CN109856517A - A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data - Google Patents
A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data Download PDFInfo
- Publication number
- CN109856517A CN109856517A CN201910247454.1A CN201910247454A CN109856517A CN 109856517 A CN109856517 A CN 109856517A CN 201910247454 A CN201910247454 A CN 201910247454A CN 109856517 A CN109856517 A CN 109856517A
- Authority
- CN
- China
- Prior art keywords
- fault
- convolutional neural
- neural networks
- discharge
- frequency
- 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.)
- Granted
Links
Landscapes
- Testing Relating To Insulation (AREA)
Abstract
The invention discloses a kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data, comprising: is down to the audible continuous sound wave frequency signal of human ear for continuous ultrasound frequency signal is sampled；The continuously frame frequency of sound wave signal of one setting time length of interception；The mel-frequency cepstrum coefficient of frame frequency of sound wave signal is extracted as fault discharge feature to be identified；The fault discharge feature to be identified of extraction is sent into CNN convolutional neural networks, the fault grader of CNN convolutional neural networks output category layer is entered through the analysis of CNN convolutional neural networks；CNN convolutional neural networks identify fault discharge feature to be identified and export fault discharge type to be identified according to the fault grader to be formed is learnt to known fault discharge characteristic in advance.The present invention directly carries out pattern learning and identification to fault type with convolutional neural networks CNN, improves the accuracy rate of identification, reduces or avoid manual intervention.
Description
Technical field
The present invention relates to fault diagnosis method of discrimination more particularly to a kind of extra-high voltage equipment Partial Discharge Detections
The method of discrimination of data.
Background technique
What the normal operation as high-voltage switch gear and transformer in electric system was directly related to entire electric system can
By operation, its failure can be effectively prevented by detecting high-voltage switch gear and the shelf depreciation of transformer in time.It puts prolonged part
Electricity accumulation will cause a series of physical chemical reaction of high-tension apparatus, aggravate insulation damages, so as to cause equipment fault.It puts part
Electricity condition detection is to ensure the important means of high-tension apparatus reliability service, and partial discharges fault identification is the core of partial discharge detection
Link.
Publication number CN105203936A discloses that " a kind of power cable shelf depreciation defect type based on spectrum analysis is sentenced
Other method ", this method pass through extraction discharge defect spectrum signature and the defect type spectral feature data library established in advance
It is compared analysis and determines shelf depreciation defect type, process is to determine shelf depreciation defect class by comparison similarity
Type, since comparison similarity is determined by preset threshold, the accuracy rate of this method is uncertain, and works as similarity not
Determining shelf depreciation defect type is required manual intervention when meeting the requirements, and there are problems that excessive personnel intervene.
Summary of the invention
The purpose of the present invention is to propose to a kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data, pass through extraction office
Put the characteristic quantity of ultrasonic signal, pattern-recognition directly carried out to fault type with convolutional neural networks (CNN), rather than with pair
Shelf depreciation defect type is determined than similarity, improves the accuracy rate of identification, reduces or avoid manual intervention.
To achieve the goals above, the technical scheme is that
A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data, including continuously acquire local fault electric discharge and generate
Ultrasound frequency signal, the method for discrimination includes:
Step 1: being down to the audible continuous sound wave frequency signal of human ear for continuous ultrasound frequency signal is sampled；
Step 2: continuously intercepting the frame frequency of sound wave signal of a setting time length；
Step 3: extracting the mel-frequency cepstrum coefficient of frame frequency of sound wave signal as fault discharge feature to be identified；
Step 4: the fault discharge feature to be identified of extraction is sent into CNN convolutional neural networks, through CNN convolutional Neural net
Network analysis enters the fault grader of CNN convolutional neural networks output category layer；
Step 5: CNN convolutional neural networks are according to learning the fault grader to be formed to known fault discharge characteristic in advance,
It identifies fault discharge feature to be identified and exports fault discharge type to be identified；
Wherein: the CNN convolutional neural networks are according to learning the failure modes to be formed to known fault discharge characteristic in advance
Device is that each type sampling of known various faults electric discharge type is separated a variety of ultrasound frequency signal samples in advance, presses
It executes according to the above-mentioned first step, the sequence of second step, and in the third step falls the mel-frequency for extracting frame frequency of sound wave signal
Spectral coefficient carries out study as known fault discharge characteristic and forms known fault grader.
Scheme is further: in the frame frequency of sound wave signal of one setting time length of the continuous interception, consecutive frame sound
Frequency signal partly overlaps.
Scheme is further: the time span is 20 milliseconds to 40 milliseconds.
Scheme is further: the ultrasound frequency signal for obtaining local fault electric discharge generation is the sampling frequency by 1MHz
Rate acquisition obtains, then, each ultrasonic signal is down-sampled to the audible continuous sound wave frequency of 80kHz frequency formation human ear
Rate signal.
Scheme is further: at that time not Shi Bai when, adjustment consecutive frame frequency of sound wave signal section overlapping overlapping percentages,
And 1 is added to a pre-set identification accumulator that repeats, it then returns to third step and re-recognizes until repeating identification accumulator
Reach preset value, reaches after preset value still recognition failures repeating identification accumulator, export unidentified failure frequency of sound wave letter
Number and alarm, manual intervention determines fault discharge type, if it is new fault discharge type then by CNN convolutional neural networks
Habit forms new fault type channel, if it is existing fault discharge type, then supplements CNN convolutional neural networks and corresponds to the event
Hinder the identification feature in electric discharge type channel.
Scheme is further: CNN convolutional neural networks analysis uses L2 regularization, and to prevent over-fitting the case where goes out
It is existing.
Scheme is further: the fault grader uses Softmax classifier.
Scheme is further: in study of the CNN convolutional neural networks to known fault discharge characteristic, including one kind
The method for verifying CNN convolutional neural networks identification fault discharge feature accuracy rate, process is:
Step 1: each type sampling of known various faults electric discharge type is separated a variety of ultrasound frequency signal samples
This, it is sampled to be down to the audible continuous sound wave frequency signal of human ear；
Step 2: continuously intercepting the frame frequency of sound wave signal of a setting time length；
Step 3: extract frame frequency of sound wave signal mel-frequency cepstrum coefficient be used as known fault discharge characteristic, and general
The known fault discharge characteristic signal collection of a variety of ultrasound frequency signal samples of each fault discharge type is divided into training set
And test set；
Step 4: training set fault discharge feature is sent into CNN convolutional neural networks, analyzed through CNN convolutional neural networks
Form the known fault grader of CNN convolutional neural networks output category layer；
Step 5: test set fault discharge feature is sent into CNN convolutional neural networks, analyzed through CNN convolutional neural networks
Into the fault grader of CNN convolutional neural networks output category layer；
Step 6: the known fault grader that CNN convolutional neural networks are formed through the 4th step, identification test set failure is put
Electrical feature simultaneously exports test set fault discharge type；
Step 7: obtaining CNN compared with exporting test set fault discharge type according to known test set fault type
Convolutional neural networks identify fault discharge feature accuracy rate.
The present invention uses mel-frequency cepstrum coefficient as characteristic quantity, directly by the way that ultrasonic signal is reduced to acoustic signals
It connects and pattern learning and identification is carried out to fault type with convolutional neural networks (CNN), rather than determine office with comparison similarity
Portion's discharge defect type, improves the accuracy rate of identification, reduces or avoid manual intervention.The volume that deep learning network uses
Product neural network (CNN) is able to reflect the substantive characteristics of former data, is more advantageous to classification problem.Depth is in feature extraction, event
There is application outstanding in the directions such as barrier classification and prediction, have preferably identification to strive for rate and high efficiency.
The present invention is described in detail with reference to the accompanying drawings and examples.
Detailed description of the invention
Fig. 1 is MFCC feature extraction flow chart of the present invention.
Fig. 2 is CNN structural schematic diagram of the present invention.
Fig. 3 is the flow chart that the method accuracy rate is verified in the present invention.
Specific embodiment
Ultrasound examination is one of most important non-electro-detection technology of shelf depreciation.Part occurs in inside electric appliance to put
When electric, charge and steeper current impulse can be generated, so that gas moment is heated in the region that shelf depreciation occurs and then expands,
Fierce shock, the effect of approximation explosion occurs.After electric discharge, the original gas cooling of expanded by heating, area reduction and extensive
Again to original volume.This volume harmomegathus variation generated due to partial discharge causes the moment density variation of medium, generates pressure
Reeb and impulse form, i.e. ultrasonic wave.Ultrasonic detection method does not contact electrical equipment, can avoid electromagnetic interference, not shadow
Ring the normal operation of equipment.
The feature extraction of voice signal is a significant challenge, because it is unlike the other kinds of data such as text and image
So directly.The present embodiment is applied to down-sampled ultrasonic signal according to several different feature extracting methods and compares
Performance.These methods can extract important feature, such as mel-frequency cepstrum coefficient (MFCC) from each data, spectrogram,
Spectrum contrast and tone centroid feature.
In recent years, deep learning network has important application in every field.Wherein, convolutional neural networks (CNN) by
In its such as local sensing field, the characteristics such as parameter sharing and pond and be widely used in image recognition；Recurrent neural network (RNN)
The time serial message that sample can be remembered, in natural language processing, speech recognition, the fields such as handwriting recognition have extremely important
Application value；The disappearance of gradient when in order to overcome neural net layer to increase, sigmoid function is by ReLU, maxout and other transmitting
Function replaces, and forms the citation form of deep neural network today (DNN).The present embodiment application CNN model identifies part
Discharge mode, and recognition accuracy and performance compared with two kinds of deep learning models of DNN and RNN.
For this purpose, as a kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data of the present embodiment, including continuously obtain
The ultrasound frequency signal for taking local fault electric discharge to generate, the method for discrimination include:
Step 1: being down to the audible continuous sound wave frequency signal of human ear for continuous ultrasound frequency signal is sampled；
Step 2: continuously intercepting the frame frequency of sound wave signal of a setting time length；
Step 3: extracting the mel-frequency cepstrum coefficient of frame frequency of sound wave signal as fault discharge feature to be identified；
Step 4: the fault discharge feature to be identified of extraction is sent into CNN convolutional neural networks, through CNN convolutional Neural net
Network analysis enters the fault grader of CNN convolutional neural networks output category layer；
Step 5: CNN convolutional neural networks are according to learning the fault grader to be formed to known fault discharge characteristic in advance,
It identifies fault discharge feature to be identified and exports fault discharge type to be identified；
Wherein: the CNN convolutional neural networks are according to learning the failure modes to be formed to known fault discharge characteristic in advance
Device is that each type sampling of known various faults electric discharge type is separated a variety of ultrasound frequency signal samples in advance, presses
It executes according to the above-mentioned first step, the sequence of second step, and in the third step falls the mel-frequency for extracting frame frequency of sound wave signal
Spectral coefficient carries out study as known fault discharge characteristic and forms known fault grader.
Known various faults electric discharge type can pass through digital PD meter, oscillograph and Portable partial discharge detector
Obtain such as tip corona defect, floating potential defect, grounding electrode creeping discharge, high voltage plane discharge, internal discharge and close
The data sample of the insulation fault type of ground electrode electric discharge.Certainly, data can also be acquired by other means, however it is not limited to
It is acquired and is obtained by partial discharge simulation experiment, and insulation fault type is also not limited to tip corona defect, suspend electricity
Position defect, grounding electrode creeping discharge, high voltage plane discharge, internal discharge and near-earth electrode discharge defect, technology in the art
Personnel can be configured according to the concrete condition of embodiment, therefore, here, repeating no more.
Wherein: in the frame frequency of sound wave signal of one setting time length of the continuous interception, consecutive frame frequency of sound wave letter
It number partly overlaps；The time span therein is 20 milliseconds to 40 milliseconds.
A preferred embodiment in embodiment this be: it is described to obtain the ultrasound frequency signal that local fault electric discharge generates and be
It is obtained by the sample frequency acquisition of 1MHz, it is then, each ultrasonic signal is down-sampled audible to 80kHz frequency formation human ear
The continuous sound wave frequency signal arrived, is certainly not limited to this.
In embodiment: at that time not Shi Bai when, the overlapping percentages of adjustment consecutive frame frequency of sound wave signal section overlapping are and right
One pre-set identification accumulator that repeats adds 1, then returns to third step and re-recognizes until repetition identification accumulator reaches
Preset value reaches after preset value still recognition failures repeating identification accumulator, exports unidentified knocking noise frequency signal simultaneously
Alarm, manual intervention determine fault discharge type, then learn shape by CNN convolutional neural networks if it is new fault discharge type
The fault type channel of Cheng Xin then supplements CNN convolutional neural networks and corresponds to the failure and put if it is existing fault discharge type
The identification feature of electric channel type.
In embodiment: in study of the CNN convolutional neural networks to known fault discharge characteristic, including a kind of verifying
The method that CNN convolutional neural networks identify fault discharge feature accuracy rate, process is:
Step 1: each type sampling of known various faults electric discharge type is separated a variety of ultrasound frequency signal samples
This, it is sampled to be down to the audible continuous sound wave frequency signal of human ear；
Step 2: continuously intercepting the frame frequency of sound wave signal of a setting time length；
Step 3: extract frame frequency of sound wave signal mel-frequency cepstrum coefficient be used as known fault discharge characteristic, and general
The known fault discharge characteristic signal collection of a variety of ultrasound frequency signal samples of each fault discharge type is divided into training set
And test set；
Step 4: training set fault discharge feature is sent into CNN convolutional neural networks, analyzed through CNN convolutional neural networks
Form the known fault grader of CNN convolutional neural networks output category layer；
Step 5: test set fault discharge feature is sent into CNN convolutional neural networks, analyzed through CNN convolutional neural networks
Into the fault grader of CNN convolutional neural networks output category layer；
Step 6: the known fault grader that CNN convolutional neural networks are formed through the 4th step, identification test set failure is put
Electrical feature simultaneously exports test set fault discharge type；
Step 7: obtaining CNN compared with exporting test set fault discharge type according to known test set fault type
Convolutional neural networks identify fault discharge feature accuracy rate.
Above-described embodiment deep learning network uses convolutional neural networks CNN.Deep learning model is available more
Representational characteristic is able to reflect the substantive characteristics of former data, is more advantageous to classification problem.Depth is mentioned in feature
Take, there is application outstanding in the directions such as failure modes and prediction, there is preferably identification to strive for rate and high efficiency.
Convolutional neural networks CNN is a kind of well known technology, and structure is as shown in Fig. 2, include input layer, convolutional layer, Chi Hua
Layer, full articulamentum and output layer；CNN usually has several convolution sum pond layer alternate applications, and a convolutional layer is followed by one
Pond layer, and a pond layer is followed by a convolutional layer, and so on.Convolutional layer is for extracting data characteristics, pond layer
Carried out for the output to convolutional layer down-sampled --- compression dimensionality reduction.In CNN, convolution pondization is usually made of three steps.
In the first step, several convolution are executed parallel, it is linearly movable to generate one group.In second step, each linear activation is executed
Nonlinear activation function.In the third step, the output of figure layer is modified using pond function.Output category is used as in the present embodiment
The fault grader of layer uses Softmax classifier.
CNN convolutional neural networks analysis processing in using stochastic gradient descent method to the sparse noise reduction self-encoding encoder of depth into
Row training, to be iterated update to its parameter, obtains the most optimized parameter.
CNN convolutional neural networks analysis uses L2 regularization, the case where to prevent over-fitting.
By mel-frequency cepstrum coefficient (Mel Frequencies Cepstral Coefficient, MFCC) in embodiment
Input as deep learning network, in which: the calculating of MFCC characteristic quantity includes following five steps as shown in Figure 1:
Step 1: short frame (Framing)；
The processing of voice signal is completed in the short time interval of referred to as frame, size usually 20 to 40 milliseconds it
Between.In addition, the data between two consecutive frames change too much since frame divides in order to prevent, need to be overlapped one of consecutive frame
Point.In the present embodiment, voice signal is turned to 20-40 milliseconds of short frame by frame, and the 50% of every frame frame adjacent thereto is overlapped.
Step 2:FFT and the cyclic graph estimation for calculating every frame power spectrum；
The Fast Fourier Transform (FFT) (FFT) that frame is calculated with formula (1), uses Si(k) it indicates, wherein h (n) is the sample n long Chinese
Bright window, N are the sample sizes in every frame, and K is the length of FFT.The power Spectral Estimation P of frame based on cyclic graphi(k) by formula
(2) it provides:
Step 3:Mel filtering；
Mel filter group is one group of 20-40 triangular filter, applied to the cyclic graph power Spectral Estimation from step 2,
According to formula (3), voice signal is converted into Mel frequency spectrum using Mel filter group:
Step 4: taking logarithm (Logarithm)；
The logarithm S'(k of all filter bank energies from step 3 is calculated according to formula (4)).The mesh of the logarithm step
Be compress shelf depreciation voice signal frequency spectrum dynamic range:
Step 5: carrying out discrete cosine transform (DCT).In order to obtain final MFCC feature relevant parameter, to pair of step 4
Number energy is handled using discrete cosine transform (DCT).The major reason for selecting DCT herein is because DCT is in processing frequency spectrum
Have unique advantage when component: the spectrum component difference of different frequency is more significant, the correlation and contiguity between ingredient compared with
It is weak.DCT expression formula is as follows:
Wherein C (n) is final MFCC feature, and n=1,2 ..., L, L are the orders of MFCC.
Fig. 3 is the flow chart for verifying the method accuracy rate, in which:
Step 100: obtaining the ultrasonic signal sample of the shelf depreciation of the characterization several insulation fault type of extra-high voltage equipment
This, carries out down-sampled pretreatment to ultrasonic signal, the frequency that human ear can be heard is dropped to, so that human ear direct feeling is locally put
Electricity and later period extract data characteristics amount；
Step 200: the MFCC characteristic quantity (extracting method of the partial discharge voice signal after extracting noise reduction by spectrum analysis technique
It is such as aforementioned)；
Step 300: data characteristics vector set being divided into training set and test set, utilizes feature vector training set training depth
Learning model CNN enables model to learn the characteristic information of different electric discharge types, obtains network optimized parameter.
Step 400: every group of test set shelf depreciation ultrasound signal signatures amount is inputted into trained deep learning respectively
MODEL C NN, output category result is to obtain the defect type of extra-high voltage equipment.
The present embodiment is acquired ultrasonic signal using the sample rate of 1MHz and saves mass data for every kind of partial discharge type
Sample, in total 3960 data samples.The composition of data is as shown in table 1.Each primary data sample continues 0.5 second and includes
500,000 values.Each ultrasonic signal is down-sampled to 80kHz frequency, this also means that the time of ultrasonic signal extends
12.5 (1M/80k) times --- it was extended to from 0.5 second 6.25 seconds, such human ear can directly hear partial discharge sound.
Table 1
In step 300, the characteristic data set extracted is divided into training set and test set, such as 3960 groups of phases in total
Pulse train data are differentiated in position, and 3564 groups of data (90% data) therein are divided into training set, remaining 396 groups of numbers at random
Deep learning network RNN, DNN and CNN are trained respectively using training set according to being then test set when training, calculate sample
The output of data calculates output and the error of sample label, using stochastic gradient descent method to the parameter of deep learning network into
Row iteration updates, and obtains the most optimized parameter.Finally verify whether three kinds of deep learning networks train completion by test set.Its
In, CNN uses level 2 volume lamination and pond layer, uses 3 layers of hidden layer as the DNN model compared.Three kinds of deep learning networks
Input layer is configured as MFCC characteristic quantity, and activation primitive uses Sigmod function, and output category layer uses Softmax classifier.
In the training process, occurs the case where over-fitting in order to prevent, using L2 regularization.
In step 400, the spy of the local discharge signal of extra-high voltage equipment to be identified is determined using MFCC as input
Vector is levied, carries out pattern-recognition using trained deep learning network.
In order to verify this case extra-high voltage equipment Partial Discharge Data method of discrimination recognition effect, will using this case three
Kind deep learning model carries out identification comparison, and comparing result is listed in table 2.
Table 2
As can be seen from Table 2, using the differentiation of the extra-high voltage equipment shelf depreciation diagnostic data based on deep learning network
Method has good discrimination and more preferably recognition performance, is highly suitable in actual application to high voltage equipment insulation
Failure is identified.Wherein, CNN recognition correct rate proposed by the present invention be apparently higher than as comparison algorithm RNN and DNN just
True rate.
In conclusion the method for discrimination of extra-high voltage equipment shelf depreciation diagnostic data can be to part described in the present embodiment
The ultrasonic signal that discharges carries out effective fault identification, so as to timely and effectively obtain the insulation fault feelings of extra-high voltage equipment
Condition is removed a hidden danger in time, avoids the generation of major accident, has directive significance for the security maintenance of extra-high voltage equipment.This reality
The discrimination for applying the method for discrimination of example is higher, and recognition performance is more excellent.
It should be noted that prior art part is not limited to given by present specification in protection scope of the present invention
Embodiment, all prior arts not contradicted with the solution of the present invention, including but not limited to first patent document, formerly
Public publication, formerly openly use etc., it can all be included in protection scope of the present invention.
In addition, it should also be noted that, institute in the combination of each technical characteristic and unlimited this case claim in this case
Combination documented by the combination or specific embodiment of record, all technical characteristics documented by this case can be to appoint
Where formula is freely combined or is combined, unless generating contradiction between each other.
The above embodiment is a preferred embodiment of the present invention, but embodiments of the present invention are not by above-described embodiment
Limitation, other any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention,
It should be equivalent substitute mode, be included within the scope of the present invention.
Claims (8)
1. a kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data, including continuously acquiring local fault electric discharge generation
Ultrasound frequency signal, which is characterized in that the method for discrimination includes:
Step 1: being down to the audible continuous sound wave frequency signal of human ear for continuous ultrasound frequency signal is sampled；
Step 2: continuously intercepting the frame frequency of sound wave signal of a setting time length；
Step 3: extracting the mel-frequency cepstrum coefficient of frame frequency of sound wave signal as fault discharge feature to be identified；
Step 4: the fault discharge feature to be identified of extraction is sent into CNN convolutional neural networks, through CNN convolutional neural networks point
Analysis enters the fault grader of CNN convolutional neural networks output category layer；
Step 5: CNN convolutional neural networks are identified according to the fault grader to be formed is learnt to known fault discharge characteristic in advance
Fault discharge feature to be identified simultaneously exports fault discharge type to be identified；
Wherein:
The CNN convolutional neural networks are prior according to the fault grader to be formed is learnt to known fault discharge characteristic in advance
The each type sampling of known various faults electric discharge type is separated into a variety of ultrasound frequency signal samples, according to above-mentioned first
Step, second step sequence execute, and in the third step using extract frame frequency of sound wave signal mel-frequency cepstrum coefficient as
Known fault discharge characteristic carries out study and forms known fault grader.
2. the method according to claim 1, wherein the frame sound wave of one setting time length of the continuous interception
In frequency signal, the overlapping of consecutive frame frequency of sound wave signal section.
3. method according to claim 1 or 2, which is characterized in that the time span is 20 milliseconds to 40 milliseconds.
4. the method according to claim 1, wherein the ultrasonic wave frequency for obtaining local fault electric discharge and generating
Rate signal is obtained by the sample frequency acquisition of 1MHz, then, is formed each ultrasonic signal is down-sampled to 80kHz frequency
The audible continuous sound wave frequency signal of human ear.
5. according to the method described in claim 2, it is characterized in that, at that time not Shi Bai when, adjust consecutive frame frequency of sound wave signal
Partly overlapping overlapping percentages, and 1 is added to a pre-set identification accumulator that repeats, it then returns to third step and knows again
Not until repeating identification accumulator reaches preset value, reach after preset value still recognition failures, output repeating identification accumulator
Unidentified knocking noise frequency signal is simultaneously alarmed, and manual intervention determines fault discharge type, if it is new fault discharge type
Then learnt to form new fault type channel by CNN convolutional neural networks, if it is existing fault discharge type, then be supplemented
CNN convolutional neural networks correspond to the identification feature of the fault discharge channel type.
6. the method according to claim 1, wherein the CNN convolutional neural networks analysis use L2 regularization,
The appearance of the case where to prevent over-fitting.
7. the method according to claim 1, wherein the fault grader uses Softmax classifier.
8. the method according to claim 1, wherein discharging in the CNN convolutional neural networks known fault
In the study of feature, a kind of method including verifying CNN convolutional neural networks identification fault discharge feature accuracy rate, process
It is:
Step 1: each type sampling of known various faults electric discharge type is separated into a variety of ultrasound frequency signal samples,
It is sampled to be down to the audible continuous sound wave frequency signal of human ear；
Step 2: continuously intercepting the frame frequency of sound wave signal of a setting time length；
Step 3: extract the mel-frequency cepstrum coefficient of frame frequency of sound wave signal as known fault discharge characteristic, and it will be each
The known fault discharge characteristic signal collection of a variety of ultrasound frequency signal samples of kind fault discharge type is divided into training set and survey
Examination collection；
Step 4: training set fault discharge feature is sent into CNN convolutional neural networks, analyze to be formed through CNN convolutional neural networks
The known fault grader of CNN convolutional neural networks output category layer；
Step 5: test set fault discharge feature is sent into CNN convolutional neural networks, analyzes and enter through CNN convolutional neural networks
The fault grader of CNN convolutional neural networks output category layer；
Step 6: the known fault grader that CNN convolutional neural networks are formed through the 4th step, identification test set fault discharge is special
It levies and exports test set fault discharge type；
Step 7: obtaining CNN convolution compared with exporting test set fault discharge type according to known test set fault type
Neural network recognization fault discharge feature accuracy rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910247454.1A CN109856517B (en) | 2019-03-29 | 2019-03-29 | Method for distinguishing partial discharge detection data of extra-high voltage equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910247454.1A CN109856517B (en) | 2019-03-29 | 2019-03-29 | Method for distinguishing partial discharge detection data of extra-high voltage equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109856517A true CN109856517A (en) | 2019-06-07 |
| CN109856517B CN109856517B (en) | 2021-02-23 |
Family
ID=66902373
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910247454.1A Active CN109856517B (en) | 2019-03-29 | 2019-03-29 | Method for distinguishing partial discharge detection data of extra-high voltage equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109856517B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110161389A (en) * | 2019-06-12 | 2019-08-23 | 国网河北省电力有限公司电力科学研究院 | A kind of Electric Power Equipment Insulation defect identification method and AEVB self-encoding encoder |
| CN110161388A (en) * | 2019-06-10 | 2019-08-23 | 上海交通大学 | A kind of the fault type recognition method and its system of high-tension apparatus |
| CN110244205A (en) * | 2019-07-12 | 2019-09-17 | 西南石油大学 | Cable accessory Recognition of Partial Discharge based on convolutional neural networks under a kind of small data quantity |
| CN110334948A (en) * | 2019-07-05 | 2019-10-15 | 上海交通大学 | Power equipment shelf depreciation Severity method and system based on characteristic quantity prediction |
| CN111524523A (en) * | 2020-04-26 | 2020-08-11 | 中南民族大学 | Instrument and equipment state detection system and method based on voiceprint recognition technology |
| CN111932493A (en) * | 2020-06-28 | 2020-11-13 | 北京国网富达科技发展有限责任公司 | Power distribution network partial discharge ultrasonic detection method and system |
| CN111929542A (en) * | 2020-07-03 | 2020-11-13 | 北京国网富达科技发展有限责任公司 | Power equipment diagnosis method and system |
| CN112034312A (en) * | 2020-08-07 | 2020-12-04 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Power equipment insulation defect mode identification method |
| CN112051493A (en) * | 2020-09-17 | 2020-12-08 | 海南电网有限责任公司琼海供电局 | Hidden danger type identification method and device for power distribution network equipment |
| CN112581940A (en) * | 2020-09-17 | 2021-03-30 | 国网江苏省电力有限公司信息通信分公司 | Discharging sound detection method based on edge calculation and neural network |
| CN112784983A (en) * | 2021-01-28 | 2021-05-11 | 邱戴飞 | Identity information prediction model training method and device based on deep neural network |
| CN114019310A (en) * | 2021-11-08 | 2022-02-08 | 国网湖南省电力有限公司 | Method, equipment and medium for prejudging open circuit fault of energy storage converter |
| CN114113943A (en) * | 2021-11-25 | 2022-03-01 | 广东电网有限责任公司广州供电局 | Transformer partial discharge detection system, method and equipment based on current and ultrasonic signals |
| CN118130984A (en) * | 2024-05-10 | 2024-06-04 | 山东博通节能科技有限公司 | Cable partial discharge fault real-time monitoring method based on data driving |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106546892A (en) * | 2016-11-10 | 2017-03-29 | 华乘电气科技（上海）股份有限公司 | The recognition methodss of shelf depreciation ultrasonic audio and system based on deep learning |
| KR101822829B1 (en) * | 2017-08-11 | 2018-01-29 | 문경훈 | Method of identifying partial discharge and noise of switchgear using machine learning |
| CN108231067A (en) * | 2018-01-13 | 2018-06-29 | 福州大学 | Sound scenery recognition methods based on convolutional neural networks and random forest classification |
| CN108573225A (en) * | 2018-03-30 | 2018-09-25 | 国网天津市电力公司电力科学研究院 | A kind of local discharge signal mode identification method and system |
| CN108896878A (en) * | 2018-05-10 | 2018-11-27 | 国家电网公司 | A kind of detection method for local discharge based on ultrasound |
| CN109324270A (en) * | 2018-08-07 | 2019-02-12 | 国网山东省电力公司淄博供电公司 | A kind of high-tension switch cabinet intelligent online monitoring system |
| CN109444682A (en) * | 2018-11-02 | 2019-03-08 | 国网四川省电力公司广安供电公司 | The construction method of partial discharge of switchgear diagnostic system based on multi-information fusion |
-
2019
- 2019-03-29 CN CN201910247454.1A patent/CN109856517B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106546892A (en) * | 2016-11-10 | 2017-03-29 | 华乘电气科技（上海）股份有限公司 | The recognition methodss of shelf depreciation ultrasonic audio and system based on deep learning |
| KR101822829B1 (en) * | 2017-08-11 | 2018-01-29 | 문경훈 | Method of identifying partial discharge and noise of switchgear using machine learning |
| CN108231067A (en) * | 2018-01-13 | 2018-06-29 | 福州大学 | Sound scenery recognition methods based on convolutional neural networks and random forest classification |
| CN108573225A (en) * | 2018-03-30 | 2018-09-25 | 国网天津市电力公司电力科学研究院 | A kind of local discharge signal mode identification method and system |
| CN108896878A (en) * | 2018-05-10 | 2018-11-27 | 国家电网公司 | A kind of detection method for local discharge based on ultrasound |
| CN109324270A (en) * | 2018-08-07 | 2019-02-12 | 国网山东省电力公司淄博供电公司 | A kind of high-tension switch cabinet intelligent online monitoring system |
| CN109444682A (en) * | 2018-11-02 | 2019-03-08 | 国网四川省电力公司广安供电公司 | The construction method of partial discharge of switchgear diagnostic system based on multi-information fusion |
Non-Patent Citations (2)
| Title |
|---|
| MENG-HUIWANG等: "《Application of extension neural network algorithm and chaos synchronization detection method to partial discharge diagnosis of power capacitor》", 《MEASUREMENT》 * |
| 刘兵等: "《基于卷积神经网络的变压器局部放电模式识别》", 《高压电器》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110161388B (en) * | 2019-06-10 | 2021-04-06 | 上海交通大学 | Fault type identification method and system of high-voltage equipment |
| CN110161388A (en) * | 2019-06-10 | 2019-08-23 | 上海交通大学 | A kind of the fault type recognition method and its system of high-tension apparatus |
| CN110161389A (en) * | 2019-06-12 | 2019-08-23 | 国网河北省电力有限公司电力科学研究院 | A kind of Electric Power Equipment Insulation defect identification method and AEVB self-encoding encoder |
| CN110334948B (en) * | 2019-07-05 | 2023-04-07 | 上海交通大学 | Power equipment partial discharge severity evaluation method and system based on characteristic quantity prediction |
| CN110334948A (en) * | 2019-07-05 | 2019-10-15 | 上海交通大学 | Power equipment shelf depreciation Severity method and system based on characteristic quantity prediction |
| CN110244205A (en) * | 2019-07-12 | 2019-09-17 | 西南石油大学 | Cable accessory Recognition of Partial Discharge based on convolutional neural networks under a kind of small data quantity |
| CN111524523A (en) * | 2020-04-26 | 2020-08-11 | 中南民族大学 | Instrument and equipment state detection system and method based on voiceprint recognition technology |
| CN111932493A (en) * | 2020-06-28 | 2020-11-13 | 北京国网富达科技发展有限责任公司 | Power distribution network partial discharge ultrasonic detection method and system |
| CN111932493B (en) * | 2020-06-28 | 2024-06-07 | 北京国网富达科技发展有限责任公司 | Power distribution network partial discharge ultrasonic detection method and system |
| WO2022001289A1 (en) * | 2020-06-28 | 2022-01-06 | 北京国网富达科技发展有限责任公司 | Power distribution network partial discharge ultrasonic test method and system |
| CN111929542B (en) * | 2020-07-03 | 2023-05-26 | 北京国网富达科技发展有限责任公司 | Power equipment diagnosis method and system |
| CN111929542A (en) * | 2020-07-03 | 2020-11-13 | 北京国网富达科技发展有限责任公司 | Power equipment diagnosis method and system |
| CN112034312A (en) * | 2020-08-07 | 2020-12-04 | 中国南方电网有限责任公司超高压输电公司检修试验中心 | Power equipment insulation defect mode identification method |
| CN112581940A (en) * | 2020-09-17 | 2021-03-30 | 国网江苏省电力有限公司信息通信分公司 | Discharging sound detection method based on edge calculation and neural network |
| CN112051493A (en) * | 2020-09-17 | 2020-12-08 | 海南电网有限责任公司琼海供电局 | Hidden danger type identification method and device for power distribution network equipment |
| CN112784983A (en) * | 2021-01-28 | 2021-05-11 | 邱戴飞 | Identity information prediction model training method and device based on deep neural network |
| CN114019310A (en) * | 2021-11-08 | 2022-02-08 | 国网湖南省电力有限公司 | Method, equipment and medium for prejudging open circuit fault of energy storage converter |
| CN114019310B (en) * | 2021-11-08 | 2024-02-20 | 国网湖南省电力有限公司 | Open-circuit fault pre-judging method, equipment and medium for energy storage converter |
| CN114113943A (en) * | 2021-11-25 | 2022-03-01 | 广东电网有限责任公司广州供电局 | Transformer partial discharge detection system, method and equipment based on current and ultrasonic signals |
| CN118130984A (en) * | 2024-05-10 | 2024-06-04 | 山东博通节能科技有限公司 | Cable partial discharge fault real-time monitoring method based on data driving |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109856517B (en) | 2021-02-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109856517A (en) | A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data | |
| EP3699614B1 (en) | Method and system of partial discharge recognition for diagnosing electrical networks | |
| CN106405339A (en) | Power transmission line fault reason identification method based on high and low frequency wavelet feature association | |
| Mazroua et al. | PD pattern recognition with neural networks using the multilayer perceptron technique | |
| CN109901031A (en) | Signal De-noising Method, information data processing terminal for local discharge signal | |
| CN106680576B (en) | Power distribution network internal overvoltage recognition methods based on piecemeal time-frequency spectrum and deep learning algorithm | |
| CN110534118A (en) | Transformer/reactor method for diagnosing faults based on Application on Voiceprint Recognition and neural network | |
| CN109343046A (en) | Radar gait recognition method based on multifrequency multiple domain deep learning | |
| CN106782505A (en) | A kind of method based on electric discharge voice recognition high-tension switch cabinet state | |
| CN102982351A (en) | Porcelain insulator vibrational acoustics test data sorting technique based on back propagation (BP) neural network | |
| CN108761287A (en) | Transformer partial discharge type identification method | |
| Ambikairajah et al. | Spectral features for the classification of partial discharge signals from selected insulation defect models | |
| CN106326918A (en) | Multiscale energy feature linear recognition method for partial discharge ultrahigh frequency signal of transformer | |
| CN114325256A (en) | Power equipment partial discharge identification method, system, equipment and storage medium | |
| Banno et al. | Partial discharge source classification for switchgears with transient earth voltage sensor using convolutional neural network | |
| CN109116196A (en) | A kind of power cable fault discharging sound intelligent identification Method | |
| CN117419915A (en) | Motor fault diagnosis method for multi-source information fusion | |
| Yang et al. | Detection of Glottal Closure Instants from Speech Signals: A Convolutional Neural Network Based Method. | |
| Wang et al. | Partial discharge detection with convolutional neural networks | |
| Mitiche et al. | Deep residual neural network for EMI event classification using bispectrum representations | |
| Illias et al. | Classification of multiple partial discharge sources in dielectric insulation material using Cepstrum analysis–artificial neural network | |
| CN112581940A (en) | Discharging sound detection method based on edge calculation and neural network | |
| CN105606967A (en) | DC single wire corona discharge audible noise analysis method | |
| Das et al. | A Novel Deep-Learning Framework to Identify and Locate Single and Multiple Partial Discharge Events | |
| Sukma et al. | Determination of type of partial discharge in cubicle-type gas insulated switchgear (C-GIS) using artificial neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |