CN114186589A - Superconducting cable partial discharge mode identification method based on residual error network Resnet50 - Google Patents
Superconducting cable partial discharge mode identification method based on residual error network Resnet50 Download PDFInfo
- Publication number
- CN114186589A CN114186589A CN202111493810.1A CN202111493810A CN114186589A CN 114186589 A CN114186589 A CN 114186589A CN 202111493810 A CN202111493810 A CN 202111493810A CN 114186589 A CN114186589 A CN 114186589A
- Authority
- CN
- China
- Prior art keywords
- partial discharge
- superconducting cable
- model
- training
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 77
- 238000012360 testing method Methods 0.000 claims abstract description 49
- 238000009826 distribution Methods 0.000 claims abstract description 24
- 238000012795 verification Methods 0.000 claims abstract description 24
- 238000001514 detection method Methods 0.000 claims abstract description 22
- 238000007781 pre-processing Methods 0.000 claims abstract description 17
- 230000006870 function Effects 0.000 claims description 25
- 230000007547 defect Effects 0.000 claims description 15
- 238000009413 insulation Methods 0.000 claims description 12
- 238000012567 pattern recognition method Methods 0.000 claims description 12
- 238000003909 pattern recognition Methods 0.000 claims description 9
- 238000013526 transfer learning Methods 0.000 claims description 9
- 238000010606 normalization Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 230000003321 amplification Effects 0.000 claims description 5
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000002922 simulated annealing Methods 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 3
- 238000005339 levitation Methods 0.000 claims 1
- 238000004880 explosion Methods 0.000 abstract description 4
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 abstract 1
- 230000009286 beneficial effect Effects 0.000 abstract 1
- 230000000052 comparative effect Effects 0.000 description 11
- 238000000605 extraction Methods 0.000 description 9
- 230000008859 change Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000004422 calculation algorithm Methods 0.000 description 4
- 230000008034 disappearance Effects 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 210000005036 nerve Anatomy 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000010355 oscillation Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000000725 suspension Substances 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000006731 degradation reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/2431—Multiple classes
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- 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)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Human Resources & Organizations (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Economics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Tourism & Hospitality (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- General Business, Economics & Management (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Software Systems (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Entrepreneurship & Innovation (AREA)
- Signal Processing (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Primary Health Care (AREA)
- Testing Relating To Insulation (AREA)
Abstract
The invention discloses a superconducting cable partial discharge mode identification method based on a residual error network Resnet50, which comprises the following steps: the method comprises the following steps: collecting different types of partial discharge fault signals of the superconducting cable by using a partial discharge detection system, and using the collected fault signals to construct a PRPD spectrogram data set of the phase distribution mode of the superconducting cable fault; step two: classifying PRPD spectrograms in a data set according to different discharge types, dividing the PRPD spectrograms into a training set, a verification set and a test set according to a certain proportion, and then preprocessing images of the PRPD spectrograms; step three: and constructing a Resnet50 network model, taking the training samples and the testing samples as input, and training and testing the network model. The deeper network can be trained by using the residual block, the problems of gradient loss and gradient explosion generated along with the deepening of the network depth are effectively solved, and particularly, the method shows good performance when the deeper network is trained, and is beneficial to realizing the mode identification of the fault of the superconducting cable.
Description
Technical Field
The invention relates to the technical field of superconducting cable fault identification, in particular to a superconducting cable partial discharge mode identification method based on a residual error network Resnet 50.
Background
The type of the partial discharge is different, the corresponding insulation defect is also different, and if the characteristic of the partial discharge can be determined, the insulation condition of the power superconducting cable can be better judged. To know the insulation condition inside the power superconducting cable, the most effective solution is to detect the partial discharge of the superconducting cable and perform pattern recognition on the discharge type. According to long-term experience conclusion, the output and the change of the partial discharge signal of the power superconducting cable can reflect the insulation condition of the superconducting cable. Therefore, the mode identification can effectively judge the insulation state and the damage degree of the power superconducting cable and provide more scientific information for accurately evaluating the insulation performance and the operation condition of the superconducting cable.
With the advent of computers and artificial intelligence technology, pattern recognition was rapidly developed into a subject in the 60 s and has been widely used in many fields. At present, the pattern recognition method is continuously developing towards the direction of intellectualization, and higher requirements are put forward on the self-adaptive capacity, the learning capacity, the fault-tolerant capacity and the like of the system. Since the 90 s, the pattern recognition method is gradually popularized and applied to the field of partial discharge, and compared with the traditional method of judging the type of the partial discharge by depending on the actual engineering experience and theoretical knowledge of relevant experts, the pattern recognition method obviously improves the scientificity and effectiveness of a recognition result.
