CN106782505A - A kind of method based on electric discharge voice recognition high-tension switch cabinet state - Google Patents
A kind of method based on electric discharge voice recognition high-tension switch cabinet state Download PDFInfo
- Publication number
- CN106782505A CN106782505A CN201710092394.1A CN201710092394A CN106782505A CN 106782505 A CN106782505 A CN 106782505A CN 201710092394 A CN201710092394 A CN 201710092394A CN 106782505 A CN106782505 A CN 106782505A
- Authority
- CN
- China
- Prior art keywords
- state
- switch cabinet
- tension switch
- electric discharge
- sample
- 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 39
- 238000012549 training Methods 0.000 claims abstract description 43
- 239000000203 mixture Substances 0.000 claims abstract description 31
- 238000000605 extraction Methods 0.000 claims abstract description 12
- 230000008569 process Effects 0.000 claims abstract description 12
- 238000012544 monitoring process Methods 0.000 claims abstract description 11
- 238000013461 design Methods 0.000 claims abstract description 9
- 238000004364 calculation method Methods 0.000 claims abstract description 6
- 239000000284 extract Substances 0.000 claims abstract description 5
- 239000013598 vector Substances 0.000 claims description 11
- 238000004458 analytical method Methods 0.000 claims description 10
- 238000001228 spectrum Methods 0.000 claims description 8
- 238000009432 framing Methods 0.000 claims description 7
- 239000011159 matrix material Substances 0.000 claims description 7
- 238000005315 distribution function Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000001914 filtration Methods 0.000 claims description 4
- 230000004044 response Effects 0.000 abstract description 4
- 238000009413 insulation Methods 0.000 description 15
- 230000015556 catabolic process Effects 0.000 description 8
- 238000001514 detection method Methods 0.000 description 8
- 230000006870 function Effects 0.000 description 8
- 238000012360 testing method Methods 0.000 description 5
- 238000003745 diagnosis Methods 0.000 description 4
- 230000008030 elimination Effects 0.000 description 3
- 238000003379 elimination reaction Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000007257 malfunction Effects 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 230000005856 abnormality Effects 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 230000005684 electric field Effects 0.000 description 2
- 239000012212 insulator Substances 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000000803 paradoxical effect Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 101100290837 Bacillus subtilis (strain 168) metAA gene Proteins 0.000 description 1
- 108010089143 GMM2 Proteins 0.000 description 1
- 241000180177 Listrocephalos corona Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004888 barrier function Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000005474 detonation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 101150003180 metB gene Proteins 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 230000009131 signaling function Effects 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/02—Feature extraction for speech recognition; Selection of recognition unit
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
- G10L15/063—Training
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/08—Speech classification or search
- G10L15/14—Speech classification or search using statistical models, e.g. Hidden Markov Models [HMMs]
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/20—Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/45—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of analysis window
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Computational Linguistics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Probability & Statistics with Applications (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Testing Relating To Insulation (AREA)
Abstract
The invention discloses the method based on electric discharge voice recognition high-tension switch cabinet state:The electric discharge voice signal of step 1, respectively collection high-tension switch cabinet corona state and electrion state;Step 2, respectively to electric discharge voice signal pre-process, obtain corresponding training sample;Step 3, extraction training sample short-time energy relevant feature parameters;Step 4, extraction training sample MFCC parameters;Step 5, design grader;Step 6, parameter training is carried out to Gaussian mixture model;Step 7, monitoring high-tension switch cabinet, the electric discharge sound to monitoring pre-process and obtain sample to be identified, extract the short-time energy relevant feature parameters and MFCC parameters of sample to be identified, when short-time energy exceedes threshold value, then into step 8;The probable value of step 8, calculating sample to be identified under Gaussian mixture model;Step 9, high-tension switch cabinet state, including normal state, corona state and electrion state are judged according to result of calculation.Reliability is higher, response speed faster, it is more intelligent.
Description
Technical field
The present invention relates to a kind of method based on electric discharge voice recognition high-tension switch cabinet state.
Background technology
Into 21 century, the leading position of electricity market has gradually been transferred to buyer's market, user from seller's market
To the desired value more and more higher of the safety and reliability of electric power system.High-tension switch cabinet be capital equipment in power distribution network it
One, it is widely used.But the insulation annex in high-tension switch cabinet can under prolonged operation and other abnormal conditions
Can cause the generation of insulation fault, heating, detonation, damage be ultimately resulted in, so as to cause a series of security incident and economy
Loss, and cause the missing of users to trust.
In existing design, in order to ensure the safe and stable operation of equipment, main behave be exactly periodically or non-periodically to
The high-tension switch cabinet of line operation carries out repair based on condition of component, the content of maintenance including high-tension switch cabinet etc. the temperature of high-tension apparatus, humidity,
The state parameters such as voltage, electric current, are more to go observation by staff for insulating properties detection.This behave often consumes
Take substantial amounts of human and material resources, financial resources, and malfunction elimination is not in time, it is impossible to potential potential safety hazard is eliminated, often causes great
Security incident.
The content of the invention
Regarding to the issue above, the present invention provides a kind of method based on electric discharge voice recognition high-tension switch cabinet state, utilizes
Electric discharge sound during high-tension switch cabinet insulation breakdown, carries out real-time fault detection, it is to avoid Traditional Man detection method spends big, event
Barrier investigation not in time, be unable to the shortcoming of real-time monitoring, reliability is higher, response speed faster, realize real-time detection, in real time report
Alert, intelligence degree is higher.
Explanation of nouns:
1st, corona state:Corona discharge is local self-maintained discharge of the gas medium in non-uniform electric field.During generation corona
Surrounding them can see light, and with hiss.Corona discharge can be metastable discharge type, or not
Early stage of development during uniform electric field gap breakdown.I.e. corona state is high-tension switch cabinet insulation breakdown early stage state.
2nd, electrion state:Electrion refers to the pressure difference electric discharge of electric energy.Electrion state is that high-tension switch cabinet insulation is broken
Bad severe conditions, often with explosive sound.
3rd, short-time energy:Short-time energy is a function for measurement audio frequency signal amplitude value changes, is commonly used to characterize audio
The energy size of signal.
4、MFCC:Mel frequency cepstral coefficients.
5、GMM:Gaussian Mixture Model gauss hybrid models, or mixed Gauss model.
To realize above-mentioned technical purpose, above-mentioned technique effect is reached, the present invention is achieved through the following technical solutions:
A kind of method based on electric discharge voice recognition high-tension switch cabinet state, comprises the following steps:
The electric discharge voice signal of step 1, respectively collection high-tension switch cabinet corona state and electrion state;
Step 2, the electric discharge voice signal respectively to corona state and electrion state are pre-processed, and obtain corona state and height
Press the corresponding training sample of electric state;
Step 3, extraction training sample short-time energy relevant feature parameters;
Step 4, extraction training sample MFCC parameters;
Step 5, design grader:Model training is carried out to training sample using Gaussian mixture model, sample space is divided;
Step 6, parameter training is carried out to Gaussian mixture model;
Step 7, monitoring high-tension switch cabinet, the electric discharge sound to monitoring pre-process and obtain sample to be identified, extracts
The short-time energy relevant feature parameters and MFCC parameters of sample to be identified, when short-time energy exceedes threshold value, then into step 8;
The probable value of step 8, calculating sample to be identified under Gaussian mixture model;
Step 9, high-tension switch cabinet state, including normal state, corona state and electrion state are judged according to result of calculation.
It is preferred that, if the electric discharge voice signal of corona state or electrion state is analog signal x (t) t ∈ [0, L], L is simulation
Signal duration, unit is the second, then step 2 specifically includes following steps:
201st, to analog signal framing:
Analog signal x (t) t ∈ [0, L] is divided into M sections, every section is an analysis frame;
202nd, analog signal is sampled:
With fsThe sample rate of Hz is sampled to analog signal x (t) t ∈ [0, L], obtains x (n), n=0,1,2 ..., N,
N is the points of sampling a later frame voice signal;
203rd, bandpass filtering is carried out to sampled signal;
204th, decentralization, obtains training sample for c (n), n=0,1,2 ..., N.
It is preferred that, step 3 specifically includes following steps:
301st, windowing process is carried out to c (n):
Adding window is carried out to signal c (n) using Hamming window, Hamming window formula is as follows:
Wherein, RZZ () is rectangular window,Z is frame length, and w (n) is Hamming window formula;
302nd, short-time energy is calculated:
Wherein, EkIt is the short-time energy of kth frame voice signal, k ∈ [1, M], Z is frame length, ck(n) (k=1,2,
... M) it is that, by pretreated kth frame signal, M is totalframes;
303rd, short-time average energy is calculated
304th, short-time energy shake E is calculateds:
It is preferred that, step 4 specifically includes following steps:
401st, Fast Fourier Transform (FFT) is carried out:
Wherein, S (n) is kth frame signal ckThe discrete power spectrum of (n),
402nd, calculate S (n) and pass through X wave filter HmPerformance number after (n):
Pm(m=0,1 ..., X-1) (formula 8)
Wherein, PmIt is kth frame signal ckThe performance number of (n), HmN () is the system function of bandpass filter, Hm(n), m=
0,1,.....,X-1;N=0,1 ... .., N/2-1;
403rd, logarithm L is soughtm:
Lm=lnPm(m=0,1 ..., X-1) (formula 9)
Wherein, LmIt is to PmSeek the value of natural logrithm;
404th, to LmDiscrete cosine transform is carried out, D is obtainedm(m=0,1 ... .X-1), omit flip-flop D0, take
D1,D2,D3,......,DhUsed as MFCC cepstrum, h is constant;
405th, the first-order difference coefficient of Mel frequency cepstral coefficients is sought:
Wherein, d (k) is the first-order difference Mel frequency cepstral coefficients of kth frame signal, and D (k+i) is the Mel of (k+i) frame
Frequency cepstral coefficient, h is constant.
It is preferred that, step 5 specifically includes following steps:
501st, the probability density function of m rank Gaussian mixture models is calculated:
Wherein, P (G/ λ) is the probability density function of m rank Gaussian mixture models, and G is D dimension random vectors, wi, i=
1 ..., m is hybrid weight, is metbi(G) it is the joint gaussian probability distribution function of D dimensions;
502nd, the joint gaussian probability distribution function b of D dimensions is calculatedi(G):
Wherein, μiIt is mean vector, ΣiIt is covariance matrix;
503rd, Gaussian mixture model is constructed:
λ={ wi,μi,Σi, i=1 ..., m (formula 13)
Wherein, wiIt is hybrid weight, μiIt is mean vector, ∑iIt is covariance matrix, λ is the parameter of Gaussian mixture model.
The beneficial effects of the invention are as follows:
Using this method can avoid conventional insulator method for testing performance spend big, malfunction elimination not in time, can not eliminate
Potential potential safety hazard, the shortcoming for being unable to real-time monitoring.Voice recognition technology is incorporated into the insulation of power system high-tension apparatus
Performance detection and fault diagnosis field, make that the insulating properties detection of high-tension apparatus and fault diagnosis reliability be higher, response speed
Faster, intelligence degree is higher.Failure when high-tension apparatus runs can be timely and accurately monitored and be diagnosed to be, find in time,
And alarm.The high-tension apparatuses such as switch cubicle are monitored in real time, the security of the high-tension apparatuses such as switch cubicle is improved.
Brief description of the drawings
Fig. 1 is a kind of overall flow figure based on electric discharge voice recognition high-tension switch cabinet state of the present invention;
Fig. 2 is data prediction flow chart of the present invention;
Fig. 3 is the extraction calculation flow chart of short-time energy relevant feature parameters of the present invention;
Fig. 4 is the extraction calculation flow chart of MFCC characteristic parameters of the present invention;
Fig. 5 is the procedural block diagram of identification of the present invention based on GMM;
Fig. 6 is that high-voltage switch gear state of insulation of the present invention differentiates flow;
Fig. 7 is that high-voltage switch gear state of insulation of the present invention differentiates simulation result figure.
Specific embodiment
Technical solution of the present invention is described in further detail with specific embodiment below in conjunction with the accompanying drawings, so that ability
The technical staff in domain can be better understood from the present invention and can be practiced, but illustrated embodiment is not as to limit of the invention
It is fixed.
A kind of method based on electric discharge voice recognition high-tension switch cabinet state, as shown in figures 1 to 6, comprises the following steps:
The electric discharge voice signal of step 1, respectively collection high-tension switch cabinet corona state and electrion state:Collection high-voltage switch gear
Cabinet insulation breakdown electric discharge sound training sample, because the present invention can recognize high-tension switch cabinet normal state, corona state and electrion
Three kinds of states of state, therefore need sampling corona state and the corresponding electric discharge voice signal sample of electrion state.
Step 2, the electric discharge voice signal respectively to corona state and electrion state are pre-processed, and obtain corona state and height
Press the corresponding training sample of electric state.
It is preferred that, as shown in Figure 2:If the electric discharge voice signal of corona state or electrion state be analog signal x (t) t ∈ [0,
L], L is analog signal duration, and unit is the second, then step 2 specifically includes following steps:
201st, to analog signal framing:
Analog signal x (t) t ∈ [0, L] is divided into M sections, every section is an analysis frame, such as, by analog signal x (t) t
∈ [0, L] be divided into a length of 3 seconds at M sections acoustic segment (acoustic segment of a length of 3 seconds when audio signal is cut into when testing,
Can be other numerical value), each acoustic segment is referred to as an analysis frame, so, carries out treatment to a frame voice signal and is equivalent to
The persistent signal that feature is fixed is processed, wherein, L is analog signal duration, and unit is the second, and M is totalframes, if at every section
A length of 3 seconds, then
202nd, analog signal is sampled:
With fsThe sample rate of Hz is sampled to analog signal x (t) t ∈ [0, L], obtains x (n), n=0,1,2 ..., N,
N is the points of sampling a later frame voice signal, general, takes fs=8000Hz.
203rd, bandpass filtering is carried out to sampled signal:
Bandpass filter is that the ripple of a permission special frequency channel passes through while the equipment for shielding other frequency ranges.One preferably
Bandpass filter should have stable passband (bandpass, it is allowed to the frequency band for passing through), while all logical out-of-band frequencies of limitation
Ripple passes through.But in fact, the ideal bandpass filter without real meaning.Real wave filter cannot be adequately filtered out set
Frequency signal outside the passband of meter, the resonable region with some frequency decay of border of coming round, it is impossible to filter completely, this
Curve is referred to as roll-off slope (roll-off).Roll-off slope generally represents the attenuation degree of frequency with dB measurements.General feelings
Under condition, wave filter design be exactly this attenuation region do it is narrow as far as possible, so that the wave filter can be approached to greatest extent
The design of perfect passband.Median filter exponent number of the present invention is set to 2 ranks, fH、fLThe respectively upper and lower cut-off frequency of bandpass filter,
4000Hz and 60Hz can be respectively set to.
204th, decentralization:
The centralization component of signal is exactly its average, and the Estimation of Mean of signal is:
Wherein, u represents the average of x (n), wants to remove the centralization component of signal, it is only necessary to subtract its average, i.e.,:
C (n)=x (n)-u (formula 2)
Wherein, c (n) is the signal after decentralization, namely training sample.
Step 3, extraction training sample short-time energy relevant feature parameters:
Unlike voice signal, paradoxical discharge voice signal is a kind of non-stationary in time to paradoxical discharge voice signal
, aperiodic random signal, but the power spectrum of this signal is on a timeline continuous, is within the time short enough
What change was relatively delayed, it is possible to regard this signal as a kind of signal of short-term stationarity.This feature based on signal, pre-
The analysis method of process part selection is short time treatment method.Short-time signal analysis method can both process the signal in time domain,
The signal on frequency domain can also be processed.Time-domain analysis mainly includes to signal amplitude, short-time energy, average amplitude and short-time average
The isoparametric analysis of zero-crossing rate;Frequency-domain analysis includes the analysis to power spectrum, spectrum envelope, frequency spectrum, cepstrum coefficient etc..Early stage is adopted
The information of the electric discharge sound of collection insulation breakdown, is sampled, the treatment such as decentralization to collection information, forms training sample data
Storehouse.
For pretreated signal c (n), short-time energy relevant parameter is extracted as shown in Figure 3:
301st, windowing process is carried out to c (n):
Adding window is carried out to signal c (n) using Hamming window, Hamming window formula is as follows:
Wherein, RZZ () is rectangular window,Z is frame length, and w (n) is Hamming window formula;
302nd, short-time energy is calculated:
Wherein, EkIt is the short-time energy of kth frame voice signal, k ∈ [1, M], Z is frame length, ck(n) (k=1,2,
... M) it is that, by pretreated kth frame signal, M is totalframes;
303rd, short-time average energy is calculated
304th, short-time energy shake E is calculateds:
Step 4, extraction training sample MFCC parameters (i.e. Mel frequency cepstral coefficients):
Tone is to differentiate volume up-down, and the sound low for frequency is sounded feels its tone " low ", frequency sound high
Sound, sounds and feels its tone " height ", but tone is not proportional with the frequency of sound, in order to describe tone,
Employ Mei Er Mel scales.Loudness level is 40Phon, and tone of the frequency produced by the pure tone of 1000Hz is set to 1000Mel.Sound
Adjust fMelWith frequency fHzBetween approximate corresponding relation can be represented by below equation:
In formula, fMelIt is the perceived frequency in units of Mel, fHzIt is the actual frequency in units of Hz, by voice signal
Spectrum Conversion to perceive frequency domain in, can preferably carry out parameter extraction and identification.
For each frame voice signal function ckN (), MFCC characteristic parameter extractions are carried out using step shown in Fig. 4:
401st, Fast Fourier Transform (FFT) (FFT) is carried out:
Wherein, S (n) is kth frame signal ckThe discrete power spectrum of (n),
402nd, triangle window filtering group:
Calculate S (n) and pass through X wave filter HmPerformance number after (n):
Pm(m=0,1 ..., X-1) (formula 8)
Wherein, PmIt is kth frame signal ckThe performance number of (n), HmN () is the system function of Mel bandpass filters.
Hm(n), m=0,1 ... .., X-1;N=0,1 ... .., N/2-1
Wherein, X is the number of wave filter, the points (after zero padding) that 24, N is a frame voice signal is generally taken, in order to calculate
The convenience of FFT, it is 256,512,1024 etc. generally to take N.Wave filter is simple triangle on Mel frequency domains, and its Mel center is frequently
Rate is fm, they are equally distributed on Mel frequency axis.The Mel frequencies of two bottom points of the triangle of each wave filter point
The Mel centre frequencies of two wave filters that Deng Yu be not adjacent, the i.e. intermediate zone of the adjacent wave filter of each two is mutually overlapped.Conspicuous
Hereby in frequency, when m is smaller, adjacent hertz frequency is closely spaced, with the increase of m, adjacent fmHertz frequency interval by
It is cumulative big, in the relatively low region of frequency, fMelAnd fHzBetween to have one section be close to linear.The parameter of Mel bandpass filters is prior
Design, directly used when calculating MFCC.Parameter X takes 24, N and takes 1024 in the present invention.
403rd, logarithm L is soughtm:
Lm=lnPm(m=0,1 ..., X-1) (formula 9)
Wherein, LmIt is to PmSeek the value of natural logrithm;
404th, to LmDiscrete cosine transform (DCT) is carried out, D is obtainedm(m=0,1 ... .X-1), omit flip-flop
D0, take D1,D2,D3,......,DhUsed as MFCC cepstrum, h is constant, generally takes h=12;
405th, the first-order difference coefficient of Mel frequency cepstral coefficients is sought:
Wherein, d (k) is the first-order difference Mel frequency cepstral coefficients of kth frame signal, and D (k+i) is the Mel of (k+i) frame
Frequency cepstral coefficient, h is constant, generally takes h=12.
Step 5, design grader:It is right using Gaussian mixture model (Gaussian Mixture Model, referred to as GMM)
Training sample carries out model training, divides sample space, completes the design of grader.Grader is constructed by Gaussian mixture model,
GMM is considered as the continuously distributed hidden Markov model CDHMM that a kind of status number is 1, comprises the following steps that:
501st, the probability density function of m rank Gaussian mixture models is calculated:
Wherein, P (G/ λ) is the probability density function of m rank Gaussian mixture models, and G is the observation vector of D n dimensional vector n sequences,
Expression step 4 is extracted in the present invention Mel frequency cepstral coefficients and the first-order difference of Mel frequency cepstral coefficients, wi, i=
1 ..., m is hybrid weight, is metB is distributed per heighti(G) it is the joint gaussian probability distribution function of D dimensions;
502nd, the joint gaussian probability distribution function b of D dimensions is calculatedi(G):
Wherein, μiIt is mean vector, ΣiIt is covariance matrix;
503rd, Gaussian mixture model is constructed:
Complete Gaussian mixture model is made up of mean parameter vector, covariance matrix and hybrid weight, is expressed as:
λ={ wi,μi,Σi, i=1 ..., m (formula 13)
Wherein, wiIt is hybrid weight, μiIt is mean vector, ∑iIt is covariance matrix, λ is the parameter of Gaussian mixture model.
Step 6, parameter training is carried out to Gaussian mixture model:
The training of GMM model is exactly to give one group of training data, and the parameter lambda of model is determined according to certain criterion, the present invention
Parameter lambda is estimated using EM algorithm (Expectation Maximization, abbreviation EM).The basic thought of EM is from one
The model λ of individual initialization starts, and goes to estimate a new modelSo that(P (G/ λ) be GMM seemingly
So spend).This stylish model turns into initial model for repetitive operation next time, and the process is performed until reaching receipts repeatedly
Hold back thresholding.
By training sample c (n) in the present invention, n=0,1,2 ... N as GMM model list entries, it is specific as follows:
601st, by training sample c (n), n=0,1,2 ... N and calculate input as the list entries of Gaussian mixture model
The likelihood score P (c/ λ) of sequence:
Degree long for one group is training sequence c (n) of Q, n=0,1,2 ... Q, and the likelihood score of Gaussian mixture model is:
Because above formula is the nonlinear function of parameter lambda, it is difficult to directly seek the maximum of above formula.Therefore, EM algorithms can be used
Estimate parameter lambda.Interative computation each time, following revaluation formula ensure that the monotonic increase of model likelihood score.
602nd, revaluation hybrid weight wi ·:
Wherein, P (i/c (n), λ) is the posterior probability of component i,W in formulaiFor upper
The hybrid weight of secondary iteration, the parameter forward and backward for the ease of distinguishing revaluation, employs different shapes to same parameter herein
Formula, such as this revaluation hybrid weight wi ·It is the w of next iterationi;
603rd, revaluation mean μi ·:
604th, revaluation variance
μ in formulaiIt is the average of last iteration.
When using EM Algorithm for Training GMM, the number L of the Gaussian component of GMM model and the initial parameter of model must be first
First determine.These are difficult theoretically to derive, can only test the performance for determining selection different parameters.
Step 7, monitoring high-tension switch cabinet, the electric discharge sound to monitoring pre-process and obtain sample to be identified, extracts
The short-time energy relevant feature parameters and MFCC parameters of sample to be identified, when short-time energy exceedes threshold value, then into step 8.
The probable value of step 8, calculating sample to be identified under Gaussian mixture model.
Step 9, high-tension switch cabinet state, including normal state, corona state and electrion state are judged according to result of calculation.
In order to improve detection accuracy, it is necessary to gather multiple high-tension switch cabinet insulation breakdown electric discharge sound training samples, if
Corona state and electrion state have J respectively1、J2Individual electric discharge voice signal, after framing, then corona state and electrion state have ξ1
=J1×M、ξ2=J2× M frames, gauss hybrid models are obtained to the training of every frame signal, after the completion of finally training, are respectively obtainedParameter andParameter, as shown in figure 5, parameter GMM1 is in Fig. 5Parameter GMM2 isThe parameter is used to recognize height in follow-up identification process
Compress switch cabinet normal state, three kinds of states of corona state and electrion state.
The fault detect of high-tension switch cabinet:
To the sample identified, can refer to step 2 to be pre-processed, characteristic parameter is carried out per frame signal with reference to step 3 and 4 pairs
Extract.
The every frame after framing all calculates its likelihood value p (X/GMM on all training patterns to the sample identifiedn), should
GMM model counts the probability from corona states model respectively by the training of step 6With from electrion
Probability of stateCalculate the overall probability value P in two classifications1And P2:
Wherein, P1It is sample to be identified in the overall probability value of corona state, P2It is sample to be identified in the total general of electrion state
Rate value.
In addition, calculating the short-time average energy of sample to be identifiedE is shaken with short-time energys *。
As shown in fig. 6, high-tension switch cabinet is normal state if following condition is met:
Wherein, thr1 and thr2 are respectively the short-time average energy of setting and the threshold value of short-time energy shake.
For sample to be identified, illustrate that high-tension switch cabinet, for abnormality, now passes through if (formula 20) condition is unsatisfactory for
It is calculated overall probability value P1And P2:
If meeting following condition, high-tension switch cabinet is corona state:
P1>=thr3 (formula 21)
Wherein, thr3 is to be set in corona probability of state threshold value.
If meeting following condition, high-tension switch cabinet is electrion state:
P2>=thr4 (formula 22)
Wherein, thr4 is to be set in electrion probability of state threshold value.
If meeting following condition, high-tension switch cabinet is normal state:
That is, when high-voltage switch gear state of insulation differentiates, the short-time average energy to the sound that discharges and short-time energy first
Shake is judged.If the short-time average energy of the sound that discharges is less than thr1 and short-time energy shake is less than thr2, high to press off
Close cabinet and be in normal condition;Otherwise, high-tension switch cabinet is in abnormality, then carry out electric discharge voice signal feeding GMM model
Judge identification.
This method collects the electric discharge of high-tension switch cabinet insulation breakdown corona state by early stage to onsite application microphone location
Sound, corona state audio sample storehouse is obtained by pretreatment;Collect the discharging sound of high-tension switch cabinet insulation breakdown electrion state
Sound, by pretreatment, phonetic material cut growth 220500, frame are moved 110250 fragment, obtain electrion state by this method
Audio sample storehouse;In order to verify the validity of this method, this method also acquires footsteps, laugh, sneeze sound, rain sound and beats
5 distracters that thunder is constituted.Choose above-mentioned corona state Sample Storehouse, electrion state Sample Storehouse and by footsteps, laugh, spray
Sound is sneezed, 5 distracters that rain sound and the sound that thunders are constituted are tested, if corona state is 1, electrion state is 2, and distracter is
3.As shown in fig. 7, corona state discrimination is 100%, electrion state discrimination is 75% to test result, and interference tones discrimination is
75%.
Using this method can avoid conventional insulator method for testing performance spend big, malfunction elimination not in time, can not eliminate
Potential potential safety hazard, the shortcoming for being unable to real-time monitoring.Voice recognition technology is incorporated into the insulation of power system high-tension apparatus
Performance detection and fault diagnosis field, make that the insulating properties detection of high-tension apparatus and fault diagnosis reliability be higher, response speed
Faster, intelligence degree is higher.Failure when high-tension apparatus runs can be timely and accurately monitored and be diagnosed to be, find in time,
And alarm.The high-tension apparatuses such as switch cubicle are monitored in real time, the security of the high-tension apparatuses such as switch cubicle is improved.
The preferred embodiments of the present invention are these are only, the scope of the claims of the invention is not thereby limited, it is every to utilize this hair
Equivalent structure or equivalent flow conversion that bright specification and accompanying drawing content are made, or directly or indirectly it is used in other correlations
Technical field, be included within the scope of the present invention.
Claims (10)
1. it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that comprise the following steps:
The electric discharge voice signal of step 1, respectively collection high-tension switch cabinet corona state and electrion state;
Step 2, the electric discharge voice signal respectively to corona state and electrion state are pre-processed, and obtain corona state and height is pressed
The corresponding training sample of electric state;
Step 3, extraction training sample short-time energy relevant feature parameters;
Step 4, extraction training sample MFCC parameters;
Step 5, design grader:Model training is carried out to training sample using Gaussian mixture model, sample space is divided;
Step 6, parameter training is carried out to Gaussian mixture model;
Step 7, monitoring high-tension switch cabinet, the electric discharge sound to monitoring pre-process and obtain sample to be identified, extracts and wait to know
Very originally short-time energy relevant feature parameters and MFCC parameters, when short-time energy exceedes threshold value, then into step 8;
The probable value of step 8, calculating sample to be identified under Gaussian mixture model;
Step 9, high-tension switch cabinet state, including normal state, corona state and electrion state are judged according to result of calculation.
2. it is according to claim 1 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
If the electric discharge voice signal of corona state or electrion state is analog signal x (t) t ∈ [0, L], L is analog signal duration, unit
It it is the second, then step 2 specifically includes following steps:
201st, to analog signal framing:
Analog signal x (t) t ∈ [0, L] is divided into M sections, every section is an analysis frame;
202nd, analog signal is sampled:
With fsThe sample rate of Hz is sampled to analog signal x (t) t ∈ [0, L], obtains x (n), n=0,1, and 2 ..., N, N are to adopt
The points of sample a later frame voice signal;
203rd, bandpass filtering is carried out to sampled signal;
204th, decentralization, obtains training sample for c (n), n=0,1,2 ..., N.
3. it is according to claim 2 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
Step 3 specifically includes following steps:
301st, windowing process is carried out to c (n):
Adding window is carried out to signal c (n) using Hamming window, Hamming window formula is as follows:
Wherein, RZZ () is rectangular window,Z is frame length, and w (n) is Hamming window formula;
302nd, short-time energy is calculated:
Wherein, EkIt is the short-time energy of kth frame voice signal, k ∈ [1, M], Z is frame length, ck(n) (k=1,2 ... M) be
By pretreated kth frame signal, M is totalframes;
303rd, short-time average energy is calculated
304th, short-time energy shake E is calculateds:
4. it is according to claim 3 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
Step 4 specifically includes following steps:
401st, Fast Fourier Transform (FFT) is carried out:
Wherein, S (n) is kth frame signal ckThe discrete power spectrum of (n),
402nd, calculate S (n) and pass through X wave filter HmPerformance number after (n):
Pm(m=0,1 ..., X-1) (formula 8)
Wherein, PmIt is kth frame signal ckThe performance number of (n), HmN () is the system function of bandpass filter, Hm(n), m=0,
1,.....,X-1;N=0,1 ... .., N/2-1;
403rd, logarithm L is soughtm:
Lm=ln Pm(m=0,1 ..., X-1) (formula 9)
Wherein, LmIt is to PmSeek the value of natural logrithm;
404th, to LmDiscrete cosine transform is carried out, D is obtainedm(m=0,1 ... .X-1), omit flip-flop D0, take D1,D2,
D3,......,DhUsed as MFCC cepstrum, h is constant;
405th, the first-order difference coefficient of Mel frequency cepstral coefficients is sought:
Wherein, d (k) is the first-order difference Mel frequency cepstral coefficients of kth frame signal, and D (k+i) is the Mel frequencies of (k+i) frame
Cepstrum coefficient, h is constant.
5. it is according to claim 4 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
Step 5 specifically includes following steps:
501st, the probability density function of m rank Gaussian mixture models is calculated:
Wherein, P (G/ λ) is the probability density function of m rank Gaussian mixture models, and G is D dimension random vectors, wi, i=1 ..., m is
Hybrid weight, meetsbi(G) it is the joint gaussian probability distribution function of D dimensions;
502nd, the joint gaussian probability distribution function b of D dimensions is calculatedi(G):
Wherein, μiIt is mean vector, ΣiIt is covariance matrix;
503rd, Gaussian mixture model is constructed:
λ={ wi,μi,Σi, i=1 ..., m (formula 13)
Wherein, wiIt is hybrid weight, μiIt is mean vector, ∑iIt is covariance matrix, λ is the parameter of Gaussian mixture model.
6. it is according to claim 5 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
Parameter lambda is estimated using EM algorithm in step 6.
7. it is according to claim 6 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
Step 6 specifically includes following steps:
601st, by training sample c (n), n=0,1,2 ... N and calculate list entries as the list entries of Gaussian mixture model
Likelihood score P (c/ λ):
Degree long for one group is training sequence c (n) of Q, n=0,1,2 ... Q, and the likelihood score of Gaussian mixture model is:
602nd, revaluation hybrid weight wi ·:
Wherein, P (i/c (n), λ) is the posterior probability of component i;
603rd, revaluation mean μi ·:
604th, revaluation variance
8. it is according to claim 7 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
If corona state and electrion state have J respectively1、J2Individual electric discharge voice signal, after framing, corona state and electrion state have ξ1
=J1×M、ξ2=J2× M frames, gauss hybrid models are obtained to the training of every frame signal, after the completion of finally training, are respectively obtainedParameter andParameter.
9. it is according to claim 8 it is a kind of based on electric discharge voice recognition high-tension switch cabinet state method, it is characterised in that
In step 8, the short-time average energy of sample to be identified is calculatedE is shaken with short-time energys *;To the sample identified after framing
Its likelihood value p (X/GMM on all training patterns is all calculated per framen), and the probability from corona states model is counted respectivelyWith from electrion probability of stateCalculate the overall probability value P in two classifications1And P2:
Wherein, P1It is sample to be identified in the overall probability value of corona state, P2It is sample to be identified in the total probability of electrion state
Value.
10. a kind of method based on electric discharge voice recognition high-tension switch cabinet state according to claim 9, its feature exists
In if thr1 and thr2 are respectively the short-time average energy of setting and the threshold value of short-time energy shake, thr3 is to be set in corona
Probability of state threshold value, thr4 is to be set in electrion probability of state threshold value:
When high-tension switch cabinet state is judged, if:
Then, high-tension switch cabinet is normal state;
Otherwise, following judgements are carried out:
A) if:
P1>=thr3 (formula 21)
Then, high-tension switch cabinet is corona state;
B) if:
P2>=thr4 (formula 22)
Then, high-tension switch cabinet is electrion state;
C) if:
Then, high-tension switch cabinet is normal state.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710092394.1A CN106782505A (en) | 2017-02-21 | 2017-02-21 | A kind of method based on electric discharge voice recognition high-tension switch cabinet state |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710092394.1A CN106782505A (en) | 2017-02-21 | 2017-02-21 | A kind of method based on electric discharge voice recognition high-tension switch cabinet state |
Publications (1)
Publication Number | Publication Date |
---|---|
CN106782505A true CN106782505A (en) | 2017-05-31 |
Family
ID=58958509
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710092394.1A Pending CN106782505A (en) | 2017-02-21 | 2017-02-21 | A kind of method based on electric discharge voice recognition high-tension switch cabinet state |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106782505A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108169639A (en) * | 2017-12-29 | 2018-06-15 | 南京康尼环网开关设备有限公司 | Method based on the parallel long identification switch cabinet failure of Memory Neural Networks in short-term |
CN108303624A (en) * | 2018-01-31 | 2018-07-20 | 舒天才 | A kind of method for detection of partial discharge of switch cabinet based on voice signal analysis |
CN109029582A (en) * | 2018-08-17 | 2018-12-18 | 国网江苏省电力有限公司盐城供电分公司 | A kind of Power Transformer Faults on-line detecting system based on many kinds of parameters acquisition |
CN109116196A (en) * | 2018-07-06 | 2019-01-01 | 山东科汇电力自动化股份有限公司 | A kind of power cable fault discharging sound intelligent identification Method |
CN109900469A (en) * | 2019-03-28 | 2019-06-18 | 西安交通大学 | A kind of high-voltage circuitbreaker STRESS RELAXATION OF HELICAL SPRING fault detection means and method |
CN110456238A (en) * | 2019-07-26 | 2019-11-15 | 苏州微木智能***有限公司 | A kind of corona discharge ion source detection method and system |
CN110531736A (en) * | 2019-08-13 | 2019-12-03 | 中国航空工业集团公司西安飞行自动控制研究所 | A kind of high power motor controller failure monitoring circuit and its method |
CN110706721A (en) * | 2019-10-17 | 2020-01-17 | 南京林业大学 | Electric precipitation spark discharge identification method based on BP neural network |
CN111157092A (en) * | 2020-01-02 | 2020-05-15 | 深圳市汉德网络科技有限公司 | Vehicle-mounted weighing automatic calibration method and computer readable storage medium |
CN111914721A (en) * | 2020-07-27 | 2020-11-10 | 华中科技大学 | Machining state identification method based on linear regression and Gaussian threshold |
CN111933186A (en) * | 2020-10-12 | 2020-11-13 | 中国电力科学研究院有限公司 | Method, device and system for fault identification of on-load tap-changer |
CN112289341A (en) * | 2020-11-03 | 2021-01-29 | 国网智能科技股份有限公司 | Sound abnormity identification method and system for transformer substation equipment |
CN113689888A (en) * | 2021-07-30 | 2021-11-23 | 浙江大华技术股份有限公司 | Abnormal sound classification method, system, device and storage medium |
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 |
CN114186581A (en) * | 2021-11-15 | 2022-03-15 | 国网天津市电力公司 | Cable hidden danger identification method and device based on MFCC (Mel frequency cepstrum coefficient) and diffusion Gaussian mixture model |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101241699A (en) * | 2008-03-14 | 2008-08-13 | 北京交通大学 | A speaker identification system for remote Chinese teaching |
CN102324232A (en) * | 2011-09-12 | 2012-01-18 | 辽宁工业大学 | Method for recognizing sound-groove and system based on gauss hybrid models |
CN102426835A (en) * | 2011-08-30 | 2012-04-25 | 华南理工大学 | Method for identifying local discharge signals of switchboard based on support vector machine model |
CN102708861A (en) * | 2012-06-15 | 2012-10-03 | 天格科技(杭州)有限公司 | Poor speech recognition method based on support vector machine |
CN103531198A (en) * | 2013-11-01 | 2014-01-22 | 东南大学 | Speech emotion feature normalization method based on pseudo speaker clustering |
CN106207763A (en) * | 2016-08-11 | 2016-12-07 | 江苏亿能电气有限公司 | There is the contact box of intelligent online monitoring function |
-
2017
- 2017-02-21 CN CN201710092394.1A patent/CN106782505A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101241699A (en) * | 2008-03-14 | 2008-08-13 | 北京交通大学 | A speaker identification system for remote Chinese teaching |
CN102426835A (en) * | 2011-08-30 | 2012-04-25 | 华南理工大学 | Method for identifying local discharge signals of switchboard based on support vector machine model |
CN102324232A (en) * | 2011-09-12 | 2012-01-18 | 辽宁工业大学 | Method for recognizing sound-groove and system based on gauss hybrid models |
CN102708861A (en) * | 2012-06-15 | 2012-10-03 | 天格科技(杭州)有限公司 | Poor speech recognition method based on support vector machine |
CN103531198A (en) * | 2013-11-01 | 2014-01-22 | 东南大学 | Speech emotion feature normalization method based on pseudo speaker clustering |
CN106207763A (en) * | 2016-08-11 | 2016-12-07 | 江苏亿能电气有限公司 | There is the contact box of intelligent online monitoring function |
Non-Patent Citations (1)
Title |
---|
程青云: ""基于GMM的办公室环境下两类异常声音识别的研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108169639A (en) * | 2017-12-29 | 2018-06-15 | 南京康尼环网开关设备有限公司 | Method based on the parallel long identification switch cabinet failure of Memory Neural Networks in short-term |
CN108303624A (en) * | 2018-01-31 | 2018-07-20 | 舒天才 | A kind of method for detection of partial discharge of switch cabinet based on voice signal analysis |
CN109116196A (en) * | 2018-07-06 | 2019-01-01 | 山东科汇电力自动化股份有限公司 | A kind of power cable fault discharging sound intelligent identification Method |
CN109116196B (en) * | 2018-07-06 | 2020-09-25 | 山东科汇电力自动化股份有限公司 | Intelligent power cable fault discharge sound identification method |
CN109029582A (en) * | 2018-08-17 | 2018-12-18 | 国网江苏省电力有限公司盐城供电分公司 | A kind of Power Transformer Faults on-line detecting system based on many kinds of parameters acquisition |
CN109900469A (en) * | 2019-03-28 | 2019-06-18 | 西安交通大学 | A kind of high-voltage circuitbreaker STRESS RELAXATION OF HELICAL SPRING fault detection means and method |
CN110456238B (en) * | 2019-07-26 | 2022-01-28 | 苏州微木智能***有限公司 | Corona discharge ion source detection method and system |
CN110456238A (en) * | 2019-07-26 | 2019-11-15 | 苏州微木智能***有限公司 | A kind of corona discharge ion source detection method and system |
CN110531736A (en) * | 2019-08-13 | 2019-12-03 | 中国航空工业集团公司西安飞行自动控制研究所 | A kind of high power motor controller failure monitoring circuit and its method |
CN110706721A (en) * | 2019-10-17 | 2020-01-17 | 南京林业大学 | Electric precipitation spark discharge identification method based on BP neural network |
CN111157092A (en) * | 2020-01-02 | 2020-05-15 | 深圳市汉德网络科技有限公司 | Vehicle-mounted weighing automatic calibration method and computer readable storage medium |
CN111914721A (en) * | 2020-07-27 | 2020-11-10 | 华中科技大学 | Machining state identification method based on linear regression and Gaussian threshold |
CN111914721B (en) * | 2020-07-27 | 2024-02-06 | 华中科技大学 | Machining state identification method based on linear regression and Gaussian threshold |
CN111933186A (en) * | 2020-10-12 | 2020-11-13 | 中国电力科学研究院有限公司 | Method, device and system for fault identification of on-load tap-changer |
CN112289341A (en) * | 2020-11-03 | 2021-01-29 | 国网智能科技股份有限公司 | Sound abnormity identification method and system for transformer substation equipment |
CN113689888A (en) * | 2021-07-30 | 2021-11-23 | 浙江大华技术股份有限公司 | Abnormal sound classification method, system, device and storage medium |
CN114186581A (en) * | 2021-11-15 | 2022-03-15 | 国网天津市电力公司 | Cable hidden danger identification method and device based on MFCC (Mel frequency cepstrum coefficient) and diffusion Gaussian mixture model |
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 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106782505A (en) | A kind of method based on electric discharge voice recognition high-tension switch cabinet state | |
Secker‐Walker et al. | Time‐domain analysis of auditory‐nerve‐fiber firing rates | |
CN110534118A (en) | Transformer/reactor method for diagnosing faults based on Application on Voiceprint Recognition and neural network | |
CN109856517A (en) | A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data | |
Meddis et al. | Virtual pitch and phase sensitivity of a computer model of the auditory periphery. II: Phase sensitivity | |
CN102426835B (en) | Method for identifying local discharge signals of switchboard based on support vector machine model | |
US6483316B2 (en) | Method of diagnosing partial discharge in gas-insulated apparatus and partial discharge diagnosing system for carrying out the same | |
CN110244204A (en) | A kind of switchgear method for diagnosing faults, system and the medium of multiple characteristic values | |
CN109616140B (en) | Abnormal sound analysis system | |
CN101426168B (en) | Sounding body abnormal sound detection method and system | |
CN111814872B (en) | Power equipment environmental noise identification method based on time domain and frequency domain self-similarity | |
CN105608823B (en) | Optical fiber security method and system based on principal component analysis | |
CN108169639A (en) | Method based on the parallel long identification switch cabinet failure of Memory Neural Networks in short-term | |
CN107993648A (en) | A kind of unmanned plane recognition methods, device and electronic equipment | |
Patterson et al. | Resiude pitch as a function of component spacing | |
CN108562837A (en) | A kind of power plant's partial discharge of switchgear ultrasonic signal noise-reduction method | |
CN109001602A (en) | Shelf depreciation Severity method based on extreme learning machine algorithm | |
CN110942784A (en) | Snore classification system based on support vector machine | |
CN111239597A (en) | Method for representing electric life of alternating current contactor based on audio signal characteristics | |
CN116778956A (en) | Transformer acoustic feature extraction and fault identification method | |
Veitch et al. | A characterization of Arctic undersea noise | |
Voigt et al. | Representation of whispered vowels in discharge patterns of auditory-nerve fibers | |
CN112581940A (en) | Discharging sound detection method based on edge calculation and neural network | |
CN104730384A (en) | Power disturbance identification and localization method based on incomplete S transformation | |
CN110703080B (en) | GIS spike discharge diagnosis method, discharge degree identification method and device |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170531 |
|
RJ01 | Rejection of invention patent application after publication |