CN113406437B - Power transmission line fault detection method for generating countermeasure network based on auxiliary classification - Google Patents
Power transmission line fault detection method for generating countermeasure network based on auxiliary classification Download PDFInfo
- Publication number
- CN113406437B CN113406437B CN202110683616.3A CN202110683616A CN113406437B CN 113406437 B CN113406437 B CN 113406437B CN 202110683616 A CN202110683616 A CN 202110683616A CN 113406437 B CN113406437 B CN 113406437B
- Authority
- CN
- China
- Prior art keywords
- fault
- discriminator
- network
- picture
- training
- 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
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
- G01R31/08—Locating faults in cables, transmission lines, or networks
- G01R31/088—Aspects of digital computing
-
- 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
- G01R31/085—Locating faults in cables, transmission lines, or networks according to type of conductors in power transmission or distribution lines, e.g. overhead
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E40/00—Technologies for an efficient electrical power generation, transmission or distribution
- Y02E40/70—Smart grids as climate change mitigation technology in the energy generation sector
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
-
- 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
- Y04S10/22—Flexible AC transmission systems [FACTS] or power factor or reactive power compensating or correcting units
-
- 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
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Testing, Inspecting, Measuring Of Stereoscopic Televisions And Televisions (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention discloses a power transmission line fault detection method for generating a countermeasure network based on auxiliary classification. Relates to the technical field of fault detection of high-voltage transmission lines. The method comprises the steps that faults of different fault points and different fault types of each line in a power grid are simulated, fault characteristic vectors of each node are obtained by utilizing a small amount of voltages of nodes of a synchronous phasor measuring device through Carlenburg transformation and variational modal decomposition, and are combined to form a characteristic vector matrix to generate a whole grid gray scale map; constructing auxiliary classification generation countermeasure networks for power transmission line fault detection, optimizing parameters of the networks, and alternately training discriminators and generators of the networks by adopting a full-network gray-scale graph in fault simulation and random vectors generated by noise obeying normal distribution; the network judges the input full-network gray-scale image to be true or false, obtains the label number of the full-network gray-scale image and obtains the fault line number, thereby realizing the fault detection of each line in the power grid and achieving higher detection accuracy. The method is used for detecting the faults of the high-voltage transmission line.
Description
Technical Field
The invention relates to the technical field of fault detection of high-voltage power transmission and transformation lines.
Background
With the development of synchronized Phasor Measurement, data processing, and high-speed communication technologies in recent years, a certain number of synchronized Phasor Measurement Units (PMUs) or relay protection devices with GPS synchronization capability are deployed in power grids of various countries around the world. The deployment of these PMUs provides new possibilities for wide area fault detection that improves line fault detection capabilities. Different from the traditional protection based on single-ended measurement, the wide-area fault detection adopts accurate synchronization in time and adopts widely-distributed PMU measurement information to detect the faults of all lines in a region.
The existing transmission line fault detection scheme mostly adopts synchronous measurement information or equivalent measurement information at two ends of a line, namely, an equivalent measurement value obtained by deducing a node without a PMU from an adjacent measurement node through kirchhoff's law to judge the fault of the line. These solutions require installation or equivalent installation of PMUs at all nodes in the power system, but for economic reasons, the installation density of PMUs in the grid does not meet these requirements.
In order to solve the problem of fault detection of each line in a power grid under the condition of less or low-density arrangement of PMUs, the invention adopts Auxiliary classification generation countermeasure network (ACGAN) in artificial intelligence technology, obtains fault characteristic data by using the voltage with less PMU nodes arranged, generates a whole-network gray scale map to train the ACGAN network of each fault type, performs fault detection of each line in the power grid, realizes higher fault line accuracy under low PMU coverage and has better engineering significance.
Disclosure of Invention
The invention aims to provide a power transmission line fault detection method based on an auxiliary classification generation countermeasure network, which can effectively solve the technical problem of fault detection when a power transmission line has a fault under low PMU coverage.
The purpose of the invention is realized by the following technical scheme:
a power transmission line fault detection method based on auxiliary classification generation countermeasure network comprises the following steps:
the method comprises the following steps: building a power transmission network model through PSCAD software, selecting M bus nodes in the power transmission network, respectively arranging a synchronous Phasor Measurement Unit (PMU), dividing each line into A sections, presetting fault points on each line, respectively setting single-phase short circuit grounding, two-phase short circuit and three-phase short circuit grounding faults at each fault point, and performing fault simulation; for each fault scenario, respectively obtaining the voltages of M bus nodes with PMUs in the first cycle after the fault occurs, and uniformly taking N voltages, wherein N is 62; respectively carrying out Carnlian transformation on the voltages of M bus nodes to obtain decoupled alpha-mode components, then carrying out variation modal decomposition on the alpha-mode components to obtain IMF1, IMF2, … and IMFk inherent modal component sequences, taking the IMF1 component sequences as fault characteristic vectors of the bus nodes under the fault situation, and uniformly taking N data which are 1 multiplied by N; combining the fault characteristic vectors of the M bus nodes to form a characteristic vector matrix which is MxN, and then generating a corresponding full-network gray-scale image by using the characteristic vector matrix; processing the other fault situations in the same way to obtain a full-network gray-scale map of each fault point of each line under different fault types;
step two: marking the whole-network gray level graph of each fault point of each line under each fault type, respectively numbering the lines from 1 to P aiming at P lines in the power grid, marking each whole-network gray level icon as a corresponding line number as a label number of each whole-network gray level graph, namely marking the gray level graph to the corresponding fault line, and if all the lines are normal, giving a label number of P + 1;
step three, aiming at various fault types, respectively constructing auxiliary classification generation countermeasure networks which are added with convolution layers and generate labeled pictures, respectively training the auxiliary classification generation countermeasure networks, correctly classifying the tested full-network gray level images, and detecting fault lines;
step four: generating a countermeasure network aiming at the auxiliary classification of various fault types, and respectively sending respective training data to train; after the training of the countermeasure network is finished, respectively inputting respective test data obtained through simulation in advance, carrying out classification test on fault lines, and carrying out statistics on detection accuracy; for line faults actually occurring in a power grid, obtaining the voltage of each bus node with PMU, obtaining the fault type by using a traditional impedance phase selection element, obtaining a full-grid gray-scale image from the voltage of each bus node according to the step I, sending the full-grid gray-scale image to a discriminator, and obtaining a corresponding fault line number from a label number output by a No. 2 output layer of the discriminator if the output of the No. 1 output layer of the discriminator is true.