However, the research of the current partial discharge pattern recognition technology is still in a starting stage, on one hand, because many complex and uncertain factors influence the acquisition of partial discharge signals and interference of reasons such as the surrounding environment and the like is added, many interference signals are mixed in the partial discharge signals, and a better signal processing method needs to be found to obtain more accurate partial discharge signals; on the other hand, the information contained in the partial discharge signal is complex, and the content and the law thereof are not completely clear. Therefore, further intensive research into the partial discharge principle and practical problems thereof is also required.
Therefore, the invention provides a superconducting cable partial discharge mode identification method based on a residual error network Resnet50, which is used for identifying the superconducting cable partial discharge type by acquiring different types of PRPD spectrograms of superconducting cable partial discharge phase distribution modes, so that the partial discharge type corresponding to each spectrogram is identified. The pattern recognition method can replace the traditional work of judging the partial discharge type of the superconducting cable by means of the actual engineering experience of relevant experts in the industry, meanwhile, a Resnet50 network model is adopted, the training process of the convolutional neural network is accelerated through the idea of residual error learning, the problems of gradient disappearance and gradient explosion are effectively avoided, the model training can be completed in a short time, and the higher recognition accuracy is obtained.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides a superconducting cable partial discharge mode identification method based on a residual error network Resnet50, which is used for solving the problems of gradient disappearance and gradient explosion and ensuring good performance when a deeper network is trained.
In order to achieve the purpose, the invention adopts the following technical scheme:
a superconducting cable partial discharge mode identification method based on a residual error network Resnet50 comprises the following steps:
s1: collecting different types of fault signals of the superconducting cable by using a partial discharge detection system, and using the collected signals to construct a partial discharge phase distribution pattern PRPD spectrogram data set of the superconducting cable insulation defect;
s2: classifying PRPD spectrograms in a data set according to different fault types, preprocessing images and amplifying data, and then dividing the PRPD spectrograms into a training set, a verification set and a test set according to a certain proportion;
s3: constructing a Resnet50 residual network model by adopting a transfer learning mode, completing the configuration of each parameter of the model, taking a training sample and a verification sample as input, and training and verifying the network model;
s4: performing convolution, pooling, activation and regression operations on the network model to update network parameters, training the network model, obtaining a trained network model when the loss rate function is not reduced and the accuracy rate function is not increased, and using the obtained model for testing;
s5: and identifying the fault type of the superconducting cable by adopting an optimal Resnet50 network model, thereby constructing a Resnet 50-based superconducting cable fault mode identification system.
Further, in the step S2, the specific steps are as follows:
s21: collecting PRPD spectrograms of various different partial discharge phase distribution modes of the superconducting cable through a partial discharge detection system, and classifying the PRPD spectrograms according to partial discharge types;
s22: preprocessing the image of the collected partial discharge signal;
s23: carrying out data amplification processing on the preprocessed image;
s24: and taking the processed image as a data set of the model, and dividing the data set into a training set, a verification set and a test set according to a certain proportion.
Further, in the step S21, the partial discharge of the superconducting cable is mainly represented by three major categories of air gap discharge, suspension discharge and normal external noise interference through the PRPD spectrum.
Further, in the step S22, in order to ensure that the specifications and sizes of the data are the same and the formats are consistent, the image needs to be preprocessed, where the preprocessing manner includes an image sample normalization preprocessing method and an image pixel value preprocessing method.
In step S23, the number of partial discharge samples is increased by data expansion, and the data is expanded by image translation, chromaticity change, shading change, random cropping, or the like.
Further, in the step S24, 70% of images in the processed partial discharge phase distribution pattern PRPD spectrogram data set are extracted as a training set, 10% of images are extracted as a verification set, and 20% of images are extracted as a test set.
Further, in the step S3, the specific steps are as follows:
s31: adopting Resnet50 model in the mode of transfer learning;
s32: configuring and adjusting various parameters of the model;
s33: and taking the training sample and the verification sample as input, and training and verifying the network model.
Further, in the step S4, the specific steps are as follows:
s41: defining a loss function, and taking the cross entropy function as the loss function, wherein the calculation formula is as follows:
wherein x is a sample value which is randomly distributed, p is expected probability distribution of the sample x, and q is actual probability distribution of the sample x;
s42: defining an optimization function, and realizing self-adaptive adjustment of simulated annealing based on cosine attenuation in order to ensure that the model training can achieve the optimal speed and precision, wherein the cosine attenuation formula is as follows:
wherein alpha is0Denotes the initial learning rate, αtThe learning rate in the T iteration is represented, and T represents the total iteration number;
s43: setting parameters of model training, setting the image resolution to be 224 multiplied by 224, setting the initial learning rate to be 0.001 and setting the number of training rounds to be 100;
s44: and training the network model, obtaining the trained network model when the loss rate function is not reduced and the accuracy rate function is not increased, and using the obtained model for testing the test sample.
Further, the superconducting cable partial discharge pattern recognition system based on Resnet50 in the step S5 includes: the device comprises a partial discharge detection unit, a discharge spectrogram construction unit, a data diversity unit, a model training unit, a model testing unit and an identification unit, wherein the partial discharge detection unit is connected with the data diversity unit through the discharge spectrogram construction unit, the data diversity unit is respectively connected with the model training unit and the model testing unit, the model training unit is connected with the model testing unit, and the partial discharge detection unit is connected with the model testing unit through the identification unit.
Further, the partial discharge detection unit is used for acquiring several types of partial discharge which often occur to the superconducting cable and partial discharge signals of the typical defects; the building discharge spectrogram unit is used for building a PRPD spectrogram data set of a partial discharge phase distribution pattern of the superconducting cable; the data diversity unit is used for dividing PRPD spectrogram in the data set into a training set, a test set and a verification set according to a certain proportion; the model training unit is used for taking a training sample and a verification sample as input and training and verifying a Resnet50 network model; the model testing unit is used for testing the trained network model on a test sample, and the individual recognition rate and the average recognition rate of various partial discharge types can be obtained according to the test result; the identification unit is used for carrying out pattern identification on the fault type of the partial discharge of the superconducting cable by utilizing an optimal Resnet50 network model.
Compared with the prior art, the invention has the following advantages:
1. the method solves the problems of noise interference of various types and difficulty in judging the partial discharge type in the process of collecting the partial discharge signal of the superconducting cable, realizes the identification of the partial discharge type of the superconducting cable by adopting a Resnet50 network model, has the identification accuracy rate of more than 97 percent, meets the requirement of the identification of the partial discharge mode of the superconducting cable, and has a wide application prospect.
2. The invention provides a superconducting cable partial discharge mode identification method, which comprises the following steps: collecting different types of partial discharge signals of the superconducting cable by using a partial discharge detection system, and using the collected partial discharge signals to construct a partial discharge phase distribution pattern PRPD spectrogram data set of the superconducting cable insulation defect; classifying PRPD spectrograms in a data set according to different discharge types, preprocessing images and amplifying data, and then dividing the PRPD spectrograms into a training set, a verification set and a test set according to a certain proportion; constructing a Resnet50 residual network model by adopting a transfer learning mode, completing the configuration of each parameter of the model, taking a training sample and a verification sample as input, and training and verifying the network model; performing convolution, pooling, activation and regression operations on the network model to update network parameters, training the network model, obtaining a trained network model when the loss rate function is not reduced and the accuracy rate function is not increased, and using the obtained model for testing; and identifying the fault type of the partial discharge of the superconducting cable by adopting an optimal Resnet50 network model, thereby constructing a Resnet 50-based superconducting cable partial discharge mode identification system.
3. According to the method, the number of partial discharge samples of the superconducting cable is increased by adopting a data amplification mode, and the data is subjected to amplification processing by adopting methods such as image translation, chromaticity change, brightness change, random cutting and the like according to the internal structure of the power superconducting cable and the partial discharge signal characteristics of typical defects, so that the problem of poor learning effect caused by the small number of samples is avoided, and the effect of model training is further improved.
4. The invention adopts a learning rate parameter self-adaptive adjustment simulated annealing algorithm based on cosine attenuation, and the algorithm ensures that the model has higher convergence speed by adopting higher learning rate when the model training is started, and automatically reduces the learning rate when the model convergence reaches the vicinity of the optimal point so as to avoid oscillation, thereby effectively ensuring that the model training can reach the optimal speed and precision.
5. The invention is further improved on the basis of a Resnet50 network model, the Resnet50 network is a residual error network, and two different connection relations exist between input and output in the residual error network, wherein one is that input data are output after passing through a plurality of nerve layers, and the other is that the input data are directly connected to the output. The Resnet50 network model can effectively solve the problems of gradient explosion and gradient disappearance, and in addition, the deep structure of the network is changed into a parallel structure when the residual error network carries out forward propagation, so that even if the capacity of the deep neural network model is increased, the model cannot generate the phenomenon of overfitting to influence the training effect of the model. Meanwhile, the result shows that the application of the Resnet50 model to the classification and identification of the superconducting cable partial discharge fault types is feasible through a transfer learning method, the model training can be completed in a short time through the mode identification method, and high identification accuracy can be obtained.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.
FIG. 1 is a flow chart of the steps of the present invention;
fig. 2 is a schematic flow chart of a superconducting cable partial discharge pattern recognition system based on Resnet50 in the invention;
FIG. 3 is a schematic diagram of the operating logic of the present invention;
FIG. 4 is a graph comparing a feed-forward neural network and a residual neural network;
fig. 5 is a diagram of a Resnet50 network architecture.
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 it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
Examples
Referring to fig. 1 and 3, a superconducting cable partial discharge pattern recognition method includes the steps of:
step 101: collecting different types of partial discharge signals of the superconducting cable by using a partial discharge detection system, and using the collected partial discharge signals to construct a partial discharge phase distribution pattern PRPD spectrogram data set of the superconducting cable insulation defect;
specifically, the partial discharge phase distribution PRPD mode is a commonly used discharge mode, which is also called as a phi-q-n mode, and the mode describes a relationship between a power frequency phase phi (0 to 360 °) corresponding to a partial discharge pulse, a discharge amount q (q represents a pulse amplitude value because the discharge amount cannot be standardized during detection by an ultrahigh frequency method), and a discharge frequency n.
Step 103: classifying PRPD spectrograms in a data set according to different discharge types, preprocessing images and amplifying data, and then dividing the PRPD spectrograms into a training set, a verification set and a test set according to a certain proportion;
specifically, partial discharge of the superconducting cable is mainly represented as three types of air gap discharge, suspension discharge and normal external noise interference through a PRPD spectrogram;
the acquired original PRPD spectrogram has different sources, sizes and formats, so that the acquired original PRPD spectrogram cannot be trained in the same model, and in order to ensure that the specifications and the sizes of the data are the same and the formats are consistent, images need to be preprocessed, wherein the preprocessing method mainly comprises two preprocessing methods, namely image sample normalization and image pixel value processing;
in addition, the model training requires that the resolution of the input picture should be 224 × 224, and the resolution of the acquired original PRPD spectrogram is random, so that the sizes of the pictures need to be unified through scaling and cropping. The calculation formula of the scaling factor of the picture is as follows:
where w and h represent the width and height of the picture, respectively, and scale represents the scaling factor.
On the basis of completing the picture scaling, the picture is cropped by adopting a center cropping method, and the width and height of the cropping are set to be 224, so that the picture with the resolution of 224 × 224 is obtained. In order to effectively improve the effect of model training and also need to perform normalization processing on the zoomed and cut picture, the invention adopts a 0-mean normalization method to perform normalization processing on picture data, and the normalization formula is as follows:
wherein z represents the normalized value, X represents the original value, and X0And σ represents the mean and variance of the original values.
The image data are different, and the corresponding image formats are also different, so that in order to keep the formats of the image data consistent, the pixel values of the image need to be processed, that is, the image in the BMP format is converted into a matrix vector of three RGB channels and then stored.
More specifically, because the number of samples of the original data is small, especially, the acquisition of the PRPD spectrogram sample of the partial discharge of the superconducting cable is difficult, and the learning effect of deep learning is greatly influenced due to the small number of samples, the number of samples of the partial discharge needs to be increased by adopting a data augmentation method, and the data is augmented by adopting methods such as image translation, chromaticity change, brightness change, random clipping and the like according to the internal structure of the power superconducting cable and the characteristic of the partial discharge signal of a typical defect.
Further, 70% of images in the processed partial discharge phase distribution pattern PRPD spectrogram data set are extracted as a training set, 10% of images are extracted as a verification set, and 20% of images are extracted as a test set.
Step 105: constructing a Resnet50 residual network model by adopting a transfer learning mode, completing the configuration of each parameter of the model, taking a training sample and a verification sample as input, and training and verifying the network model;
referring to fig. 4, specifically, the Resnet network used is also called a residual neural network, and refers to a module that adds residual learning in a conventional convolutional neural network, and may be defined by the formula:
y＝F(x，{Wi})+x
where y denotes the output, F (x, { Wi }) denotes the residual portion, and x denotes the sample.
The residual error network effectively solves the problem of model degradation through Short-cut connection, and the Short-cut connection is formed by adding a jump connection on the basis of a standard feedforward convolutional neural network, so that certain convolutional layers are skipped and directly connected to output.
The predicted value H (x) passing through the two layers of feedforward neural networks is calculated as follows:
f(x)＝ReLU(bi+xin·w1)
H(x)＝ReLU(b2+f(x)·w2)
wherein x isinIs an input feature; w is a1、w2The weights of the first and second nerve layers respectively; b1、 b2Bias of the first and second nerve layers, respectively; ReLU is the activation function.
Assuming that the functional relationship between the predicted values h (x) and x in the residual network satisfies h (x) 2x, the method will be described
H(x)＝ReLU(b2+f(x)·w2)+xin
Referring to fig. 5, the network structure of the Resnet50 model consists of 49 convolutional layers and 1 fully-connected layer.
Step 107: performing convolution, pooling, activation and regression operations on the network model to update network parameters, training the network model, obtaining a trained network model when the loss rate function is not reduced and the accuracy rate function is not increased, and using the obtained model for testing;
specifically, updating the network parameters includes the following steps:
the method comprises the following steps: defining a loss function, and adopting a cross entropy function as the loss function, wherein the calculation formula is as follows:
wherein x is a sample value randomly distributed, p is an expected probability distribution of the sample x, and q is an actual probability distribution of the sample x.
Step two: and defining an optimization function, and using a learning rate parameter self-adaptive adjustment simulated annealing algorithm based on cosine attenuation in order to ensure that the model training can achieve the optimal speed and precision. By the algorithm, the convergence speed of the model can be ensured by using a larger learning rate when the model training is started, and the learning rate is automatically reduced when the model convergence reaches the vicinity of an optimal point so as to avoid oscillation, wherein the cosine attenuation formula is as follows:
wherein alpha is0Denotes the initial learning rate, αtThe learning rate at the T-th iteration is shown, and T represents the total number of iterations.
Step three: the parameters of the model training were set, the image resolution was set to 224 × 224, the initial learning rate was set to 0.001, and the number of training rounds was set to 100 rounds.
Step 109: and identifying the fault type of the partial discharge of the superconducting cable by adopting an optimal Resnet50 network model, thereby constructing a Resnet 50-based superconducting cable partial discharge mode identification system.
According to the embodiment of the invention, a partial discharge mode identification system of the superconducting cable based on Resnet50 is also provided.
Referring to fig. 2, a partial discharge pattern recognition system for a superconducting cable based on Resnet50 includes: the device comprises a partial discharge detection unit, a discharge spectrogram construction unit, a data diversity unit, a model training unit, a model testing unit and an identification unit, wherein the partial discharge detection unit is connected with the data diversity unit through the discharge spectrogram construction unit, the data diversity unit is respectively connected with the model training unit and the model testing unit, the model training unit is connected with the model testing unit, and the partial discharge detection unit is connected with the model testing unit through the identification unit.
Wherein:
the partial discharge detection unit is used for collecting several types of partial discharge which often occur in the superconducting cable and partial discharge signals of the typical defects;
the building discharge spectrogram unit is used for building a PRPD spectrogram data set of a partial discharge phase distribution pattern of the superconducting cable;
the data diversity unit is used for dividing PRPD spectrogram in the data set into a training set, a test set and a verification set according to a certain proportion;
the model training unit is used for taking a training sample and a verification sample as input and training and verifying a Resnet50 network model;
the model testing unit is used for testing the trained network model on a test sample, and the individual recognition rate and the average recognition rate of various partial discharge types can be obtained according to the test result;
and the identification unit is used for carrying out pattern identification on the fault type of the partial discharge of the superconducting cable by utilizing the optimal Resnet50 network model.
In the present embodiment, the modules and method steps of the examples described in connection with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two, and programs corresponding to the software modules and method steps can be disposed in a Random Access Memory (RAM), a memory, a Read Only Memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. To clearly illustrate this interchangeability of electronic hardware and software, various illustrative components and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as electronic hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In order to better understand the technical scheme of the invention, the following is further explained in combination with a comparative example.
Comparative example 1
A superconducting cable partial discharge mode identification method comprises the following steps:
s1: collecting different types of fault signals of the superconducting cable by using a partial discharge detection system, and using the collected signals to construct a partial discharge phase distribution pattern PRPD spectrogram data set of the superconducting cable insulation defect;
s2: classifying PRPD spectrograms in a data set according to different fault types, preprocessing images and amplifying data, and then dividing the PRPD spectrograms into a training set, a verification set and a test set according to a certain proportion;
s3: feature extraction: carrying out feature extraction on the marked partial discharge PRPD map;
s4: and training through a 3-layer BP neural network to obtain a trained network model.
In comparison with the example, in comparative example 1, on the basis of step S1 and step S2, comparative example 1 performs the "feature extraction" step, and then performs model training using a 3-layer BP network.
Comparative example 2
A superconducting cable partial discharge mode identification method comprises the following steps:
s1: collecting different types of fault signals of the superconducting cable by using a partial discharge detection system, and using the collected signals to construct a partial discharge phase distribution pattern PRPD spectrogram data set of the superconducting cable insulation defect;
s2: classifying PRPD spectrograms in a data set according to different fault types, preprocessing images and amplifying data, and then dividing the PRPD spectrograms into a training set, a verification set and a test set according to a certain proportion;
s3: feature extraction: carrying out feature extraction on the marked partial discharge PRPD map;
s4: and training through an SVM (support vector machine) to obtain a trained network model.
In comparison with the embodiment, in comparative example 2, on the basis of steps S1 and S2, the model training is performed by using an SVM support vector machine after the step of "feature extraction" is performed in comparative example 2.
Example compared with comparative example 1 and comparative example 2, the step of "feature extraction" is omitted, and the feature extraction is completed by handing over the convolutional layer, wherein the specific data of the accuracy rate is identified in the example, the comparative example 1 and the comparative example 2, and the table 1 is shown.
TABLE 1
According to the method for identifying the partial discharge mode of the superconducting cable based on the Resnet50, provided by the invention, aiming at the problem that a plurality of complex and uncertain factors exist in the identification process and the identification process is difficult to control due to the interference of ambient noise in the data acquisition process, the Resnet50 network model is adopted to carry out classification identification on the PRPD spectrogram of various types of partial discharge phase distribution modes, so that a more ideal classification identification effect can be achieved. In addition, a transfer learning method is introduced to train the model, and the pre-trained model weight is transferred to the training process of the partial discharge pattern recognition model. Because the image data set in the deep learning is huge, the convergence rate of the network model can be increased by the transfer learning, and the identification accuracy of the model can be effectively improved. The ResNet50 network introduces a residual error structure, can solve the problem of gradient disappearance caused by the over-depth of a neural network, not only enhances the capability of feature extraction, but also improves the accuracy of classification and identification, simultaneously improves the generalization capability of the model, is suitable for being used in practical engineering application, and therefore, the type and the property of the internal defect of the power superconducting cable and the defect feature can be better mastered.
Although the present invention has been described with reference to specific preferred embodiments, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention.
Claims (10)
1. A superconducting cable partial discharge mode identification method based on a residual error network Resnet50 is characterized in that the superconducting cable partial discharge mode identification method based on a residual error network Resnet50 comprises the following steps:
s1: collecting different types of fault signals of the superconducting cable by using a partial discharge detection system, and using the collected signals to construct a partial discharge phase distribution pattern PRPD spectrogram data set of the superconducting cable insulation defect;
s2: classifying PRPD spectrograms in a data set according to different fault types, preprocessing images and amplifying data, and then dividing the PRPD spectrograms into a training set, a verification set and a test set according to a certain proportion;
s3: constructing a Resnet50 residual network model by adopting a transfer learning mode, completing the configuration of each parameter of the model, taking a training sample and a verification sample as input, and training and verifying the network model;
s4: performing convolution, pooling, activation and regression operations on the network model to update network parameters, training the network model, obtaining a trained network model when the loss rate function is not reduced and the accuracy rate function is not increased, and using the obtained model for testing;
s5: and identifying the fault type of the superconducting cable by adopting an optimal Resnet50 network model, thereby constructing a Resnet 50-based superconducting cable fault mode identification system.
2. The residual network Resnet 50-based superconducting cable partial discharge pattern recognition method according to claim 1, wherein the step S2 includes the following steps:
s21: collecting PRPD spectrograms of various different partial discharge phase distribution modes of the superconducting cable through a partial discharge detection system, and classifying the PRPD spectrograms according to partial discharge types;
s22: preprocessing the image of the collected partial discharge signal;
s23: carrying out data amplification processing on the preprocessed image;
s24: and taking the processed image as a data set of the model, and dividing the data set into a training set, a verification set and a test set according to a certain proportion.
3. The residual error network Resnet 50-based superconducting cable partial discharge pattern recognition method according to claim 2, wherein in the step S21, partial discharge of the superconducting cable is mainly represented by three categories of air gap discharge, levitation discharge and normal external noise interference through a PRPD spectrogram.
4. The residual error network Resnet 50-based superconducting cable partial discharge pattern recognition method according to claim 2, wherein in the step S22, in order to ensure that the data are the same in size and format, the image needs to be preprocessed, wherein the preprocessing comprises an image sample normalization preprocessing method and an image pixel value preprocessing method.
5. The method for identifying the partial discharge mode of the superconducting cable based on the Resnet50 as claimed in claim 2, wherein in the step S23, the number of the partial discharge samples is increased by data amplification, and the data is amplified by image translation, chrominance variation, shading variation, random clipping, and the like.
6. The residual network Resnet 50-based superconducting cable partial discharge pattern recognition method according to claim 2, wherein in the step S24, 70% of images in the processed partial discharge phase distribution pattern PRPD spectrogram data set are extracted as a training set, 10% of images are extracted as a verification set, and 20% of images are extracted as a test set.
7. The residual network Resnet 50-based superconducting cable partial discharge pattern recognition method according to claim 1, wherein the step S3 includes the following steps:
s31: adopting Resnet50 model in the mode of transfer learning;
s32: configuring and adjusting various parameters of the model;
s33: and taking the training sample and the verification sample as input, and training and verifying the network model.
8. The residual network Resnet 50-based superconducting cable partial discharge pattern recognition method according to claim 1, wherein the step S4 includes the following steps:
s41: defining a loss function, and taking the cross entropy function as the loss function, wherein the calculation formula is as follows:
wherein x is a sample value which is randomly distributed, p is expected probability distribution of the sample x, and q is actual probability distribution of the sample x;
s42: defining an optimization function, and realizing self-adaptive adjustment of simulated annealing based on cosine attenuation in order to ensure that the model training can achieve the optimal speed and precision, wherein the cosine attenuation formula is as follows:
wherein alpha is0Denotes the initial learning rate, αtThe learning rate in the T iteration is represented, and T represents the total iteration number;
s43: setting parameters of model training, setting the image resolution to be 224 multiplied by 224, setting the initial learning rate to be 0.001 and setting the number of training rounds to be 100;
s44: and training the network model, obtaining the trained network model when the loss rate function is not reduced and the accuracy rate function is not increased, and using the obtained model for testing the test sample.
9. The residual error network Resnet 50-based superconducting cable partial discharge pattern recognition method according to claim 1, further comprising a Resnet 50-based superconducting cable partial discharge pattern recognition system, wherein the Resnet 50-based superconducting cable partial discharge pattern recognition system in the S5 step comprises: the device comprises a partial discharge detection unit, a discharge spectrogram construction unit, a data diversity unit, a model training unit, a model testing unit and an identification unit, wherein the partial discharge detection unit is connected with the data diversity unit through the discharge spectrogram construction unit, the data diversity unit is respectively connected with the model training unit and the model testing unit, the model training unit is connected with the model testing unit, and the partial discharge detection unit is connected with the model testing unit through the identification unit.
10. The superconducting cable partial discharge mode identification method based on the residual error network Resnet50 as claimed in claim 9, wherein the partial discharge detection unit is used for collecting several types of partial discharge that frequently occur to the superconducting cable and the partial discharge signals of these typical defects; the building discharge spectrogram unit is used for building a PRPD spectrogram data set of a partial discharge phase distribution pattern of the superconducting cable; the data diversity unit is used for dividing PRPD spectrogram in the data set into a training set, a test set and a verification set according to a certain proportion; the model training unit is used for taking a training sample and a verification sample as input and training and verifying a Resnet50 network model; the model testing unit is used for testing the trained network model on a test sample, and the individual recognition rate and the average recognition rate of various partial discharge types can be obtained according to the test result; the identification unit is used for carrying out pattern identification on the fault type of the partial discharge of the superconducting cable by utilizing an optimal Resnet50 network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111493810.1A CN114186589A (en) | 2021-12-08 | 2021-12-08 | Superconducting cable partial discharge mode identification method based on residual error network Resnet50 |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111493810.1A CN114186589A (en) | 2021-12-08 | 2021-12-08 | Superconducting cable partial discharge mode identification method based on residual error network Resnet50 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114186589A true CN114186589A (en) | 2022-03-15 |
Family
ID=80603885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111493810.1A Pending CN114186589A (en) | 2021-12-08 | 2021-12-08 | Superconducting cable partial discharge mode identification method based on residual error network Resnet50 |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114186589A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114609493A (en) * | 2022-05-09 | 2022-06-10 | 杭州兆华电子股份有限公司 | Partial discharge signal identification method with enhanced signal data |
| WO2023213332A1 (en) * | 2022-06-27 | 2023-11-09 | 上海格鲁布科技有限公司 | Separation and identification method for multi-source hybrid ultra-high frequency partial discharge diagram |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063780A (en) * | 2018-08-10 | 2018-12-21 | 国网上海市电力公司 | Partial discharge of transformer recognition methods based on particle group optimizing core neighbour's propagation algorithm |
| CN110443130A (en) * | 2019-07-01 | 2019-11-12 | 国网湖南省电力有限公司 | A kind of electric distribution network overhead wire abnormal state detection method |
| FR3088129A1 (en) * | 2018-11-07 | 2020-05-08 | Fujitsu Technology Solutions | METHOD AND SYSTEM FOR DETECTING FRAUD IN A POINT OF SALE |
| WO2020129025A1 (en) * | 2018-12-20 | 2020-06-25 | Wrnch Inc. | Method and system for detecting holding in images |
| CN112149549A (en) * | 2020-09-18 | 2020-12-29 | 国网山东省电力公司泰安供电公司 | GIS partial discharge type identification method based on depth residual error network |
| CN113449776A (en) * | 2021-06-04 | 2021-09-28 | 中南民族大学 | Chinese herbal medicine identification method and device based on deep learning and storage medium |
| US20210365342A1 (en) * | 2020-05-25 | 2021-11-25 | Wuhan University | Method and system for power equipment diagnosis based on windowed feature and hilbert visualization |
-
2021
- 2021-12-08 CN CN202111493810.1A patent/CN114186589A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063780A (en) * | 2018-08-10 | 2018-12-21 | 国网上海市电力公司 | Partial discharge of transformer recognition methods based on particle group optimizing core neighbour's propagation algorithm |
| FR3088129A1 (en) * | 2018-11-07 | 2020-05-08 | Fujitsu Technology Solutions | METHOD AND SYSTEM FOR DETECTING FRAUD IN A POINT OF SALE |
| WO2020129025A1 (en) * | 2018-12-20 | 2020-06-25 | Wrnch Inc. | Method and system for detecting holding in images |
| CN110443130A (en) * | 2019-07-01 | 2019-11-12 | 国网湖南省电力有限公司 | A kind of electric distribution network overhead wire abnormal state detection method |
| US20210365342A1 (en) * | 2020-05-25 | 2021-11-25 | Wuhan University | Method and system for power equipment diagnosis based on windowed feature and hilbert visualization |
| CN112149549A (en) * | 2020-09-18 | 2020-12-29 | 国网山东省电力公司泰安供电公司 | GIS partial discharge type identification method based on depth residual error network |
| CN113449776A (en) * | 2021-06-04 | 2021-09-28 | 中南民族大学 | Chinese herbal medicine identification method and device based on deep learning and storage medium |
Non-Patent Citations (4)
| Title |
|---|
| ZHIGUO TANG ET AL., 《2020 IEEE INTERNATIONAL CONFERENCE ON HIGH VOLTAGE ENGINEERING AND APPLICATION (ICHVE)》APPLICATION OF CONVOLUTIONAL NEURAL NETWORK TRANSFER LEARNING IN PARTIAL DISCHARGE PATTERN RECOGNITION, 15 December 2020 (2020-12-15), pages 1 - 4 * |
| 王宇龙编著: "《PyTorch深度学习入门与实战》", 30 September 2020, 中国铁道出版社, pages: 57 - 58 * |
| 金天一 等, 《电测与仪表》 基于混合放电模型的变压器内部放电联合检测方法研究, vol. 55, no. 18, 25 September 2018 (2018-09-25), pages 102 - 108 * |
| 高敬鹏: "《深度学习 卷积神经网络技术与实践》", 31 July 2020, 机械工业出版社, pages: 47 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114609493A (en) * | 2022-05-09 | 2022-06-10 | 杭州兆华电子股份有限公司 | Partial discharge signal identification method with enhanced signal data |
| CN114609493B (en) * | 2022-05-09 | 2022-08-12 | 杭州兆华电子股份有限公司 | Partial discharge signal identification method with enhanced signal data |
| WO2023213332A1 (en) * | 2022-06-27 | 2023-11-09 | 上海格鲁布科技有限公司 | Separation and identification method for multi-source hybrid ultra-high frequency partial discharge diagram |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114186589A (en) | Superconducting cable partial discharge mode identification method based on residual error network Resnet50 | |
| CN110532932B (en) | Method for identifying multi-component radar signal intra-pulse modulation mode | |
| CN111738301A (en) | Long-tail distribution image data identification method based on two-channel learning | |
| CN111006865A (en) | Motor bearing fault diagnosis method | |
| CN114092793B (en) | End-to-end biological target detection method suitable for complex underwater environment | |
| CN112818893A (en) | Lightweight open-set landmark identification method facing mobile terminal | |
| CN111222442A (en) | Electromagnetic signal classification method and device | |
| CN112347910B (en) | Signal fingerprint identification method based on multi-mode deep learning | |
| CN114387627A (en) | Small sample wireless device radio frequency fingerprint identification method and device based on depth measurement learning | |
| CN116402825A (en) | Bearing fault infrared diagnosis method, system, electronic equipment and storage medium | |
| CN115578585A (en) | Industrial image anomaly detection method, system, computer device and storage medium | |
| CN110503157B (en) | Image steganalysis method of multitask convolution neural network based on fine-grained image | |
| CN117527495A (en) | Modulation mode identification method and device for wireless communication signals | |
| CN116310644A (en) | Abnormality detection method and device based on normalized stream network and encoder | |
| CN113343796B (en) | Knowledge distillation-based radar signal modulation mode identification method | |
| CN113343924B (en) | Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network | |
| CN115034983A (en) | Image data amplification method for power transmission and transformation equipment | |
| CN115829029A (en) | Channel attention-based self-distillation implementation method | |
| CN111899161A (en) | Super-resolution reconstruction method | |
| CN113435334B (en) | Small target face recognition method based on deep learning | |
| CN111612704B (en) | Band texture noise denoising method for continuous terahertz wave image | |
| CN113222953B (en) | Natural image enhancement method based on depth gamma transformation | |
| CN114743044A (en) | Fault diagnosis method and device based on partial domain adaptation, storage medium and terminal | |
| Yang et al. | Image Recognition Based on an Improved Deep Residual Shrinkage Network | |
| CN110730280B (en) | Noise equalization method and noise removal 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 |