifier fault diagnosis method according to claim 1, wherein the step S2 includes the sub-steps of:
s21, searching K IMF components with specific sparsity through a constraint variation model, so that the estimated bandwidth sum of the components is minimum, and limiting constraint conditions to be the sum of the components and equal to an original signal; wherein, the specific sparsity refers to the percentage of zero elements in the matrix or dataset relative to the total number of elements in the dataset;
s22, obtaining each IMF component u through Hilbert transformation for obtaining K IMFs with limited bandwidths k (t) single side inter-spectral and then estimating the center frequency ω of each IMF k And its index signal
Multiplying, modulating the spectrum of the mode to the corresponding baseband, and then calculating the analytic signal gradient square norm L 2 Constructing a variation mode;
s23, introducing a penalty factor alpha and Lagrange multiplier lambda to convert a constraint variation problem into an unconstrained variation problem so as to solve the variation problem and obtain an augmented Lagrange expression;
s24, updating iteration solution saddle points by adopting an alternate direction multiplier algorithm to obtain an optimal solution so as to decompose an original signal into K IMF components.
3. The data-driven single-phase rectifier fault diagnosis method according to claim 1, wherein the step S3 includes the sub-steps of:
s31, building a 1D-CNN-based feature primary extraction module, wherein the first part is a convolution layer C1 and a convolution layer C2, and performing convolution operation to extract structural features of adjacent domains; the second part is a pooling layer S1 and S2, and downsampling operation is carried out to remove redundant information of the feature map;
s32, constructing a feature secondary extraction module based on SRU, wherein the SRU network operates in a matrix multiplication mode, and each gating structure needs to process input x through an activation function t Finally, reset the gate and cell internal state and input x t Obtain output h t The method comprises the steps of carrying out a first treatment on the surface of the The calculation formula is as follows:
f t ＝α f (W f *x t )+b f
r t ＝α r (W r *x t )+b r
h t ＝r t ⊙tanh(c t )+(1-r t )⊙x t
in the method, in the process of the invention,
is expressed linearly, f t Is forgetful door r t To reset the gate c t Representing an intrinsic state; alpha f And alpha r Respectively forget gate f And reset gate r S activation function of (2); w, W f And W is r Respectively linear representation +.>
Forgetting gate f And reset gate r Weights of (2); b f And b r Respectively indicate forgetting gate f And reset gate r Deviation of (2); the corresponding element is multiplied by; tanh () represents a hyperbolic tangent activation function in a hidden state;
s33, introducing an attention mechanism before the full connection layer;
s34, respectively taking IMFs obtained by decomposing the network side current and the direct current side voltage through VMDs as inputs of the built network model to form a double-model framework, namely a CRNN current sub-model and a CRNN voltage sub-model, and training and calculating in the current sub-model and the voltage sub-model.
4. The data-driven single-phase rectifier fault diagnosis method according to claim 3, wherein in the step S31, each signal in a normal state and a fault is transferred as an input to the CNN layer through a one-dimensional convolution filter; obtaining a characteristic diagram through convolution operation of an l-th convolution layer
Wherein k is ij l And b j l Respectively representing the weight and offset value, M, of the j-th layer convolution filter j Is the number of input feature maps that are presented,
is the characteristic diagram of the upper layer;
the pooling process after the convolution process, the maximum pooling operation formula is as follows:
wherein beta is j l And b j l Representing the weights and bias values of the max pooling layer, dowm () representing the max pooling function,
Is a characteristic diagram after pooling.
5. The method for diagnosing a fault of a single-phase rectifier based on data driving as claimed in claim 3, wherein the specific method of step S33 is as follows: firstly, carrying out global pooling on each feature map through squeeze operation, and averaging the feature maps into a real value; then carrying out an expitaton operation, wherein in the process, the dimension of the C channels is reduced and then the C channels are expanded back; and finally, taking the output of the association as the importance of each channel after feature selection, multiplying the importance of each channel to the previous features in a multiplication weighting mode, and realizing the promotion of important features and the inhibition of unimportant features.
6. The method for diagnosing a fault of a single-phase rectifier based on data driving as claimed in claim 1, wherein said step S4 is specifically as follows:
the extracted features of the two sub-models are integrated through the flat layer, the extracted feature information is converted into a tag space through the full connection layer to complete data classification, and the output of the tag space is expressed as:
wherein D is i 、b i Is a learning parameter of the full-connection layer,
input data and output data respectively;
the output data is converted into probability values of corresponding categories through softmax, and the expression is as follows:
wherein y is i I=1, 2, i.e., i is the vector y i parameter value.
