CN107219457B - Frame-type circuit breaker fault diagnosis and degree assessment method based on operation attachment electric current - Google Patents
Frame-type circuit breaker fault diagnosis and degree assessment method based on operation attachment electric current Download PDFInfo
- Publication number
- CN107219457B CN107219457B CN201710450109.9A CN201710450109A CN107219457B CN 107219457 B CN107219457 B CN 107219457B CN 201710450109 A CN201710450109 A CN 201710450109A CN 107219457 B CN107219457 B CN 107219457B
- Authority
- CN
- China
- Prior art keywords
- energy storage
- coil
- electric current
- fault
- storage motor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/327—Testing of circuit interrupters, switches or circuit-breakers
- G01R31/3277—Testing of circuit interrupters, switches or circuit-breakers of low voltage devices, e.g. domestic or industrial devices, such as motor protections, relays, rotation switches
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Testing Electric Properties And Detecting Electric Faults (AREA)
- Control Of Electric Motors In General (AREA)
Abstract
The present invention relates to frame-type circuit breaker fault diagnosises and degree assessment method based on operation attachment electric current, this method first determines whether the working stage wait diagnose and assess breaker, the working stage includes energy storage stage, combined floodgate stage and separating brake stage, then carries out fault diagnosis to energy storage stage using energy storage motor electric current, carries out fault diagnosis and scale evaluation to the stage of combined floodgate using closing coil electric current and carries out fault diagnosis and scale evaluation to the separating brake stage using opening coil electric current;Energy storage motor current signal of the detection block pantograph disconnecting switch in energy storage stage, the closing coil current signal in the combined floodgate stage and the opening coil current signal in the separating brake stage respectively, fault diagnosis is carried out in combination with multi-kernel support vector machine, when diagnosis be out of order need to carry out fault degree assessment when, the judgement of fault degree can be accurately carried out by fault degree characteristic curve.
Description
Technical field
Technical solution of the present invention is related to the fault diagnosis and scale evaluation of breaker, specifically a kind of based on operation
The frame-type circuit breaker fault diagnosis and degree assessment method of attachment electric current.
Background technique
With the rapid development of the national economy, ordinary users also start to put forward higher requirements power quality, low pressure
Power distribution network is the end power grid of power transmission network, is connected directly with users, is particularly significant between connection power transmission network and user
Tie, it is the key that electric system distributes to vast power consumer, adjusts electric energy, the O&M level of low-voltage network and strong
Kang Chengdu often directly determines that the power quality of power consumer is horizontal.Key equipment frame-type circuit breaker in low-voltage network
It is widely used at distribution system total input-wire or at important equipment, protection and control dual role is born, in power station equipment
Quantity is more, and investment is big, and operational reliability is related to the safe operation of electric system, so to key equipment in low-voltage network
One of frame-type circuit breaker study it is most important, carry out frame-type circuit breaker monitoring running state research, can send out in time
Existing breaker mechanical failure, prevents accident, while the research for carrying out frame-type circuit breaker fault diagnosis technology can be very big
Ground mitigates the labor intensity of service personnel, improves power supply reliability, automation, the level of IT application.
It is limited to detect the vibration signal and sound during breaker actuation currently, the research of circuit breaker failure diagnostic techniques
Signal completes the fault diagnosis of breaker with this, such as (Sun Laijun, Hu Xiaoguang, Ji Yanchao one kind are based on vibration to Sun Laijun
Fault Diagnosis for HV Circuit Breakers new method [J] Proceedings of the CSEE of signal, 2006,26 (6): 157-161.) with high pressure
Vibration signal under breaker is non-loaded is that fingerprint signal carries out Fault Diagnosis for HV Circuit Breakers.And breaker operator attachment electric current
Signal has many advantages, such as high stability, is easy to detect, and breaker energy storage, combined floodgate, separating brake action process are contained in synchronous signal
In bulk information, be able to reflect many fault types of attachment, become one of important detection parameters of breaker.It is existing
Research on Fault Diagnosis Technology focuses on fault location more, and, fault degree quantitative analysis energy less for the research of fault degree
More accurate maintenance mode is provided for equipment.(Chen Bin, Yan Zhaoli, Cheng Xiaobin are based on SVDD and relative distance by such as Chen Bin
Equipment fault degree predicts [J] Chinese journal of scientific instrument, 2011,32 (7): 1558-1563.) find out the vibration of rotor oscillation simulation table
The statistics variations rule of signal spectrum curve, extracts the frequency domain character that can describe signal under faulty equipment different faults degree
Relative distance is constructed, the ordered categorization of fault degree size is then carried out.Current most of researchs for fault degree are only
It is limited to the ordered categorization of fault degree, i.e., identifies the failure of fixation degree using intelligent recognition algorithm, this qualitative analysis mode
The not high constraint of recognizer accuracy cannot still be got rid of, it is easy to generate the misinterpretation to fault degree, gently then influence pair
The assurance of working state of circuit breaker, it is heavy then cause power outage and heavy economic losses.
Summary of the invention
In view of the deficiencies of the prior art, the technical problems to be solved by the present invention are: providing a kind of based on operation attachment electricity
The frame-type circuit breaker fault diagnosis and degree assessment method of stream.This method distinguishes detection block pantograph disconnecting switch in energy storage stage
Energy storage motor current signal, the closing coil current signal in the combined floodgate stage and the opening coil electric current in the separating brake stage letter
Number, in combination with multi-kernel support vector machine carry out fault diagnosis, when diagnosis be out of order need to carry out fault degree assessment when, lead to
The judgement of fault degree can accurately be carried out by crossing fault degree characteristic curve, what this method breaker operator attachment can occur
Mechanical breakdown is accurately and reliably diagnosed, while fault degree being rationally and effectively quantitatively evaluated.This method is to open circuit
Device energy storage stage carries out fault diagnosis, carries out fault diagnosis and scale evaluation to combined floodgate, separating brake stage.Energy storage motor electric current is believed
Number envelope line drawing is carried out, then extracts the material time of envelope, amplitude Characteristics parameter, in this, as local feature vectors,
Set empirical mode decomposition (the Ensemble Empirical Mode of energy storage motor current signal is sought simultaneously
Decomposition, EEMD) energy square is as global characteristics, the two combined structure fault feature vector, through Principal Component Analysis
After dimensionality reduction, it is input to multi-kernel support vector machine and carries out fault diagnosis, when being diagnosed to be breaker and being in normal condition, without appointing
Where reason, when the failure being diagnosed to be needs to repair immediately, makes alert process, notifies at maintenance personal immediately
Reason.It is special by the material time, the current amplitude that extract divide-shut brake coil current when being monitored to circuit-breaker switching on-off state
Sign and EEMD energy moment characteristics, combine and establish fault feature vector, input multicore branch after Principal Component Analysis merges dimensionality reduction
It holds vector machine and carries out the diagnosis of divide-shut brake fault type, when the failure being diagnosed to be needs to carry out fault degree qualitative assessment, then count
Point counting closing coil electric current EEMD energy square relative entropy, as fault degree evaluation index, in conjunction with the event having had built up
Barrier degree characteristic curve, and then complete the qualitative assessment of fault degree.
The present invention solve the technical problem the technical solution adopted is that, provide it is a kind of based on operation attachment electric current frame
The diagnosis of formula circuit breaker failure and degree assessment method, this method first determine whether the working stage wait diagnose and assess breaker, institute
Stating working stage includes energy storage stage, combined floodgate stage and separating brake stage, is then carried out using energy storage motor electric current to energy storage stage
Fault diagnosis carries out fault diagnosis and scale evaluation to the stage of combined floodgate using closing coil electric current and utilizes opening coil electric current pair
The separating brake stage carries out fault diagnosis and scale evaluation;Wherein,
One, fault diagnosis is carried out to energy storage stage using energy storage motor electric current, comprising the following steps:
The first step acquires energy storage motor electric current S of the breaker under different working conditioncc(t), energy storage is set in the application
4 kinds of different working conditions that motor is shared normal, transmission gear bite, spring bite and spring fall off, under every kind of working condition
Acquire r group energy storage motor current signal;
Second step is carried out noise suppression preprocessing to the energy storage motor current signal of acquisition, is obtained using mean filter Denoising Algorithm
To denoising energy storage motor current signal S 'cc(t);
Third step, by Hilbert envelope method to denoising energy storage motor current signal S 'cc(t) envelope is carried out to seek,
The amplitude A (t) of denoising energy storage motor current signal is acquired according to formula (3), A (t) is to denoise energy storage motor current signal S 'cc
(t) Hilbert envelope line;
4th step, by Hilbert envelope line to denoising energy storage motor current signal S 'cc(t) material time and electricity are carried out
It flows magnitude parameters to extract, by current maxima I when starting in energy storage motor current signal1, motor stabilizing operation electric current most
Big value I2And maximum value I2Corresponding moment t1, energy storage motor electric current continue total time t2Local fault as energy storage motor electric current
Feature obtains the local feature vectors component T that energy storage motor electric current is formed in time domain1=[I1, I2, t1, t2];
5th step, to denoising energy storage motor current signal S 'cc(t) it is special to carry out set empirical mode decomposition (EEMD) energy square
Sign is extracted, i.e., global characteristics extract;
EEMD energy Moment Feature Extraction comprises the concrete steps that:
5-1., which is determined, decomposes number M and the noise amplitude to be added, by white noise signal nm(t) it is added to point
The denoising energy storage motor current signal S ' of solutioncc(t) in, new signal x to be decomposed is obtained according to formula (5)m(t):
xm(t)=S 'cc(t)+nm(t) (5)
5-2. determines new signal x to be decomposedm(t) Local modulus maxima and minimum point, utilize cubic spline interpolation
Whole maximum points is attached, coenvelope line is formed;Similarly, cubic spline interpolation is carried out to all minimum points, obtained
To lower envelope line;
5-3. finds out the average value of all envelope points of coenvelope line and lower envelope line, is denoted as m1;
5-4. is by new signal x to be decomposedm(t) m1 is subtracted, obtains one-component h according to formula (6)1If: h1Meet IMF
Component condition, then h1It is just new signal x to be decomposedm(t) first intrinsic mode (IMF) component for meeting IMF component condition,
First intrinsic modal components for meeting IMF component condition is denoted as c1, enter step 5-6;
h1=xm(t)-m1 (6)
If 5-5. h1It is unsatisfactory for IMF component condition, and includes the odd function of other different scales, then by h1As original letter
Number, step 5-2~5-4 is repeated, h is obtained1Upper and lower envelope, calculate h1Coenvelope average value, be denoted as m11;Then sentence
Disconnected h11=h1-m11IMF component condition can be met to continue cycling through if be not met by, until h1k=h1(k-1)-m1kIt is full
Until sufficient IMF component condition;By h1kAs new signal x to be decomposedm(t) first meets the eigen mode of IMF component condition
State component, is denoted as c1;
5-6. is c1From new signal x to be decomposedm(t) it independently goes out in, obtains r according to formula (7)1:
r1=xm(t)-c1 (7)
5-7. is by r1Step 5-2~5-5 is repeated as original signal to be decomposed, obtains second IMF component c2;
5-8. repeats step 5-2~5-7, obtains new signal x to be decomposed according to formula (8)m(t) m IMF component, when
Residual components rmWhen for monotonic function, then stop repeating step 5-2~5-7, complete it is primary decompose,
At this point, new signal x to be decomposedm(t) it is indicated with formula (9);
In formula, rmIt is survival function, the average tendency of representation signal;ciFor i-th of IMF component;
The amplitudes noise signal such as new is added to the denoising energy storage motor current signal S ' decomposed by 5-9.cc
(t) in, it is repeated in step 5-2~5-8M times, obtains M group IMF vector sequence { cn,l, cn,lFirst decomposed for n-th
IMF component, n=1,2 ... ..., M, l=1,2 ... ..., m;
5-10. seeks M group IMF vector sequence { c according to formula (10)n,lFirst of IMF component M times decomposition average valueIt willAs first of IMF component of EEMD, i.e., it is by the IMF component that EEMD is obtained
5-11. is according to formula (11) to the IMF component obtained by EEMDIt carries out that energy square is taken to calculate:
In formula: Δ t is sampling time interval, n1For sampling number, k1=1,2 ..., n1;
5-12. new signal x to be decomposedm(t) it is decomposed into m IMF component, acquires m energy square E altogetherm, then energy storage motor
The global characteristics component of a vector T that electric current is formed in time-frequency domain2=[E1, E2... ..., Em];
6th step establishes energy storage motor fault feature vector, constructs energy storage motor sample dimensionality reduction eigenmatrix:
The local feature vectors component T that energy storage motor electric current is formed in time domain is acquired by the 4th step1=[I1, I2, t1,
t2] and the 5th step acquire the global characteristics component of a vector T that energy storage motor electric current is formed in time-frequency domain2=[E1, E2... ..., Em],
By two feature vector components makes fusion constructs at energy storage motor fault feature vector T, indicated with formula (12):
T=[T1 T2] (12)
The dimension of this energy storage motor fault feature vector is 4+m dimension, special to energy storage motor failure using Principal Component Analysis
It levies vector and carries out dimension-reduction treatment, the dimension of the feature vector T ' after dimensionality reduction is λ dimension;Energy storage motor dimensionality reduction under all working state
Feature vector afterwards constitutes energy storage motor sample dimensionality reduction eigenmatrix, then the dimension of energy storage motor sample dimensionality reduction eigenmatrix is 4r
×λ;
7th step constructs energy storage motor multi-category support vector machines, carries out fault diagnosis:
The total class number of the energy storage motor working condition identified is 4, and the data under some working condition are regarded as just
Class, the data under remaining working condition regard negative class as, and the energy storage motor sample dimensionality reduction eigenmatrix obtained using the 6th step is used
" one-to-many " method constructs three Sub-SVMs, and three Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function
Multi-kernel support vector machine is constituted based on kernel function and multilayer perceptron kernel function, the nuclear parameter of each kernel function is true using empirical method
It is fixed, and multicore weight coefficient is optimized using genetic algorithm, and then obtain multi-category support vector machines, pass through more classification
Support vector machines identifies the working condition of breaker energy storage motor;
When breaker energy storage motor needs to carry out fault diagnosis, energy storage motor current signal is acquired, is mentioned according to the 6th step
Energy storage motor fault feature vector is taken, then input quantity will be used as after energy storage motor fault feature vector dimensionality reduction, be input to energy storage
The working condition identification to energy storage motor can be completed in motor multi-category support vector machines;
Two, fault diagnosis and scale evaluation are carried out to the stage of combined floodgate using closing coil electric current, comprising the following steps:
The first step acquires closing coil electric current S of the breaker under different working conditionhc(t), closing coil circuit is set
Shared normal, iron core bite, mechanical structure bite, the working condition that iron core stroke is insufficient and coil voltage is different less than 5 kinds, often
R group current signal is acquired under kind working condition;
Second step, to closing line loop current Shc(t) denoising is carried out, denoising closing coil electric current S ' is obtainedhc(t), it goes
Algorithm of making an uproar is consistent with Denoising Algorithm in the processing of energy storage motor current signal;
Third step, to denoising closing coil electric current S 'hc(t) material time, current amplitude parameter extraction are carried out, is closed a floodgate
The local feature vectors component T that coil current signal is formed in time domainh1;
Closing coil Current wave-shape characteristic is analyzed, the process that iron core acts is divided into four-stage: first stage is iron
Core setting in motion, coil magnetization, but do not generate magnetic saturation always;Second stage is coil magnetic saturation, the exciting current of generation
To have the characteristics that the peaked wave of 1/4 periodic symmetry;Three phases are electric current approximation steady state stage, coil current waveform and magnetic flux
Same-phase, peak valley amplitude are constant;Fourth stage is the current waveform of auxiliary switch separation phase, and electric current shows as being reduced rapidly,
Until electric current thoroughly disappears;The time t in aforementioned four stage11、t12、t13、t14Different working conditions can be in due to breaker
And corresponding variation occurs, while closing coil electric current continues total time t15Also it can change;Since closing coil is exchange
Power voltage supply, the coil current of generation are also AC wave shape, and 3 trough I occurs altogether in closing coil electric current11、I13、I15, 2 waves
Peak I12、I14, it is hereby achieved that the local feature vectors component T that closing coil current signal is formed in time domainh1=[t11,
t12, t13, t14, t15, I11, I12, I13, I14, I15];
4th step will denoise closing coil electric current S 'hc(t) EEMD energy Moment Feature Extraction, denoising closing coil electricity are carried out
Flow S 'hc(t) it is decomposed into mhA IMF component, acquires m altogetherhA energy square, then closing coil current signal is formed in time-frequency domain
Global characteristics component of a vector Th2=[E1h, E2h... ..., Emh];Specific steps are the same as EEMD energy Moment Feature Extraction in energy storage stage
Method;
5th step establishes closing coil fault feature vector, constructs closing coil sample dimensionality reduction eigenmatrix:
By local feature vectors component T of the closing coil current signal in time domainh1=[t11, t12, t13, t14, t15, I11,
I12, I13, I14, I15] and global characteristics component of a vector T in time-frequency domainh2=[E1h, E2h... ..., Emh] fusion constructs combined floodgate
Coil fault feature vector Th=[Th1, Th2], the dimension of thus obtained closing coil fault feature vector is 10+mh, through master
The feature vector T ' of the closing coil obtained after componential analysis dimensionality reductionhDimension be λhIt ties up, the closing line under all working state
Feature vector after enclosing dimensionality reduction constitutes closing coil sample dimensionality reduction eigenmatrix, the dimension of closing coil sample dimensionality reduction eigenmatrix
For 5r × λh;
6th step constructs closing coil circuit multi-category support vector machines, carries out fault diagnosis:
The total class number of closing coil loop works state identified is 5, is dropped using above-mentioned closing coil sample
Dimensional feature matrix constructs 4 Sub-SVMs, and 4 Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function core letter
Multi-kernel support vector machine is constituted based on several and multilayer perceptron kernel function, the nuclear parameter of each kernel function is determined using empirical method,
And multicore weight coefficient is optimized using genetic algorithm, and then obtain multi-category support vector machines, pass through more classification branch
It holds vector machine to identify the working condition of breaker closing coil, constructs closing coil circuit multi-category support vector machines,
Fault diagnosis is carried out to breaker closing wire loop using closing coil circuit multi-category support vector machines;
7th step determines closing coil fault degree evaluation index:
As the denoising closing coil electric current S ' for determining input by closing coil circuit multi-category support vector machineshc(t) it is
When fault-current signal, it is diagnosed to be in closing coil loop fault and iron core stroke deficiency and coil voltage deficiency failure occurs, with
EEMD energy square, as fault degree evaluation index, carries out fault degree assessment with respect to entropy;
8th step constructs closing coil loop fault degree characteristic curve:
Every kind of failure is respectively provided with 4 kinds of different fault degrees, the iron core stroke under calculating in various degree using the 7th step
Then insufficient EEMD energy square corresponding with coil voltage deficiency failure is fitted, linearly with respect to entropy using linear function
Function is formula (16), obtains closing coil loop fault degree characteristic curve,
Y=a1x+a2 (16)
Wherein: x is fault degree, and y is EEMD energy square with respect to entropy, a1,a2For constant;
Respectively obtain less than two kinds corresponding closing coil loop fault degree of failure of iron core stroke deficiency and coil voltage
Characteristic curve carries out at denoising closing coil current signal to be assessed through second step when needing to carry out fault degree assessment
Then reason calculates the EEMD energy square of denoising closing coil current signal with respect to entropy according to the 7th step, substitutes into above-mentioned corresponding
In closing coil loop fault degree characteristic curve, you can get it corresponding closing coil fault degree quantitative values;
Three, fault diagnosis and scale evaluation are carried out to the separating brake stage using opening coil electric current, comprising the following steps:
The first step acquires opening coil electric current S of the breaker under different faults statefc(t), opening coil circuit is set
Shared working condition normal, armature resistance is abnormal, armature travel is insufficient and coil voltage is different less than 4 kinds, every kind of work shape
R group current signal is acquired under state;
Second step, to opening coil electric current Sfc(t) denoising is carried out, denoising opening coil electric current S ' is obtainedfc(t), it goes
Algorithm of making an uproar is consistent with Denoising Algorithm in the processing of energy storage motor current signal;
Third step, to denoising opening coil electric current S 'fc(t) material time, current amplitude parameter extraction are carried out, separating brake is obtained
The local feature vectors component T that coil current signal is formed in time domainf1;It analyzes opening coil electric current and passes through characteristic, generate 5
Material time point t21, t22, t23, t24, t25, there are 2 trough I altogether in opening coil electric current21、I23, 2 wave crest I22、I24, obtain
The local feature vectors component T that opening coil electric current is formed in time domainf1=[t21, t22, t23, t24, t25, I21, I22, I23,
I24];
4th step will denoise opening coil electric current S 'fc(t) EEMD energy Moment Feature Extraction, denoising opening coil electricity are carried out
Flow S 'fc(t) it is decomposed into mfA IMF component, acquires m altogetherfA energy square, then opening coil current signal is formed in time-frequency domain
Global characteristics component of a vector Tf2=[E1f, E2f... ..., Emf];
5th step establishes opening coil fault feature vector, constructs opening coil sample dimensionality reduction eigenmatrix:
By local feature vectors component T of the opening coil current signal in time domainf1=[t21, t22, t23, t24, t25, I21,
I22, I23, I24] and global characteristics component of a vector T in time-frequency domainf2=[E1f, E2f... ..., Emf] fusion constructs opening coil
Fault feature vector Tf=[Tf1, Tf2], the dimension of thus obtained closing coil fault feature vector is 9+mf, through principal component point
Feature vector T ' after the dimensionality reduction obtained after analysis method dimensionality reductionfDimension be λfDimension;After opening coil dimensionality reduction under all working state
Feature vector constitute opening coil sample dimensionality reduction eigenmatrix, the dimension of closing coil sample dimensionality reduction eigenmatrix be 4r ×
λf;
6th step constructs opening coil circuit multi-category support vector machines, carries out fault diagnosis:
The total class number of opening coil loop works state identified is 4, is dropped using above-mentioned opening coil sample
Dimensional feature matrix constructs 3 Sub-SVMs, and 3 Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function core letter
Multi-kernel support vector machine is constituted based on several and multilayer perceptron kernel function, the nuclear parameter of each kernel function is determined using empirical method,
And multicore weight coefficient is optimized using genetic algorithm, and then obtain multi-category support vector machines, pass through more classification branch
It holds vector machine to identify the working condition of breaker open operation coil, constructs opening coil circuit multi-category support vector machines,
Fault diagnosis is carried out to breaker open operation wire loop by opening coil circuit multi-category support vector machines;
7th step determines opening coil fault degree evaluation index:
As the denoising closing coil electric current S ' for determining input by opening coil circuit multi-category support vector machineshc(t) it is
When fault-current signal, it is diagnosed to be in opening coil loop fault and armature travel deficiency and coil voltage deficiency failure occurs, with
EEMD energy square relative entropy carries out fault degree assessment as fault degree evaluation index;
8th step constructs opening coil loop fault degree characteristic curve, every kind of failure is respectively provided with 4 kinds of different failures
Degree, the corresponding EEMD energy square of armature travel deficiency and coil voltage deficiency failure under being calculated in various degree using the 7th step
Opposite entropy, is then fitted using linear function, and linear function is formula (17), and it is special to obtain opening coil loop fault degree
Linearity curve;
Y=b1x+b2 (17)
Wherein: x is fault degree, and y is EEMD energy square with respect to entropy, b1, b2For constant;
It is completed by the opening coil circuit multi-category support vector machines of above-mentioned building to occurring in opening coil circuit
Normally, armature resistance is abnormal, armature travel is insufficient, less than 4 kinds working conditions of coil voltage are diagnosed;When the failure being diagnosed to be
To need to continue fault degree assessment, after calculating denoising to be assessed when armature travel is insufficient or coil voltage deficiency
Opening coil electric current EEMD energy square relative entropy, substitutes into the opening coil loop fault degree characteristic curve of above-mentioned foundation, i.e.,
The achievable qualitative assessment to opening coil fault degree.
Compared with prior art, the beneficial effects of the present invention are:
The invention proposes the processing modes of the ac signal based on operation attachment, not with traditional DC analysis mode
Together, operation attachment AC signal is electric signal, and stability is more preferable, and it is any subtle more accurately to reflect that operation attachment generates
Variation, and not vulnerable to external interference, feature extraction mode diversification, due to the existing fault diagnosis for breaker both for
In high-voltage circuitbreaker, operates various coils in attachment and use direct current supply, the research object frame-type open circuit in the present invention
Device is low-voltage circuit breaker, is powered for alternating voltage;Multicore supporting vector of the fault diagnosis mode Selection utilization based on genetic algorithm
Machine efficiently solves the problems, such as the Selection of kernel function of conventional single-core support vector machines, and fault diagnosis reliability is higher, has stronger
Engineering practicability;Propose fault degree quantitative evaluating method, rather than simple judgement failure whether there is or not and fault location, together
When overcome the problem of existing fault degree qualitative analysis inaccuracy.
Present invention substantive distinguishing features outstanding are: for frame-type circuit breaker fault diagnosis and the practical need of scale evaluation
It wants, constructs completely circuit breaker failure diagnosis and scale evaluation mode based on operation attachment electric current.Using non-intrusion type
The measurement method operation attachment electric current sensitive to circuit-breaker status variation detects, and can conveniently and efficiently utilize breaker signal
Acquisition system;When handling current signal, by the way of ac signal, feature extraction mode diversification, by office
Portion's feature is combined with global characteristics effectively prevents the existing problem to fault signature deficiency in direct current signal processing;Utilize base
In multi-kernel support vector machine (the Multiple-Kernel Learning of genetic algorithm (GeneticAlgorithm, GA)
Support Vector Machine, MKL-SVM) carry out fault diagnosis, effective solution monokaryon support vector machines
The Selection of kernel function and parameter of (SupportVector Machine, SVM) determine problem;For breaker energy storage, close a floodgate and
The problem of needing to continue fault degree assessment in the separating brake stage on the basis of having diagnosed and being out of order, to different failure structures
Different fault degree characteristic curves has been made, according to this fault degree characteristic curve, the qualitative assessment of corresponding failure can be completed,
The problem of fault location can only be carried out in fault diagnosis technology is not only solved, also solving existing fault degree research cannot
The problem of quantitative analysis.
The present invention is based on the marked improvements of the frame-type circuit breaker fault diagnosis of operation attachment electric current and degree assessment method
It is:
(1) the method for the present invention not only identifies circuit breaker failure type, also to breaker occur fault degree into
Row Quantitative Analysis can help service personnel preferably to grasp the mechanical breakdown degree of injury of breaker, and effectively guidance is set
Standby maintenance.
(2) the method for the present invention is to include the current signal of abundant mechanical property information during breaker actuation as data
Source, the electrical cable of the energy storage motor of breaker and divide-shut brake coil, which is passed through current sensor, can be completed to electric current
The measurement of signal does not need to destroy breaker body structure, realizes the state-detection to frame-type circuit breaker non-intrusion type.
(3) current signal used in the method for the present invention is AC signal, can not only carry out time-domain analysis, can be with
Time and frequency domain characteristics analysis is carried out, i.e., the fault signature extraction of current signal material time, amplitude, embodies electric current in completion time domain
The local feature of signal, while EEMD energy Moment Feature Extraction in time-frequency domain is carried out, the global characteristics of current signal are embodied, it will
Local feature is combined with global characteristics, can preferably reflect the faint variation of current signal.
(4) the method for the present invention uses work of the multi-kernel support vector machine to breaker that weight coefficient is determined based on genetic algorithm
Identified solve in monokaryon support vector machines when sample size is very big, sample characteristics include Heterogeneous Information, sample as state
The problem of good recognition effect cannot be obtained whens unevenness distribution etc., improve the accuracy of intelligent diagnostics algorithm.
(5) during fault degree characteristic curve constructs, different fault degrees requires to establish under each failure
Fault degree evaluation index is described, fault degree evaluation index be also by current signal under different faults degree into
Row feature extraction obtains, unified using EEMD energy square relative entropy as fault degree evaluation index, the fault degree in the present invention
Evaluation index is able to reflect the difference between the probability distribution of two signal energy spectrum.
(6) the method for the present invention is by establishing fault degree spy to the relationship between fault degree and fault degree characteristic index
Linearity curve, the fault degree characteristic curve can the failure to the arbitrary extent in fault degree scope of assessment carry out assessment point
Analysis, compared in the research of current fault degree only for specific fault degree carries out orderly identification, by it is qualitative switch to it is quantitative,
Fault degree assessment accuracy, which has, to be obviously improved.
Detailed description of the invention
Fig. 1 is that the present invention a kind of frame-type circuit breaker fault diagnosis and degree assessment method based on operation attachment electric current are total
Body flow chart.
Fig. 2 is the energy storage motor current signal waveform figure in embodiment 1 under different working condition.
Fig. 3 is that energy storage motor current signal denoises front and back comparison diagram under normal condition in embodiment 1.
Fig. 4 is energy storage motor current signal envelope line drawing figure under normal condition in embodiment 1.
Fig. 5 is that energy storage motor current signal EEMD schemes under normal condition in embodiment 1.
Fig. 6 is the fault diagnosis result figure based on energy storage motor current signal in embodiment 1.
Fig. 7 is the closing coil current signal waveform figure in embodiment 2 under different working condition.
Fig. 8 is that closing coil current signal denoises front and back comparison diagram under normal condition in embodiment 2.
Fig. 9 is that closing coil current signal EEMD schemes under normal condition in embodiment 2.
Figure 10 is the fitness curve based on genetic algorithm optimizing parameter in embodiment 2.
Figure 11 is genetic algorithm optimization weight coefficient variation tendency in embodiment 2.
Figure 12 is closing coil failure of the current degree characteristic curve in embodiment 2.
Figure 13 is the opening coil current signal waveform figure in embodiment 3 under different working condition.
Figure 14 is that opening coil current signal denoises front and back comparison diagram under normal condition in embodiment 3.
Figure 15 is that opening coil current signal EEMD schemes under normal condition in embodiment 3.
Figure 16 is opening coil failure of the current degree characteristic curve in embodiment 3.
Specific embodiment
The present invention is further described with the present embodiment with reference to the accompanying drawing, but does not want in this, as to the application right
Ask the restriction of protection scope.
The present invention provides a kind of (simple based on the frame-type circuit breaker fault diagnosis for operating attachment electric current and degree assessment method
Title method), this method first determines whether the working stage wait diagnose and assess breaker, the working stage include energy storage stage,
Then combined floodgate stage and separating brake stage carry out fault diagnosis to energy storage stage using energy storage motor electric current, utilize closing coil electricity
Stream fault diagnosis and scale evaluation are carried out to the stage of combined floodgate and using opening coil electric current to separating brake stage progress fault diagnosis and
Scale evaluation;Wherein,
One, fault diagnosis is carried out to energy storage stage using energy storage motor electric current, comprising the following steps:
The first step acquires energy storage motor electric current S of the breaker under different working conditioncc(t), energy storage is set in the application
4 kinds of different working conditions that motor is shared normal, transmission gear bite, spring bite and spring fall off, under every kind of working condition
Acquire r group energy storage motor current signal;
Second step is carried out noise suppression preprocessing to the energy storage motor current signal of acquisition, is obtained using mean filter Denoising Algorithm
To denoising energy storage motor current signal S 'cc(t);
The specific algorithm step of the mean filter Denoising Algorithm is according to (Wu Jianhua, Li Chisheng, the Zhou Weixing such as Wu Jianhua
Mean filter [J] University Of Nanchang journal (engineering version), 1998,20 (1): 32-35. compared with the denoising performance of mean filter) it is public
The mean filter Denoising Algorithm step opened carries out calculating analysis;
Third step, by Hilbert envelope method to denoising energy storage motor current signal S 'cc(t) envelope is carried out to seek,
Since Hilbert envelope method is the envelope of time-domain signal absolute value, it extracts modulated signal from signal, analyzes modulated signal
Variation, it is simple and effective and be chiefly used in engineer application, therefore select Hilbert envelope method to denoising energy storage motor in the application
The extraction of current signal progress envelope;
Hilbert envelope method comprises the concrete steps that:
3-1. is to denoising energy storage motor current signal S 'cc(t) Hilbert transform is carried out, hubert transformed signal is obtainedFor formula (1):
In formula: τ is the time shifting distinguished with time t.
3-2. is by denoising energy storage motor current signal S 'cc(t) and its hubert transformed signalAccording to formula (2) structure
At denoising energy storage motor current signal S 'cc(t) analytic signal S 'zcc(t):
Wherein j indicates that hubert transformed signal is the imaginary part of analytic signal;
Then analytic signal S 'zcc(t) amplitude A (t) is indicated with formula (3):
The phase of analytic signalIt is indicated with formula (4):
It is then to denoise energy storage motor current signal S ' according to the amplitude A (t) that formula (3) acquirescc(t) Hilbert envelope
Line;
4th step, by Hilbert envelope line to denoising energy storage motor current signal S 'cc(t) material time and electricity are carried out
It flows magnitude parameters to extract, obtains the local feature vectors component T that energy storage motor electric current is formed in time domain1;
The most direct nuance of current signal of the breaker under different faults state is embodied in time domain, is carried out to it
Fault signature extraction is most important, and Hilbert envelope line is the absolute value envelope in time domain, has very to fault signature is extracted
Big superiority.By analyzing energy storage motor electric current operation characteristic, energy storage motor is first with the progress of biggish starting current
Starting, motor stabilizing is run later, and when energy storage is near completion, switch dynamic/static contact is just contacted, and load increases suddenly, electric current
It rises with it, reaches maximum value when stable operation, then limit switch acts, and electric current is cut off.It therefore, can be by energy storage motor
Current maxima I when starting in current signal1, motor stabilizing operation current maxima I2And maximum value I2Corresponding moment t1、
Energy storage motor electric current continues total time t2The local fault feature formed in time domain as energy storage motor electric current;
Wherein about the extraction to energy storage motor electric current material time and magnitude parameters, referring to Sun Yinshan etc., (Sun Yinshan opens
The such as Wen Tao, Zhang Yiming high-voltage circuit-breaker switching on-off coil current signal feature extraction and failure are in method of discrimination research [J] high
Piezoelectricity device, 2015,51 (9): 134-139.) utilize advantage of the wavelet transformation in terms of abrupt climatic change, it may be convenient to it detects
The amplitude of current signal is mutated and frequency discontinuity.The part formed in time domain using the available energy storage motor electric current of the method
Feature vector components makes T1=[I1, I2, t1, t2];
5th step, to denoising energy storage motor current signal S 'cc(t) set empirical mode decomposition (Ensemble is carried out
Empirical Mode Decomposition, EEMD) energy Moment Feature Extraction, i.e. global characteristics extraction;
Set empirical mode decomposition is carried out to denoising energy storage motor current signal first, then seeks each component after decomposing
Energy square.For the more comprehensively faint variation of reflection energy storage motor current signal, in addition to extracting in time domain other than local feature,
Increasing global characteristics the slight change of energy storage motor electric current can pass through feature vector under the more three-dimensional state by different faults
It embodies.
EEMD algorithm is a kind of signal processing method based on noise auxiliary proposed on the basis of empirical mode decomposition,
It is zero this principle according to the average statistical of uncorrelated sequence in statistical theory, and noise is added into signal to be decomposed, by
It is random sequence in noise, frequency-flat is distributed, and the addition of continuing noise makes discrete signal become continuous signal, and noise makes original
While discrete change in beginning signal is continuous, and because average statistical be zero, can cancel out each other its influence to signal, it in addition to
Except the advantages of retaining empirical mode decomposition, modal overlap can be effectively suppressed, improve the accuracy of signal analysis;It is more using EEMD
Energy storage motor current signal S ' will be denoised by differentiating analysiscc(t) multilayer decomposition is carried out, this unconspicuous signal frequency feature can be made
It is showed in the form of significant energy variation in several subspaces of different resolution, so as to extract reflection system
The global characteristics information of operating status.Traditional frequency band energy feature does not account for each decomposition frequency band rise time axis distribution
The characteristics of, the characteristic parameter that may cause extraction not can accurately reflect the feature of energy storage motor current signal, thus in the application
Energy feature is showed in the form of energy square, EEMD is combined with energy square and carries out global fault's feature in time-frequency domain
It seeks.
EEMD energy Moment Feature Extraction comprises the concrete steps that:
5-1., which is determined, decomposes number M and the noise amplitude to be added, by white noise signal nm(t) it is added to point
The denoising energy storage motor current signal S ' of solutioncc(t) in, new signal x to be decomposed is obtained according to formula (5)m(t):
xm(t)=S 'cc(t)+nm(t) (5)
5-2. determines new signal x to be decomposedm(t) Local modulus maxima and minimum point, utilize cubic spline interpolation
Whole maximum points is attached, coenvelope line is formed;Similarly, cubic spline interpolation is carried out to all minimum points, obtained
To lower envelope line;
5-3. finds out the average value of all envelope points of coenvelope line and lower envelope line, is denoted as m1;
5-4. is by new signal x to be decomposedm(t) m1 is subtracted, obtains one-component h according to formula (6)1If: h1Meet IMF
Component condition, then h1It is just new signal x to be decomposedm(t) first intrinsic mode (IMF) component for meeting IMF component condition,
First intrinsic modal components for meeting IMF component condition is denoted as c1, enter step 5-6;
h1=xm(t)-m1 (6)
The IMF component condition are as follows: 1) in entire signal length, both extreme point number and zero crossing number must phases
Deng or at most differ one;2) coenvelope of the signal and the mean value of lower envelope are necessarily equal to zero at any data point;
If 5-5. h1It is unsatisfactory for IMF component condition, and includes the odd function of other different scales, then by h1As original letter
Number, step 5-2~5-4 is repeated, h is obtained1Upper and lower envelope, calculate h1Coenvelope average value, be denoted as m11;Then sentence
Disconnected h11=h1-m11IMF component condition can be met to continue cycling through if be not met by, until h1k=h1(k-1)-m1kIt is full
Until sufficient IMF component condition;By h1kAs new signal x to be decomposedm(t) first meets the eigen mode of IMF component condition
State component, is denoted as c1;
5-6. is c1From new signal x to be decomposedm(t) it independently goes out in, obtains r according to formula (7)1:
r1=xm(t)-c1 (7)
5-7. is by r1Step 5-2~5-5 is repeated as original signal to be decomposed, obtains second IMF component c2;
5-8. repeats step 5-2~5-7, obtains new signal x to be decomposed according to formula (8)m(t) m IMF component, when
Residual components rmWhen for monotonic function, then stop repeating step 5-2~5-7, complete it is primary decompose,
At this point, new signal x to be decomposedm(t) it is indicated with formula (9);
In formula, rmIt is survival function, the average tendency of representation signal;ciFor i-th of IMF component;
The amplitudes noise signal such as new is added to the denoising energy storage motor current signal S ' decomposed by 5-9.cc
(t) in, it is repeated in step 5-2~5-8M times, obtains M group IMF vector sequence { cn,l, cn,lFirst decomposed for n-th
IMF component, n=1,2 ... ..., M, l=1,2 ... ..., m;
5-10. seeks M group IMF vector sequence { c according to formula (10)n,lFirst of IMF component M times decomposition average valueIt willAs first of IMF component of EEMD, i.e., it is by the IMF component that EEMD is obtained
5-11. is according to formula (11) to the IMF component obtained by EEMDIt carries out that energy square is taken to calculate:
In formula: Δ t is sampling time interval, n1For sampling number, k1=1,2 ..., n1;
5-12. new signal x to be decomposedm(t) it is decomposed into m IMF component, acquires m energy square E altogetherm, then energy storage motor
The global characteristics component of a vector T that electric current is formed in time-frequency domain2=[E1, E2... ..., Em];
6th step establishes energy storage motor fault feature vector, constructs energy storage motor sample dimensionality reduction eigenmatrix: passing through the 4th
Step acquires the local feature vectors component T that energy storage motor electric current is formed in time domain1=[I1, I2, t1, t2] and the 5th step acquire storage
The global characteristics component of a vector T that energy current of electric is formed in time-frequency domain2=[E1, E2... ..., Em], by two feature vectors point
Fusion constructs are measured into energy storage motor fault feature vector T, are indicated with formula (12):
T=[T1 T2] (12)
The dimension of this energy storage motor fault feature vector is 4+m dimension, in order to reduce consequent malfunction diagnosis support vector machines instruction
Practice and the time of diagnosis, while reducing influence of the existence of redundant to fault diagnosis accuracy rate in fault feature vector, in the application
Dimension-reduction treatment is carried out to fault feature vector using principal component analysis (principlecomponent analysis, PCA) method,
The specific steps of Principal Component Analysis are with reference to (finger of the such as Han little Hai, Zhang Yaohui, Sun Fujun based on principal component analysis such as children Han
Mark the Sichuan Weight Determination [J] war industry institute, 2012,33 (10): 124-126) to the step in Principal Component Analysis research
It carries out, the dimension of the feature vector T ' after dimensionality reduction is λ dimension.
When constructing energy storage motor multi-category support vector machines, need to construct the progress of energy storage motor sample dimensionality reduction eigenmatrix
It trains, energy storage motor shares 4 kinds of different working conditions and identifies in the application, acquires r under every kind of working condition
Energy storage motor current signal is organized, the dimension after the feature vector dimensionality reduction that every group of energy storage motor current signal extracts is λ, then is formed
The dimension of energy storage motor sample dimensionality reduction eigenmatrix is 4r × λ, with this energy storage motor sample dimensionality reduction eigenmatrix to energy storage motor
Multi-category support vector machines are trained;
7th step constructs energy storage motor multi-category support vector machines, carries out fault diagnosis:
The total class number of the energy storage motor working condition identified is 4, and the data under some working condition are regarded as just
Class, the data under remaining working condition regard negative class as, and the energy storage motor sample dimensionality reduction eigenmatrix obtained using the 6th step is used
" one-to-many " method constructs three Sub-SVMs, and three Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function
Multi-kernel support vector machine is constituted based on kernel function and multilayer perceptron kernel function, the nuclear parameter of each kernel function is true using empirical method
It is fixed, and multicore weight coefficient is optimized using genetic algorithm, and then obtain multi-category support vector machines, pass through more classification
Support vector machines identifies the working condition of breaker energy storage motor;The application selects genetic algorithm to multicore supporting vector
Weight coefficient in machine training process optimizes, and genetic algorithm has obtained in monokaryon support vector machines parameter optimization extensively
Application, generalization ability is stronger, can be effectively prevented and causes the problem of classification inaccuracy because regularisation parameter selection is not smart
Occur, weight coefficient be one dynamic adjustment process, one group of weight coefficient of every generation, will linear combination at a multicore into
Row test sample obtains nicety of grading, iteration adjustment weight coefficient, until optimal value is found in search;
Constructing energy storage motor multi-category support vector machines, specific step is as follows:
A7-1. the feature vector under normal operating conditions in sample dimensionality reduction eigenmatrix is classified as positive class, remaining work shape
Feature vector under state is classified as negative class, constructs normal Sub-SVM;By transmission gear bite in sample dimensionality reduction eigenmatrix
Feature vector under working condition is classified as positive class, and the feature vector under remaining working condition is classified as negative class, constructs transmission gear card
Puckery Sub-SVM;Feature vector under spring bite working condition in sample dimensionality reduction eigenmatrix is classified as positive class, remaining
Feature vector under working condition is classified as negative class, constructs spring bite Sub-SVM;If above-mentioned 3 Sub-SVMs are equal
It cannot effectively identify, then energy storage motor is in spring and falls off working condition;Wherein multi-kernel support vector machine, i.e. Sub-SVM
Construction step and Ye Hui (unmanned plane sensor fault diagnosis system research of the Ye Hui based on small echo Yu multiple supporting vectors
[D] Nanjing Aero-Space University, 2014.) step used when using based on genetic algorithm optimization multi-kernel support vector machine
Unanimously.
A7-2. to above-mentioned 3 Sub-SVMs be respectively adopted in sample dimensionality reduction eigenmatrix corresponding feature vector into
Row is trained, and rule of thumb method is arranged Polynomial kernel function parameter d in training process, and gaussian radial basis function parameter is using intersection
Verification mode obtains, and multilayer perceptron kernel functional parameter takes the derivative of sample classification number.Determine that multicore is supported in conjunction with genetic algorithm
The core weight coefficient of vector machine, obtains optimal output node, reaches the operating mode recognition accuracy of Sub-SVM most
Height to get arrive the energy storage motor multi-category support vector machines based on breaker energy storage motor current signal.
When breaker energy storage motor needs to carry out fault diagnosis, energy storage motor current signal is acquired, is mentioned according to the 6th step
Energy storage motor fault feature vector is taken, then input quantity will be used as after energy storage motor fault feature vector dimensionality reduction, be input to step
The energy storage motor multi-category support vector machines that A7-2 is established, can be completed the identification to energy storage motor working condition;
Two, fault diagnosis and scale evaluation are carried out to the stage of combined floodgate using closing coil electric current, comprising the following steps:
The first step acquires closing coil electric current S of the breaker under different working conditionhc(t), it sets and closes a floodgate in the application
Shared normal, the iron core bite of wire loop, mechanical structure bite, the work that iron core stroke is insufficient and coil voltage is different less than 5 kinds
Make state, r group current signal is acquired under every kind of working condition;
Second step, to closing line loop current Shc(t) denoising is carried out, denoising closing coil electric current S ' is obtainedhc(t), it goes
Algorithm of making an uproar is consistent with Denoising Algorithm in the processing of energy storage motor current signal;
Third step, to denoising closing coil electric current S 'hc(t) material time, current amplitude parameter extraction are carried out, is closed a floodgate
The local feature vectors component T that coil current signal is formed in time domainh1;
Closing coil Current wave-shape characteristic is analyzed, the process point by iron core movement can be four-stage: first stage is
Iron core setting in motion, coil magnetization, but do not generate magnetic saturation always;Second stage is coil magnetic saturation, the excitation electricity of generation
Stream is the peaked wave with 1/4 periodic symmetry;Three phases are electric current approximation steady state stage, coil current waveform and magnetic
It is constant to collude phase, peak valley amplitude;Fourth stage is the current waveform of auxiliary switch separation phase, and electric current shows as subtracting rapidly
It is small, until electric current thoroughly disappears;The time t in aforementioned four stage11、t12、t13、t14Different work can be in due to breaker
State and occur to change accordingly, while closing coil electric current continues total time t15Also it can change;Since closing coil is
Alternating voltage power supply, the coil current of generation is also AC wave shape, and 3 trough I occurs altogether in closing coil electric current11、I13、I15, 2
A wave crest I12、I14, it is hereby achieved that the local feature vectors component T that closing coil current signal is formed in time domainh1=
[t11, t12, t13, t14, t15, I11, I12, I13, I14, I15];
4th step will denoise closing coil electric current S 'hc(t) EEMD energy Moment Feature Extraction, denoising closing coil electricity are carried out
Flow S 'hc(t) it is decomposed into mhA IMF component, acquires m altogetherhA energy square, then closing coil current signal is formed in time-frequency domain
Global characteristics component of a vector Th2=[E1h, E2h... ..., Emh];Specific steps are the same as the EEMD energy square in energy storage motor current processing
The method of feature extraction;
5th step establishes closing coil fault feature vector, constructs closing coil sample dimensionality reduction eigenmatrix: by closing line
Local feature vectors component T of the loop current signal in time domainh1=[t11, t12, t13, t14, t15, I11, I12, I13, I14, I15] and
Global characteristics component of a vector T in time-frequency domainh2=[E1h, E2h... ..., Emh] fusion constructs closing coil fault feature vector
Th=[Th1, Th2], the dimension of thus obtained closing coil fault feature vector is 10+mh, after Principal Component Analysis dimensionality reduction
Obtained feature vector T 'hDimension be λhDimension;
Meanwhile when constructing closing coil circuit multi-category support vector machines, it is special to need to construct closing coil sample dimensionality reduction
Sign matrix is trained, and closing coil circuit shares 5 kinds of different working conditions and diagnoses in the application, every kind of state
Under acquire r group closing line loop current signal, dimension is after the dimensionality reduction of the feature vector of every group of closing line loop current signal extraction
λh, the feature vector after closing coil dimensionality reduction under all working state constitutes closing coil sample dimensionality reduction eigenmatrix, closes a floodgate
The dimension of coil sample dimensionality reduction eigenmatrix is 5r × λh, with this closing coil sample dimensionality reduction eigenmatrix to closing coil circuit
Multi-category support vector machines are trained;
6th step constructs closing coil circuit multi-category support vector machines, carries out fault diagnosis:
The total class number of closing coil loop works state identified is 5, is dropped using above-mentioned closing coil sample
Dimensional feature matrix constructs 4 Sub-SVMs, and 4 Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function core letter
Multi-kernel support vector machine is constituted based on several and multilayer perceptron kernel function, the nuclear parameter of each kernel function is determined using empirical method,
And multicore weight coefficient is optimized using genetic algorithm, and then obtain multi-category support vector machines, pass through more classification branch
It holds vector machine to identify the working condition of breaker closing coil, constructs closing coil circuit multi-category support vector machines,
The closing coil circuit multi-category support vector machines can be used for the fault diagnosis to breaker closing wire loop;Specific steps are same
The building process of energy storage motor multi-category support vector machines;
7th step determines closing coil fault degree evaluation index:
As the denoising closing coil electric current S ' for determining input by closing coil circuit multi-category support vector machineshc(t) it is
When fault-current signal, that is, it is diagnosed to be in closing coil loop fault and iron core stroke deficiency and coil voltage deficiency failure occurs,
Then continue fault degree assessment, the application is asked it needs to be determined that fault degree evaluation index by closing coil current signal
The index value out, in conjunction with subsequent foundation fault degree characteristic curve can quantitative analysis show that the event occurs for closing coil circuit
The fault degree of barrier.The application is using the opposite entropy of EEMD energy square as fault degree evaluation index;Fault degree evaluation index
Current signal under using normal operating conditions calculates the EEMD of current signal under fault-current signal and normal condition as benchmark
For energy square with respect to entropy, opposite entropy is smaller, indicates that two signal distributions similarity degrees are higher, conversely, then difference is bigger;
Steps are as follows with respect to the calculating of entropy for EEMD energy square:
B7-1. by acquiring denoising closing coil electric current S ' in the 4th stephc(t) mhThe m of a IMF componenthA energy square, i.e.,
E1h, E2h... ..., Emh;Then m is found out according to formula (13)hThe summation E of a energy squareH:
EH=E1+E2+…+Emh (13)
B7-2. kth being found out according to formula (14) ' a energy square accounts for the specific gravity of energy square summation are as follows:
Then denoise closing coil electric current S 'hc(t) energy square distribution character matrix are as follows: P=(p1,p2,…,pk′,…,
pmh);
It similarly calculates and denoises closing coil current signal S ' under normal conditionhz(t) energy square distribution character matrix are as follows: P '
=(p '1,p′2,…,p′k′,…,p′mh);
B7-3. it obtains denoising closing coil current signal S ' under malfunction by formula (15)hc(t) go down with normal condition
Make an uproar closing coil current signal S 'hz(t) EEMD energy square relative entropy:
8th step constructs closing coil loop fault degree characteristic curve.In fault degree analytic process, some failures
There are monotonic relationshi between scale evaluation index and fault degree, referred to as dullness fault signature, that is, the fault degree constructed is commented
Index is estimated with the increase of fault degree and dullness becomes larger or becomes smaller, dull fault signature is able to reflect out and fault severity level
Between monotonic trend, for fault severity level assessment intuitive, simple information is provided.
Every kind of failure is respectively provided with 4 kinds of different fault degrees, calculates the iron core under different faults degree using the 7th step
Then stroke deficiency and the corresponding EEMD energy square of coil voltage deficiency failure are fitted with respect to entropy using linear function,
Linear function is formula (16), obtains closing coil loop fault degree characteristic curve,
Y=a1x+a2 (16)
Wherein: x is fault degree, and y is EEMD energy square with respect to entropy, a1, a2For constant.
Respectively obtain less than two kinds corresponding closing coil loop fault degree of failure of iron core stroke deficiency and coil voltage
Characteristic curve carries out at denoising closing coil current signal to be assessed through second step when needing to carry out fault degree assessment
Reason, (fault degree assessment refers to the opposite entropy of the EEMD energy square for then calculating denoising closing coil current signal according to the 7th step
Mark), it substitutes into above-mentioned corresponding closing coil loop fault degree characteristic curve, corresponding closing coil failure journey that you can get it
Spend quantitative values;
By the closing coil circuit multi-category support vector machines of above-mentioned building, can be completed to closing coil circuit just
Often, less than 5 kinds iron core bite, mechanical structure bite, iron core stroke deficiency, coil voltage working conditions are diagnosed;Work as diagnosis
When haveing closing coil circuit and iron core stroke deficiency and coil voltage deficiency failure occur, need to continue fault degree assessment,
EEMD energy square by calculating denoising closing coil current signal to be assessed substitutes into the closing coil of above-mentioned foundation with respect to entropy
In loop fault degree characteristic curve, the qualitative assessment to fault degree can be completed;
Three, fault diagnosis and scale evaluation are carried out to the separating brake stage using opening coil electric current, comprising the following steps:
The first step acquires opening coil electric current S of the breaker under different faults statefc(t), separating brake is set in the application
The shared working condition normal, armature resistance is abnormal, armature travel is insufficient and coil voltage is different less than 4 kinds of wire loop, often
R group current signal is acquired under kind working condition;
Second step, to opening coil electric current Sfc(t) denoising is carried out, denoising opening coil electric current S ' is obtainedfc(t), it goes
Algorithm of making an uproar is consistent with Denoising Algorithm in the processing of energy storage motor current signal.
Third step, to denoising opening coil electric current S 'fc(t) material time, current amplitude parameter extraction are carried out, separating brake is obtained
The local feature vectors component T that coil current signal is formed in time domainf1;It analyzes opening coil electric current and passes through characteristic, generate 5
Material time point t21, t22, t23, t24, t25, there are 2 trough I altogether in opening coil electric current21、I23, 2 wave crest I22、I24, obtain
The local feature vectors component T that opening coil electric current is formed in time domainf1=[t21, t22, t23, t24, t25, I21, I22, I23,
I24];
4th step will denoise opening coil electric current S 'fc(t) EEMD energy Moment Feature Extraction, denoising opening coil electricity are carried out
Flow S 'fc(t) it is decomposed into mfA IMF component, acquires m altogetherfA energy square, then opening coil current signal is formed in time-frequency domain
Global characteristics component of a vector Tf2=[E1f, E2f... ..., Emf];Specific steps are the same as EEMD energy Moment Feature Extraction in energy storage stage
Method;
5th step establishes opening coil fault feature vector, constructs opening coil sample dimensionality reduction eigenmatrix: by separating brake line
Local feature vectors component T of the loop current signal in time domainf1=[t21, t22, t23, t24, t25, I21, I22, I23, I24] and when
Global characteristics component of a vector T in frequency domainf2=[E1f, E2f... ..., Emf] fusion constructs opening coil fault feature vector Tf=
[Tf1, Tf2], the dimension of thus obtained closing coil fault feature vector is 9+mf, obtained after Principal Component Analysis dimensionality reduction
Feature vector T 'fDimension be λfDimension;Feature vector after opening coil dimensionality reduction under all working state constitutes opening coil sample
This dimensionality reduction eigenmatrix, the dimension of closing coil sample dimensionality reduction eigenmatrix are 4r × λf;
6th step constructs opening coil circuit multi-category support vector machines, carries out fault diagnosis:
The total class number of opening coil loop works state identified is 4, is dropped using above-mentioned opening coil sample
Dimensional feature matrix constructs 3 Sub-SVMs, and 3 Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function core letter
Multi-kernel support vector machine is constituted based on several and multilayer perceptron kernel function, the nuclear parameter of each kernel function is determined using empirical method,
And multicore weight coefficient is optimized using genetic algorithm, and then obtain multi-category support vector machines, pass through more classification branch
It holds vector machine to identify the working condition of breaker open operation coil, constructs opening coil circuit multi-category support vector machines,
The opening coil circuit multi-category support vector machines can be used for the fault diagnosis to breaker open operation wire loop;Specific steps are same
The building process of energy storage motor multi-category support vector machines;
7th step determines opening coil fault degree evaluation index:
As the denoising closing coil electric current S ' for determining input by opening coil circuit multi-category support vector machineshc(t) it is
When fault-current signal, that is, it is diagnosed to be in opening coil loop fault and armature travel deficiency and coil voltage deficiency failure occurs,
Then continue fault degree assessment, using EEMD energy square relative entropy as fault degree evaluation index, specific EMD energy square phase
Step is calculated with respect to entropy with EMD energy square described in closing coil fault degree evaluation index to the calculating step of entropy;
8th step constructs opening coil loop fault degree characteristic curve, every kind of failure is respectively provided with 4 kinds of different failures
Degree calculates armature travel deficiency and the corresponding EEMD energy of coil voltage deficiency failure under different faults degree using the 7th step
Square is measured with respect to entropy, is then fitted using linear function, linear function is formula (17), obtains opening coil loop fault journey
Spend characteristic curve;
Y=b1x+b2 (17)
Wherein: x is fault degree, and y is EEMD energy square with respect to entropy, b1, b2For constant;
It is completed by the opening coil circuit multi-category support vector machines of above-mentioned building to occurring in opening coil circuit
Normally, armature resistance is abnormal, armature travel is insufficient, less than 4 kinds working conditions of coil voltage are diagnosed;When the failure being diagnosed to be
To need to continue fault degree assessment, after calculating denoising to be assessed when armature travel is insufficient or coil voltage deficiency
Opening coil electric current EEMD energy square relative entropy, substitutes into the opening coil loop fault degree characteristic curve of above-mentioned foundation, i.e.,
The achievable qualitative assessment to opening coil fault degree.
In breaker closing coil current circuit by the relationship construction between fault degree and EEMD energy square relative entropy
The fault degree characteristic curve of iron core stroke deficiency and coil voltage deficiency failure is respectively as follows:
Iron core stroke is insufficient: y1=0.17989x+0.5502;
Coil voltage is insufficient: y2=0.87358x-0.5279;
The armature travel deficiency and the insufficient failure journey of coil voltage in breaker open operation coil current circuit constructed
Degree characteristic curve is respectively as follows:
Armature travel is insufficient: y3=0.86404x+0.40975;
Coil voltage is insufficient: y4=1.41578x-0.16885.
The working stage of frame-type circuit breaker in the present invention includes energy storage stage, combined floodgate stage and separating brake stage, energy storage
Stage, combined floodgate stage and separating brake stage are three different processes, successively carry out in breaker work, need in the application method
The failure to occur respectively to the three above stage is diagnosed and is assessed respectively, i.e., breaker is using energy storage motor electric current to storage
It can carry out fault diagnosis the stage, fault diagnosis and scale evaluation are carried out to the stage of combined floodgate using closing coil electric current and utilize separating brake
Coil current carries out fault diagnosis and scale evaluation to the separating brake stage.
The present invention is carried out when building is based on above-mentioned multi-category support vector machines using more classification methods of " one-to-many ",
It is calculated simply, is able to solve the multiple classification problems needed in fault diagnosis;If the total class number identified is G,
F class regards positive class as, remaining G-1 class regards negative class as, constructs multiple two classification Sub-SVMs with this and forms more points
Class support vector machines.
Support vector machines, its determined generalization ability of the kernel function of support vector machines and popularization energy are used in the present invention
Power, the difference of kernel function, the sample distribution being mapped in high-dimensional feature space is just different, influences on nicety of grading very big.Using
Multicore can improve the performance of kernel function, i.e., replace single kernel function using the linear combination of multiple single kernel functions, can
Obtain performance more preferably than monokaryon model.Polynomial kernel function (Polynomial Kernel is used in the application
Function), gaussian radial basis function (Radial Basis Function, RBF) and multilayer perceptron kernel function
(SigmoidKernel Function) as mostly with the kernel function of support vector machines because Polynomial kernel function have it is good
Global nature has very strong extrapolability, and RBF kernel function local property is fine, and it is very strong inside to push away ability i.e. learning ability, more
Layer perceptron kernel function derives from neural network, is suitable for deep learning, therefore, in order to establish an existing preferable extrapolation energy
Power has the multi-kernel support vector machine of preferable learning ability again, can carry out linear combination using these three kernel functions.Multi-kernel function
In linear combination, the weight coefficient before all monokaryons determines the model of multi-kernel function, the study for support vector machines
Ability, Generalization Ability have a great impact, therefore need to optimize weight coefficient, and the basic point of optimization is to solve secondary rule
The problem of drawing, when in the training process of support vector machines using the genetic algorithm based on real coding, weight coefficient is that dynamic is adjusted
Whole, one group of weight coefficient of every generation, linear combination is tested at a monokaryon and obtains nicety of grading, iteration adjustment weight coefficient,
Genetic algorithm illustrates the adaptive process of nature and manual system.
Embodiment illustrated in fig. 1 shows a kind of frame-type circuit breaker failure based on operation attachment electric current provided by the invention
The overall procedure of diagnosis and degree assessment method is: the operation attachment electricity under acquisition characterization frame-type circuit breaker different working condition
Stream signal (including divide-shut brake coil current signal, energy storage motor current signal) → use Mean Filtering Algorithm respectively to various works
Make material time and current amplitude parameter that current signal under state carries out noise suppression preprocessing → seek current signal, when composition
Empirical Mode is gathered in domain local feature vectors component (wherein energy storage motor current signal needs first to carry out envelope extraction) → utilization
State decomposition algorithm decomposes the current signal after denoising, and calculates the energy square of each component, constitutes time-frequency domain global characteristics vector point
Amount → constituted using principal component analysis (PCA) method to by time domain local feature vectors component and time-frequency domain global characteristics component of a vector
Fault feature vector carry out dimension-reduction treatment → using the feature vector after dimensionality reduction as input quantity and be input to more class Support Vectors
Working condition identification, fault diagnosis are carried out in machine (MKL-SVM), opening coil includes normal, abnormal (the mandril resistance of armature resistance
It is abnormal), the working condition that armature travel is insufficient and coil voltage is different less than 4 kinds, closing coil include normal, iron core bite,
Mechanical structure bite, the working condition that iron core stroke is insufficient and coil voltage is different less than 5 kinds, energy storage motor electric current include just
Often, transmission gear bite, spring bite and spring fall off 4 kinds of different working conditions → when to breaker carry out divide-shut brake coil return
When the fault diagnosis of road, if closing coil loop diagnostics have iron core stroke deficiency and coil voltage deficiency failure, opening coil diagnosis
When haveing armature travel deficiency and coil voltage deficiency failure, need to continue fault degree assessment, at this time to the line after denoising
Divide-shut brake loop current signal directly carries out the calculating of EEMD energy square relative entropy, and substituting into fault degree characteristic curve can be obtained failure
Degree quantitative values.In above-mentioned operating mode identification process, when energy storage motor is diagnosed to be the failure of any abnormal condition, it is both needed to
Maintenance personal is notified to immediately treat;It is directly logical when going out iron core bite and mechanical structure bite to closing coil loop diagnostics
Know that maintenance personal immediately treats, does not need to carry out fault degree assessment at this time;When being diagnosed to be armature to opening coil loop fault
When resistance exception, need directly maintenance personal to be notified to handle;During above-mentioned three kinds, it is diagnosed to be energy storage motor, closes a floodgate
When wire loop and opening coil circuit are in normal condition, do not need to carry out any processing.
Embodiment 1
The present embodiment is using the frame-type circuit breaker of model DW15-1600 as subjects, the breaker mechanical structure
Visualization is strong, and suitable for the power distribution network that rated operational voltage is 380V, 50Hz, rated current 1600A, switching mechanism is bullet
Spring operating mechanism.Energy storage motor is reluctance synchronous motor, and voltage rating AC380V, starting torque is not less than 8Nm, revolving speed
For 750r/min.Fault simulation mode in energy storage stage are as follows: simulated using the energy-stored spring of the fixed certain length of hard wire disconnected
The corrosion of road device energy-stored spring and caused by spring bite, timber wedge is placed between transmission gear mimic-disconnecting switch, transmission gear occurs
Jam faults since the spring probability of happening that falls off is smaller and more extreme, therefore only simulate a spring dropping situations, that is, remove
A piece energy-stored spring.
Using the circuit breaker failure diagnostic method based on energy storage motor electric current to the frame-type under above-mentioned simulated failure state
Breaker carries out fault diagnosis, the specific steps are as follows:
The first step uses the energy storage motor electric current in the frame-type circuit breaker fault detection system acquisition breaker energy storage stage
Signal, the sample frequency that data collecting card is arranged is 20kHz/s, sampling time 2s, and acquisition breaker is normal, transmission gear card
Puckery, spring bite and spring fall off the energy storage motor current signal S under 4 kinds of different working conditionscc(t), under every kind of working condition
30 groups of current signals are acquired, wherein 15 groups of training for multi-category support vector machines, 15 groups are supported for testing more classification
The classification accuracy of vector machine, i.e. r=15;
Typical energy storage motor current signal under 4 kinds of working conditions is as shown in Fig. 2, respectively be normal, transmission gear
Bite, spring bite and spring fall off, it can be seen from the figure that energy storage motor current signal is the friendship that peak valley amplitude gradually changes
Signal is flowed, the energy storage motor current signal under different conditions only can not observe difference with naked eyes.
Second step, to the energy storage motor current signal S of acquisitioncc(t) noise suppression preprocessing is carried out using Mean Filtering Algorithm, obtained
To denoising energy storage motor current signal S 'cc(t);
Use the filter window of mean filter Denoising Algorithm for 200.It is with the energy storage motor current signal under normal condition
Example, the effect for denoising front and back are as shown in Figure 3.
Third step, to denoising energy storage motor current signal S 'cc(t) Hilbert envelope line is carried out to seek;With normal condition
Under energy storage motor current signal for, Hilbert envelope line is as shown in Figure 4.
4th step, by Hilbert envelope line to denoising energy storage motor current signal S 'cc(t) material time and electricity are carried out
Magnitude parameters are flowed to extract;I.e. to the starting current maximum value I in energy storage motor current signal1, electric current stable operation maximum value I2And
Maximum value I2Corresponding moment t1, electric current continues total time t2It extracts, it is special to obtain part in the time domain of energy storage motor current signal
Levy component of a vector.To material time and current amplitude parameter in the time domain of the energy storage motor current signal under the different conditions of acquisition
The results are shown in Table 1 for feature extraction.
Material time and current amplitude parameter attribute extract result in energy storage motor current signal time domain under 1 different conditions of table
5th step carries out set empirical mode decomposition energy Moment Feature Extraction to energy storage motor current signal, i.e., global special
Sign is extracted;
By taking the energy storage motor current signal under normal condition as an example, set empirical mode decomposition is carried out, each IMF after decomposition
Component is decomposed into 16 components, therefore time and frequency domain characteristics as shown in figure 5, each component signal is sequentially distributed from high frequency to low frequency altogether
The dimension of component of a vector is 16 dimensions, i.e. m=16 carries out energy Moment Feature Extraction, different conditions according to formula (11) to each component
The results are shown in Table 2 for EEMD energy Moment Feature Extraction in lower energy storage motor current signal time-frequency domain.
EEMD energy moment characteristics (unit/J) in energy storage motor current signal time-frequency domain under 2 different conditions of table
EEMD energy moment characteristics (continued) in energy storage motor current signal time-frequency domain under 2 different conditions of table
6th step establishes energy storage motor fault feature vector, constructs energy storage motor sample dimensionality reduction eigenmatrix;
The local feature vectors component T formed in time domain by the available energy storage motor electric current of the 4th step1=[I1,
I2, t1, t2], dimension 4, the global characteristics vector point formed in time-frequency domain by the available energy storage motor electric current of the 5th step
Measure T2=[E1, E2... ..., E16], dimension 16, then energy storage motor fault feature vector T=[T1,T2] dimension be 20.Setting
The principal component percentage of Principal Component Analysis is 95%, then the dimension after Principal Component Analysis dimensionality reduction is 8.
It needs to construct sample dimensionality reduction eigenmatrix simultaneously and carries out the training of energy storage motor fault diagnosis model and test, energy storage electricity
Machine shares 4 kinds of different states and is diagnosed, and 30 groups of signals are acquired under every kind of state, wherein 15 groups are used to train, 15
Group is for testing, and dimension of the feature vector that each energy storage motor current signal extracts after dimensionality reduction is 8, then the training sample formed
This dimensionality reduction eigenmatrix dimension is 60 × 8, and test sample dimensionality reduction eigenmatrix dimension is 60 × 8.
7th step constructs energy storage motor multi-category support vector machines, carries out fault diagnosis;
The training that multi-category support vector machines are carried out using the sample dimensionality reduction eigenmatrix that the 6th step obtains, utilizes test specimens
This dimensionality reduction eigenmatrix is for the test to multi-category support vector machines classification accuracy.3 are constructed using " one-to-many " method
Sub-SVM, 3 Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function and multilayer perceptron core letter
Based on number, and core weighting parameter is optimized using genetic algorithm, and then obtain multi-category support vector machines, it is more by this
Category support vector machines test the classification accuracy of breaker.Penalty factor and core ginseng in gaussian radial basis function nuclear parameter
Number σ is respectively [2-15..., 215], [2-7,…,27] obtained using cross validation mode, after optimization C=1.2578, σ=
Rule of thumb method is set as d=3 by 0.3185, Polynomial kernel function parameter d, and multilayer perceptron kernel functional parameter γ takes sample classification
Several derivatives, i.e. γ=0.25.Core weight coefficient is used to be optimized in the genetic algorithm of [0,1] interior real coding form,
Optimize during determining weight coefficient, population quantity 20, it is 100 that heredity, which terminates the number of iterations,.Utilize 60 test samples
Energy storage motor fault diagnosis model is tested, for the ease of comparing, the setting of monokaryon function parameter and multi-kernel support vector machine
Parameter setting is identical, while multi-kernel support vector machine is carried out fault diagnosis to single features and fusion feature respectively, each to identify
Rate is as shown in table 3.Fault diagnosis based on multi-kernel support vector machine and fusion feature (local feature and global characteristics combine)
Result figure is as shown in Figure 6.
3 energy storage motor electric current difference kernel function support vector machine discrimination of table
Embodiment 1 shows to can be good at energy storage motor electric current being applied to open circuit by the method proposed in the application
Device fault diagnosis, and accuracy rate of diagnosis is higher.
Embodiment 2
The present embodiment is using the frame-type circuit breaker of model DW15-1600 as subjects, the breaker mechanical structure
Visualization is strong, and suitable for the power distribution network that rated operational voltage is 380V, 50Hz, rated current 1600A, switching mechanism is bullet
Spring operating mechanism.Switching electromagnet is armoured type solenidal electromagnet in closing coil circuit, and voltage 380V, 50Hz, the number of turns is
2130 circles, enamel covered wire is through 0.25mm.Fault simulation mode in closing coil circuit are as follows: in the dynamic iron core of solenoid electromagnet
Thin wire simulation iron core jam faults are filled between axle portion and coil;By pushing rod slide to rise elliptic space internal plug in tripping
Enter small wood analog mechanical structure jam faults;The insufficient fault degree simulation of iron core stroke at the top of dynamic iron core by increasing not
Rubber pad (1.0mm) with quantity realizes that rubber pad addition quantity is more, and fault degree is bigger;The insufficient difference of coil voltage
Fault degree adjusts voltage by electric voltage regulator and completes.It is as shown in table 4 that concrete mode is arranged in closing coil fault degree:
The setting of 4 closing coil fault degree of table
Wherein: the insufficient distance of iron core stroke is the difference with standard iron core stroke, when closing iron core stroke it is insufficient away from
When from being greater than 4.0mm, breaker closing failure;U1For closing coil voltage rating, it is reliable which releases energy electromagnet
Operation voltage need to be greater than 80%U1;Fault severity level: d1 < d2 < d3 < d4.
It is diagnosed using the circuit breaker failure based on closing coil electric current and degree assessment method is to above-mentioned simulated failure state
Under frame-type circuit breaker carry out fault diagnosis and scale evaluation, the specific steps are as follows:
The first step uses the closing coil electric current in the frame-type circuit breaker fault detection system acquisition breaker closing stage
Signal, the sample frequency that data collecting card is arranged is 20kHz/s, sampling time 0.07s.Circuit-breaker switching on-off coil is acquired to return
Road is in normal, iron core bite, mechanical structure bite, iron core stroke is insufficient, the closing coil under less than 5 kinds states of coil voltage
Current signal data Shc(t), 30 groups of current signals are acquired under every kind of working condition, wherein 15 groups are used for more class Support Vectors
The training of machine, 15 groups for testing the classification accuracy of multi-category support vector machines, i.e. r=15;Typical case under 5 kinds of working conditions
Closing coil current signal as shown in fig. 7, respectively be normal, iron core bite, mechanical structure bite, iron core stroke it is insufficient,
Coil voltage deficiency failure, electricity when wherein the closing coil current signal of iron core stroke not foot be fault degree is 4.0mm
Flow waveform, the closing coil current signal of coil voltage foot be not fault degree is 80%U1When current waveform, Cong Tuzhong
As can be seen that closing coil current signal is the AC signal that peak valley amplitude gradually changes, 3 troughs, 2 wave crests are shared.
Second step, to the closing coil current signal S of acquisitionhc(t) noise suppression preprocessing is carried out using Mean Filtering Algorithm, obtained
To denoising closing coil current signal S 'hc(t);The filter window of mean filter Denoising Algorithm is 50.With the conjunction under normal condition
For lock coil current signal, the effect for denoising front and back is as shown in Figure 8.
Third step, to denoising closing coil electric current S 'hc(t) material time is carried out, current amplitude parameter attribute extracts.I.e. pair
The time t in 4 stages11, t12, t13, t14Continue total time t with electric current15, 3 trough I of electric current11、I13、I15, 2 wave crests
I12、I14It extracts, material time and current amplitude parameter attribute such as table in closing coil current signal time domain under different conditions
Shown in 5.
Material time and current amplitude parameter attribute in closing coil current signal time domain under 5 different conditions of table
4th step carries out set empirical mode decomposition energy Moment Feature Extraction to closing line loop current signal, i.e., global special
Sign is extracted;By taking the closing coil current signal under normal condition as an example, EEMD is carried out, each component after decomposition is as shown in figure 9, altogether
It is decomposed into 11 components, i.e. mh=11, therefore feature vector components makes dimension is 11 dimensions in time-frequency domain, to each component according to formula
(11) energy Moment Feature Extraction is carried out, EEMD energy moment characteristics such as table 6 in closing coil current signal time-frequency domain under different conditions
It is shown.
EEMD energy moment characteristics (unit/J) in switching current signal time-frequency domain under 6 different conditions of table
5th step, the foundation of fault feature vector;Through the available closing coil current signal of third step in time domain
The local feature vectors component dimension of formation is 10, passes through the available closing coil current signal shape in time-frequency domain of the 4th step
At global characteristics component of a vector dimension be 11, then the dimension of fault feature vector be 21.Set Principal Component Analysis it is main at
Dividing percentage is 95%, then the dimension after Principal Component Analysis dimensionality reduction is 9.
It needs to construct sample dimensionality reduction eigenmatrix simultaneously and carries out the training of closing coil loop fault diagnostic model, closing coil
Circuit shares 5 kinds of different states and is diagnosed, and 30 groups of signals are acquired under every kind of state, wherein 15 groups are used to train,
15 groups are used to test, and the dimension after the feature vector dimensionality reduction that every group of current signal extracts is 9, then the training sample dimensionality reduction formed is special
The dimension for levying matrix is 75 × 9, and the dimension of test sample dimensionality reduction eigenmatrix is 75 × 9.
6th step constructs closing coil circuit multi-category support vector machines and is used for state recognition with the 7th step of embodiment 1.
Wherein in gaussian radial basis function parameter penalty factor and nuclear parameter σ respectively [2-15..., 215], [2-7,…,27] use
Cross validation mode obtains, and C=0.3711, σ=1.6352 are obtained after optimization;Rule of thumb method is arranged Polynomial kernel function parameter d
For d=3;Multilayer perceptron kernel functional parameter γ takes the derivative of sample classification number, i.e. γ=0.2.List based on different kernel functions
The fault diagnosis discrimination of kernel support vectors machine and the multi-kernel support vector machine based on different characteristic is as shown in table 7.
7 closing coil electric current difference kernel function SVM discrimination of table
During determining weight coefficient using GA algorithm optimization, population quantity 20, heredity terminates the number of iterations and is
100, Figure 10 be the fitness curve based on genetic algorithm optimizing parameter.Evolution of Population is best to or so 30 generations as seen from the figure
Fitness value is just basically stable at 92.0%.
Figure 11 is the variation tendency that GA algorithm carries out each kernel function weight coefficient in weight coefficient optimization process.It can be seen that
For each weight coefficient there are the process of a dynamic adjustment, gaussian radial basis function core remains at higher weight, polynomial kernel power
Value shows the trend of rising, but finally tends towards stability, and multilayer perceptron core constantly learns in preceding 30 generation, and final weight coefficient is steady
It is scheduled on 0.21 or so, algorithm convergence when population iterates to or so 30 generations exports weight coefficient optimal value.
7th step calculates fault-current signal entropy opposite with the EEMD of normal current signal;When the closing line being diagnosed to be
When there is iron core stroke deficiency and coil voltage deficiency failure in circle loop fault, need to continue fault degree assessment, i.e.,
Calculate the EEMD energy square relative entropy of different faults degree current signal under iron core stroke deficiency and coil voltage deficiency failure
Value.Its EEMD energy square relative entropy with normal switching current signal is calculated by formula (15), table 8 is that 4 are arranged in this application
Carried out respectively under kind of fault degree EEMD energy square with respect to entropy calculate as a result, every kind of result is 5 test averagings of income
Value.
EEMD energy square relative entropy under 8 combined floodgate different faults degree of table
8th step constructs closing coil loop fault degree characteristic curve.Become using linear function to fault degree variation
Gesture is fitted, and constructs fault degree characteristic curve with this, and fitting result is as shown in figure 12, matching correlation 98% with
On.Iron core stroke deficiency and the fault characteristic curve of coil voltage deficiency fitting are respectively as follows:
Iron core stroke is insufficient: y1=0.17989x+0.5502;
Coil voltage is insufficient: y2=0.87358x-0.5279;
2 show that the multi-category support vector machines of the application building can be efficiently accomplished to breaker closing through this embodiment
The fault diagnosis of wire loop, by the fault degree evaluation index EEMD energy square relative entropy of building, in conjunction with the failure of fitting
Degree characteristic curve, the achievable qualitative assessment to closing coil loop fault degree.
Embodiment 3
The present embodiment is using the frame-type circuit breaker of model DW15-1600 as subjects, the breaker mechanical structure
Visualization is strong, and suitable for the power distribution network that rated operational voltage is 380V, 50Hz, rated current 1600A, switching mechanism is bullet
Spring operating mechanism.Tripping electromagnet is beating type electromagnet, voltage 380V, 50Hz, the number of turns 4650 in opening coil circuit
Circle, enamel covered wire is through 0.21mm.Opening coil loop fault analog form are as follows: tie down mandril push jack, simulation top using elastic string
Bar suffered resistance in stroke is abnormal;Since tripping electromagnet is beating type electromagnet, armature travel is less than combined floodgate electromagnetism
Iron, in order to construct the failure of the different degrees of grade of sufficient amount, therefore the hard paper by pasting different number on armature
(0.1mm) simulates armature travel deficiency failure, and stickup quantity is more, and fault degree is bigger;Coil voltage deficiency fault simulation is closed
Brake cable circle mode is identical, i.e., the insufficient different faults degree of coil voltage adjusts voltage by electric voltage regulator and completes.Closing line
It is as shown in table 9 to enclose fault degree setting concrete mode:
The setting of 9 separating brake fault degree of table
Wherein: the insufficient distance of armature travel is the difference with standard armature travel, when separating brake armature travel it is insufficient away from
When from being greater than 2.0mm, breaker open operation failure;U2For opening coil voltage rating, the model breaker shunt release can
70%U need to be greater than by operation voltage2;Fault severity level: d1 < d2 < d3 < d4.
It is diagnosed using the circuit breaker failure based on opening coil electric current and degree assessment method is to above-mentioned simulated failure state
Under frame-type circuit breaker carry out fault diagnosis and scale evaluation, the specific steps are as follows:
The first step uses the opening coil electric current in the frame-type circuit breaker fault detection system acquisition breaker open operation stage
Signal;Circuit-breaker switching on-off wire loop is acquired in normal, armature resistance is abnormal, armature travel is insufficient, coil voltage is less than 4
Opening coil current signal data S under kind statefc(t), 30 groups of current signals are acquired under every kind of working condition, wherein 15 groups
For the training of multi-category support vector machines, 15 groups for testing the classification accuracy of multi-category support vector machines, i.e. r=15;
Typical closing coil current signal under 4 kinds of working conditions is as shown in figure 13, respectively is normal, armature resistance
It is abnormal, armature travel is insufficient, coil voltage deficiency failure, wherein the opening coil current signal of armature travel not foot is failure
The opening coil current signal of current waveform when degree is 2.0mm, coil voltage foot be not fault degree is 72%U2When
Current waveform share 2 from the waveforms it can be seen that opening coil current signal is the AC signal that gradually changes of peak valley amplitude
A trough, 2 wave crests.
Second step, to the opening coil current signal S of acquisitionfc(t) noise suppression preprocessing is carried out using Mean Filtering Algorithm, obtained
To denoising opening coil current signal S 'fc(t);The filter window of mean filter Denoising Algorithm is 50.With point under normal condition
For lock coil current signal, the effect for denoising front and back is as shown in figure 14.
Third step, to denoising opening coil electric current S 'fc(t) material time, current amplitude parameter extraction are carried out.I.e. to 4
The time t in stage21, t22, t23, t24Continue total time t with opening coil electric current25, 2 trough I of electric current21、I23, 2 wave crests
I22、I24It extracts, material time and current amplitude parameter attribute be such as in the opening coil current signal time domain under different conditions
Shown in table 10.
Material time and current amplitude parameter attribute in separating brake current signal time domain under 10 different conditions of table
4th step carries out set empirical mode decomposition energy Moment Feature Extraction to opening coil current signal;With normal shape
For opening coil current signal under state, EEMD decomposition is carried out, each component after decomposition is as shown in figure 15, is decomposed into 11 altogether
Component, i.e., m at this timef=11, energy Moment Feature Extraction is carried out according to formula (11) to each component, opening coil electricity under different conditions
It is as shown in table 11 to flow EEMD energy moment characteristics in signal time-frequency domain.
EEMD energy moment characteristics (unit/J) in separating brake current signal time-frequency domain under 11 different conditions of table
5th step, the foundation of fault feature vector.It is formed in time domain by the available opening coil electric current of third step
Local feature vectors component dimension be 9, the overall situation formed in time-frequency domain by the available opening coil electric current of the 4th step
Feature vector components makes dimension is 11, then the dimension of fault feature vector is 20.Set the principal component percentage of Principal Component Analysis
It is 95%, then the dimension after principal component analysis dimensionality reduction is 8.
It needs to construct sample dimensionality reduction eigenmatrix simultaneously and carries out the training of opening coil loop fault diagnostic model and test, point
Lock wire loop shares 4 kinds of different states and is diagnosed, and 30 groups of signals are acquired under every kind of state, wherein 15 groups are used for
Training, 15 groups are used to test, and the dimension after the feature vector dimensionality reduction that every group of current signal extracts is 8, then the training sample formed
Dimensionality reduction eigenmatrix dimension is 60 × 8, and test sample dimensionality reduction eigenmatrix dimension is 60 × 8.
6th step constructs opening coil circuit multi-category support vector machines and is used for state recognition with the 6th step of embodiment 2.
Wherein in gaussian radial basis function parameter penalty factor and nuclear parameter σ respectively [2-15..., 215], [2-7,…,27] use
Cross validation mode obtains, and C=1.3367, σ=0.1524 are obtained after optimization;Rule of thumb method is arranged Polynomial kernel function parameter d
For d=3;Multilayer perceptron kernel functional parameter γ takes the derivative of sample classification number, i.e. γ=0.25.List based on different kernel functions
The fault diagnosis discrimination of kernel support vectors machine and the multi-kernel support vector machine based on different characteristic is as shown in table 12.
12 opening coil electric current difference kernel function SVM discrimination of table
GA algorithm population iteration to 40 generations or so parameter optimal adaptation degree during optimization reaches 93.3%, multicore power
Value coefficient also tends towards stability.
7th step calculates opening coil fault-current signal entropy opposite with the EEMD of normal current signal.When being diagnosed to be
Opening coil loop fault in when there is armature travel deficiency and coil voltage deficiency failure, need to continue fault degree
Assessment.The EEMD energy square relative entropy of opening coil current signal and normal switching current signal, table 13 are calculated by formula (15)
EEMD energy square is carried out under 4 kinds of fault degrees respectively with respect to entropy calculating as a result, every kind of result is equal to be arranged in this application
For 5 test averaging of income values.
EEMD energy square relative entropy under 13 combined floodgate different faults degree of table
8th step constructs opening coil loop fault degree characteristic curve.Become using linear function to fault degree variation
Gesture is fitted, and constructs fault degree characteristic curve with this, and fitting result is as shown in figure 16, matching correlation 99% with
On.Armature travel deficiency and the fault characteristic curve of coil voltage deficiency fitting are respectively as follows:
Armature travel is insufficient: y3=0.86404x+0.40975;
Coil voltage is insufficient: y4=1.41578x-0.16885;
3 verifyings show that the multi-category support vector machines of the application building can be efficiently accomplished to breaker through this embodiment
The diagnosis of opening coil loop fault, by the fault degree evaluation index EEMD energy square relative entropy of building, in conjunction with fitting
Fault degree characteristic curve, the achievable qualitative assessment to opening coil loop fault degree.
Above-mentioned steps are all made of LabVIEW and MATLAB software realization.
LabVIEW and MATLAB software used in above-mentioned the present embodiment is known to those skilled in the art
's.
Percentage in examples detailed above is numerical percentage.
The present invention does not address place and is suitable for the prior art.
Claims (2)
1. a kind of frame-type circuit breaker fault diagnosis and degree assessment method, this method based on operation attachment electric current first determines whether
Wait diagnose and assess the working stage of breaker, the working stage includes energy storage stage, combined floodgate stage and separating brake stage, then
Fault diagnosis is carried out to energy storage stage using energy storage motor electric current, fault diagnosis is carried out to the stage of combined floodgate using closing coil electric current
And scale evaluation and fault diagnosis and scale evaluation are carried out to the separating brake stage using opening coil electric current;Wherein,
One, fault diagnosis is carried out to energy storage stage using energy storage motor electric current, comprising the following steps:
The first step acquires energy storage motor electric current S of the breaker under different working conditioncc(t), setting energy storage motor is shared just
Often, transmission gear bite, spring bite and spring fall off 4 kinds of different working conditions, and the storage of r group is acquired under every kind of working condition
It can motor current signal;
Second step is carried out noise suppression preprocessing to the energy storage motor current signal of acquisition, is gone using mean filter Denoising Algorithm
Make an uproar energy storage motor current signal S 'cc(t);
Third step, by Hilbert envelope method to denoising energy storage motor current signal S 'cc(t) envelope is carried out to seek, according to
Formula (3) acquires the amplitude A (t) of denoising energy storage motor current signal, and A (t) is to denoise energy storage motor current signal S 'cc(t)
Hilbert envelope line;
4th step, by Hilbert envelope line to denoising energy storage motor current signal S 'cc(t) material time and electric current width are carried out
Value parameter is extracted, by current maxima I when starting in energy storage motor current signal1, motor stabilizing operation current maxima I2
And maximum value I2Corresponding moment t1, energy storage motor electric current continue total time t2As the local fault feature of energy storage motor electric current, obtain
The local feature vectors component T formed in time domain to energy storage motor electric current1=[I1, I2, t1, t2];
5th step, to denoising energy storage motor current signal S 'cc(t) set empirical mode decomposition EEMD energy moment characteristics are carried out to mention
It takes, i.e., global characteristics extract;
Set empirical mode decomposition EEMD energy Moment Feature Extraction comprises the concrete steps that:
5-1., which is determined, decomposes number M and the noise amplitude to be added, by white noise signal nm(t) it is added to and to be decomposed
Denoise energy storage motor current signal S 'cc(t) in, new signal x to be decomposed is obtained according to formula (5)m(t):
xm(t)=S 'cc(t)+nm(t) (5)
5-2. determines new signal x to be decomposedm(t) Local modulus maxima and minimum point, will be whole using cubic spline interpolation
Maximum point be attached, formed coenvelope line;Similarly, cubic spline interpolation is carried out to all minimum points, obtains lower packet
Winding thread;
5-3. finds out the average value of all envelope points of coenvelope line and lower envelope line, is denoted as m1;
5-4. is by new signal x to be decomposedm(t) m1 is subtracted, obtains one-component h according to formula (6)1If: h1Meet intrinsic mode
IMF component condition, then h1It is just new signal x to be decomposedm(t) the first intrinsic mode for meeting intrinsic mode IMF component condition
IMF component, first intrinsic modal components for meeting intrinsic mode IMF component condition are denoted as c1, enter step 5-6;
h1=xm(t)-m1 (6)
If 5-5. h1It is unsatisfactory for intrinsic mode IMF component condition, and includes the odd function of other different scales, then by h1As original
Beginning signal repeats step 5-2~5-4, obtains h1Upper and lower envelope, calculate h1Coenvelope average value, be denoted as m11;So
After judge h11=h1-m11Intrinsic mode IMF component condition can be met to continue cycling through if be not met by, until h1k
=h1(k-1)-m1kUntil meeting intrinsic mode IMF component condition;By h1kAs new signal x to be decomposedm(t) first is full
The intrinsic modal components of the intrinsic mode IMF component condition of foot, are denoted as c1;
5-6. is c1From new signal x to be decomposedm(t) it independently goes out in, obtains r according to formula (7)1:
r1=xm(t)-c1 (7)
5-7. is by r1Step 5-2~5-5 is repeated as original signal to be decomposed, obtains second intrinsic mode IMF component c2;
5-8. repeats step 5-2~5-7, obtains new signal x to be decomposed according to formula (8)m(t) m intrinsic mode IMF components,
As residual components rmWhen for monotonic function, then stop repeating step 5-2~5-7, complete it is primary decompose,
At this point, new signal x to be decomposedm(t) it is indicated with formula (9);
In formula, rmIt is survival function, the average tendency of representation signal;ciFor i-th of intrinsic mode IMF component;
The amplitudes noise signal such as new is added to the denoising energy storage motor current signal S ' decomposed by 5-9.cc(t) in,
It is repeated in step 5-2~5-8 M times, obtains the intrinsic mode IMF vector sequence { c of M groupn,l, cn,lThe l decomposed for n-th
A intrinsic mode IMF component, n=1,2 ... ..., M, l=1,2 ... ..., m;
5-10. seeks the intrinsic mode IMF vector sequence { c of M group according to formula (10)n,lFirst of intrinsic mode IMF component M times point
The average value of solutionIt willAs first of intrinsic mode IMF component of set empirical mode decomposition EEMD, i.e., by set warp
Testing the intrinsic mode IMF component that mode decomposition EEMD is obtained is
5-11. is according to formula (11) to the intrinsic mode IMF component obtained by set empirical mode decomposition EEMDIt carries out taking energy
Square calculates:
In formula: Δ t is sampling time interval, n1For sampling number, k1=1,2 ..., n1;
5-12. new signal x to be decomposedm(t) m intrinsic mode IMF components are decomposed into, acquire m energy square E altogetherm, then energy storage
The global characteristics component of a vector T that current of electric is formed in time-frequency domain2=[E1, E2... ..., Em];
6th step establishes energy storage motor fault feature vector, constructs energy storage motor sample dimensionality reduction eigenmatrix:
The local feature vectors component T that energy storage motor electric current is formed in time domain is acquired by the 4th step1=[I1, I2, t1, t2] and
5th step acquires the global characteristics component of a vector T that energy storage motor electric current is formed in time-frequency domain2=[E1, E2... ..., Em], by two
A feature vector components makes fusion constructs are indicated at energy storage motor fault feature vector T with formula (12):
T=[T1 T2] (12)
The dimension of this energy storage motor fault feature vector be 4+m dimension, using Principal Component Analysis to energy storage motor fault signature to
Amount carries out dimension-reduction treatment, and the dimension of the feature vector T ' after dimensionality reduction is λ dimension;After energy storage motor dimensionality reduction under all working state
Feature vector constitutes energy storage motor sample dimensionality reduction eigenmatrix, then the dimension of energy storage motor sample dimensionality reduction eigenmatrix is 4r × λ;
7th step constructs energy storage motor multi-category support vector machines, carries out fault diagnosis:
The total class number of the energy storage motor working condition identified is 4, and the data under some working condition are regarded as positive class,
Data under remaining working condition regard negative class as, the energy storage motor sample dimensionality reduction eigenmatrix obtained using the 6th step, using " a pair of
It is more " method three Sub-SVMs of building, three Sub-SVMs are again with Polynomial kernel function, gaussian radial basis function
With constitute multi-kernel support vector machine based on multilayer perceptron kernel function, the nuclear parameter of each kernel function is determined using empirical method, and
Multicore weight coefficient is optimized using genetic algorithm, and then obtains multi-category support vector machines, is supported by more classification
Vector machine identifies the working condition of breaker energy storage motor;
When breaker energy storage motor needs to carry out fault diagnosis, energy storage motor current signal is acquired, extracts and stores up according to the 6th step
Then energy electrical fault feature vector will be used as input quantity, be input to energy storage motor after energy storage motor fault feature vector dimensionality reduction
The working condition identification to energy storage motor can be completed in multi-category support vector machines;
Two, fault diagnosis and scale evaluation are carried out to the stage of combined floodgate using closing coil electric current, comprising the following steps:
The first step acquires closing coil electric current S of the breaker under different working conditionhc(t), setting closing coil circuit is shared
Normally, iron core bite, mechanical structure bite, the working condition that iron core stroke is insufficient and coil voltage is different less than 5 kinds, every kind of work
Make to acquire r group current signal under state;
Second step, to closing line loop current Shc(t) denoising is carried out, denoising closing coil electric current S ' is obtainedhc(t), denoising is calculated
Method is consistent with Denoising Algorithm in the processing of energy storage motor current signal;
Third step, to denoising closing coil electric current S 'hc(t) material time, current amplitude parameter extraction are carried out, closing coil is obtained
The local feature vectors component T that current signal is formed in time domainh1;
Closing coil Current wave-shape characteristic is analyzed, the process that iron core acts is divided into four-stage: first stage is opened for iron core
Begin to move, coil magnetization, but does not generate magnetic saturation always;Second stage is coil magnetic saturation, and the exciting current of generation is tool
There is the peaked wave of 1/4 periodic symmetry feature;Three phases are electric current approximation steady state stage, coil current waveform and the same phase of magnetic flux
Position, peak valley amplitude are constant;Fourth stage is the current waveform of auxiliary switch separation phase, and electric current shows as being reduced rapidly, until
Electric current thoroughly disappears;The time t in aforementioned four stage11、t12、t13、t14It can be sent out since breaker is in different working conditions
Raw corresponding variation, while closing coil electric current continues total time t15Also it can change;Since closing coil is alternating voltage
Power supply, the coil current of generation is also AC wave shape, and 3 trough I occurs altogether in closing coil electric current11、I13、I15, 2 wave crest I12、
I14, it is hereby achieved that the local feature vectors component T that closing coil current signal is formed in time domainh1=[t11, t12, t13,
t14, t15, I11, I12, I13, I14, I15];
4th step will denoise closing coil electric current S 'hc(t) set empirical mode decomposition EEMD energy Moment Feature Extraction is carried out, is gone
Make an uproar closing coil electric current S 'hc(t) it is decomposed into mhA intrinsic mode IMF component, acquires m altogetherhA energy square, then closing coil electric current
The global characteristics component of a vector T that signal is formed in time-frequency domainh2=[E1h, E2h... ..., Emh];Specific steps are the same as in energy storage stage
Gather the method for empirical mode decomposition EEMD energy Moment Feature Extraction;
5th step establishes closing coil fault feature vector, constructs closing coil sample dimensionality reduction eigenmatrix:
By local feature vectors component T of the closing coil current signal in time domainh1=[t11, t12, t13, t14, t15, I11, I12,
I13, I14, I15] and global characteristics component of a vector T in time-frequency domainh2=[E1h, E2h... ..., Emh] fusion constructs closing coil
Fault feature vector Th=[Th1, Th2], the dimension of thus obtained closing coil fault feature vector is 10+mh, through principal component
The closing coil feature vector T ' obtained after analytic approach dimensionality reductionhDimension be λhIt ties up, the closing coil dimensionality reduction under all working state
Rear feature vector constitutes closing coil sample dimensionality reduction eigenmatrix, the dimension of closing coil sample dimensionality reduction eigenmatrix be 5r ×
λh;
6th step constructs closing coil circuit multi-category support vector machines, carries out fault diagnosis:
The total class number of closing coil loop works state identified is 5, special using above-mentioned closing coil sample dimensionality reduction
Levy matrix, construct 4 Sub-SVMs, 4 Sub-SVMs again with Polynomial kernel function, gaussian radial basis function and
Multi-kernel support vector machine is constituted based on multilayer perceptron kernel function, the nuclear parameter of each kernel function is determined using empirical method, and is adopted
Multicore weight coefficient is optimized with genetic algorithm, and then obtains multi-category support vector machines, by the more classification support to
Amount machine identifies the working condition of breaker closing coil, constructs closing coil circuit multi-category support vector machines, uses
Closing coil circuit multi-category support vector machines carry out fault diagnosis to breaker closing wire loop;
7th step determines closing coil fault degree evaluation index:
As the denoising closing coil electric current S ' for determining input by closing coil circuit multi-category support vector machineshcIt (t) is failure
It when current signal, is diagnosed to be in closing coil loop fault and iron core stroke deficiency and coil voltage deficiency failure occurs, with set
Empirical mode decomposition EEMD energy square, as fault degree evaluation index, carries out fault degree assessment with respect to entropy;
Steps are as follows with respect to the calculating of entropy for the set empirical mode decomposition EEMD energy square of closing coil:
B7-1. by acquiring denoising closing coil electric current S ' in the 4th stephc(t) mhThe m of a intrinsic mode IMF componenthA energy
Square, i.e. E1h, E2h... ..., Emh;Then m is found out according to formula (13)hThe summation E of a energy squareH:
B7-2. kth being found out according to formula (14) ' a energy square accounts for the specific gravity of energy square summation are as follows:
Then denoise closing coil electric current S 'hc(t) energy square distribution character matrix are as follows: P=(p1,p2,…,pk′,…,pmh);
It similarly calculates and denoises closing coil current signal S ' under normal conditionhz(t) energy square distribution character matrix are as follows: P '=
(p′1,p′2,…,p′k′,…,p′mh);
B7-3. failure is obtained by formula (15) denoise closing coil current signal S 'hc(t) and closing coil is denoised under normal condition
Current signal S 'hz(t) set empirical mode decomposition EEMD energy square relative entropy:
8th step constructs closing coil loop fault degree characteristic curve:
Every kind of failure is respectively provided with 4 kinds of different fault degrees, the iron core stroke under calculating in various degree using the 7th step is insufficient
Set empirical mode decomposition EEMD energy square corresponding with coil voltage deficiency failure with respect to entropy, then using linear function into
Row fitting, linear function are formula (16), obtain closing coil loop fault degree characteristic curve,
Y=a1x+a2 (16)
Wherein: x is fault degree, and y is set empirical mode decomposition EEMD energy square with respect to entropy, a1, a2For constant;
Respectively obtain the corresponding closing coil loop fault degree characteristic of less than two kinds failures of iron core stroke deficiency and coil voltage
Curve carries out denoising to closing coil current signal to be assessed through second step, so when needing to carry out fault degree assessment
The set empirical mode decomposition EEMD energy square of denoising closing coil current signal is calculated according to the 7th step afterwards with respect to entropy, is substituted into
In above-mentioned corresponding closing coil loop fault degree characteristic curve, you can get it, and corresponding closing coil fault degree is quantitative
Value;
Three, fault diagnosis and scale evaluation are carried out to the separating brake stage using opening coil electric current, comprising the following steps:
The first step acquires opening coil electric current S of the breaker under different faults statefc(t), setting opening coil circuit is shared
Normally, the working condition that armature resistance is abnormal, armature travel is insufficient and coil voltage is different less than 4 kinds, under every kind of working condition
Acquire r group current signal;
Second step, to opening coil electric current Sfc(t) denoising is carried out, denoising opening coil electric current S ' is obtainedfc(t), denoising is calculated
Method is consistent with Denoising Algorithm in the processing of energy storage motor current signal;
Third step, to denoising opening coil electric current S 'fc(t) material time, current amplitude parameter extraction are carried out, opening coil is obtained
The local feature vectors component T that current signal is formed in time domainf1;It analyzes opening coil electric current and passes through characteristic, generate 5 keys
Time point t21, t22, t23, t24, t25, there are 2 trough I altogether in opening coil electric current21、I23, 2 wave crest I22、I24, obtain separating brake
The local feature vectors component T that coil current is formed in time domainf1=[t21, t22, t23, t24, t25, I21, I22, I23, I24];
4th step will denoise opening coil electric current S 'fc(t) set empirical mode decomposition EEMD energy Moment Feature Extraction is carried out, is gone
Opening coil of making an uproar electric current S 'fc(t) it is decomposed into mfA intrinsic mode IMF component, acquires m altogetherfA energy square, then opening coil electric current
The global characteristics component of a vector T that signal is formed in time-frequency domainf2=[E1f, E2f... ..., Emf];
5th step establishes opening coil fault feature vector, constructs opening coil sample dimensionality reduction eigenmatrix:
By local feature vectors component T of the opening coil current signal in time domainf1=[t21, t22, t23, t24, t25, I21, I22,
I23, I24] and global characteristics component of a vector T in time-frequency domainf2=[E1f, E2f... ..., Emf] fusion constructs opening coil failure
Feature vector Tf=[Tf1, Tf2], the dimension of thus obtained closing coil fault feature vector is 9+mf, through Principal Component Analysis
The feature vector T ' obtained after dimensionality reductionfDimension be λfDimension;The feature vector structure after opening coil dimensionality reduction under all working state
At opening coil sample dimensionality reduction eigenmatrix, the dimension of closing coil sample dimensionality reduction eigenmatrix is 4r × λf;
6th step constructs opening coil circuit multi-category support vector machines, carries out fault diagnosis:
The total class number of opening coil loop works state identified is 4, special using above-mentioned opening coil sample dimensionality reduction
Levy matrix, construct 3 Sub-SVMs, 3 Sub-SVMs again with Polynomial kernel function, gaussian radial basis function and
Multi-kernel support vector machine is constituted based on multilayer perceptron kernel function, the nuclear parameter of each kernel function is determined using empirical method, and is adopted
Multicore weight coefficient is optimized with genetic algorithm, and then obtains multi-category support vector machines, by the more classification support to
Amount machine identifies the working condition of breaker open operation coil, constructs opening coil circuit multi-category support vector machines, passes through
Opening coil circuit multi-category support vector machines carry out fault diagnosis to breaker open operation wire loop;
7th step determines opening coil fault degree evaluation index:
As the denoising opening coil electric current S ' for determining input by opening coil circuit multi-category support vector machinesfcIt (t) is event
It when hindering current signal, is diagnosed to be in opening coil loop fault and armature travel deficiency and coil voltage deficiency failure occurs, with collection
Empirical mode decomposition EEMD energy square relative entropy is closed as fault degree evaluation index, carries out fault degree assessment;
Steps are as follows with respect to the calculating of entropy for the set empirical mode decomposition EEMD energy square of opening coil:
By acquiring denoising opening coil electric current S ' in the 4th stepfc(t) mfThe m of a intrinsic mode IMF componentfA energy square, i.e.,
E1f, E2f... ..., Emf;Then according toFind out mfThe summation E of a energy squareF;
Then basisK '=1f,2f,…,mfFinding out kth ' a energy square accounts for the specific gravity of energy square summation;
Then denoise opening coil electric current S 'fc(t) energy square distribution character matrix are as follows: P=(p1f,p2f,…,pk′,…,pmf);
It similarly calculates and denoises opening coil current signal S ' under normal conditionfc(t) energy square distribution character matrix are as follows: P '=
(p′1f,p′2f,…,p′k′,…,p′mf);
Finally byObtain failure denoising opening coil current signal S 'fc
(t) and under normal condition closing coil current signal S ' is denoisedfc(t) set empirical mode decomposition EEMD energy square relative entropy;
8th step constructs opening coil loop fault degree characteristic curve, every kind of failure is respectively provided with 4 kinds of different failure journeys
Degree, the corresponding set empirical modal of armature travel deficiency and coil voltage deficiency failure under being calculated in various degree using the 7th step
EEMD energy square is decomposed with respect to entropy, is then fitted using linear function, linear function is formula (17), obtains opening coil
Loop fault degree characteristic curve;
Y=b1x+b2 (17)
Wherein: x is fault degree, and y is set empirical mode decomposition EEMD energy square with respect to entropy, b1, b2For constant;
By the opening coil circuit multi-category support vector machines of above-mentioned building complete to occur in opening coil circuit it is normal,
Armature resistance is abnormal, armature travel is insufficient, less than 4 kinds working conditions of coil voltage are diagnosed;When the failure being diagnosed to be is rank
It when iron stroke is insufficient or coil voltage deficiency, needs to continue fault degree assessment, passes through separating brake after calculating denoising to be assessed
Coil current set empirical mode decomposition EEMD energy square relative entropy, the opening coil loop fault degree for substituting into above-mentioned foundation are special
In linearity curve, the qualitative assessment to opening coil fault degree can be completed.
2. the frame-type circuit breaker fault diagnosis and scale evaluation side according to claim 1 based on operation attachment electric current
Method, it is characterised in that specific step is as follows for building energy storage motor multi-category support vector machines:
A7-1. the feature vector under normal operating conditions in sample dimensionality reduction eigenmatrix is classified as positive class, under remaining working condition
Feature vector be classified as negative class, construct normal Sub-SVM;Transmission gear bite in sample dimensionality reduction eigenmatrix is worked
Feature vector under state is classified as positive class, and the feature vector under remaining working condition is classified as negative class, building transmission gear bite
Support vector machines;Feature vector under spring bite working condition in sample dimensionality reduction eigenmatrix is classified as positive class, remaining work
Feature vector under state is classified as negative class, constructs spring bite Sub-SVM;If above-mentioned 3 Sub-SVMs cannot
Effectively identification, then energy storage motor is in spring and falls off working condition;
A7-2. corresponding feature vector in sample dimensionality reduction eigenmatrix is respectively adopted to above-mentioned 3 Sub-SVMs to instruct
Practice, rule of thumb method is arranged Polynomial kernel function parameter d in training process, and gaussian radial basis function parameter uses cross validation
Mode obtains, and multilayer perceptron kernel functional parameter takes the derivative of sample classification number;Multicore supporting vector is determined in conjunction with genetic algorithm
The core weight coefficient of machine, obtains optimal output node, so that the operating mode recognition accuracy of Sub-SVM is reached highest, i.e.,
Obtain the energy storage motor multi-category support vector machines based on breaker energy storage motor current signal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710450109.9A CN107219457B (en) | 2017-06-15 | 2017-06-15 | Frame-type circuit breaker fault diagnosis and degree assessment method based on operation attachment electric current |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710450109.9A CN107219457B (en) | 2017-06-15 | 2017-06-15 | Frame-type circuit breaker fault diagnosis and degree assessment method based on operation attachment electric current |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107219457A CN107219457A (en) | 2017-09-29 |
CN107219457B true CN107219457B (en) | 2019-05-31 |
Family
ID=59949541
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710450109.9A Active CN107219457B (en) | 2017-06-15 | 2017-06-15 | Frame-type circuit breaker fault diagnosis and degree assessment method based on operation attachment electric current |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107219457B (en) |
Families Citing this family (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107860587B (en) * | 2017-11-08 | 2019-07-02 | 南京康尼机电股份有限公司 | Train-door system sub-health state early warning based on multi-feature fusion and diagnostic method |
CN108303253B (en) * | 2017-12-06 | 2019-10-18 | 华南理工大学 | Bearing initial failure recognition methods based on long short-term memory Recognition with Recurrent Neural Network |
CN108614940A (en) * | 2018-05-02 | 2018-10-02 | 湖南大学 | A kind of permanent magnet synchronous motor performance degradation assessment method and system |
CN108872772B (en) * | 2018-07-12 | 2020-06-16 | 武汉科技大学 | MMC submodule open-circuit fault detection method based on hybrid nuclear support tensor machine |
CN109031114A (en) * | 2018-09-29 | 2018-12-18 | 华南理工大学 | A kind of modeling of spring actuator mechanism circuit-breaker and method for diagnosing faults |
CN110208688A (en) * | 2019-05-09 | 2019-09-06 | 石家庄科林电气股份有限公司 | A kind of failure pre-judging method of breaker energy storage motor |
CN110174612A (en) * | 2019-06-10 | 2019-08-27 | 珠海东帆科技有限公司 | Low-voltage circuit breaker fault early-warning detection method |
CN110887644A (en) * | 2019-10-28 | 2020-03-17 | 深圳供电局有限公司 | Load switch gear state detection method and system |
CN111289888B (en) * | 2019-12-11 | 2022-04-08 | 嘉兴恒创电力集团有限公司博创物资分公司 | High-voltage circuit breaker state detection and fault diagnosis method based on big data technology |
CN111289892B (en) * | 2020-04-07 | 2022-07-01 | 南京机电职业技术学院 | High-voltage circuit breaker fault diagnosis method based on PCA-SVM |
CN112924208B (en) * | 2021-02-01 | 2023-04-11 | 上海三一重机股份有限公司 | Excavator fault diagnosis system and diagnosis method |
CN114578195A (en) * | 2022-03-03 | 2022-06-03 | 海南电网有限责任公司儋州供电局 | IEMD and GA-BP based online intelligent diagnosis method for insulation fault of switching-on and switching-off coils of ring main unit switch |
CN115862681B (en) * | 2023-03-01 | 2023-05-12 | 深圳市新凯来技术有限公司 | Rotor rub-impact fault acoustic signal diagnosis method and device and electronic equipment |
CN116754934B (en) * | 2023-05-22 | 2024-02-23 | 杭州轨物科技有限公司 | Mechanical characteristic fault diagnosis method for high-voltage circuit breaker |
CN116484308B (en) * | 2023-06-25 | 2023-09-29 | 火眼科技(天津)有限公司 | Data acquisition method based on edge self-adaptive calculation |
CN117131336B (en) * | 2023-10-26 | 2024-01-30 | 深圳市欧康精密技术有限公司 | Data processing method for electronic equipment connector |
CN117725388B (en) * | 2024-02-07 | 2024-05-03 | 国网山东省电力公司枣庄供电公司 | Adjusting system and method aiming at ground fault information |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0442726A (en) * | 1990-06-06 | 1992-02-13 | Meidensha Corp | Ground fault indicator for distribution line |
CN2430804Y (en) * | 2000-06-01 | 2001-05-16 | 西安科技学院 | Intelligent monitoring protector for breaker and motor fault |
CN103292989A (en) * | 2013-06-17 | 2013-09-11 | 广州供电局有限公司 | Breaker mechanical characteristic performance testing method and system |
CN105445657A (en) * | 2015-11-26 | 2016-03-30 | 国家电网公司 | Breaker operating mechanism state diagnosis method based on grey relational analysis |
CN105606997A (en) * | 2016-02-26 | 2016-05-25 | 国家电网公司 | Mechanical fault diagnosis method of high voltage breaker operation mechanism for electric power system |
CN205539382U (en) * | 2016-04-05 | 2016-08-31 | 山东科技大学 | High voltage circuit breaker status monitoring device |
-
2017
- 2017-06-15 CN CN201710450109.9A patent/CN107219457B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPH0442726A (en) * | 1990-06-06 | 1992-02-13 | Meidensha Corp | Ground fault indicator for distribution line |
CN2430804Y (en) * | 2000-06-01 | 2001-05-16 | 西安科技学院 | Intelligent monitoring protector for breaker and motor fault |
CN103292989A (en) * | 2013-06-17 | 2013-09-11 | 广州供电局有限公司 | Breaker mechanical characteristic performance testing method and system |
CN105445657A (en) * | 2015-11-26 | 2016-03-30 | 国家电网公司 | Breaker operating mechanism state diagnosis method based on grey relational analysis |
CN105606997A (en) * | 2016-02-26 | 2016-05-25 | 国家电网公司 | Mechanical fault diagnosis method of high voltage breaker operation mechanism for electric power system |
CN205539382U (en) * | 2016-04-05 | 2016-08-31 | 山东科技大学 | High voltage circuit breaker status monitoring device |
Non-Patent Citations (2)
Title |
---|
基于电机电流分析的万能式断路器机械故障诊断;孙曙光 等;《仪器仪表学报》;20170430;第38卷(第4期);第952-960页 |
框架式断路器关键附件电气参数检测与试验技术研究;孙曙光 等;《电测与仪表》;20160510;第53卷(第9期);第112-119页 |
Also Published As
Publication number | Publication date |
---|---|
CN107219457A (en) | 2017-09-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107219457B (en) | Frame-type circuit breaker fault diagnosis and degree assessment method based on operation attachment electric current | |
CN106482937B (en) | A kind of monitoring method of mechanical state of high-voltage circuit breaker | |
CN107101813B (en) | A kind of frame-type circuit breaker mechanical breakdown degree assessment method based on vibration signal | |
CN106017879A (en) | Universal circuit breaker mechanical fault diagnosis method based on feature fusion of vibration and sound signals | |
CN110221200A (en) | A kind of omnipotent breaker attachment method for diagnosing faults based on deep learning | |
CN105259471B (en) | A kind of three-dimensional fault-line selecting method based on accidental resonance and transient current signal | |
Strachan et al. | Providing decision support for the condition-based maintenance of circuit breakers through data mining of trip coil current signatures | |
CN101799368B (en) | Electromechanical device nonlinear failure prediction method | |
CN105891707A (en) | Opening-closing fault diagnosis method for air circuit breaker based on vibration signals | |
CN105868770B (en) | A kind of Fault Diagnosis for HV Circuit Breakers method based on unsupervised learning model | |
CN109061463A (en) | A kind of monitoring of mechanical state of high-voltage circuit breaker and method for diagnosing faults | |
Yang et al. | Condition evaluation for opening damper of spring operated high-voltage circuit breaker using vibration time-frequency image | |
CN103116090B (en) | Three-phrase pulse-width modulation (PWM) rectifier fault diagnosis method based on wavelet packet analysis and support vector machine | |
CN106443316A (en) | Power transformer winding deformation state multi-information detection method and device | |
CN105547465A (en) | Transformer vibration signal winding state feature extraction method | |
CN109753951A (en) | A kind of OLTC method for diagnosing faults based on instantaneous energy entropy and SVM | |
Li et al. | Condition assessment of power transformers using a synthetic analysis method based on association rule and variable weight coefficients | |
CN110320467A (en) | A kind of low-voltage direct circuit breaker failure diagnostic method | |
CN106203382A (en) | A kind of excitation surge current based on kernel function extreme learning machine and fault current recognition methods | |
CN109784310A (en) | Panel switches mechanical breakdown feature extracting method based on CEEMDAN and weighting time-frequency entropy | |
CN103513212A (en) | BH curve characteristic CT state recognition and unsaturation degree computing method based on reconstruction | |
CN110174612A (en) | Low-voltage circuit breaker fault early-warning detection method | |
CN110703151A (en) | Transformer fault diagnosis method based on vibration blind source separation and Bayesian model | |
CN112557966A (en) | Transformer winding looseness identification method based on local mean decomposition and support vector machine | |
CN104713714A (en) | High-voltage circuit breaker operating characteristic analysis method based on grid multi-density clustering |
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 |