A transmission line fault detection method based on auxiliary classification generation countermeasure network is characterized in that: the method for obtaining the full-network gray-scale map of each fault point of each line under different fault types comprises the following specific operations:
1) karenbauer transformation
Three-phase voltage U to one bus nodea、Ub、UcThe carrousel transformation is performed as follows:
obtaining alpha module component UαBeta modulus component U β0 modulus component U0；
2) Variational modal decomposition
Only adopting an alpha-mode component as a measurement signal, carrying out variation modal decomposition on the alpha-mode component to obtain IMF1, IMF2, … and IMFk inherent modal component sequences, selecting an IMF1 component which has most signal essential characteristics as a fault component, and uniformly taking N data as fault characteristic vectors of the bus node, wherein the N data is 1 multiplied by N;
3) forming a feature vector matrix
Combining the fault characteristic vectors of M bus nodes to form an original characteristic vector matrix E which is MxN;
4) generating a full-screen grayscale map
Normalizing the original characteristic vector matrix E to be between 0 and 1 to obtain the characteristic vector matrix E2：
E2(i,j)＝(E(i,j)-min)/(max-min) (2)
Where E (i, j) is the ith row and jth column element of the original eigenvector matrix E, E2(i, j) is the eigenvector matrix E2Row i and column j of (1), min is the minimum of all elements in E, and max is the maximum of all elements in E;
the feature vector matrix E2And multiplying each element by 255, rounding, converting to 0-255, namely a gray value, and generating a full-network gray map H according to the gray value data, wherein the size of the full-network gray map H is MxN, and the extension name of the full-network gray map H is jpg.
A transmission line fault detection method based on auxiliary classification generation countermeasure network is characterized in that: the method for training the confrontation network generated by each auxiliary classification comprises the following specific operations:
each countermeasure network comprises a generator and a discriminator, so that the generator can generate a simulated generated picture, and meanwhile, the discriminator can accurately judge the generated picture, a fault picture and a label thereof;
1) structure of generator and discriminator
The generator structure comprises an input layer, a full connection layer, 3 convolution layers and an output layer;
an input layer: inputting a random vector generated by noise that follows a normal distribution;
full connection layer: connecting the input to dimensions of 32 multiplied by 3 multiplied by 15, and modifying the input to a characteristic layer pattern with the pixel height of 3, the width of 15 and the channel number of 32 by using a Reshape function during output;
1, a convolutional layer: the number of channels is 64, the number of convolution kernels is 3, the activation function is Relu, and the step length is 1;
2, a convolutional layer: the number of channels is 128, the number of convolution kernels is 3, the height and width of a feature layer which is once expanded by the last sampling are respectively 6 and 30, and the activation function is Relu;
layer 3: the number of channels is 64, the number of convolution kernels is 3, the height and width of a feature layer which is once expanded by the last sample are respectively 12 and 60, and the activation function is Relu;
an output layer: the number of channels is 1, the convolution layer is used as picture output, the number of convolution kernels is 3, and the activation function is tanh;
the discriminator structure comprises an input layer, 4 convolution layers and 2 output layers;
an input layer: the pixel height 12, width 60, channel number of the input picture are 1;
1, a convolutional layer: the number of channels is 16, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 2, and the dropout parameter is 0.4;
2, a convolutional layer: the number of channels is 32, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 2, and the dropout parameter is 0.4;
layer 3: the number of channels is 64, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 1, and the dropout parameter is 0.4;
the 4 th convolution layer: the number of channels is 128, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 1, and the dropout parameter is 0.4;
output layer 1: the dimension is 1, the activation function is sigmoid, and the authenticity of the picture is judged;
output layer 2: the dimensionality is P, the activation function is softmax, and the category of the picture is judged;
when the 1 st output layer of the discriminator judges that the picture is true, the 2 nd output layer outputs the label number, and the label number obtains the corresponding fault line number;
2) network training process
Data import: because the designed generator output picture and the discriminator input picture have pixel height and width of 12 and 60 respectively, and the gray level image generated by extracting the characteristics has pixel height of 12 and width of 62, and the gray level image needs to be cut, considering that the original data for the generated gray level image is instantaneous three-phase voltage and has periodic continuity, the gray level image is cut, and the [0,60] part is taken to be provided for training and testing of the confrontation network;
training process: generating a random vector by using the noise which is subjected to normal distribution as the input of a generator, and generating a generated picture; when the generator hopes to provide the generated picture for the discriminator, the output picture of the discriminator is true and a corresponding label is output; when the discriminator hopes to input a real fault picture, the discriminator judges that the fault picture is true and the corresponding label number, and when a generated picture is input, the discriminator judges that the fault picture is false; training by adopting a cross entropy loss function until the network is converged;
generating an antagonistic network, generally training a discriminator first, predicting an input picture of the discriminator before training, and training a generator first if a prediction result is correct; otherwise, training the discriminator; in the training of the minimization loss function, the problem that gradient disappears due to the fact that one side suppresses the other side occurs, and optimization is carried out by using a random gradient descent method;
the learning rate is 0.0002, Adam is used by an optimizer, the training times are 10000, and cross entropy parameters are used by a loss function;
3) parameter optimization strategy for discriminator
a. Strategy for alternate training of generator and discriminator
When the generator and the arbiter are alternately trained, the training of the arbiter is shielded when the generator is trained;
in the process of predicting the input picture of the discriminator, when the discriminator can better identify the picture, the training of the discriminator is interrupted;
b. processing and mapping a gray-scale image H input by a discriminator to a matrix M between-1 and-1HThe activation function is convenient to exert the maximum effect, namely, each element H (i, j) in the gray-scale image H is mapped from 0-255 to-1, as follows:
MH(i,j)＝(H(i,j)-127.5)/127.5 (3)
c. adjusting an alpha parameter of an activation function Leakyrelu of the discriminator to be 0.2, so that a gradient exists in a negative half shaft;
d. before data is input into an activation function, adding a standardized function BatchNormal, wherein the momentum parameter of the standardized function BatchNormal is 0.8, so that the data is scaled to the interval of the activation function, and the activation function achieves the maximum effect on data processing;
e. a dropout function is added to the discriminator to inactivate part of the neurons and prevent the discriminator from overfitting.
Compared with the prior art, the advantages and effects are as follows: the invention utilizes the voltage with less PMU nodes, obtains the fault characteristic data of each bus node by simulating different fault types and faults of different fault points of each line, combines to form a characteristic vector matrix, generates a full-network gray-scale map under each fault situation, trains respective confrontation network ACGAN aiming at each fault type, completes the fault detection of each line in a power grid, realizes higher fault line detection accuracy under low PMU coverage, and has the following advantages:
1) the invention only utilizes the voltage data of each PMU node, adopts voltage signals and has simple acquisition;
2) by respectively training and testing the countermeasure networks of the fault types, the invention can achieve higher fault line accuracy under the condition of low PMU coverage rate;
3) the invention carries out normalization processing on the characteristic vector matrix of the whole network, emphasizes the difference of a fault line relative to a normal line, and de-emphasizes the absolute value of voltage, so that the invention is insensitive to transition resistance.
Drawings
FIG. 1 is a schematic diagram of a generator according to the present invention;
FIG. 2 is a schematic structural diagram of an arbiter according to the present invention;
FIG. 3 is a schematic diagram of an IEEE 39 node system with 12 PMUs deployed;
FIG. 4 is a flow chart of a method of the present invention;
FIG. 5 is a graph showing the variation of the line recognition accuracy of the discriminator with the training times
FIG. 6 is a graph showing the variation of the line identification loss value of the discriminator with the training times
Detailed Description
The invention is described in further detail below with reference to the figures and specific examples.
As shown in fig. 3 of an IEEE 39 node system with 12 PMUs, a transmission line fault detection method based on an assisted classification generation countermeasure network (ACGAN) includes the steps of:
the method comprises the following steps: building a power transmission network model through PSCAD software, selecting M bus nodes in the power transmission network, respectively arranging a synchronous Phasor Measurement Unit (PMU), dividing each line into A sections, presetting fault points on each line, respectively setting single-phase short circuit grounding, two-phase short circuit and three-phase short circuit grounding faults at each fault point, and performing fault simulation; for each fault scenario, respectively obtaining the voltages of M bus nodes with PMUs in the first cycle after the fault occurs, and uniformly taking N voltages, wherein N is 62; respectively carrying out Carnlian transformation on the voltages of M bus nodes to obtain decoupled alpha-mode components, then carrying out variation modal decomposition on the alpha-mode components to obtain IMF1, IMF2, … and IMFk inherent modal component sequences, taking the IMF1 component sequences as fault characteristic vectors of the bus nodes under the fault situation, and uniformly taking N data which are 1 multiplied by N; combining the fault characteristic vectors of the M bus nodes to form a characteristic vector matrix which is MxN, and then generating a corresponding full-network gray-scale image by using the characteristic vector matrix; processing the other fault situations in the same way to obtain a full-network gray-scale map of each fault point of each line under different fault types;
step two: marking the whole-network gray level graph of each fault point of each line under each fault type, respectively numbering the lines from 1 to P aiming at P lines in the power grid, marking each whole-network gray level icon as a corresponding line number as a label number of each whole-network gray level graph, namely marking the gray level graph to the corresponding fault line, and if all the lines are normal, giving a label number of P + 1;
step three, aiming at various fault types, respectively constructing auxiliary classification generation countermeasure networks which are added with convolution layers and generate labeled pictures, respectively training the auxiliary classification generation countermeasure networks, correctly classifying the tested full-network gray level images, and detecting fault lines;
step four: generating a countermeasure network aiming at the auxiliary classification of various fault types, and respectively sending respective training data to train; after the training of the countermeasure network is finished, respectively inputting respective test data obtained through simulation in advance, carrying out classification test on fault lines, and carrying out statistics on detection accuracy; for line faults actually occurring in a power grid, obtaining the voltage of each bus node with PMU, obtaining the fault type by using a traditional impedance phase selection element, obtaining a full-grid gray-scale image from the voltage of each bus node according to the step I, sending the full-grid gray-scale image to a discriminator, and obtaining a corresponding fault line number from a label number output by a No. 2 output layer of the discriminator if the output of the No. 1 output layer of the discriminator is true.