7. The data-driven single-phase rectifier fault diagnosis method according to claim 6, wherein in the step S4, the confusion matrix and the T-SNE are visualized as visualization tools for analyzing the recognition result of the test sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310222390.6A CN115935244B (en) | 2023-03-09 | 2023-03-09 | Single-phase rectifier fault diagnosis method based on data driving |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310222390.6A CN115935244B (en) | 2023-03-09 | 2023-03-09 | Single-phase rectifier fault diagnosis method based on data driving |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115935244A CN115935244A (en) | 2023-04-07 |
| CN115935244B true CN115935244B (en) | 2023-05-09 |
Family
ID=85822055
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310222390.6A Active CN115935244B (en) | 2023-03-09 | 2023-03-09 | Single-phase rectifier fault diagnosis method based on data driving |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115935244B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117235653B (en) * | 2023-11-15 | 2024-03-12 | 深圳市盛格纳电子有限公司 | Power connector fault real-time monitoring method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2017099103A (en) * | 2015-11-20 | 2017-06-01 | 株式会社東芝 | Rotary rectifier fault detector and rotary rectifier fault detection method |
| CN110957762A (en) * | 2019-12-11 | 2020-04-03 | 中南大学 | Fault classification method, system and storage medium for three-level traction converter |
| CN111162684A (en) * | 2020-02-04 | 2020-05-15 | 北京科技大学 | Voltage-sensor-free power prediction control method for three-phase voltage type PWM rectifier |
| CN112051480A (en) * | 2020-08-06 | 2020-12-08 | 国网安徽省电力有限公司电力科学研究院 | Neural network power distribution network fault diagnosis method and system based on variational modal decomposition |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105093132B (en) * | 2015-06-30 | 2018-05-22 | 许昌许继软件技术有限公司 | A kind of heavy-duty rectifier open fault rapid on-line diagnosis method |
| CN106160533B (en) * | 2016-08-12 | 2018-04-27 | 大连理工大学 | A kind of pulse rectifier sensor fault fault tolerant control method based on sliding mode observer |
| CN110133423B (en) * | 2019-05-23 | 2020-08-07 | 西南交通大学 | Open-circuit fault diagnosis method for switching tube of single-phase cascaded H-bridge rectifier |
| CN112782614A (en) * | 2019-11-11 | 2021-05-11 | 株洲中车时代电气股份有限公司 | Fault early warning method and device of converter based on multi-information fusion |
| CN111308392B (en) * | 2020-03-06 | 2021-03-16 | 西南交通大学 | Method for diagnosing IGBT open-circuit fault of single-phase cascaded NPC rectifier |
| CN112731051B (en) * | 2020-12-18 | 2021-10-08 | 武汉大学 | Single-ended fault positioning method and system for high-voltage direct-current transmission line of hybrid network |
| CN112798982B (en) * | 2020-12-29 | 2021-12-14 | 武汉大学 | Model-based open-circuit fault diagnosis method and system for three-phase converter power tube |
| CN114841553A (en) * | 2022-04-27 | 2022-08-02 | 西安热工研究院有限公司 | Fault diagnosis method, system, equipment and storage medium for generator rotating rectifier |
| CN115659133A (en) * | 2022-10-26 | 2023-01-31 | 国电电力云南新能源开发有限公司 | Wind power converter fault diagnosis method |
-
2023
- 2023-03-09 CN CN202310222390.6A patent/CN115935244B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2017099103A (en) * | 2015-11-20 | 2017-06-01 | 株式会社東芝 | Rotary rectifier fault detector and rotary rectifier fault detection method |
| CN110957762A (en) * | 2019-12-11 | 2020-04-03 | 中南大学 | Fault classification method, system and storage medium for three-level traction converter |
| CN111162684A (en) * | 2020-02-04 | 2020-05-15 | 北京科技大学 | Voltage-sensor-free power prediction control method for three-phase voltage type PWM rectifier |
| CN112051480A (en) * | 2020-08-06 | 2020-12-08 | 国网安徽省电力有限公司电力科学研究院 | Neural network power distribution network fault diagnosis method and system based on variational modal decomposition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115935244A (en) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103761372B (en) | A kind of multi-electrical level inverter Fault Diagnosis Strategy based on pivot analysis with more classification Method Using Relevance Vector Machines | |
| CN112710969A (en) | Open-circuit fault diagnosis method for switching tube of single-phase half-bridge five-level inverter | |
| Yuan et al. | Open-circuit fault diagnosis of NPC inverter based on improved 1-D CNN network | |
| CN112435142B (en) | Power load identification method and load power utilization facility knowledge base construction method thereof | |
| Lin et al. | Parametric fault diagnosis based on fuzzy cerebellar model neural networks | |
| CN115935244B (en) | Single-phase rectifier fault diagnosis method based on data driving | |
| Jia et al. | DC-DC Buck circuit fault diagnosis with insufficient state data based on deep model and transfer strategy | |
| CN113673768A (en) | Wind power probability prediction model based on VMD and CNN-LSTM | |
| Zhu et al. | Aiming to Complex Power Quality Disturbances: A Novel Decomposition and Detection Framework | |
| CN115035336B (en) | GASF and CNN-based three-level T-type inverter fault diagnosis method | |
| CN105426923A (en) | Semi-supervised classification method and system | |
| CN112465250B (en) | Power load prediction method, device, computer equipment and storage medium | |
| Jiang et al. | Adaptive Sparse Attention Wavelet Network for the Robust Open-circuit Fault Diagnosis in PMSM Drives | |
| Dang et al. | seq2graph: discovering dynamic dependencies from multivariate time series with multi-level attention | |
| CN116664870A (en) | Power quality disturbance identification method based on transfer learning | |
| Shang et al. | Effinformer: A deep-learning-based data-driven modeling of DC–DC bidirectional converters | |
| CN116566061A (en) | Grid-connected inverter system stability on-line monitoring method and system | |
| CN116129212A (en) | Power load prediction method and system based on graph neural network | |
| CN115019129A (en) | Double-active-bridge converter fault diagnosis method and system based on time sequence imaging and image fusion | |
| CN114019310A (en) | Method, equipment and medium for prejudging open circuit fault of energy storage converter | |
| Wan et al. | A review of power electronic parametric fault diagnosis methods | |
| CN105654128A (en) | Kernel norm regularized low-rank coding-based fan blade image fault identification method | |
| CN118611059B (en) | Offshore wind power prediction method of attention mechanism double-channel neural network | |
| CN111913461A (en) | Distributed chemical process monitoring method based on regularization GCCA model | |
| Toussi et al. | An Application of Feature Engineering and Machine Learning Algorithms on Condition Monitoring of SiC Converters |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |