CN117113180A - Direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network - Google Patents
Direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network Download PDFInfo
- Publication number
- CN117113180A CN117113180A CN202311062882.XA CN202311062882A CN117113180A CN 117113180 A CN117113180 A CN 117113180A CN 202311062882 A CN202311062882 A CN 202311062882A CN 117113180 A CN117113180 A CN 117113180A
- Authority
- CN
- China
- Prior art keywords
- neural network
- convolutional neural
- vmd
- layer
- imf
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 47
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 26
- 238000013528 artificial neural network Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 48
- 238000011176 pooling Methods 0.000 claims description 26
- 230000008569 process Effects 0.000 claims description 22
- 230000001052 transient effect Effects 0.000 claims description 22
- 238000004422 calculation algorithm Methods 0.000 claims description 14
- 238000001914 filtration Methods 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 7
- 238000013178 mathematical model Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 239000013598 vector Substances 0.000 claims description 5
- 241000255581 Drosophila <fruit fly, genus> Species 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 4
- 230000006835 compression Effects 0.000 claims description 3
- 238000007906 compression Methods 0.000 claims description 3
- 230000009022 nonlinear effect Effects 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 2
- 230000008447 perception Effects 0.000 claims description 2
- 239000000284 extract Substances 0.000 abstract description 6
- 238000012952 Resampling Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000010606 normalization Methods 0.000 description 4
- 238000001514 detection method Methods 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000002441 reversible effect Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 101100499229 Mus musculus Dhrsx gene Proteins 0.000 description 1
- 230000003213 activating effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000002759 z-score normalization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/081—Locating faults in cables, transmission lines, or networks according to type of conductors
- G01R31/086—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution networks, i.e. with interconnected conductors
-
- 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
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/088—Aspects of digital computing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The application relates to a direct current distribution network high-resistance fault identification method based on VMD and a convolutional neural network, which decomposes an input signal into a plurality of intrinsic mode components IMF through changing a modal decomposition VMD, wherein each IMF comprises specific frequency range and amplitude information, the convolutional neural network can better combine the time-frequency characteristic provided by the VMD and the convolutional operation of CNN, extract richer and meaningful characteristics, improve the accuracy and robustness of fault identification, solve the problems of more parameters, large calculated amount, overfitting and the like of the neural network through an acceptance module in the convolutional network, improve the generalization capability of the model, and keep better performance when facing new data.
Description
Technical Field
The application relates to the field of power system fault detection, in particular to a direct current distribution network high-resistance fault identification method based on a VMD and a convolutional neural network.
Background
With the large-scale sudden rise of the power grid data and the great improvement of the computing capacity, the intelligent algorithm of the artificial neural network shows great superiority, with the enhancement of the network depth, the dimension reduction and the processing capacity of the data are further improved, and the information useful for fault discrimination can be accurately, effectively and automatically extracted from the influence of complex external and internal environments.
For example, CN111598166a "single-phase earth fault classification method and system based on principal component analysis and Softmax function" discloses "single-phase earth fault classification method and system based on principal component analysis and Softmax function, the method firstly obtains fault components of bus three-phase voltage, three-phase current, zero-sequence voltage and zero-sequence current for one to two weeks after fault occurs, then performs singular value decomposition on a data matrix formed by the fault components, performs principal component decomposition on a covariance matrix, and extracts fault feature values and normalized feature vectors; reducing the dimension of the fault feature vector matrix to obtain an output matrix; inputting the output matrix into a softmax classifier, and training a fault characteristic quantity training sample; and estimating a probability value to realize fault type identification. Based on the method provided by the application, the application also provides a classification system. The method is based on actual power distribution network fault recording data, realizes the classification of single-phase earth faults by using a fault equipment classification method, establishes a more accurate and faster data classification model ", but has the problems of more parameters, large calculated amount and easy overfitting along with the increase of network depth and width.
Therefore, how to improve the accuracy of fault identification on the basis of solving the problem of network model overfitting becomes a popular research direction of fault identification of the current power distribution network.
Disclosure of Invention
Based on the technical problems, the application provides a direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network, which comprises the following specific steps:
collecting a positive transient voltage time domain waveform signal of a fault point of a direct current distribution network system, and preprocessing the positive transient voltage time domain waveform signal to obtain a preprocessed signal;
VMD decomposition is carried out on the preprocessed signals to obtain intrinsic mode components IMF;
constructing a convolutional neural network classifier, training the convolutional neural network classifier by using the IMF to obtain a trained convolutional neural network classifier, and performing fault recognition and classification on data to be recognized by using the trained convolutional neural network classifier.
Preferably, the preprocessing of the positive transient voltage time domain waveform signal to obtain a preprocessed signal is specifically filtering and denoising the positive transient voltage time domain waveform signal to obtain a preprocessed signal, wherein the filtering is expressed as follows:
V filtered (t)＝H filter (V(t))；
wherein V (t) is a positive transient voltage time domain waveform signal, H filter () V is the transfer function of the filter filtered (t) is a filtered waveform;
denoising by using a wavelet denoising algorithm, wherein the denoising is expressed as follows:
W＝wavelet transform (V filtered (t))；
wherein V is filtered (t) is a filtered waveform, wavelet transform () As a wavelet transform function, W is a wavelet coefficient;
T＝threshold estimate (W)；
in the threshold estimate () T represents a threshold value as a function of estimating the threshold value from the statistical characteristics of the wavelet coefficients;
W denoised ＝threshold process (W,T)；
in the threshold process () W as a function of processing wavelet coefficients according to a threshold denoised The processed wavelet coefficients;
V denoised (t)＝inverse_wavelet transform (W_denoised)；
in the reverse_wavelet transform () As an inverse wavelet transform function, V denoised And (t) is a denoised waveform signal, i.e. a pre-processed signal.
Preferably, VMD decomposition is performed on the preprocessed signal to obtain IMF, which is expressed as:
V denoised (t)＝∑[u k (t)+r(t)]；
wherein u is k (t) is the kth IMF, and r (t) is the remainder;
wherein, the VMD decomposition constraint is expressed as:
in the formula, { u k The k IMF sets obtained by decomposition are { omega } k The frequency center set corresponding to each IMF is shown, j is the imaginary part of the complex number, delta (t) is the dirac distribution,is the partial derivative of the variable t;
introducing a quadratic penalty term alpha and a Lagrangian algorithm multiplier lambda (t) to convert the constraint problem into an unconstrained problem, and calculating the unconstrained problem by using an alternating direction multiplier method to obtain k IMFs, wherein the unconstrained problem is expressed as the following formula:
preferably, the VMD decomposition of the preprocessed signal to obtain the IMF further includes optimizing parameters k and α in the VMD decomposition process by using a drosophila optimization algorithm.
Preferably, the constructed convolutional neural network classifier comprises a convolutional layer, a pooling layer, an acceptance module, a Dropout layer and a full connection layer, the IMF is used as input data, local perception is carried out on the IMF data by using a convolutional kernel, time sequence features are extracted, downsampling and compression are carried out on the time sequence features through pooling operation, the acceptance module is used for capturing combined features of the IMF, higher-level time sequence features are extracted from the combined features through the convolutional layer and the pooling layer which are connected with the acceptance module, the Dropout layer is used for reducing the risk of overfitting, the features mapped through the Dropout layer are converted into one-dimensional vectors, and fault recognition is carried out through the full connection layer.
Preferably, the acceptance module comprises 3 parallel 1x1 convolution kernels, wherein two 1x1 convolution kernels are respectively connected with 1 3x3 convolution kernels and 1x 5 convolution kernels and are used for extracting features of different scales, the acceptance module further comprises a 3x3 maximum value pooling layer, and the 3x3 maximum value pooling layer outputs to a new 1x1 convolution kernel and is used for carrying out downsampling on the extracted features of different scales, filtering and combining to obtain output features.
Preferably, the convolutional neural network classifier introduces a nonlinear property function, namely an activation function, into a convolutional layer and a fully connected layer, and is expressed as follows:
f(x)＝max(0,x)；
the pooling layer adopts maximum pooling, and the mathematical model expression is:
where down () is the pool sampling function; beta is the network multiplicative bias;
the mathematical model of the convolutional neural network is expressed as:
in the method, in the process of the application,is the output of the jth neuron of the first layer; />Is the input to the layer 1, i neuron; m is M j Is an input feature map; l is a layer-1 network; omega is a weight matrix; />Is the bias of the jth neural network on the first layer;
the loss function of the convolutional neural network classifier is a cross entropy function and expressed as:
where n is the total number of samples of the input data, t is the predicted value, and y is the actual value.
Compared with the prior art, the application has the beneficial effects that:
1. the application provides a direct current distribution network high-resistance fault identification method based on a VMD and a convolutional neural network, which can decompose a fault signal into a plurality of modal components by using a VMD decomposition technology, wherein each component contains specific frequency range and amplitude information, so that the convolutional neural network can better combine the time-frequency characteristic provided by the VMD and the convolutional operation of the CNN, extract richer and meaningful characteristics and improve the accuracy and the robustness of fault identification;
2. the application provides a direct current distribution network high-resistance fault identification method based on a VMD and a convolutional neural network, which reduces the parameter quantity and the calculation complexity of the neural network by using an acceptance module, and the acceptance module can effectively reduce the parameter quantity in the network, improve the calculation efficiency, simultaneously capture the characteristics of a plurality of scales and layers, and simultaneously reduce the calculation complexity while maintaining higher performance by combining and characteristic splicing of a plurality of convolutional kernels.
3. The application provides a direct current distribution network high-resistance fault identification method based on a VMD and a convolutional neural network, which provides richer and diversified feature representation by using a VMD decomposition and acceptance module in an overall architecture, reduces the sensitivity of a model to partial noise or redundant features, thereby effectively reducing the occurrence of over-fitting, improving the generalization capability of the overall model of the convolutional neural network classifier, and keeping better performance in the face of new data.
Drawings
FIG. 1 is a schematic flow diagram of a method according to an embodiment of the present application;
FIG. 2 is a diagram of a 10kv flexible DC distribution network architecture in an embodiment of the application;
FIG. 3 is a model block diagram of a convolutional neural network classifier in an embodiment of the present application;
FIG. 4 is a model diagram of the acceptance module in an embodiment of the application.
Detailed Description
The following description of the embodiments of the present application will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Example 1
In the embodiment of the application, a PSCAD/EMTDC simulation platform is adopted to build a + -10 kV direct current distribution network structure as shown in figure 4, the voltage of an alternating current side is 10kV, a delta/Yn type grounding mode through a large resistor is adopted by an alternating current side transformer, the system frequency is 50Hz, the reactance of a bridge arm is 10mH, the capacitance of a submodule is 4500uF, the number of the submodules is 50, a cable is connected with an alternating current power grid through a multi-level converter (MMC), the system is a small current grounding system, when a single-pole grounding fault occurs in a direct current circuit, the fault current does not have a grounding loop, the current of the direct current circuit is still a rated value, the zero potential of the system shifts, the voltage of a grounding pole circuit drops to 0, the voltage of the non-grounding pole circuit rises to twice the original voltage, the interelectrode voltage remains unchanged, and the system can still operate for two hours after the single-pole grounding fault occurs.
Compared with a monopolar grounding fault, the interelectrode short-circuit fault of the direct current distribution network is more serious, the fault can cause current to rise sharply, the voltage of the positive electrode and the negative electrode of the grounding electrode is reduced to 0 rapidly, and the interelectrode voltage is also reduced to 0. If the fault cannot be timely removed after the converter is locked, the system will continue to maintain in the state, and the safety of the distribution network equipment is damaged. The zero sequence voltage component in the asymmetric fault of the alternating current side can cause power frequency common mode fluctuation of the positive and negative voltage of the direct current side, and the transient characteristic of the power frequency common mode fluctuation is similar to the situation of the direct current high-resistance grounding fault. Therefore, it is necessary to identify and maintain the fault of the direct current distribution network in time, and the application provides a direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network, comprising the following specific steps:
s1, acquiring a positive transient voltage time domain waveform signal of a fault point of a direct current distribution network system, and preprocessing the positive transient voltage time domain waveform signal to obtain a preprocessed signal;
collecting a positive transient voltage time domain waveform signal of a fault point when a built 10kV flexible direct current distribution network structure performs fault test, and preprocessing the positive transient voltage time domain waveform signal to obtain a preprocessed signal, wherein:
s11, signal preprocessing
Preprocessing the positive transient voltage time domain waveform signal to obtain a preprocessed signal, specifically filtering and denoising the positive transient voltage time domain waveform signal to obtain a preprocessed signal, wherein the filtering is expressed as follows in a formula:
V filtered (t)＝H filter (V(t))；
wherein V (t) is a positive transient voltage time domain waveform signal, H filter () V is the transfer function of the filter filtered (t) is a filtered waveform;
denoising by using a wavelet denoising algorithm, wherein the denoising is expressed as follows:
W＝wavelet transform (V filtered (t))；
wherein V is filtered (t) is a filtered waveform, wavelet transform () As a wavelet transform function, W is a wavelet coefficient;
T＝threshold estimate (W)；
in the threshold estimate () T represents a threshold value as a function of estimating the threshold value from the statistical characteristics of the wavelet coefficients;
W denoised ＝threshold process (W,T)；
in the threshold process () W as a function of processing wavelet coefficients according to a threshold denoised The processed wavelet coefficients;
V denoised (t)＝inverse_wavelet transform (W_denoised)；
in the reverse_wavelet transform () As an inverse wavelet transform function, V denoised (t) is a denoised waveform signal, i.e. a pre-processed signal;
it should be noted that, in this embodiment, the signal is preprocessed by denoising and filtering, which is not limited to denoising and filtering, and the following signal preprocessing may be performed according to the needs of the user by performing trend removal, resampling and normalization as well:
trending: removing trend components of the positive transient voltage time domain waveform signals by using linear fitting, polynomial fitting and other methods to eliminate the influence of long-term change, obtaining trend functions of the positive transient voltage time domain waveform signals, and subtracting the trend functions from the original data;
resampling: resampling the positive transient voltage time domain waveform signal according to actual needs to adjust sampling rate or time resolution, wherein common resampling methods include linear interpolation, nearest neighbor interpolation, spline interpolation and the like;
normalization: the positive transient voltage time domain waveform signal is normalized to make the value range fall within a specific range, so that the subsequent processing is convenient, and common normalization methods include maximum and minimum normalization, Z-score normalization and the like.
S2, performing VMD decomposition on the preprocessed signals to obtain intrinsic mode components IMF;
VMD is a time-frequency analysis algorithm, through redefining a signal capable of adjusting amplitude and frequency, the original signal is disassembled into a series of intrinsic mode signals (Intrinsic Mode Function, IMF), the variational problem is constructed and solved, useful components in the frequency domain are extracted, modal aliasing and end-point effects are overcome, the algorithm has certain anti-interference capability, fault signals can be comprehensively decomposed, hidden characteristic information in the signals is obtained, and the optimal solution of the variational problem is obtained.
There are two constraints on the VMD algorithm: (1) the sum of the modes is equal to the input signal f; under the constraint, the optimal solution of the model is searched iteratively to obtain the center frequency and bandwidth of each decomposition component. (2) The sum of the estimated bandwidths of the center frequencies of the eigenmode functions uk (t) is minimized by constructing and solving the variation problem.
In this embodiment, VMD decomposition is performed on the preprocessed signal to obtain IMF, which is expressed as:
V denoised (t)＝∑[u k (t)+r(t)]；
wherein u is k (t) is the kth IMF, and r (t) is the remainder;
for each IMF, computing [ u ] by Hilbert transform k The analytical signal of (t) is expressed as:
by estimating the center frequency omega of each analysis signal k Multiplying the single-sided spectrum obtained by the above formula by an exponential term signal, thereby converting the spectrum of each analysis signal into a baseband, and expressing the baseband as:
demodulating the analysis signal through Gaussian smoothing, preventing overfitting, estimating the bandwidth of each IMF, and finally obtaining VMD decomposition constraint conditions, wherein the VMD decomposition constraint conditions are expressed as follows:
in the formula, { u k The k IMF sets obtained by decomposition are { omega } k The frequency center set corresponding to each IMF is represented by j, the imaginary part of the complex number is represented by delta (t), the dirac distribution is represented by theta t Is the partial derivative of the variable t;
introducing a quadratic penalty term alpha and a Lagrangian algorithm multiplier lambda (t) to convert the constraint problem into an unconstrained problem, and calculating the unconstrained problem by using an alternating direction multiplier method to obtain k IMFs, wherein the unconstrained problem is expressed as the following formula:
solving the unconstrained problem by using an alternate direction multiplier method, thereby decomposing to obtain k IMFs, and obtaining u in the decomposition process k ，ω k The updated expression for λ is as follows:
sequentially according to the above formulaLoop iteration solution, in the iteration process, u of each IMF is continuously updated k ，ω k Outputting k IMFs until the three parameters lambda meet the discrimination precision;
preferably, the performing VMD decomposition on the preprocessed signals to obtain IMFs further includes optimizing parameters k and α in the VMD decomposition process by using a drosophila optimization algorithm, specifically, calculating an approximate entropy value of each IMF to serve as an objective function, and optimizing the parameters k and α by using iterative optimization capability of the drosophila optimization algorithm to find an optimal solution.
S3, constructing a convolutional neural network classifier, training the convolutional neural network classifier by using the IMF to obtain a trained convolutional neural network classifier, and performing fault recognition and classification on data to be identified by using the trained convolutional neural network classifier;
CNN is a supervised machine learning, and has been widely used in image recognition, object detection and fault recognition, and the main learning process is divided into a forward Propagation (Forward Propagation, FP) process and a reverse parameter update (BP), where the forward Propagation mainly includes a convolution layer (Convolution layer), a Pooling layer (Pooling layer), and a full-link layer (process layer), and the basic model structure is shown in fig. 1, and this process can implement extraction and pre-classification of the pre-processed signal, and the reverse parameter update can compare the pre-classified result with the expected result, automatically adjust the learnable parameters of the model, and implement accurate classification of the fault class;
in this embodiment, the constructed convolutional neural network classifier includes a convolutional layer, a pooling layer, an acceptance module, a dropoff layer and a full connection layer, takes an IMF as input data, uses a convolutional kernel to perform local sensing on the IMF data, extracts time sequence features, performs downsampling and compression on the time sequence features through pooling operation, the acceptance module is used for capturing combined features of the IMF, extracts higher-level time sequence features from the combined features through the convolutional layer and the pooling layer connected to the acceptance module, and the dropoff layer is used for reducing the risk of overfitting, mapping the features passing through the dropoff layer into one-dimensional vectors, and performing fault recognition through the full connection layer;
s31, an acceptance module;
an increase in network depth or width can cause two problems with convolutional networks: 1) Parameters required to be trained by the network are increased continuously in the process of increasing the network layer number, and the problem of fitting is inevitably brought; 2) As required training parameters increase, the model training speed also decreases, so that the convolution model is difficult to apply to actual engineering; accordingly, in the embodiment, an acceptance module is introduced into the convolutional neural network, and the core idea of the module is to combine different convolutional layers together in a parallel manner;
preferably, the acceptance module comprises 3 parallel 1x1 convolution kernels, wherein two 1x1 convolution kernels are respectively connected with 1 3x3 convolution kernels and 1x 5 convolution kernels and are used for extracting features of different scales, the acceptance module further comprises a 3x3 maximum value pooling layer, and the 3x3 maximum value pooling layer outputs to a new 1x1 convolution kernel and is used for carrying out downsampling on the extracted features of different scales, filtering and combining to obtain output features; the acceptance module increases the depth and the width of the network, reduces the dimension of data, converts the full-connection structure into sparse connection, effectively reduces the parameter quantity and remarkably improves the model accuracy;
s32, activating functions and pooling layers;
preferably, the convolutional neural network classifier introduces a nonlinear property function, namely an activation function, into a convolutional layer and a fully connected layer, and is expressed as follows:
f(x)＝max(0,x)；
if the input is greater than 0, the input value is directly returned, if the output is less than or equal to 0, the input value is returned to 0, and compared with the Tanh and Sigmoid functions which are frequently used, the Relu can accelerate the training speed of the model, reduce the calculation difficulty, has strong robustness and solves the gradient vanishing problem to a certain extent;
the pool sampling layer extracts local features, can detect the same features at different positions, has better space and structural invariance, and commonly has two sampling modes of maximum value pooling and average pooling, and in the embodiment, the maximum value pooling is adopted, and the mathematical model expression is as follows:
where down () is the pool sampling function; beta is the network multiplicative bias; after pool sampling, the characteristics of the sampling layer and the convolution layer are kept unchanged in quantity, but the size is reduced by n times;
after multiple convolution pooling, the neuron weights are connected by adopting a full connection layer, and the probability of each output is placed in [0,1] by an activation function, so that data classification is carried out on different characteristics;
s33, expressing a mathematical model of the convolutional neural network as a formula:
in the method, in the process of the application,is the output of the jth neuron of the first layer; />Is the input to the layer 1, i neuron; m is M j Is an input feature map; l is a layer-1 network; omega is a weight matrix; />Is the bias of the jth neural network on the first layer;
s34, loss function
For classification problems, to minimize the loss function of the model, the accuracy of the model is improved as much as possible, and therefore, the choice of the loss function is important. Common loss functions include a root mean square error function, an average absolute error function, a cross entropy cost function and the like, and the cross entropy function is selected as the loss function in the embodiment, and the loss function is expressed as follows:
wherein n is the total number of samples of the input data, t is a predicted value, and y is an actual value; in the back propagation process, the iteration process is continuously updated by a common gradient descent method, and the first derivative is obtained for the above formula, so that the network learnable parameters can be adjusted, and the method is as follows:
wherein ω' is the updated weight; b' is the updated bias; omega is the weight that is not updated; b is the non-updated bias; η is a learning rate parameter used to control the step size of the weight update.
The foregoing description is only illustrative of the present application and is not intended to limit the scope of the application, and all equivalent structures or equivalent processes or direct or indirect application in other related technical fields are included in the scope of the present application.
Claims (7)
1. The direct current distribution network high-resistance fault identification method based on the VMD and the convolutional neural network is characterized by comprising the following specific steps of:
collecting a positive transient voltage time domain waveform signal of a fault point of a direct current distribution network system, and preprocessing the positive transient voltage time domain waveform signal to obtain a preprocessed signal;
VMD decomposition is carried out on the preprocessed signals to obtain intrinsic mode components IMF;
constructing a convolutional neural network classifier, training the convolutional neural network classifier by using the IMF to obtain a trained convolutional neural network classifier, and performing fault recognition and classification on data to be recognized by using the trained convolutional neural network classifier.
2. The method for identifying a direct current distribution network high-resistance fault based on a VMD and a convolutional neural network according to claim 1, wherein the preprocessing of the positive transient voltage time domain waveform signal to obtain a preprocessed signal is specifically filtering and denoising the positive transient voltage time domain waveform signal to obtain a preprocessed signal, wherein the filtering is expressed as:
V filtered (t)＝H filter (V(t))；
wherein V (t) is a positive transient voltage time domain waveform signal, H filter () V is the transfer function of the filter filtered (t) is a filtered waveform;
denoising by using a wavelet denoising algorithm, wherein the denoising is expressed as follows:
W＝wavelet transform (V filtered (t))；
wherein V is filtered (t) is a filtered waveform, wavelet transform () As a wavelet transform function, W is a wavelet coefficient;
T＝threshold estimate (W)；
in the threshold estimate () T represents a threshold value as a function of estimating the threshold value from the statistical characteristics of the wavelet coefficients;
W denoised ＝threshold process (W,T)；
in the threshold process () W as a function of processing wavelet coefficients according to a threshold denoised The processed wavelet coefficients;
V denoised (t)＝inverse_wavelet transform (W_denoised)；
in the reverse_wavelet transform () As an inverse wavelet transform function, V denoised And (t) is a denoised waveform signal, i.e. a pre-processed signal.
3. The method for identifying the direct current distribution network high-resistance fault based on the VMD and the convolutional neural network according to claim 2, wherein the pre-processed signal is subjected to VMD decomposition to obtain the IMF, and the IMF is expressed as the following formula:
V denoised (t)＝∑[u k (t)+r(t)]；
wherein u is k (t) is the kth IMF, and r (t) is the remainder;
wherein, the VMD decomposition constraint is expressed as:
in the formula, { u k The k IMF sets obtained by decomposition are { omega } k The frequency center set corresponding to each IMF is shown, j is the imaginary part of the complex number, delta (t) is the dirac distribution,is the partial derivative of the variable t;
introducing a quadratic penalty term alpha and a Lagrangian algorithm multiplier lambda (t) to convert the constraint problem into an unconstrained problem, and calculating the unconstrained problem by using an alternating direction multiplier method to obtain k IMFs, wherein the unconstrained problem is expressed as the following formula:
4. the method for identifying direct current distribution network high-resistance faults based on VMD and convolutional neural network as claimed in claim 3, wherein said performing VMD decomposition on the preprocessed signals to obtain IMFs further comprises optimizing parameters k and alpha in the VMD decomposition process by using a Drosophila optimization algorithm.
5. The direct current distribution network high-resistance fault identification method based on the VMD and the convolutional neural network according to claim 1, wherein the constructed convolutional neural network classifier comprises a convolutional layer, a pooling layer, an acceptance module, a Dropout layer and a full connection layer, the IMF is used as input data, local perception is carried out on the IMF data by using a convolutional kernel, time sequence features are extracted, downsampling and compression are carried out on the time sequence features through pooling operation, the acceptance module is used for capturing combined features of the IMF, higher-level time sequence features are extracted on the combined features through the convolutional layer and the pooling layer which are connected with the acceptance module, the Dropout layer is used for reducing fitting risk, feature mapping through the Dropout layer is converted into one-dimensional vectors, and fault identification is carried out through the full connection layer.
6. The method for identifying direct current distribution network high-resistance faults based on VMD and convolutional neural network according to claim 5, wherein said acceptance module comprises 3 parallel 1x1 convolution kernels, wherein two 1x1 convolution kernels are respectively connected with 1 3x3 convolution kernels and 1x 5 convolution kernels for extracting features of different scales, said acceptance module further comprises a 3x3 maximum pooling layer, said 3x3 maximum pooling layer outputs to a new 1x1 convolution kernel for downsampling the extracted features of different scales, filtering and combining to obtain output features.
7. The method for identifying the direct current distribution network high-resistance fault based on the VMD and the convolutional neural network according to claim 5, wherein the convolutional neural network classifier introduces a nonlinear property function, namely an activation function, into a convolutional layer and a full connection layer, and is expressed as:
f(x)＝max(0,x)；
the pooling layer adopts maximum pooling, and the mathematical model expression is:
where down () is the pool sampling function; beta is the network multiplicative bias;
the mathematical model of the convolutional neural network is expressed as:
in the method, in the process of the application,is the output of the jth neuron of the first layer; />Is the input to the layer 1, i neuron; m is M j Is an input feature map; l is a layer-1 network; omega is a weight matrix; />Is the bias of the jth neural network on the first layer;
the loss function of the convolutional neural network classifier is a cross entropy function and expressed as:
where n is the total number of samples of the input data, t is the predicted value, and y is the actual value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311062882.XA CN117113180A (en) | 2023-08-22 | 2023-08-22 | Direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311062882.XA CN117113180A (en) | 2023-08-22 | 2023-08-22 | Direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117113180A true CN117113180A (en) | 2023-11-24 |
Family
ID=88804972
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311062882.XA Pending CN117113180A (en) | 2023-08-22 | 2023-08-22 | Direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117113180A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117872039A (en) * | 2024-03-13 | 2024-04-12 | 国网山东省电力公司鱼台县供电公司 | Line fault location method and system based on improved RBF network |
| CN117929952A (en) * | 2024-03-21 | 2024-04-26 | 国网(山东)电动汽车服务有限公司 | Novel arc fault detection method for electric automobile charging pile |
| CN118152786A (en) * | 2024-05-10 | 2024-06-07 | 中国矿业大学 | Knowledge graph-based equipment fault auxiliary decision-making method, system and storage medium |
-
2023
- 2023-08-22 CN CN202311062882.XA patent/CN117113180A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117872039A (en) * | 2024-03-13 | 2024-04-12 | 国网山东省电力公司鱼台县供电公司 | Line fault location method and system based on improved RBF network |
| CN117872039B (en) * | 2024-03-13 | 2024-05-31 | 国网山东省电力公司鱼台县供电公司 | Line fault location method and system based on improved RBF network |
| CN117929952A (en) * | 2024-03-21 | 2024-04-26 | 国网(山东)电动汽车服务有限公司 | Novel arc fault detection method for electric automobile charging pile |
| CN117929952B (en) * | 2024-03-21 | 2024-05-28 | 国网(山东)电动汽车服务有限公司 | Novel arc fault detection method for electric automobile charging pile |
| CN118152786A (en) * | 2024-05-10 | 2024-06-07 | 中国矿业大学 | Knowledge graph-based equipment fault auxiliary decision-making method, system and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117113180A (en) | Direct current distribution network high-resistance fault identification method based on VMD and convolutional neural network | |
| CN114755745B (en) | Hail weather identification and classification method based on multi-channel depth residual shrinkage network | |
| CN109635928B (en) | Voltage sag reason identification method based on deep learning model fusion | |
| CN111307453B (en) | Transmission system fault diagnosis method based on multi-information fusion | |
| WO2021068454A1 (en) | Method for identifying energy of micro-energy device on basis of bp neural network | |
| WO2021212891A1 (en) | Fault arc signal detection method using convolutional neural network | |
| CN110829417B (en) | Electric power system transient stability prediction method based on LSTM double-structure model | |
| CN108333468B (en) | The recognition methods of bad data and device under a kind of active power distribution network | |
| CN112051480A (en) | Neural network power distribution network fault diagnosis method and system based on variational modal decomposition | |
| CN112557826A (en) | Ship electric power system fault diagnosis method | |
| CN102279358A (en) | MCSKPCA based neural network fault diagnosis method for analog circuits | |
| CN112070104A (en) | Main transformer partial discharge identification method | |
| CN112881942A (en) | Abnormal current diagnosis method and system based on wavelet decomposition and empirical mode decomposition | |
| CN114118150A (en) | Power distribution network single-phase earth fault line selection method and system | |
| CN116679161A (en) | Power grid line fault diagnosis method, equipment and medium | |
| CN114169377A (en) | G-MSCNN-based fault diagnosis method for rolling bearing in noisy environment | |
| CN112949142A (en) | ECT image reconstruction method based on deep neural network | |
| CN115204231A (en) | Digital human-computer interface cognitive load assessment method based on EEG (electroencephalogram) multi-dimensional features | |
| CN115238796A (en) | Motor imagery electroencephalogram signal classification method based on parallel DAMSCN-LSTM | |
| CN113705405B (en) | Nuclear pipeline fault diagnosis method | |
| CN110703006A (en) | Three-phase power quality disturbance detection method based on convolutional neural network | |
| CN112151067B (en) | Digital audio tampering passive detection method based on convolutional neural network | |
| CN116973677A (en) | Distribution network single-phase earth fault line selection method based on cavity convolution and attention mechanism | |
| CN115144211A (en) | Fault detection method and device based on discrete wavelet transform and gated cyclic unit | |
| CN114252725A (en) | HHT and ResNet 18-based single-phase earth fault type comprehensive identification method |
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 |