A transmission line fault detection method based on auxiliary classification generation countermeasure network is characterized in that: the method for obtaining the full-network gray-scale map of each fault point of each line under different fault types comprises the following specific operations:
1) karenbauer transformation
Three-phase voltage U to one bus nodea、Ub、UcThe carrousel transformation is performed as follows:
obtaining alpha module component UαBeta modulus component U β0 modulus component U0；
2) Variational modal decomposition
Only adopting an alpha-mode component as a measurement signal, carrying out variation modal decomposition on the alpha-mode component to obtain IMF1, IMF2, … and IMFk inherent modal component sequences, selecting an IMF1 component which has most signal essential characteristics as a fault component, and uniformly taking N data as fault characteristic vectors of the bus node, wherein the N data is 1 multiplied by N;
3) forming a feature vector matrix
Combining the fault characteristic vectors of M bus nodes to form an original characteristic vector matrix E which is MxN;
4) generating a full-screen grayscale map
Normalizing the original characteristic vector matrix E to be between 0 and 1 to obtain the characteristic vector matrix E2：
E2(i,j)＝(E(i,j)-min)/(max-min) (2)
Where E (i, j) is the ith row and jth column element of the original eigenvector matrix E, E2(i, j) is the eigenvector matrix E2Row i and column j of (1), min is the minimum of all elements in E, and max is the maximum of all elements in E;
the feature vector matrix E2And multiplying each element by 255, rounding, converting to 0-255, namely a gray value, and generating a full-network gray map H according to the gray value data, wherein the size of the full-network gray map H is MxN, and the extension name of the full-network gray map H is jpg.
A transmission line fault detection method based on auxiliary classification generation countermeasure network is characterized in that: the method for training the confrontation network generated by each auxiliary classification comprises the following specific operations:
each countermeasure network comprises a generator and a discriminator, so that the generator can generate a simulated generated picture, and meanwhile, the discriminator can accurately judge the generated picture, a fault picture and a label thereof;
1) structure of generator and discriminator
The generator structure comprises an input layer, a full connection layer, 3 convolution layers and an output layer;
an input layer: inputting a random vector generated by noise that follows a normal distribution;
full connection layer: connecting the input to dimensions of 32 multiplied by 3 multiplied by 15, and modifying the input to a characteristic layer pattern with the pixel height of 3, the width of 15 and the channel number of 32 by using a Reshape function during output;
1, a convolutional layer: the number of channels is 64, the number of convolution kernels is 3, the activation function is Relu, and the step length is 1;
2, a convolutional layer: the number of channels is 128, the number of convolution kernels is 3, the height and width of a feature layer which is once expanded by the last sampling are respectively 6 and 30, and the activation function is Relu;
layer 3: the number of channels is 64, the number of convolution kernels is 3, the height and width of a feature layer which is once expanded by the last sample are respectively 12 and 60, and the activation function is Relu;
an output layer: the number of channels is 1, the convolution layer is used as picture output, the number of convolution kernels is 3, and the activation function is tanh;
the discriminator structure comprises an input layer, 4 convolution layers and 2 output layers;
an input layer: the pixel height 12, width 60, channel number of the input picture are 1;
1, a convolutional layer: the number of channels is 16, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 2, and the dropout parameter is 0.4;
2, a convolutional layer: the number of channels is 32, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 2, and the dropout parameter is 0.4;
layer 3: the number of channels is 64, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 1, and the dropout parameter is 0.4;
the 4 th convolution layer: the number of channels is 128, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 1, and the dropout parameter is 0.4;
output layer 1: the dimension is 1, the activation function is sigmoid, and the authenticity of the picture is judged;
output layer 2: the dimensionality is P, the activation function is softmax, and the category of the picture is judged;
when the 1 st output layer of the discriminator judges that the picture is true, the 2 nd output layer outputs the label number, and the label number obtains the corresponding fault line number;
2) network training process
Data import: because the designed generator output picture and the discriminator input picture have pixel height and width of 12 and 60 respectively, and the gray level image generated by extracting the characteristics has pixel height of 12 and width of 62, and the gray level image needs to be cut, considering that the original data for the generated gray level image is instantaneous three-phase voltage and has periodic continuity, the gray level image is cut, and the [0,60] part is taken to be provided for training and testing of the confrontation network;
training process: generating a random vector by using the noise which is subjected to normal distribution as the input of a generator, and generating a generated picture; when the generator hopes to provide the generated picture for the discriminator, the output picture of the discriminator is true and a corresponding label is output; when the discriminator hopes to input a real fault picture, the discriminator judges that the fault picture is true and the corresponding label number, and when a generated picture is input, the discriminator judges that the fault picture is false; training by adopting a cross entropy loss function until the network is converged;
generating an antagonistic network, generally training a discriminator first, predicting an input picture of the discriminator before training, and training a generator first if a prediction result is correct; otherwise, training the discriminator; in the training of the minimization loss function, the problem that gradient disappears due to the fact that one side suppresses the other side occurs, and optimization is carried out by using a random gradient descent method;
the learning rate is 0.0002, Adam is used by an optimizer, the training times are 10000, and cross entropy parameters are used by a loss function;
3) parameter optimization strategy for discriminator
a. Strategy for alternate training of generator and discriminator
When the generator and the arbiter are alternately trained, the training of the arbiter is shielded when the generator is trained;
in the process of predicting the input picture of the discriminator, when the discriminator can better identify the picture, the training of the discriminator is interrupted;
b. processing and mapping a gray-scale image H input by a discriminator to a matrix M between-1 and-1HThe activation function is convenient to exert the maximum effect, namely, each element H (i, j) in the gray-scale image H is mapped from 0-255 to-1, as follows:
MH(i,j)＝(H(i,j)-127.5)/127.5 (3)
c. adjusting an alpha parameter of an activation function Leakyrelu of the discriminator to be 0.2, so that a gradient exists in a negative half shaft;
d. before data is input into an activation function, adding a standardized function BatchNormal, wherein the momentum parameter of the standardized function BatchNormal is 0.8, so that the data is scaled to the interval of the activation function, and the activation function achieves the maximum effect on data processing;
e. a dropout function is added to the discriminator to inactivate part of the neurons and prevent the discriminator from overfitting.
Examples of the design
For an IEEE 39 node system, which has 39 bus nodes, as shown in FIG. 3, G1-G10 respectively represent generator nodes, T1-T12 respectively represent transformers, B1～B39Represents a bus node, L1～L33Respectively shows the circuit, has built its electric wire netting model with PSCAD, has arranged synchrophasor measuring device PMU on 12 bus nodes wherein, and M is 12, has had PMU to represent the bus node of the rectangle frame of PMU characters in the side in FIG. 3 promptly, includes bus node B3、B5、B8、B11、B14、B16、 B19、B23、B25、B27、B29、B39。
Four line faults of three-phase short circuit, single-phase grounding, two-phase grounding and two-phase short circuit are simulated respectively. And constructing auxiliary classification of each fault type to generate a countermeasure network, and then setting and optimizing parameters of the network, wherein the learning rate lambda is set to be 0.0002 in a classifier, and the training times are 10000 times.
Because the generation of the countermeasure network is alternate training, firstly, the input picture of the discriminator needs to be predicted, if the judgment result is correct, the parameter is fixed, the generator is trained firstly, and if the judgment result is wrong, the parameter of the generator needs to be fixed firstly, and the discriminator needs to be trained. Setting 54 samples taken for each batch to train, defining loss functions of a generator and a discriminator, wherein the generator calculates loss by collecting the condition of judging that the generated pictures and labels are false when being input, the discriminator outputs the judged true pictures and labels thereof when needing to input the true pictures, outputs the judged false pictures and corresponding labels thereof when inputting the generated pictures, and calculates loss when judging errors occur. Both the generator and the arbiter use Adam optimizers and are trained using a stochastic gradient descent method.
Taking 15 lines of an IEEE 39 node system, setting a fault position for each line every 10% of the length, and totally 9 fault points, simulating 4 fault types of single-phase grounding, two-phase short-circuit and three-phase short-circuit, and the conditions of 0.01 omega and 300 omega 2 grounding resistances, namely 6 fault scenarios at one fault point, wherein 15 multiplied by 6 multiplied by 9 is the data of 810 groups in total, wherein 80% (648) data sets are used for training auxiliary classification of the 4 fault types to generate a countermeasure network, and the other 20% (162) data sets are used for testing the performance of the network.
1) And analyzing the identification correctness of different fault types. For the line A phase grounding fault, fig. 5 is a conversion curve of the line identification accuracy of the discriminator along with the training times, fig. 6 is a conversion curve of the line identification loss value of the discriminator along with the training times, after 7000 times of training, the identification accuracy and the loss value of the discriminator reach stable values, and the generator generates 40000 generated pictures in total. And then, respectively carrying out simulation verification on the two-phase short circuit grounding, the two-phase short circuit and the three-phase short circuit fault, and finally obtaining the line fault detection accuracy of the auxiliary classified generation countermeasure network under the same PMU coverage as shown in Table 1.
Table 1 line fault detection accuracy of IEEE 39 node system for generating countermeasure network by auxiliary classification under 4 fault types
As can be seen from table 1, for low coverage rate of only 12 nodes with PMUs, for 4 fault types, the constructed assisted classification generation countermeasure network has better identification capability, and the identification accuracy is higher than 95%. After statistical analysis of all fault types, the total accuracy of fault identification of 15 lines reaches 96.68%, and the identification rates of 4 fault types have no obvious difference.
2) And (4) carrying out line fault identification correctness analysis under different PMU coverage rates. For the case of different PMU coverage, comparing the a-phase ground fault with the case of installing 8 PMU nodes, i.e., reducing nodes 8, 19, 29, and 39, installing 10 PMU nodes, i.e., reducing nodes 8 and 29, and obtaining the line fault detection accuracy rates of the nodes as shown in table 2.
Table 2 IEEE 39 node system assisted classification of line fault detection accuracy for generation countermeasure networks at different PMU coverage
As can be seen from table 2, with the increase of the number of nodes of the PMU, the accuracy of line fault detection of the auxiliary classified generation countermeasure network is improved to a certain extent.
3) And analyzing the fault detection accuracy of different grounding transition resistors. For the ground fault conditions, different ground resistance conditions of low resistance and high resistance were analyzed, and the line fault detection accuracy is shown in table 3.
TABLE 3 line fault detection accuracy of IEEE 39 node system for generation of countermeasure network by auxiliary classification under different ground resistances
As can be seen from table 3, the identification of the secondary classification generation countermeasure network is not affected by the transition resistance variations in the case of different ground transition resistances.
Finally, it is noted that: the above-mentioned embodiments are only examples of the present invention, and it is a matter of course that those skilled in the art can make modifications and variations to the present invention, and it is considered that the present invention is protected by the modifications and variations if they are within the scope of the claims of the present invention and their equivalents.
Claims (3)
1. A power transmission line fault detection method based on auxiliary classification generation countermeasure network comprises the following steps:
the method comprises the following steps: building a power transmission network model through PSCAD software, selecting M bus nodes in the power transmission network, respectively arranging a synchronous Phasor Measurement Unit (PMU), dividing each line into A sections, presetting fault points on each line, respectively setting single-phase short circuit grounding, two-phase short circuit and three-phase short circuit grounding faults at each fault point, and performing fault simulation; for each fault scenario, respectively obtaining the voltages of M bus nodes with PMUs in the first cycle after the fault occurs, and uniformly taking N voltages, wherein N is 62; respectively carrying out Carnlian transformation on the voltages of M bus nodes to obtain decoupled alpha-mode components, then carrying out variation modal decomposition on the alpha-mode components to obtain IMF1, IMF2, … and IMFk inherent modal component sequences, taking the IMF1 component sequences as fault characteristic vectors of the bus nodes under the fault situation, and uniformly taking N data which are 1 multiplied by N; combining the fault characteristic vectors of the M bus nodes to form a characteristic vector matrix which is MxN, and then generating a corresponding full-network gray-scale image by using the characteristic vector matrix; processing the other fault situations in the same way to obtain a full-network gray-scale map of each fault point of each line under different fault types;
step two: marking the whole-network gray level graph of each fault point of each line under each fault type, respectively numbering the lines from 1 to P aiming at P lines in the power grid, marking each whole-network gray level icon as a corresponding line number as a label number of each whole-network gray level graph, namely marking the gray level graph to the corresponding fault line, and if all the lines are normal, giving a label number of P + 1;
step three, aiming at various fault types, respectively constructing auxiliary classification generation countermeasure networks which are added with convolution layers and generate labeled pictures, respectively training the auxiliary classification generation countermeasure networks, correctly classifying the tested full-network gray level images, and detecting fault lines;
step four: generating a countermeasure network aiming at the auxiliary classification of various fault types, and respectively sending respective training data to train; after the training of the countermeasure network is finished, respectively inputting respective test data obtained through simulation in advance, carrying out classification test on fault lines, and carrying out statistics on detection accuracy; for line faults actually occurring in a power grid, obtaining the voltage of each bus node with PMU, obtaining the fault type by using a traditional impedance phase selection element, obtaining a full-grid gray-scale image from the voltage of each bus node according to the step I, sending the full-grid gray-scale image to a discriminator, and obtaining a corresponding fault line number from a label number output by a No. 2 output layer of the discriminator if the output of the No. 1 output layer of the discriminator is true.
2. The power transmission line fault detection method based on the auxiliary classification generation countermeasure network of claim 1, characterized in that: the method for obtaining the full-network gray-scale map of each fault point of each line under different fault types comprises the following specific operations:
1) karenbauer transformation
Three-phase voltage U to one bus nodea、Ub、UcThe carrousel transformation is performed as follows:
obtaining alpha module component UαBeta modulus component Uβ0 modulus component U0；
2) Variational modal decomposition
Only adopting an alpha-mode component as a measurement signal, carrying out variation modal decomposition on the alpha-mode component to obtain IMF1, IMF2, … and IMFk inherent modal component sequences, selecting an IMF1 component which has most signal essential characteristics as a fault component, and uniformly taking N data as fault characteristic vectors of the bus node, wherein the N data is 1 multiplied by N;
3) forming a feature vector matrix
Combining the fault characteristic vectors of M bus nodes to form an original characteristic vector matrix E which is MxN;
4) generating a full-screen grayscale map
Normalizing the original characteristic vector matrix E to be between 0 and 1 to obtain the characteristic vector matrix E2：
E2(i,j)＝(E(i,j)-min)/(max-min) (2)
Where E (i, j) is the ith row and jth column element of the original eigenvector matrix E, E2(i, j) is the eigenvector matrix E2Row i and column j of (1), min is the minimum of all elements in E, and max is the maximum of all elements in E;
the feature vector matrix E2And multiplying each element by 255, rounding, converting to 0-255, namely a gray value, and generating a full-network gray map H according to the gray value data, wherein the size of the full-network gray map H is MxN, and the extension name of the full-network gray map H is jpg.
3. The power transmission line fault detection method based on the auxiliary classification generation countermeasure network of claim 1, characterized in that: the method for training the confrontation network generated by each auxiliary classification comprises the following specific operations:
each countermeasure network comprises a generator and a discriminator, so that the generator can generate a simulated generated picture, and meanwhile, the discriminator can accurately judge the generated picture, a fault picture and a label thereof;
1) structure of generator and discriminator
The generator structure comprises an input layer, a full connection layer, 3 convolution layers and an output layer;
an input layer: inputting a random vector generated by noise that follows a normal distribution;
full connection layer: connecting the input to dimensions of 32 multiplied by 3 multiplied by 15, and modifying the input to a characteristic layer pattern with the pixel height of 3, the width of 15 and the channel number of 32 by using a Reshape function during output;
1, a convolutional layer: the number of channels is 64, the number of convolution kernels is 3, the activation function is Relu, and the step length is 1;
2, a convolutional layer: the number of channels is 128, the number of convolution kernels is 3, the height and width of a feature layer which is once expanded by the last sampling are respectively 6 and 30, and the activation function is Relu;
layer 3: the number of channels is 64, the number of convolution kernels is 3, the height and width of a feature layer which is once expanded by the last sample are respectively 12 and 60, and the activation function is Relu;
an output layer: the number of channels is 1, the convolution layer is used as picture output, the number of convolution kernels is 3, and the activation function is tanh;
the discriminator structure comprises an input layer, 4 convolution layers and 2 output layers;
an input layer: the pixel height 12, width 60, channel number of the input picture are 1;
1, a convolutional layer: the number of channels is 16, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 2, and the dropout parameter is 0.4;
2, a convolutional layer: the number of channels is 32, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 2, and the dropout parameter is 0.4;
layer 3: the number of channels is 64, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 1, and the dropout parameter is 0.4;
the 4 th convolution layer: the number of channels is 128, the number of convolution kernels is 3, the activation function is Leakyrelu, the step length is 1, and the dropout parameter is 0.4;
output layer 1: the dimension is 1, the activation function is sigmoid, and the authenticity of the picture is judged;
output layer 2: the dimensionality is P, the activation function is softmax, and the category of the picture is judged;
when the 1 st output layer of the discriminator judges that the picture is true, the 2 nd output layer outputs the label number, and the label number obtains the corresponding fault line number;
2) network training process
Data import: because the designed generator output picture and the discriminator input picture have pixel height and width of 12 and 60 respectively, and the gray level image generated by extracting the characteristics has pixel height of 12 and width of 62, and the gray level image needs to be cut, considering that the original data for the generated gray level image is instantaneous three-phase voltage and has periodic continuity, the gray level image is cut, and the [0,60] part is taken to be provided for training and testing of the confrontation network;
training process: generating a random vector by using the noise which is subjected to normal distribution as the input of a generator, and generating a generated picture; when the generator hopes to provide the generated picture for the discriminator, the output picture of the discriminator is true and a corresponding label is output; when the discriminator hopes to input a real fault picture, the discriminator judges that the fault picture is true and the corresponding label number, and when a generated picture is input, the discriminator judges that the fault picture is false; training by adopting a cross entropy loss function until the network is converged;
generating an antagonistic network, generally training a discriminator first, predicting an input picture of the discriminator before training, and training a generator first if a prediction result is correct; otherwise, training the discriminator; in the training of the minimization loss function, the problem that gradient disappears due to the fact that one side suppresses the other side occurs, and optimization is carried out by using a random gradient descent method;
the learning rate is 0.0002, Adam is used by an optimizer, the training times are 10000, and cross entropy parameters are used by a loss function;
3) parameter optimization strategy for discriminator
a. Strategy for alternate training of generator and discriminator
When the generator and the arbiter are alternately trained, the training of the arbiter is shielded when the generator is trained;
in the process of predicting the input picture of the discriminator, when the discriminator can better identify the picture, the training of the discriminator is interrupted;
b. processing and mapping a gray-scale image H input by a discriminator to a matrix M between-1 and-1HThe activation function is convenient to exert the maximum effect, namely, each element H (i, j) in the gray-scale image H is mapped from 0-255 to-1, as follows:
MH(i,j)＝(H(i,j)-127.5)/127.5 (3)
c. adjusting an alpha parameter of an activation function Leakyrelu of the discriminator to be 0.2, so that a gradient exists in a negative half shaft;
d. before data is input into an activation function, adding a standardized function BatchNormal, wherein the momentum parameter of the standardized function BatchNormal is 0.8, so that the data is scaled to the interval of the activation function, and the activation function achieves the maximum effect on data processing;
e. a dropout function is added to the discriminator to inactivate part of the neurons and prevent the discriminator from overfitting.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110683616.3A CN113406437B (en) | 2021-06-21 | 2021-06-21 | Power transmission line fault detection method for generating countermeasure network based on auxiliary classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110683616.3A CN113406437B (en) | 2021-06-21 | 2021-06-21 | Power transmission line fault detection method for generating countermeasure network based on auxiliary classification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113406437A CN113406437A (en) | 2021-09-17 |
| CN113406437B true CN113406437B (en) | 2022-03-08 |
Family
ID=77681827
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110683616.3A Active CN113406437B (en) | 2021-06-21 | 2021-06-21 | Power transmission line fault detection method for generating countermeasure network based on auxiliary classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113406437B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114501525B (en) * | 2022-01-28 | 2024-02-02 | 东南大学 | Wireless network interruption detection method based on condition generation countermeasure network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109325985A (en) * | 2018-09-18 | 2019-02-12 | 上海联影智能医疗科技有限公司 | MR image reconstruction method, apparatus and computer readable storage medium |
| CN111027439A (en) * | 2019-12-03 | 2020-04-17 | 西北工业大学 | SAR target recognition method for generating countermeasure network based on auxiliary classification |
| CN111308270A (en) * | 2020-03-06 | 2020-06-19 | 西南交通大学 | Power transmission line fault detection method based on node fault injection current |
| CN111337243A (en) * | 2020-02-27 | 2020-06-26 | 上海电力大学 | ACGAN-based wind turbine generator planet wheel gearbox fault diagnosis method |
| WO2020172838A1 (en) * | 2019-02-26 | 2020-09-03 | 长沙理工大学 | Image classification method for improvement of auxiliary classifier gan |
| CN111751763A (en) * | 2020-06-08 | 2020-10-09 | 武汉大学 | Power transformer winding fault diagnosis method based on GSMallat-NIN-CNN network |
| CN111864706A (en) * | 2020-08-19 | 2020-10-30 | 剑科云智(深圳)科技有限公司 | Fault early warning and relay protection system of power distribution network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10825219B2 (en) * | 2018-03-22 | 2020-11-03 | Northeastern University | Segmentation guided image generation with adversarial networks |
-
2021
- 2021-06-21 CN CN202110683616.3A patent/CN113406437B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109325985A (en) * | 2018-09-18 | 2019-02-12 | 上海联影智能医疗科技有限公司 | MR image reconstruction method, apparatus and computer readable storage medium |
| WO2020172838A1 (en) * | 2019-02-26 | 2020-09-03 | 长沙理工大学 | Image classification method for improvement of auxiliary classifier gan |
| CN111027439A (en) * | 2019-12-03 | 2020-04-17 | 西北工业大学 | SAR target recognition method for generating countermeasure network based on auxiliary classification |
| CN111337243A (en) * | 2020-02-27 | 2020-06-26 | 上海电力大学 | ACGAN-based wind turbine generator planet wheel gearbox fault diagnosis method |
| CN111308270A (en) * | 2020-03-06 | 2020-06-19 | 西南交通大学 | Power transmission line fault detection method based on node fault injection current |
| CN111751763A (en) * | 2020-06-08 | 2020-10-09 | 武汉大学 | Power transformer winding fault diagnosis method based on GSMallat-NIN-CNN network |
| CN111864706A (en) * | 2020-08-19 | 2020-10-30 | 剑科云智(深圳)科技有限公司 | Fault early warning and relay protection system of power distribution network |
Non-Patent Citations (2)
| Title |
|---|
| Research on power grid fault diagnosis based on artificial fish swarm algorithm;Chen Liang et al.;《2017 2nd International Conference on Power and Renewable Energy》;20180621;第1047-1051页 * |
| 基于有限PMU 的电网故障在线识别算法研究;童晓阳 等;《电力系统保护与控制》;20161001;第44卷(第19期);第49-55页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113406437A (en) | 2021-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108107324B (en) | Power distribution network fault positioning method based on deep convolutional neural network | |
| CN109873610B (en) | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network | |
| CN109932627A (en) | A kind of localization method and system of GIS pressure test shelf depreciation | |
| CN110829417B (en) | Electric power system transient stability prediction method based on LSTM double-structure model | |
| Tong et al. | Detection and classification of transmission line transient faults based on graph convolutional neural network | |
| CN110619479B (en) | Fault judgment model modeling method and fault judgment method for photovoltaic module | |
| CN111679158A (en) | Power distribution network fault identification method based on synchronous measurement data similarity | |
| CN101477172A (en) | Analogue circuit fault diagnosis method based on neural network | |
| JP3058564B2 (en) | Transmission line failure section and failure mode evaluation method | |
| CN114414942A (en) | Power transmission line fault identification classifier, identification method and system based on transient waveform image identification | |
| CN107037327A (en) | Partial discharges fault judges feature extracting method and decision method | |
| CN108400595A (en) | A kind of voltage dip Stochastic prediction method considering new energy output correlation | |
| CN111614066B (en) | Automatic setting method and system for relay protection setting value of power distribution network | |
| CN108562821A (en) | A kind of method and system determining Single-phase Earth-fault Selection in Distribution Systems based on Softmax | |
| CN110726898A (en) | Power distribution network fault type identification method | |
| CN113406437B (en) | Power transmission line fault detection method for generating countermeasure network based on auxiliary classification | |
| CN110794254B (en) | Power distribution network fault prediction method and system based on reinforcement learning | |
| CN115932484A (en) | Method and device for identifying and ranging faults of power transmission line and electronic equipment | |
| Shiddieqy et al. | Power line transmission fault modeling and dataset generation for ai based automatic detection | |
| CN113610119B (en) | Method for identifying power transmission line development faults based on convolutional neural network | |
| CN113158781B (en) | Lightning trip type identification method | |
| CN111062569A (en) | Low-current fault discrimination method based on BP neural network | |
| CN107462810A (en) | A kind of fault section location method suitable for active power distribution network | |
| Wang et al. | The cable fault diagnosis for XLPE cable based on 1DCNNs-BiLSTM network | |
| CN113884818A (en) | LSTM-based power distribution network fault traveling wave arrival time accurate estimation 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |