CN113030789A - Series arc fault diagnosis and line selection method based on convolutional neural network - Google Patents
Series arc fault diagnosis and line selection method based on convolutional neural network Download PDFInfo
- Publication number
- CN113030789A CN113030789A CN202110386461.7A CN202110386461A CN113030789A CN 113030789 A CN113030789 A CN 113030789A CN 202110386461 A CN202110386461 A CN 202110386461A CN 113030789 A CN113030789 A CN 113030789A
- Authority
- CN
- China
- Prior art keywords
- fault
- diagnosis
- neural network
- current
- arc
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/50—Testing of electric apparatus, lines, cables or components for short-circuits, continuity, leakage current or incorrect line connections
- G01R31/58—Testing of lines, cables or conductors
- G01R31/60—Identification of wires in a multicore cable
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- 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
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Emergency Protection Circuit Devices (AREA)
Abstract
The invention belongs to the field of fault arc diagnosis, and particularly relates to a series fault arc diagnosis and line selection method based on a convolutional neural network, which comprises the following steps of: 1) developing a series type fault arc experiment by using a series type fault arc experiment system; 2) obtaining main circuit current signals when arc faults occur on different branches and different phases, and directly taking the main circuit current signals as diagnosis model samples after classification, segmentation and standardization processing; 3) constructing a convolutional neural network model, and establishing a series fault arc diagnosis and line selection model based on the convolutional neural network through sample training; 4) and analyzing the serial arc fault diagnosis and line selection effects of the convolutional neural network model by comparing the accuracy and the loss function value, the online test speed and the accuracy of the optimized classification result. Compared with the prior art, the diagnosis of the arc fault and the selection of the fault branch are realized through the main circuit current under the condition of not analyzing the arc fault characteristics.
Description
Technical Field
The invention belongs to the field of fault arc diagnosis, and particularly relates to a series type arc fault diagnosis and line selection method based on a convolutional neural network.
Background
When a distribution line or electrical equipment runs in a severe environment for a long time, poor contact of a contact point can be caused to cause a series type arc fault, and the series type arc fault has concealment, randomness and complexity and is always lack of a mature detection method. In an industrial power distribution system, a plurality of motors are connected in parallel. When the series type fault arc occurs, if the branch circuit with the fault arc can be detected through the main circuit current, the investment of fault arc protection can be saved, and the time for overhauling and maintaining is reduced.
In recent years, a large amount of research is carried out by domestic and foreign scholars aiming at the problem of series arc fault diagnosis, and the research is mainly divided into the following aspects: firstly, constructing a feature vector according to physical characteristics of arc sound, arc light, electromagnetic radiation intensity and the like of the series arc fault, and realizing arc fault diagnosis by combining a mode identification method; and secondly, constructing a characteristic vector according to the electric signal characteristics of the current and the load terminal voltage of the loop where the series arc fault is located, and realizing the arc fault diagnosis by combining a mode identification method. The method is suitable for diagnosing the series arc fault characteristics in the closed space. The accuracy, typicality and comprehensiveness of the selected features influence the accuracy of the series arc fault identification.
The method comprises the steps of firstly, developing a series arc fault experiment of an industrial system, obtaining main circuit current signals when arc faults occur on different branches and phases, and directly taking the main circuit current signals as diagnosis model samples after classification, segmentation and standardization processing; then, constructing a convolutional neural network model, and establishing a convolutional neural network series type arc fault diagnosis model through sample training; carrying out optimization analysis on the online classification result of the network model through differential processing; and verifying the serial arc fault diagnosis and line selection effect of the model through accuracy, loss function values, online test speed and optimized classification results. Research results show that the convolutional neural network has certain theoretical reference value for the series arc fault diagnosis and fault line selection effects of the industrial system and the power generation arc fault circuit breaker.
Disclosure of Invention
In view of the above technical problems, the present invention provides a method for diagnosing and selecting a line of a series arc fault based on a convolutional neural network, which is based on a series arc fault experimental system, and the system comprises: the device comprises a three-phase alternating current power supply, a circuit breaker, a voltage transformer, a current transformer, an arc generator, a data acquisition card and an experimental load.
The three-phase alternating current power supply, the circuit breaker, the arc generator and the experimental load are connected in sequence; a primary coil of a current transformer is connected with the main loop in series, and a primary coil of a voltage transformer is connected with two ends of the arc generator in parallel; signals collected by the current transformer and the voltage transformer are processed and then are sent to a computer by a data acquisition card; the experimental load comprises a three-phase asynchronous motor M1 and a frequency converter thereof, and a three-phase asynchronous motor M2.
The method comprises the following steps.
And 2, dividing the main line current into 10 types according to working conditions, standardizing the data by adopting a standard deviation method, taking a main line current signal time sequence of a period after the standardization as a diagnosis model sample, and establishing a diagnosis sample library.
And 3, carrying out model architecture on the Convolutional Neural Network (CNN) and the long-short term memory network (LSTM).
And 4, performing 200 rounds of training on the CNN, the LSTM and the common neural network (ANN) model, and testing by using training set data and a test set. After 200 rounds of training, the accuracy rate of CNN and LSTM training sets is close to 96%, and the accuracy rate of test sets is close to 94%; the loss value of the CNN training set is close to 0.1, the loss value of the test set is close to 0.25, the loss value of the LSTM training set is close to 0.1, and the loss value of the test set is close to 0.2. The accuracy rate of ANN test is lower than 85%, and the convergence rate is obviously lower than that of the deep learning method.
And 5, implanting the three network models into raspberries to obtain online test time for performing multi-class identification on the test samples by the three models.
And 6, optimizing the classification results of the three neural network models by adopting a difference method to obtain the accuracy of the classification results before and after optimization.
And 7, comprehensively measuring the online diagnosis and line selection effects of the series fault arc according to the aspects of the test accuracy, the loss function value, the online test time and the optimized classification accuracy: the accuracy and the real-time performance of the CNN on the series arc fault diagnosis and the line selection of the industrial system are superior to those of the LSTM and the ANN.
The specific experimental steps of the series arc fault experiment are as follows: the 380V three-phase alternating current power supply drives two parallel three-phase alternating current asynchronous motors M1 and M2. Wherein M1 is 11KW motor, can select through change over switch and pass through converter operation and two kinds of operating modes of direct operation, and M1 can pass through magnetic particle brake and adjust load current. M2 is a 7.5KW motor, and is in no-load operation. The experimental system works under the experimental conditions that an ABC phase of a main circuit, an ABC phase of a branch circuit where the M1 motor is located, an ABC phase of a branch circuit where the M2 motor is located have an arc fault and the ABC phase runs normally without the fault. During the experiment, the moving contact and the static contact are well contacted to simulate a normal state, and the moving contact and the static contact are separated by controlling the moving contact driven by the stepping motor, so that a series type arc fault is generated. And storing the three-phase current data and the arc voltage data of the main circuit to a computer through a data acquisition card.
The diagnosis model sample is specifically divided into: the current coherent path fault, the current phase M1 motor branch fault, the current phase M2 motor branch fault, the leading coherent path fault, the leading phase M1 motor branch fault, the leading phase M2 motor branch fault, the lagging coherent path fault, the lagging phase M1 motor branch fault, the lagging phase M2 motor branch fault and the normal 10 types respectively correspond to the category labels 0-9, and the network output result of the diagnosis model also corresponds to the category labels 10. Wherein the number of the samples is 50289, each sample comprises a periodic current signal time sequence, namely 784 sampling points, 90% of the samples are randomly extracted for training the series arc fault diagnosis network model, and the rest 10% of the samples are used for verifying the network model.
The convolutional neural network model architecture comprises three convolutional layers, two maximum pooling layers and two full-connection layers. One set of data in the training set is a periodic time series of main current signals, 784 points, which are arranged into a 28 × 28 matrix as the input of the convolutional neural network. Each layer of the model is specifically as follows: the first convolution layer conv2d contains 32 convolution kernels of 3 × 3 size, 32 groups of 26 × 26 matrices are obtained after convolution operation, and the output data size is 13 × 13 and the number of channels is 32 after maximum pooling. The second convolution layer conv2d _1 contains 64 convolution kernels with the size of 3 × 3, and after convolution operation, 64 groups of 11 × 11 matrixes are obtained, and after maximum pooling, the output data size is 5 × 5, and the number of channels is 64. The third convolution layer conv2d _2 also contains 64 convolution kernels with the size of 3 × 3, the size of output data is 3 × 3, the number of channels is 64, and 64 groups of 3 × 3 matrixes are obtained after the convolution kernels go through flatten, wherein all the convolution layers use relu as an activation function, and the 3 rd convolution layer is spliced into a vector with the length of 576 and input into the first fully-connected layer dense. The number of nodes is 64, the activation function is relu, the second layer fully-connected layer dense _1 and the number of nodes is 10, and softmax is used as the activation function for classified output.
The long-short term memory network model architecture specifically comprises: a data group consisting of 784 points in the training set is divided into 28 time steps, and the length of the input data vector at each moment is 28 points. The first layer of the network is a fully connected layer dense, and the number of nodes is 128. The second layer is long short term memory unit LSTM, hidden node 128, activation function tanh, sequence input single output. The last level dense _1 is a fully connected level, has 10 nodes, and adopts softmax as an activation function. The network has a total of 136,586 parameters.
The invention has the beneficial effects.
The invention provides a series arc fault diagnosis and line selection method based on a convolutional neural network, which realizes the diagnosis of arc faults and the selection of fault branches through main circuit current under the condition of not performing characteristic analysis on the arc faults, and opens up a new idea for the diagnosis of the arc faults under the condition of multi-load parallel connection.
Drawings
FIG. 1 is a circuit diagram of a series type fault arc experimental system of the present invention.
Fig. 2(a) is a comparison graph of the current phase fault current of the circuit under different working conditions when the M1 of the invention does not have a frequency converter.
FIG. 2(b) is a comparative graph of phase current of the dry circuit lead under different operating conditions without the inverter of the M1 of the present invention.
FIG. 2(c) is a comparative graph of the current of the dry circuit lagging phase under different operating conditions when the M1 of the present invention is not provided with a frequency converter.
Fig. 2(d) is a partial enlarged diagram comparing the main circuit current under different working conditions without the frequency converter of the M1 of the present invention.
FIG. 3(a) is a comparison graph of the current fault phase current of the main circuit under different working conditions when the M1 frequency converter is provided.
FIG. 3(b) is a comparative graph of phase current of the lower circuit leading under different working conditions when the M1 frequency converter is provided.
FIG. 3(c) is a comparative graph of the current of the delay phase of the dry circuit under different working conditions when the M1 frequency converter is provided.
FIG. 3(d) is a partial enlarged view of the current of the shunt lagging phase under different operating conditions when the M1 frequency converter is provided.
FIG. 4 is a model architecture diagram of the convolutional neural network of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The invention provides a series type arc fault diagnosis and line selection method based on a convolutional neural network, which is based on a series type arc fault experiment system, and comprises the following steps: the device comprises a three-phase alternating current power supply, a circuit breaker, a voltage transformer, a current transformer, an arc generator, a data acquisition card and an experimental load.
The series fault arc experiment system is a three-phase alternating current power supply series fault arc experiment system. In the experimental system, as shown in fig. 1, a 380V three-phase ac power supply drives two parallel three-phase ac asynchronous motors M1 and M2. Wherein the rated power of M1 is 11KW, can select to take converter operation and direct operation two kinds of modes through change over switch to load current is adjusted through magnetic particle brake. M2 rated power is 7.5KW, no-load running. P1-P9 are access points of the arc fault generator, AF is the arc fault generator, C1, C2 and C3 are current sensors, V1 is a voltage sensor, and three-phase current and arc voltage of the main circuit are stored in a computer through a data acquisition card.
During the experiment of the series fault arc experiment system, the moving contact and the static contact are well contacted to simulate a normal state, and the moving contact and the static contact are separated by controlling the moving contact driven by the stepping motor to generate the series fault arc. And storing the three-phase current data and the arc voltage data of the main circuit to a computer through a data acquisition card.
Different experimental conditions of the series type fault arc experimental system are shown in table 1.
Table 1 experimental protocol.
The series type fault arc experiment system analyzes the experiment result of the series type arc fault of the experiment group 10-18 through the main circuit A phase current, and is shown in figure 2. As can be seen from fig. 2(a), when the phase a, i.e., the current phase, has a fault, the loop current at the position of the fault is larger, and the trunk current is smaller. As can be seen from fig. 2(b), when the C-phase, i.e., the leading-phase, fails, the rule that the larger the loop current at the position of the arc fault, the smaller the amplitude of the main-line current is still satisfied. However, when the leading phase of the motor load branch fails, the amplitude of the main circuit current is between the amplitude of the normal operation current and the amplitude of the main circuit fault current, and the influence of the amplitude of the fault branch current is basically avoided. As can be seen from fig. 2(c), when the B phase, i.e. the lagging phase, fails, the loop current is larger at the position of the failure, and the trunk current is larger. When the M2B phase, namely the lagging phase, has an arc fault, the main circuit current is already smaller than the normal operation current. The data of fig. 2(a) -2 (c) 0.013s-0.018s are partially enlarged to obtain fig. 2(d), and it is found that the larger the loop current at the fault position is, the larger the harmonic content in the main line is.
The series type fault arc experimental system analyzes experimental data of experimental groups 46-54, and the result is shown in fig. 3. As can be seen from FIG. 3, due to the influence of the nonlinearity of the frequency converter, the amplitudes of the current double peaks are similar when the M1 band frequency converter operates normally. When an arc fault occurs in the main circuit or the M1 branch circuit, the bimodal amplitude difference of the current waveform is increased, and the bimodal peak difference is maximum when the M1 branch circuit has a fault; when the M2 branch circuit has a fault, the M2 does not carry out frequency converter operation, so that the difference of the peak value of the current phase current of the main circuit is not obviously increased, and is only the superposition of the normal waveform of M1 and the fault characteristic of M2. The comparison shows that when the main circuit of the leading phase and the main circuit of the lagging phase and the branch circuit of M1 have arc faults, the peak value of the main circuit current is obviously increased compared with the current phase fault and normal operation, and the increase amplitude of the leading phase is larger. Meanwhile, the harmonic content in the trunk line is increased no matter what kind of faults occur.
According to the method, the main circuit current is divided into 10 types according to working conditions, the data is subjected to standardization processing by adopting a standard deviation method, a main circuit current signal time sequence in a period after the standardization processing is used as a diagnosis model sample, and a diagnosis sample library is established. The specific classification of the diagnostic sample library and the corresponding label are shown in table 2.
Table 2 sample label correspondence table.
The number of the samples is 50289, each sample comprises a periodic current signal time sequence, namely 784 sampling points, 90% of the samples are randomly extracted for training the network model, and the rest 10% of the samples are used for verifying the network model.
The invention constructs a convolutional neural network model, a group of data in a training set is a trunk current signal time sequence of one period, 784 points are total, and the trunk current signal time sequence is arranged into a 28-by-28 matrix as the input of the convolutional neural network. The model architecture is shown in FIG. 4, which includes three convolutional layers, two max-pooling layers, and two full-link layers. Each layer of the model is specifically as follows: the first convolution layer conv2d contains 32 convolution kernels of 3 × 3 size, 32 groups of 26 × 26 matrices are obtained after convolution operation, and the output data size is 13 × 13 and the number of channels is 32 after maximum pooling. The second convolution layer conv2d _1 contains 64 convolution kernels with the size of 3 × 3, and after convolution operation, 64 groups of 11 × 11 matrixes are obtained, and after maximum pooling, the output data size is 5 × 5, and the number of channels is 64. The third convolution layer conv2d _2 also contains 64 convolution kernels with the size of 3 × 3, the size of output data is 3 × 3, the number of channels is 64, and 64 groups of 3 × 3 matrixes are obtained after the convolution kernels go through flatten, wherein all the convolution layers use relu as an activation function, and the 3 rd convolution layer is spliced into a vector with the length of 576 and input into the first fully-connected layer dense. The number of nodes is 64, the activation function is relu, the second layer fully-connected layer dense _1 and the number of nodes is 10, and softmax is used as the activation function for classified output.
The calculation formula of the convolutional layer is shown as the formula (1).
Wherein
For the input vector of the convolutional layer, the vector is input,
is as follows
Go to the first
The convolutional layer of the row outputs a vector,
in order to activate the function(s),
to be connected to
A weight matrix of convolution kernels of the individual feature maps,
is the offset vector of the feature map.
The parameters of the convolutional neural network model proposed by the present invention are shown in table 3.
Table 3 convolutional neural network model parameters.
The architecture of the long-term and short-term memory network model specifically comprises the following steps: a data group consisting of 784 points in the training set is divided into 28 time steps, and the length of the input data vector at each moment is 28 points. The first layer of the network is a fully connected layer dense, and the number of nodes is 128. The second layer is long short term memory unit LSTM, hidden node 128, activation function tanh, sequence input single output. The last level dense _1 is a fully connected level, has 10 nodes, and adopts softmax as an activation function. The network has a total of 136,586 parameters.
The parameters of the long and short term memory network model proposed by the present invention are shown in table 4.
Table 4 long short term memory network model parameters.
The invention performs 200 rounds of training on CNN, LSTM and common neural network (ANN) models, and tests by using training set data and a test set. After 200 rounds of training, the accuracy rate of CNN and LSTM training sets is close to 96%, and the accuracy rate of test sets is close to 94%; the loss value of the CNN training set is close to 0.1, the loss value of the test set is close to 0.25, the loss value of the LSTM training set is close to 0.1, and the loss value of the test set is close to 0.2. The accuracy rate of ANN test is lower than 85%, and the convergence rate is obviously lower than that of the deep learning method.
According to the invention, three network models are implanted into the raspberry to obtain the online test time for performing multi-class identification on the test sample by the three models, and the online test time is specifically shown in Table 5.
Table 5 network model versus sample test time.
The classification results of the three neural network models are optimized by adopting a difference method, and the accuracy of the obtained classification results before and after optimization is shown in table 6.
Table 6 comparison of accuracy of classification results before and after optimization.
The method comprehensively measures the effects of the CNN on the series arc fault diagnosis and line selection of the industrial system according to the aspects of the test accuracy, the loss function value, the on-line test time and the optimized classification accuracy, and the accuracy and the real-time performance of the method are superior to those of the LSTM and the ANN.
Claims (5)
1. A series arc fault diagnosis and line selection method based on a convolutional neural network is based on a series arc fault experiment system, and the system comprises: the system comprises a three-phase alternating current power supply, a circuit breaker, a voltage transformer, a current transformer, an arc generator, a data acquisition card and an experimental load; the three-phase alternating current power supply, the circuit breaker, the arc generator and the experimental load are connected in sequence; a primary coil of a current transformer is connected with the main loop in series, and a primary coil of a voltage transformer is connected with two ends of the arc generator in parallel; signals collected by the current transformer and the voltage transformer are processed and then are sent to a computer by a data acquisition card; the experimental load comprises a three-phase asynchronous motor M1 and a frequency converter thereof, and a three-phase asynchronous motor M2;
the method is characterized by comprising the following steps:
step 1, performing a series arc fault experiment by using a series arc fault experiment system, acquiring main circuit arc fault current under different experiment conditions, and performing primary analysis on the main circuit current;
step 2, dividing the main line current into 10 types according to working conditions, standardizing data by adopting a standard deviation method, taking a main line current signal time sequence of a period after the standardization as a diagnosis model sample, and establishing a diagnosis sample library;
step 3, model architecture is carried out on the Convolutional Neural Network (CNN) and the long-short term memory network (LSTM);
step 4, performing 200 rounds of training on the CNN, the LSTM and the common neural network (ANN) model, and testing by using training set data and a test set; after 200 rounds of training, the accuracy rate of CNN and LSTM training sets is close to 96%, and the accuracy rate of test sets is close to 94%; the loss value of the CNN training set is close to 0.1, the loss value of the test set is close to 0.25, the loss value of the LSTM training set is close to 0.1, and the loss value of the test set is close to 0.2; the accuracy rate of ANN test is lower than 85%, and the convergence rate is obviously lower than that of the deep learning method;
step 5, implanting the three network models into raspberries to obtain online test time for performing multi-class identification on the test samples by the three models;
step 6, optimizing the classification results of the three neural network models by adopting a difference method to obtain the accuracy of the classification results before and after optimization;
and 7, comprehensively measuring the online diagnosis and line selection effects of the series fault arc according to the aspects of the test accuracy, the loss function value, the online test time and the optimized classification accuracy: the accuracy and the real-time performance of the CNN on the series arc fault diagnosis and the line selection of the industrial system are superior to those of the LSTM and the ANN.
2. The convolutional neural network-based series arc fault diagnosis and line selection method as claimed in claim 1, wherein the series arc fault experiment comprises the specific experimental steps of: the 380V three-phase alternating current power supply drives two parallel three-phase alternating current asynchronous motors M1 and M2; wherein M1 is 11KW motor, can select through the change-over switch and run through the frequency converter and run two kinds of running modes directly, M1 can adjust the load current through the magnetic particle brake; m2 is a 7.5KW motor, and is in no-load operation; the experimental system works under the experimental conditions that an ABC phase of a main circuit, an ABC phase of a branch circuit where the M1 motor is located, an ABC phase of a branch circuit where the M2 motor is located have an arc fault and the branch circuit runs normally without the fault; during the experiment, the moving contact and the static contact are well contacted to simulate a normal state, and the moving contact and the static contact are separated by controlling the moving contact driven by the stepping motor to generate a series arc fault; and storing the three-phase current data and the arc voltage data of the main circuit to a computer through a data acquisition card.
3. The convolutional neural network-based series arc fault diagnosis and line selection method as claimed in claim 1, wherein the diagnosis model samples are specifically divided into: the method comprises the following steps that a current coherent path fault, a current phase M1 motor branch fault, a current phase M2 motor branch fault, an advanced coherent path fault, an advanced phase M1 motor branch fault, an advanced phase M2 motor branch fault, a lagging coherent path fault, a lagging phase M1 motor branch fault, a lagging phase M2 motor branch fault and a normal class 10 are respectively corresponding to class labels 0-9, and a network output result of a diagnosis model also corresponds to the class labels 10; wherein the number of the samples is 50289, each sample comprises a periodic current signal time sequence, namely 784 sampling points, 90% of the samples are randomly extracted for training the series arc fault diagnosis network model, and the rest 10% of the samples are used for verifying the network model.
4. The convolutional neural network-based series arc fault diagnosis and line selection method as claimed in claim 1, wherein the convolutional neural network model architecture comprises three convolutional layers, two max-pooling layers, two full-connection layers; one group of data in the training set is a trunk circuit current signal time sequence of one period, 784 points in total are arranged into a 28-by-28 matrix which is used as the input of the convolutional neural network; each layer of the model is specifically as follows: the first layer of convolution layer conv2d contains 32 convolution kernels with the size of 3 × 3, 32 groups of 26 × 26 matrixes are obtained after convolution operation, the output data size is 13 × 13 after maximum pooling, and the number of channels is 32; the second convolution layer conv2d _1 contains 64 convolution kernels with the size of 3 × 3, 64 groups of 11 × 11 matrixes are obtained after convolution operation, the output data size is 5 × 5 after maximum pooling is carried out, and the number of channels is 64; the third convolution layer conv2d _2 also comprises 64 convolution kernels with the size of 3 × 3, the size of output data is 3 × 3, the number of channels is 64, 64 groups of 3 × 3 matrixes are obtained after the convolution kernels pass through flatten, wherein all the convolution layers use relu as an activation function, and the 3 rd convolution layer is spliced into a vector with the length of 576 and input into the first fully-connected layer dense; the number of nodes is 64, the activation function is relu, the second layer fully-connected layer dense _1 and the number of nodes is 10, and softmax is used as the activation function for classified output.
5. The convolutional neural network-based series arc fault diagnosis and line selection method as claimed in claim 1, wherein the long-short term memory network model architecture specifically comprises: dividing a data group consisting of 784 points in a training set into 28 time steps, wherein the length of a data vector input at each moment is 28 points; the first layer of the network is a full connection layer dense, and the number of nodes is 128; the second layer is a long-short term memory unit LSTM, the hidden node is 128, the activation function is tanh, and the sequence is input and output; the last layer dense _1 is a full connection layer, has 10 nodes and adopts softmax as an activation function; the network has a total of 136,586 parameters.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110386461.7A CN113030789A (en) | 2021-04-12 | 2021-04-12 | Series arc fault diagnosis and line selection method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110386461.7A CN113030789A (en) | 2021-04-12 | 2021-04-12 | Series arc fault diagnosis and line selection method based on convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113030789A true CN113030789A (en) | 2021-06-25 |
Family
ID=76456398
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110386461.7A Pending CN113030789A (en) | 2021-04-12 | 2021-04-12 | Series arc fault diagnosis and line selection method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113030789A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114282608A (en) * | 2021-12-22 | 2022-04-05 | 国网安徽省电力有限公司 | Hidden fault diagnosis and early warning method and system for current transformer |
| CN115014571A (en) * | 2022-05-25 | 2022-09-06 | 汕头大学 | Submarine cable risk event identification system and method |
| CN116990648A (en) * | 2023-09-26 | 2023-11-03 | 北京科技大学 | Fault arc detection method based on one-dimensional cavity convolutional neural network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080157781A1 (en) * | 2006-12-27 | 2008-07-03 | General Electric Company | Methods and systems for detecting series arcs in electrical systems |
| CN108646149A (en) * | 2018-04-28 | 2018-10-12 | 国网江苏省电力有限公司苏州供电分公司 | Fault electric arc recognition methods based on current characteristic extraction |
| CN110488161A (en) * | 2019-07-23 | 2019-11-22 | 南京航空航天大学 | A kind of detection of multi-load series arc faults and localization method |
| CN111123048A (en) * | 2019-12-23 | 2020-05-08 | 温州大学 | Series fault arc detection device and method based on convolutional neural network |
| CN111239569A (en) * | 2020-03-12 | 2020-06-05 | 杭州电弧数据科技有限公司 | Arc fault detection method, device, equipment and storage medium |
| CN111458599A (en) * | 2020-04-16 | 2020-07-28 | 福州大学 | Series arc fault detection method based on one-dimensional convolutional neural network |
-
2021
- 2021-04-12 CN CN202110386461.7A patent/CN113030789A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080157781A1 (en) * | 2006-12-27 | 2008-07-03 | General Electric Company | Methods and systems for detecting series arcs in electrical systems |
| CN108646149A (en) * | 2018-04-28 | 2018-10-12 | 国网江苏省电力有限公司苏州供电分公司 | Fault electric arc recognition methods based on current characteristic extraction |
| CN110488161A (en) * | 2019-07-23 | 2019-11-22 | 南京航空航天大学 | A kind of detection of multi-load series arc faults and localization method |
| CN111123048A (en) * | 2019-12-23 | 2020-05-08 | 温州大学 | Series fault arc detection device and method based on convolutional neural network |
| CN111239569A (en) * | 2020-03-12 | 2020-06-05 | 杭州电弧数据科技有限公司 | Arc fault detection method, device, equipment and storage medium |
| CN111458599A (en) * | 2020-04-16 | 2020-07-28 | 福州大学 | Series arc fault detection method based on one-dimensional convolutional neural network |
Non-Patent Citations (4)
| Title |
|---|
| 刘艳丽等: "串联型故障电弧的实验研究", 《辽宁工程技术大学学报（自然科学版）工程科技Ⅰ辑 》 * |
| 周新城等: "基于1D-CNN的串联故障电弧检测", 《消防科学与技术》 * |
| 孙曙光等: "基于一维卷积神经网络的低压万能式断路器附件故障诊断", 《电工技术学报》 * |
| 褚若波等: "基于多层卷积神经网络的串联电弧故障检测方法", 《电网技术 工程科技Ⅱ辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114282608A (en) * | 2021-12-22 | 2022-04-05 | 国网安徽省电力有限公司 | Hidden fault diagnosis and early warning method and system for current transformer |
| CN115014571A (en) * | 2022-05-25 | 2022-09-06 | 汕头大学 | Submarine cable risk event identification system and method |
| CN116990648A (en) * | 2023-09-26 | 2023-11-03 | 北京科技大学 | Fault arc detection method based on one-dimensional cavity convolutional neural network |
| CN116990648B (en) * | 2023-09-26 | 2023-12-19 | 北京科技大学 | Fault arc detection method based on one-dimensional cavity convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113030789A (en) | Series arc fault diagnosis and line selection method based on convolutional neural network | |
| CN110398663B (en) | Flexible direct current power grid fault identification method based on convolutional neural network | |
| CN114021433B (en) | Construction method and application of dominant instability mode identification model of power system | |
| CN109782091A (en) | Multi-electrical level inverter method for diagnosing faults based on depth small echo extreme learning machine | |
| CN108898182A (en) | A kind of MMC method for diagnosing faults based on core pivot element analysis and support vector machines | |
| CN111553112A (en) | Power system fault identification method and device based on deep belief network | |
| CN111007429A (en) | ANFIS-based synchronous phase modulator short-circuit fault identification method and system | |
| CN110068776A (en) | Three-level inverter open-circuit fault diagnostic method based on Support Vector Machines Optimized | |
| CN114330571B (en) | Power system dominant instability mode identification method and system based on transfer learning | |
| Yanan et al. | Synchronous condenser fault diagnosis method based on RBF neutral network | |
| Abed et al. | Fault diagnosis of brushless DC motor for an aircraft actuator using a neural wavelet network | |
| CN113991652B (en) | Data-driven multi-output calculation method for short-circuit current of distribution network containing IIDG | |
| CN114089181B (en) | XG-Boost-based MMC switching tube open-circuit fault detection method | |
| An et al. | IGBT open circuit fault diagnosis method for a modular multilevel converter based on PNN-MD | |
| Obi et al. | An extended ann-based high speed accurate transmission line fault location for double phase to earth fault on non-directground | |
| Hatata et al. | Neural Network Based Fault Detector and Classifier for Synchronous Generator Stator Windings. | |
| Fan et al. | Transmission Line Fault Type Identification based on Transfer Learning and Deep Learning | |
| Thapa et al. | A machine learning-based approach to detection of fault locations in transmission networks | |
| Hu et al. | An Open Circuit Faults Diagnosis Method Based on XG-Boost Used in MMC | |
| Song et al. | An improved negative selection algorithm based identification model for distribution network operating state with small samples | |
| Ramadoss et al. | Ensemble machine learning approach to identify excitation failure in synchronous generators | |
| Wang et al. | Fuzzy model based on-line stator winding turn fault detection for induction motors | |
| CN118040679B (en) | Disturbance and fault identification method for in-phase power supply system | |
| CN113791294B (en) | Extra-high voltage direct current transmission system commutation failure tracing method based on edge cloud cooperation | |
| Liwei et al. | An Identification Method of Fault Type Based on GWO-SVM for Distribution Network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210625 |
|
| WD01 | Invention patent application deemed withdrawn after publication |