CN114218988A - Method for identifying unidirectional ground fault feeder line under unbalanced samples - Google Patents
Method for identifying unidirectional ground fault feeder line under unbalanced samples Download PDFInfo
- Publication number
- CN114218988A CN114218988A CN202111521526.0A CN202111521526A CN114218988A CN 114218988 A CN114218988 A CN 114218988A CN 202111521526 A CN202111521526 A CN 202111521526A CN 114218988 A CN114218988 A CN 114218988A
- Authority
- CN
- China
- Prior art keywords
- sample
- model
- generator
- feeder line
- discriminator
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
- G06F2218/16—Classification; Matching by matching signal segments
- G06F2218/18—Classification; Matching by matching signal segments by plotting the signal segments against each other, e.g. analysing scattergrams
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24137—Distances to cluster centroïds
- G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The invention provides a single-phase earth fault feeder line identification method under unbalanced samples, which aims at the problems of single-phase earth fault data shortage and unbalance of a power distribution network and can provide sufficient and balanced training data for a data driving model so as to carry out more accurate single-phase earth fault feeder line identification work. The conditional generation confrontation network semi-supervised learning characteristic is utilized, the generator generates single-phase earth fault samples with data distribution characteristics consistent with real data distribution characteristics through game confrontation training of the generator and the discriminator, fault feeder waveform data and sound feeder waveform data are balanced, the number of training set sample bases is increased, and the identification accuracy of the data-driven fault feeder identification method is improved.
Description
Technical Field
The invention relates to a method for identifying a unidirectional earth fault feeder line under unbalanced samples, aims at the problems of single-phase earth fault data shortage and unbalance of a power distribution network, and belongs to the technical field of comprehensive processing of earth faults of the power distribution network.
Background
The distribution network plays a crucial role in the power system as the system network closest to the user terminal in the power system. With the increasing of cable feeders, the capacitance current of the line to the ground is increased continuously, and the system is easy to generate ground faults. Due to reasons such as lightning, bird damage, tree growth, equipment faults and the like, the faults of the power distribution network frequently occur, wherein the single-phase ground faults account for about 80 percent of the total faults. When a single-phase earth fault occurs in the power distribution network, the fault current characteristic is weakened due to the fact that arc suppression coils are used, the difficulty of detecting a fault feeder line is increased, if the fault feeder line cannot be detected in time and the fault can be eliminated, serious economic loss can be caused, even accidents such as personal casualties and the like can be caused, and therefore the fact that the single-phase earth fault feeder line is identified quickly is very important for clearing the fault of the power distribution network and recovering power supply.
With the wide application of a data driving technology in single-phase earth fault line selection of a power distribution network, a patent with application number of CN201710419091.6 discloses a single-phase earth fault positioning method based on data processing of the power distribution network, and particularly discloses a method for acquiring massive real-time data of the power distribution network through equipment such as a power distribution terminal, a fault indicator and a smart meter, establishing a real-time data analysis platform of the power distribution network based on Storm clusters, designing a data processing topological structure fusing multiple single-phase earth fault positioning technologies, outputting and storing results according to criteria of different single-phase earth fault positioning technologies. However, in the field of fault feeder detection by using the data driving technology in the above scheme, because the field single-phase earth fault case is difficult to obtain, and in addition, the samples of the fault feeder are far less than those of a sound feeder, the imbalance phenomenon among data sample classes is obvious. Assuming that the power distribution network is provided with N feeders, when a single-phase earth fault occurs to a certain feeder, the ratio of the waveform quantity of the fault feeder to the waveform quantity of the healthy feeder is 1: n-1, the fault feeder waveform data is much less than the robust feeder waveform data. Most data sets used for identifying a single-phase earth fault feeder of a power distribution network have a serious problem of unbalanced sample types, when a data drive identification model is trained by using a sample unbalanced data set, the network tends to learn characteristics of most types of samples, the identification capability of few types of samples is weak, and the identification accuracy is low, namely the data drive identification model tends to simply judge the samples as sound feeders when identifying input feeder waveforms, and the fault feeder cannot be correctly identified, so that the grounding point fault condition is further worsened. Therefore, the existing data-driven fault line selection method has the problems of over-training fitting, unsatisfactory field application effect and the like in an unbalanced small sample scene, so that the high reliability of power supply of a power distribution network is difficult to ensure.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a single-phase earth fault feeder line identification method for generating a countermeasure network based on a conditional expression under unbalanced samples, which can provide sufficient and balanced training data for a data driving model, thereby carrying out more accurate single-phase earth fault feeder line identification work.
The technical scheme of the invention is as follows:
the invention provides a method for identifying a single-phase earth fault feeder line under unbalanced samples, which is characterized by comprising the following steps:
s1, screening single-phase ground fault cases on site, collecting first half-wave zero-sequence current signals of first sections of feeder lines of a main station with faults, selecting zero-sequence current data with fault information missing and over-high noise components, and constructing an original database;
s2, imaging the time sequence signals, transforming the time sequence signals in the original database into a time-frequency spectrum gray-scale image, and constructing a sample training set to be enhanced;
s3, constructing a conditional expression-based generation countermeasure network CGAN model; the conditional generation confrontation network CGAN model consists of a generator model and a discriminator model;
s4, inputting the to-be-enhanced sample training set obtained in the step S2 into a CGAN model for model training, obtaining a trained CGAN model through the circulating game training of a generator and a discriminator, and storing the model;
s5, the CGAN model saved in the step S4 is used for enhancing the original database of the single-phase earth fault constructed in the step S1, the quantity difference between the fault feeder line and the non-fault feeder line is balanced, and the quantity of training set samples is increased;
and S6, taking the data set in the original database after being enhanced as a training set of the data driving model, and then classifying by using a test set which is not subjected to data enhancement to obtain a classification result which can indicate that the waveform data belongs to a fault feeder or a sound feeder.
Further, in the step S1, the first half-wave zero-sequence current signal of each feeder line is a one-dimensional timing signal including 100 sampling points, the sampling frequency of the first half-wave zero-sequence current signal is 10kHz, and the sampling time is 0.1S.
Further, the imaging of the time sequence signal in step S2 is specifically to obtain a two-dimensional time-frequency spectrum gray scale map by performing continuous wavelet transform on the original one-dimensional time sequence signal; the characteristic dimension of the time-frequency spectrum gray-scale map is 64 multiplied by 64.
Further, the conditional generation of the confrontation network CGAN model in step S3 is based on the originally generated confrontation network GAN, and introduces the class label of the sample as an additional information condition in the construction process of the generator and the discriminator to guide the generation process of the sample.
Further, the generator network comprises a 1-layer fully-connected layer, a 4-layer deconvolution layer and a 4-layer batch normalization layer, wherein the hidden layer uses a Relu activation function, and the output layer uses a tanh activation function; the discriminator network comprises 5 convolutional layers, 5 batch normalization layers and 1 full-connection layer, and the activation functions are LeakyRelu and Sigmoid activation functions.
Further, the specific step of inputting the to-be-enhanced sample training set into the CGAN model for model training in step S4 includes:
a1, initializing generator and discriminator network parameter;
a2, fixing a generator network, and performing 100 rounds of pre-training on the discriminator by using the same number of sample training sets to be enhanced and the generator generated samples, so as to improve the discrimination capability of the discriminator;
a3, circularly training a generator and a discriminator until the generator and the discriminator reach the Nash equilibrium point of the generator and the discriminator, generating a time-frequency spectrum sample distribution which is consistent with the real sample distribution, and storing model parameters after training;
a4, performing One-Hot coding on the state type of the waveform sample, splicing random Gaussian noise with the characteristic dimension of 2 and the dimension of 100 and label information of each label information, inputting the trained model, and mapping the model into a specific type generation sample through a generator network.
Further, the generator G and the discriminator D adopt a stochastic gradient descent method to optimize the model, and the optimized parameters comprise weight omega and bias b of the neural network linear layer of the generator G and the discriminator D.
Further, when initializing the network parameters of the generator G and the discriminator D in the step a2, the learning rate is set to 0.0002, the number of iterations is 4000, a small-batch training mode is adopted, the number of batch iteration samples is set to 128, and the Adam algorithm is selected as the optimization algorithm.
When initializing the network parameters of the generator and the discriminator in the step A1, the learning rate is set to 0.0002, the iteration times is 4000, a small-batch training mode is adopted, the number of batch iteration samples is set to 128, and the Adam algorithm is selected as the optimization algorithm.
Further, the data driving model in step S6 is a convolutional neural network CNN, which includes 3 convolutional layers, 2 pooling layers, and 1 fully-connected layer, and the fully-connected layer splicing vector passes through a softmax activation function to obtain a fault feeder identification result.
Further, the classification performance of the convolutional neural network CNN adopts four measurement criteria of identification accuracy, precision, recall rate and F1 measurement as evaluation indexes.
Compared with the prior art, the invention has the beneficial effects that:
1. the method for identifying the one-way ground fault feeder line under the unbalanced samples realizes the enhancement of the identification data of the one-way ground fault feeder line by utilizing the conditional generation countermeasure network CGAN, increases the number of the training set samples, can obtain reliable and sufficient labeled samples by utilizing the existing data, does not need a large amount of manpower and material resources to obtain the data and the labeled data, and increases the identification accuracy of the one-way ground fault feeder line under the small sample scene by utilizing the number of the training set samples.
2. The method for identifying the one-way grounding fault feeder line under the unbalanced sample realizes that the conditional generation confrontation network CGAN model is used for generating the specified type generation sample which is consistent with the actual single-phase grounding fault feeder line data distribution, further realizes the balance of the number of the fault feeder line samples and the number of the sound feeder line samples, solves the unbalanced problem between the fault feeder line data and the sound feeder line data in the single-phase grounding fault feeder line identification, and further improves the identification accuracy of the single-phase grounding fault feeder line of the data driving model under the unbalanced scene.
3. According to the method for identifying the single-phase earth fault feeder line under the unbalanced sample condition, the countermeasure network CGAN is generated by utilizing the conditional expression to obtain the enhanced data sets with sufficient quantity and balanced categories, the enhanced data sets are used as the training sets of the convolutional neural network classification model, the problem of overfitting can be solved, the identification performance of the classification model on the single-phase earth fault feeder line is comprehensively improved, and the high reliability of power supply of a power distribution network is guaranteed.
Drawings
FIG. 1 is a flow chart of the present invention. .
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments.
Referring to fig. 1, the invention provides a method for identifying a single-phase ground fault feeder line under sample imbalance, which specifically comprises the following steps:
s1, screening single-phase ground fault cases on site, collecting first half-wave zero-sequence current signals of first sections of feeder lines of a main station with faults, starting to collect first half-wave zero-sequence current signals of the first sections of the feeder lines after the single-phase ground faults occur, wherein the sampling frequency is 10kHz, and the sampling time is 0.1S, so that the first half-wave signals of the first sections of the zero-sequence currents of the feeder lines are one-dimensional time sequence signals containing 100 sampling points, zero-sequence current data with fault information missing and over-high noise components are selected, and an original database is constructed;
s2, imaging the zero sequence current data, namely the time sequence signals, obtained in the step S1, transforming the zero sequence current data, namely the one-dimensional time sequence signals, into a two-dimensional time-frequency spectrum gray-scale map through continuous wavelet transformation, and constructing a sample training set to be enhanced; wherein, let the one-dimensional time sequence signal be x (t), and the result of the continuous wavelet transform of x (t) bePsi is a mother wavelet function, a is a scale parameter, and b is a displacement parameter; the one-dimensional time sequence signal is converted into a time-frequency spectrum gray-scale map containing rich time-frequency information after continuous wavelet transformation, and the characteristic dimension is 64 multiplied by 64;
s3, constructing a conditional generation confrontation network CGAN model, wherein the conditional generation confrontation network CGAN model consists of a generator model and a discriminator model; on the basis of the original generation of the confrontation network GAN, a conditional generation confrontation network CGAN model guides the generation process of a sample by introducing a class label of the sample as an additional information condition in the construction process of a generator G and a discriminator D, so that the defect that the original generation of the confrontation network GAN cannot control the generation of a data mode is overcome;
s4, inputting the to-be-enhanced sample training set obtained in the step S2 into a CGAN model for model training, enhancing the single-phase earth fault feeder data through the circulating game training of a generator G and a discriminator D, obtaining a trained CGAN model, and storing the model; after the label and the noise are spliced, inputting a generated sample obtained in the trained CGAN model;
wherein, through the cyclic game training of the generator G and the discriminator DThe objective function of the exercise is:in the formula, E [. cndot]Representing a desired operation; c is a class label of the sample; max represents that when the input is a real sample, the discriminator D maximizes the real sample discrimination probability; min represents the probability that the generator G ensures that the minimum generated sample is judged as the 'generated sample' when the input is the generated sample;
s5, the CGAN model saved in the step S4 is used for enhancing the original database of the single-phase earth fault constructed in the step S1, the quantity difference between the fault feeder line and the non-fault feeder line is balanced, and the quantity of training set samples is increased;
and S6, taking the data set in the original database after being enhanced as a training set of the data driving model, and then classifying by using a test set which is not subjected to data enhancement to obtain a classification result which can indicate that the waveform data belongs to a fault feeder or a sound feeder.
Further, the generator network in step S3 includes a 1-layer fully-connected layer, a 4-layer deconvolution layer, and a 4-layer batch normalization layer, where the hidden layer uses a Relu activation function, and the output layer uses a tanh activation function, and the expression isWherein x is input data of the tan h activation function layer, y is output of the tan h activation function layer, namely, a time-frequency spectrum gray-scale map generated by the G network of the generator, and the data range of the time-frequency spectrum gray-scale map belongs to [ -1,1](ii) a The discriminator network comprises 5 convolutional layers, 5 batch normalization layers and 1 full-connection layer, the activation functions are LeakyRelu and Sigmoid activation functions, and the corresponding activation function expressions are y ═ max (0.01x, x) andwherein x is input data of the activation function layer, y is output of the activation function layer, namely authenticity judgment score of the network of the discriminator D for the input sample, and the data range thereof belongs to [0,1 ]];
In the training process of generating the anti-network CGAN model by the conditional expression, the output distribution of each layer of network gradually approaches the upper limit and the lower limit of the nonlinear activation function along with the updating of parameters, so that the problem of gradient disappearance is caused, and the convergence speed is reduced. In order to solve the above problem, a batch normalization layer is added after the deconvolution layer of the generator G and the convolution layer of the discriminator D, so that the input of the network can fall into the linear region of the nonlinear activation function, and the convergence speed of the conditional generation confrontation network CGAN model is increased.
Further, the specific step of inputting the to-be-enhanced sample training set into the CGAN model for model training in step S4 includes:
a1, initializing a generator G and a discriminator D, setting the learning rate to be 0.0002, setting the iteration number to be 4000, adopting a small-batch training mode, setting the number of batch iteration samples to be 128, and selecting an Adam algorithm as an optimization algorithm;
a2, fixing a generator G network, and performing 100 rounds of pre-training on a discriminator D by using the same number of sample training sets to be enhanced and the generator G generated samples, so as to improve the discrimination capability of the discriminator D;
a3, a cyclic training generator G and a discriminator D, firstly fixing the generator G, updating the discriminator D3 times, then fixing the parameters of the discriminator D, updating the generator G1 times, finishing the conditional generation confrontation network CGAN model training, at the moment, the generator G and the discriminator D reach the Nash equilibrium point of the generator G and the discriminator D, the distribution of the generated time-frequency spectrum samples is consistent with the distribution of the real samples, the discriminator D is difficult to distinguish whether the input data comes from the generated samples or the real samples, the output scalar value is close to 0.5, and after the training is finished, the model parameters are stored;
the generator G and the discriminator D adopt a random gradient descent method to optimize the model, the optimized parameters comprise weight omega and bias b of a neural network linear layer of the generator G and the discriminator D, and meanwhile, the hyper-parameters such as learning rate, iteration times, batch training quantity and the like are finely adjusted according to the quality of a generated sample so as to obtain an optimal conditional generation confrontation network model;
a4, performing One-Hot coding on the state type of the waveform sample, splicing the random Gaussian noise with the dimension of 2 and the dimension of 100 with the label information, inputting the spliced label information into the trained model, mapping the spliced label information into a specified type of generated sample through a generator network, and enabling the generated sample dimension to be consistent with the dimension of a real time frequency spectrum gray level graph; the discriminator D is a binary classifier, the input sample consists of a generation sample and a real sample, the output is a scalar, the size of the scalar reflects the source of the input sample, and the larger the value of the scalar is, the higher the probability that the input sample is derived from the real sample is.
Further, the data driving model in step S6 is a convolutional neural network CNN, which includes 3 convolutional layers, 2 pooling layers, and 1 fully-connected layer, the fully-connected layer splicing vector obtains a fault feeder identification result through a softmax activation function, and in order to be able to comprehensively and effectively evaluate the classification performance of the convolutional neural network CNN under the condition that the data set has sample class imbalance, four measurement standards of identification Accuracy (Accuracy), Precision (Precision), Recall rate (Recall), and F1 are selected as evaluation indexes, and the calculation formula is as follows:
in the formula, TP represents a positive sample with correct prediction, TN represents a negative sample with correct prediction, FP represents a positive sample with wrong prediction, and FN represents a negative sample with wrong prediction;
in order to verify the effectiveness of the method provided by the invention, a method which does not use the method of the invention and directly uses a convolutional neural network CNN for identification is selected as a comparison method; in order to ensure the reliability of the test result, the structure of the convolutional neural network CNN in the method provided by the invention is consistent with that of a comparison method, and the adopted test set is also the same; wherein the experimental results on the unbalanced data set are shown in table 1;
table 1 experimental results on unbalanced data set
The classification results in table 1 show that the inter-class imbalance of the data samples results in a low identification accuracy of CNN to a few classes of samples, namely, the fault feeder waveform, but the method of the present invention expands the sample data set by using the high quality fault feeder sample generated by CGAN, and solves the problem of inter-class imbalance, so that the comprehensive identification performance of the implementation method provided by the present invention is significantly better than that of the comparison method, and the identification accuracy, recall rate and F1 accuracy are respectively 4.816%, 29.8% and 13.3% higher;
meanwhile, in order to verify the identification performance of the method under a small sample identification scene, a spectrum gray-scale image generated by a conditional generation countermeasure network CGAN is used for expanding an original training set on the premise of not changing the proportion of the number of fault feeder waveforms to the number of sound feeder waveforms, and the number of samples in the training set is increased. The comparison method does not expand the original training set, and the recognition result is shown in table 2;
TABLE 2 recognition Performance under Small sample recognition scenarios
As can be seen from table 2, the recognition accuracy, precision, recall rate and F1 precision of the method provided by the present invention are all higher than those of the comparison method, which proves that the conditional expression provided by the present invention is used to generate the confrontation network CGAN enhanced data set training CNN, thereby solving the problem of insufficient number of samples in the training set, improving the recognition performance of the data-driven model, and having strong generalization capability.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by the present specification, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (10)
1. A single-phase earth fault feeder line identification method under sample imbalance is characterized by comprising the following steps:
s1, screening single-phase ground fault cases on site, collecting first half-wave zero-sequence current signals of first sections of feeder lines of a main station with faults, selecting zero-sequence current data with fault information missing and over-high noise components, and constructing an original database;
s2, imaging the time sequence signals, transforming the time sequence signals in the original database into a time-frequency spectrum gray-scale image, and constructing a sample training set to be enhanced;
s3, constructing a conditional expression-based generation countermeasure network CGAN model; the conditional generation confrontation network CGAN model consists of a generator model and a discriminator model;
s4, inputting the to-be-enhanced sample training set obtained in the step S2 into a CGAN model for model training, obtaining a trained CGAN model through the circulating game training of a generator and a discriminator, and storing the model; after the label and the noise are spliced, inputting a generated sample obtained in the trained CGAN model;
s5, the CGAN model saved in the step S4 is used for enhancing the original database of the single-phase earth fault constructed in the step S1, the quantity difference between the fault feeder line and the non-fault feeder line is balanced, and the quantity of training set samples is increased;
and S6, taking the data set in the original database after being enhanced as a training set of the data driving model, and then classifying by using a test set which is not subjected to data enhancement to obtain a classification result which can indicate that the waveform data belongs to a fault feeder or a sound feeder.
2. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 1, wherein the method comprises the following steps: in the step S1, the first half-wave zero-sequence current signal of each feeder line is a one-dimensional timing signal including 100 sampling points, the sampling frequency of the first half-wave zero-sequence current signal is 10kHz, and the sampling time is 0.1S.
3. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 1, wherein the method comprises the following steps: the step S2 of imaging the time sequence signal is to obtain a two-dimensional time-frequency spectrum gray scale image from the original one-dimensional time sequence signal through continuous wavelet transform; the characteristic dimension of the time-frequency spectrum gray-scale map is 64 multiplied by 64.
4. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 1, wherein the method comprises the following steps: the conditional generation of the confrontation network CGAN model in step S3 is to introduce the class label of the sample as an additional information condition in the construction process of the generator and the discriminator on the basis of the originally generated confrontation network GAN to guide the generation process of the sample.
5. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 4, wherein the method comprises the following steps: the generator network comprises a 1-layer full-connection layer, a 4-layer deconvolution layer and a 4-layer batch normalization layer, wherein the hidden layer uses a Relu activation function, and the output layer uses a tanh activation function; the discriminator network comprises 5 convolutional layers, 5 batch normalization layers and 1 full-connection layer, and the activation functions are LeakyRelu and Sigmoid activation functions.
6. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 1, wherein the method comprises the following steps: the specific steps of inputting the to-be-enhanced sample training set into the CGAN model for model training and generating the sample in step S4 include:
a1, initializing generator and discriminator network parameter;
a2, fixing a generator network, and performing 100 rounds of pre-training on the discriminator by using the same number of sample training sets to be enhanced and the generator generated samples, so as to improve the discrimination capability of the discriminator;
a3, circularly training a generator and a discriminator until the generator and the discriminator reach the Nash equilibrium point of the generator and the discriminator, generating a time-frequency spectrum sample distribution which is consistent with the real sample distribution, and storing model parameters after training;
a4, performing One-Hot coding on the state type of the waveform sample, splicing random Gaussian noise with the characteristic dimension of 2 and the dimension of 100 and label information of each label information, inputting the trained model, and mapping the model into a specific type generation sample through a generator network.
7. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 6, wherein the method comprises the following steps: the generator and the discriminator adopt a random gradient descent method to optimize the model, and the optimized parameters comprise the weight omega and the bias b of the neural network linear layer of the generator and the discriminator.
8. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 7, wherein the method comprises the following steps: when initializing the network parameters of the generator and the discriminator in the step A1, the learning rate is set to 0.0002, the iteration times is 4000, a small-batch training mode is adopted, the number of batch iteration samples is set to 128, and the Adam algorithm is selected as the optimization algorithm.
9. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 1, wherein the method comprises the following steps: the data driving model in the step S6 is a convolutional neural network CNN, and includes 3 convolutional layers, 2 pooling layers, and 1 fully-connected layer, and the fully-connected layer splicing vector passes through the softmax activation function to obtain a fault feeder line recognition result.
10. The method for identifying the single-phase ground fault feeder line under the condition of sample imbalance according to claim 1, wherein the method comprises the following steps: the classification performance of the convolutional neural network CNN adopts four measurement standards of identification accuracy, precision, recall rate and F1 measurement as evaluation indexes.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111521526.0A CN114218988A (en) | 2021-12-13 | 2021-12-13 | Method for identifying unidirectional ground fault feeder line under unbalanced samples |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111521526.0A CN114218988A (en) | 2021-12-13 | 2021-12-13 | Method for identifying unidirectional ground fault feeder line under unbalanced samples |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114218988A true CN114218988A (en) | 2022-03-22 |
Family
ID=80701596
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111521526.0A Pending CN114218988A (en) | 2021-12-13 | 2021-12-13 | Method for identifying unidirectional ground fault feeder line under unbalanced samples |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114218988A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116432534A (en) * | 2023-04-18 | 2023-07-14 | 中海石油(中国)有限公司上海分公司 | Data-driven TOC sample prediction method |
CN117743947A (en) * | 2024-02-20 | 2024-03-22 | 烟台哈尔滨工程大学研究院 | Intelligent cabin fault diagnosis method and medium under small sample |
-
2021
- 2021-12-13 CN CN202111521526.0A patent/CN114218988A/en active Pending
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116432534A (en) * | 2023-04-18 | 2023-07-14 | 中海石油(中国)有限公司上海分公司 | Data-driven TOC sample prediction method |
CN117743947A (en) * | 2024-02-20 | 2024-03-22 | 烟台哈尔滨工程大学研究院 | Intelligent cabin fault diagnosis method and medium under small sample |
CN117743947B (en) * | 2024-02-20 | 2024-04-30 | 烟台哈尔滨工程大学研究院 | Intelligent cabin fault diagnosis method and medium under small sample |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP3832553A1 (en) | Method for identifying energy of micro-energy device on basis of bp neural network | |
CN109164343B (en) | Transformer fault diagnosis method based on characteristic information quantization and weighted KNN | |
CN113033837A (en) | Artificial intelligence fault identification system and method based on transient waveform of power transmission line | |
CN114218988A (en) | Method for identifying unidirectional ground fault feeder line under unbalanced samples | |
CN110070102B (en) | Method for establishing sequence-to-sequence model for identifying power quality disturbance type | |
WO2008016109A1 (en) | Learning data set optimization method for signal identification device and signal identification device capable of optimizing the learning data set | |
CN110726898A (en) | Power distribution network fault type identification method | |
CN113884844A (en) | Transformer partial discharge type identification method and system | |
CN111400540A (en) | Singing voice detection method based on extrusion and excitation residual error network | |
CN111382803A (en) | Feature fusion method based on deep learning | |
CN112434743A (en) | Fault identification method based on GIL metal particle partial discharge time domain waveform image | |
CN113889198A (en) | Transformer fault diagnosis method and equipment based on oil chromatogram time-frequency domain information and residual error attention network | |
CN115471712A (en) | Learning method for generating zero sample based on visual semantic constraint | |
CN111144462A (en) | Unknown individual identification method and device for radar signals | |
CN109145685A (en) | Fruits and vegetables EO-1 hyperion quality detecting method based on integrated study | |
CN114511774A (en) | Ship target comprehensive identification method, medium and system | |
CN116561562B (en) | Sound source depth optimization acquisition method based on waveguide singular points | |
CN116594057B (en) | Earthquake early warning method and device based on deep learning and edge calculation | |
CN111796173B (en) | Partial discharge pattern recognition method, computer device, and storage medium | |
CN115409317A (en) | Transformer area line loss detection method and device based on feature selection and machine learning | |
CN113657149A (en) | Electric energy quality analysis and identification method based on deep learning | |
CN113762131A (en) | Method and terminal for detecting deformation position of oil casing double-layer pipe column of high-sulfur-content gas field | |
CN107492384B (en) | Voice emotion recognition method based on fuzzy nearest neighbor algorithm | |
CN112001449A (en) | Ship electric power system fault identification method | |
CN112163613A (en) | Rapid identification method for power quality disturbance |
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 |