CN110096927A - Contactor diagnostic method and diagnostic system based on particle group optimizing support vector machines - Google Patents
Contactor diagnostic method and diagnostic system based on particle group optimizing support vector machines Download PDFInfo
- Publication number
- CN110096927A CN110096927A CN201810090128.XA CN201810090128A CN110096927A CN 110096927 A CN110096927 A CN 110096927A CN 201810090128 A CN201810090128 A CN 201810090128A CN 110096927 A CN110096927 A CN 110096927A
- Authority
- CN
- China
- Prior art keywords
- svc
- optimal
- particle
- data
- parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Abstract
A kind of contactor diagnostic method and its diagnostic system based on particle group optimizing support vector machines, method is the following steps are included: acquire the vibration signal and coil current signal of the contactor dynamic iron core under different working condition；FIR low pass noise reduction and FFT butterfly frequency-domain transform are carried out to vibration signal, current signal, wait until comprehensive quality features parameter；Obtained characteristic parameter is normalized；It is main at constituents extraction that obtained normalization characteristic vector is subjected to PCA；It establishes preliminary SVC sorter model and continues to optimize SVC sorter model using optimization particle swarm algorithm, obtain optimized parameter C and g, obtain optimal SVC sorter model；It is trained model and test data；Characteristic parameter, condition diagnosing result are imported in expert system, carry out live signal show, the processing such as Image Rendering, condition diagnosing and classification, the inquiry of historical data.
Description
Technical field
The present invention relates to fault diagnosis technology field, especially a kind of contactor based on particle group optimizing support vector machines
Diagnostic method and diagnostic system.
Background technique
Contactor is widely used in all kinds of circuits as the electric appliance for being frequently switched on and cut-offfing all kinds of circuits.Contactor
Failure can cause serious power grid accident, even result in power system collapse, cause serious economic loss.Therefore its working condition
And performance seems particularly important.In contactor operational process, core vibration signal and divide-shut brake coil current signal contain
The a large amount of status information of contactor can be diagnosed to be the contact moving part card of contactor by analysis electric current and vibration signal
The various faults features such as stagnant, coil voltage is insufficient, electromagnet core foreign matter.Therefore pass through the vibration signal and electric current to contactor
Signal carry out fault diagnosis and classification be very it is necessary to.
Currently, the state-detection of contactor is divided into offline and two kinds of online classification with classification method.Classify offline and is mainly
Whether artificial detection electromagnetic mechanism, contact part, arc quenching system, power pack are working properly, backward in technique, inefficiency,
As a result unreliable, and influence electric system normal work.And traditional on-line monitoring method mainly monitors probe of contactor and closes
Opening velocity, number etc. are closed, there are the shortcomings such as anti-interference is poor, stability is poor, result inaccuracy.
In recent years, with the continuous growth of the energy and electricity needs, strong smart grid is constructed as main trend.Shape
State monitoring and fault diagnosis are the important links in smart grid, can be obtained in real time in the case where not influencing power equipment operation
Operating status is taken, maximizing the benefits is obtained.Intelligent algorithm is introduced into fault diagnosis field by some scholars, as neural network is calculated
Method.But the algorithm has the shortcomings that need that great amount of samples data, to be easily trapped into local optimum, inefficiency, Generalization Ability insufficient,
Accurately diagnostic result is hardly resulted in practical application.
By being analyzed above it is found that existing method for diagnosing faults has respective deficiency, therefore it is effective to be badly in need of one kind
Contactor condition diagnosing and classification method.
Disclosed above- mentioned information are used only for enhancing the understanding to background of the present invention in the background section, it is thus possible to
Information comprising not constituting the prior art known to a person of ordinary skill in the art in home.
Summary of the invention
In view of the above problems, the purpose of the present invention is to provide a kind of contactors based on particle group optimizing support vector machines
Diagnostic method and diagnostic system, this method effectively overcome that small sample, dimension be high, non-linear, local optimum, man-machine interaction
The bottlenecks such as difference have ensured the speed and precision of condition diagnosing and classification, and have had good human-computer interaction function.
The purpose of the present invention is be achieved by the following technical programs.
One aspect of the present invention, a kind of contactor diagnostic method based on particle group optimizing support vector machines, processing side
Method includes the following steps,
First step acquires the vibration signal of the contactor dynamic iron core under different working condition and the current signal of coil；
Second step carries out FIR low pass noise reduction to vibration signal and current signal and FFT butterfly frequency-domain transform obtains feature
Parameter；
Third step, characteristic parameter, which is normalized, obtains normalization characteristic vector；
Four steps, normalization characteristic vector PCA is main at constituents extraction；
5th step establishes SVC sorter model；
6th step continues to optimize SVC sorter model using optimization particle swarm algorithm, obtains optimized parameter C and g,
Obtain optimal SVC sorter model；
7th step, is trained model and test data；
8th step, by characteristic parameter import optimal SVC sorter model obtain live signal show, Image Rendering and/
Or condition diagnosing and classification.
In the described method, further include following steps:
Second step, the FIR low pass filter for carrying out FIR filtering noise reduction use rectangular window function, and sample frequency is
70kHZ, cutoff frequency 15kHz, filter order 65, FFT butterfly frequency-domain transform choose 4096 points and carry out the change of butterfly FFT frequency domain
Get signal spectrum figure in return.
In the described method, third step, by the control of different dimension different dimensions data under same referential, formulaWherein: x is characterized parameter set；X is the parameter sample after normalization；xminIt is whole numbers
According to the minimum value of concentration；xmaxIt is the maximum value that total data is concentrated.
In the described method, normalization characteristic vector is found out the covariance square of sample using PCA algorithm by four steps
The characteristic value and feature vector of battle array, i.e., each Wesy are arranged in the contribution rate for distinguishing total data according to contribution margin size
Compositional model vector selects main Cheng Chengfen, and wherein covariance formula is as follows
Cn×n=(ci,j,ci,j=cov (Dimi,Dimj)), wherein xi,yiThe stochastic variable of two dimensions,It is two
The average of variable of a dimension, cov (x, y) indicate the covariance of two stochastic variables of X and Y, Cn×nIndicate the association side of n dimension data
Difference, Dim indicate that array dimension, n indicate that the array sample dimension, i are i-th in n sample dimension.
In the described method, the 5th step passes through nonlinear mapping functionData are mapped to higher dimensional space,
Hyperplane is established, optimal classification surface formula is released and then mapping function is found out by Lagrange optimization, optimal core is obtained by experiment
Function and SVC disaggregated model, wherein SVC disaggregated model is as follows:
Optimal kernel function expression formula is as follows: K (x, xi)=exp (- γ | | x-xi||2), γ > 0, wherein i=1,2 ..., p
It is the constant that user specifies；ξ is slack variable；P be to learning data group number, w is weight vectors；B is biasing；ξ is
Slack variable (ξ > 0) indicates the fault-tolerance of data classification；C is penalty factor (C > 0), is punished to control error sample
The degree penalized；α1,α2,…,αpIt is non-negative Lagrange's multiplier, wherein αp>0；Sample (x1,x2,…,xp) be supporting vector with
Determine decision boundary；Y (x) is class categories.
In the described method, the 6th step simulates flock of birds predation, initializes a group random particles, generates first
Spatial position and speed for population；Particle is constantly iterated, and seeks optimal solution, and in each iteration, particle can all update
Two extreme values, one is particle optimal solution itself, the other is population total optimization solution, particle optimal solution itself is referred to as individual pole
Value, population total optimization solution is referred to as global extremum, optimizes and revises particle inertia weight, and when the number of iterations is small, dynamic increases used
Property weight, when the number of iterations increases, dynamic reduces inertia weight, while constantly judging whether to meet iteration and presetting extreme value precision
Or iteration maximum times optimize kernel functional parameter g and penalty factor, and then obtain most if satisfied, then converging to optimal value
Excellent SVC disaggregated model.
In the described method, second step carries out FIR low pass noise reduction and FFT butterfly to vibration signal and current signal
Frequency-domain transform obtains the characteristic parameter of following different conditions: closed state electric current period 1 peak value, closed state electric current second
Cycle peak, closed state electric current period 3 peak value, closed state electric current period 1 valley, closed state electric current second week
Phase valley, closed state electric current period 3 valley, steady state current period 1 peak value, steady state current period 1
Valley, starting current maximum frequency, stabling current maximum frequency, closed state dynamic iron core vibration maximum frequency and stable state are dynamic
Core vibration maximum frequency.
According to another aspect of the present invention, the contactor diagnosis described in a kind of implementation based on particle group optimizing support vector machines
The diagnostic system of method includes
Data acquisition device acquires the vibration signal of the contactor dynamic iron core under different working condition and the electric current of coil
Signal；
Data pre-processing unit carries out FIR low pass noise reduction and FFT butterfly frequency-domain transform to vibration signal and current signal
Obtain characteristic parameter；
Normalizing unit is normalized characteristic parameter and obtains normalization characteristic vector；Normalization can be by different dimensions
Different dimensions data control under same referential, and then simplify and calculate, reduce magnitude.
PCA processing unit, device is by normalization characteristic vector PCA processing and completes main at constituents extraction, effectively reduces
Redundancy and noise reduce overfitting, greatly increase the efficiency of data analysis；
SVC classifier establishes SVC sorter model；
Particle group optimizing unit continues to optimize SVC sorter model using optimization particle swarm algorithm, obtains optimal SVC
Sorter model, particle swarm optimization algorithm realize the optimal solution of complex space by individual cooperation and competition；
Training unit, training pattern and test data；
Display interface, characteristic parameter import optimal SVC sorter model obtain live signal show, Image Rendering, shape
State diagnosis and classification and/or the inquiry of historical data.
In the diagnostic system, diagnostic system is set on the server, and the server includes processor, hard disk, interior
It deposits, bus and the communication port for being interacted with data acquisition device.
In the diagnostic system, the hard disk of server include can fast reading and writing SDD hard disk and pluggable SD card
Mobile read-write equipment.
The advantage of the invention is that restrained effectively noise using signal processing algorithms such as FIR, FFT, enhance useful
Signal energy is extracted comprehensive validity feature information from nonlinear properties.Using normalizing algorithm, PCA and POS algorithm,
It is extracted main at signal, simplified calculating process, optimization training pattern, realization complex space optimal solution.It is special by exploitation host computer
Family's system is realized the functions such as historical data, fault inquiry, live signal, malfunction coefficient, signal waveform drafting, is had good
Human-computer interaction interface.Detection of the invention, classification accuracy can achieve 99% or more.
The above description is only an overview of the technical scheme of the present invention, in order to make technological means of the invention clearer
Understand, reach the degree that those skilled in the art can be implemented in accordance with the contents of the specification, and in order to allow the present invention
Above and other objects, features and advantages can be more clearly understood, illustrated below with a specific embodiment of the invention
Explanation.
Detailed description of the invention
By reading the detailed description in hereafter preferred embodiment, various other advantages and benefits of the present invention
It will become apparent to those of ordinary skill in the art.Figure of description only for the purpose of illustrating preferred embodiments,
And it is not to be construed as limiting the invention.It should be evident that drawings discussed below is only some embodiments of the present invention,
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings
Other attached drawings.And throughout the drawings, identical component is presented with like reference characters.
In the accompanying drawings:
Fig. 1 is the contactor diagnostic method according to an embodiment of the invention based on particle group optimizing support vector machines
Step schematic diagram；
Fig. 2 is the contactor diagnostic method according to an embodiment of the invention based on particle group optimizing support vector machines
Workflow schematic diagram；
Fig. 3 is the contactor diagnostic method according to an embodiment of the invention based on particle group optimizing support vector machines
Model foundation flow diagram；
Fig. 4 is the contactor diagnostic method according to an embodiment of the invention based on particle group optimizing support vector machines
System Establishing process schematic diagram；
Fig. 5 is the structural schematic diagram of diagnostic system according to an embodiment of the invention；
Fig. 6 is the display interface schematic diagram of diagnostic system according to an embodiment of the invention.
Below in conjunction with drawings and examples, the present invention will be further explained.
Specific embodiment
The specific embodiment that the present invention will be described in more detail below with reference to accompanying drawings.Although being shown in attached drawing of the invention
Specific embodiment, it being understood, however, that may be realized in various forms the present invention without that should be limited by embodiments set forth here
System.It is to be able to thoroughly understand the present invention on the contrary, providing these embodiments, and can be complete by the scope of the present invention
Be communicated to those skilled in the art.
It should be noted that having used some vocabulary in the specification and claims to censure specific components.Ability
Field technique personnel it would be appreciated that, technical staff may call the same component with different nouns.This specification and right
It is required that not in such a way that the difference of noun is as component is distinguished, but with the difference of component functionally as differentiation
Criterion."comprising" or " comprising " as mentioned throughout the specification and claims are an open language, therefore should be solved
It is interpreted into " including but not limited to ".Specification subsequent descriptions are to implement better embodiment of the invention, so the description be with
For the purpose of the rule of specification, the range that is not intended to limit the invention.Protection scope of the present invention is when the appended right of view
It is required that subject to institute's defender.
In order to facilitate understanding of embodiments of the present invention, further by taking specific embodiment as an example below in conjunction with attached drawing to be solved
Explanation is released, and each attached drawing does not constitute the restriction to the embodiment of the present invention.
In order to better understand, the step of Fig. 1 is the contactor diagnostic method based on particle group optimizing support vector machines is shown
It is intended to, as shown in Figure 1, a kind of contactor diagnostic method based on particle group optimizing support vector machines includes the following steps,
First step S1 acquires the vibration signal of the contactor dynamic iron core under different working condition and the electric current letter of coil
Number；
Second step S2 carries out FIR low pass noise reduction to vibration signal and current signal and FFT butterfly frequency-domain transform obtains spy
Levy parameter；
Third step S3, characteristic parameter, which is normalized, obtains normalization characteristic vector；
Four steps S4, normalization characteristic vector PCA is main at constituents extraction；
5th step S5 establishes SVC sorter model；
6th step S6, using optimization particle swarm algorithm continue to optimize SVC sorter model, obtain optimized parameter C and
G obtains optimal SVC sorter model；
7th step S7, is trained model and test data；
8th step S8, by characteristic parameter import optimal SVC sorter model obtain live signal show, Image Rendering
And/or condition diagnosing and classification.
In the preferred embodiment of the method for the invention, second step S2 carries out the FIR low pass of FIR filtering noise reduction
Filter uses rectangular window function, sample frequency 70kHZ, cutoff frequency 15kHz, filter order 65, the change of FFT butterfly frequency domain
It changes 4096 points progress butterfly FFT frequency-domain transforms of selection and obtains signal spectrum figure.
In the preferred embodiment of the method for the invention, third step S3, by different dimension different dimensions data controls
System is under same referential, formulaWherein: x is characterized parameter set；X is normalization
Parameter sample afterwards；xminIt is the minimum value that total data is concentrated；xmaxIt is the maximum value that total data is concentrated.
In the preferred embodiment of the method for the invention, four steps S4, using PCA algorithm by normalization characteristic to
Amount finds out the characteristic value and feature vector of the covariance matrix of sample, i.e., each Wesy presses in the contribution rate for distinguishing total data
It carries out rearranging pattern vector according to contribution margin size, selects main Cheng Chengfen, wherein covariance formula is as followsCn×n=(ci,j,ci,j=cov (Dimi,Dimj)), wherein xi,
yiThe stochastic variable of two dimensions,For the average of variable of two dimensions, cov (x, y) indicates two stochastic variables of X and Y
Covariance, Cn×nIndicate the covariance of n dimension data, Dim indicates that array dimension, n indicate that the array sample dimension, i are n sample
I-th in this dimension.
In the preferred embodiment of the method for the invention, the 5th step S5 passes through nonlinear mapping function?
Data are mapped to higher dimensional space, establish hyperplane, release optimal classification surface formula and then find out mapping letter by Lagrange optimization
Number, obtains optimal kernel function and SVC disaggregated model by experiment, wherein SVC disaggregated model is as follows:
Optimal kernel function expression formula
It is as follows: K (x, xi)=exp (- γ | | x-xi||2), γ > 0, wherein i=1,2 ..., p are the constants that user specifies；ξ is loose
Variable；P be to learning data group number, w is weight vectors；B is biasing；ξ is slack variable ξ > 0, indicates data point
The fault-tolerance of class；C is penalty factor > 0, to control the degree to error sample punishment；α1,α2,…,αpIt is non-negative glug
Bright day multiplier, wherein αp>0；Sample (x1,x2,…,xp) it is supporting vector to determine decision boundary；Y (x) is class categories.
In the preferred embodiment of the method for the invention, the 6th step S6 simulates flock of birds predation, initialization one
Group's random particles generate spatial position and the speed of first generation population；Particle is constantly iterated, and seeks optimal solution, every time
In iteration, particle can all update two extreme values, and one is particle optimal solution itself, the other is population total optimization solution, particle
Optimal solution itself is referred to as individual extreme value, and population total optimization solution is referred to as global extremum, optimizes and revises particle inertia weight, work as iteration
Secondary a few hours, dynamic increase inertia weight, and when the number of iterations increases, dynamic reduces inertia weight, while constantly judging whether
Meet iteration and preset extreme value precision or iteration maximum times, if satisfied, then converge to optimal value, i.e. optimization kernel functional parameter g and
Penalty factor, and then obtain optimal SVC disaggregated model.
In the preferred embodiment of the method for the invention, second step S2 carries out vibration signal and current signal
FIR low pass noise reduction and FFT butterfly frequency-domain transform obtain the characteristic parameter of following different conditions: closed state electric current period 1 peak
Value, closed state electric current second round peak value, closed state electric current period 3 peak value, closed state electric current period 1 paddy
Value, closed state electric current second round valley, closed state electric current period 3 valley, steady state current period 1 peak
Value, steady state current period 1 valley, starting current maximum frequency, stabling current maximum frequency, closed state dynamic iron core
It vibrates maximum frequency and stable state dynamic iron core vibrates maximum frequency.
In one embodiment, Fig. 2 is according to an embodiment of the invention based on particle group optimizing support vector machines
The workflow schematic diagram of contactor diagnostic method, as shown in Fig. 2, a kind of be based on particle group optimizing support vector machines and expert
The contactor condition diagnosing and classification method of system, comprising the following steps: step 1: the contactor under acquisition different working condition is dynamic
The vibration signal and coil current signal of iron core；Step 2: FIR low pass noise reduction and FFT butterfly are carried out to vibration signal, current signal
Shape frequency-domain transform obtains comprehensive quality features parameter；Step 3: the characteristic parameter that step 2 is obtained is normalized；Step
Rapid four: it is main at constituents extraction that the normalization characteristic vector that step 3 is obtained carries out PCA；Step 5: preliminary SVC classifier is established
Model；Step 6: SVC sorter model is continued to optimize using optimization particle swarm algorithm, optimized parameter C and g is obtained, obtains most
Excellent SVC sorter model；Step 7: model and test data are trained；Step 8: by characteristic parameter, condition diagnosing result
Import in expert system, carry out live signal show, the processing such as Image Rendering, condition diagnosing and classification, the inquiry of historical data.This
The characteristics of invention, also resides in, and FIR filters noise reduction and the processing of FFT frequency-domain transform in step 1.FIR low pass filter uses rectangle
Window function, coefficient are generated by MATLAB, sample frequency 70kHZ, cutoff frequency 15kHz, filter order 65.FFT frequency domain becomes
It changes by length formula N=2MIt is found that choosing 4096 points carries out butterfly FFT frequency-domain transform, signal spectrum figure is obtained.Of the invention
Feature is that the normalized in step 3 can control different dimension different dimensions data under same referential, in turn
Simplify and calculates, reduces magnitude.Formula is as followsIn formula: x is characterized parameter set；X: after normalization
Parameter sample；xminIt is the minimum value that total data is concentrated；xmaxIt is the maximum value that total data is concentrated.The features of the present invention is also
It is, in step 4, carries out PCA processing, that is, the characteristic value and feature vector for finding out the covariance matrix of sample is i.e. per one-dimensional
For distinguishing the contribution rate of total data.It is arranged according to contribution margin size.Compositional model vector selects main Cheng Chengfen.Have
Effect ground reduces redundancy and noise, reduces overfitting, greatly increases the efficiency of data analysis.Wherein covariance formula is such as
Under:
X in formulai,yiThe stochastic variable of two dimensions,For the average of variable of two dimensions.The characteristics of invention, also exists
SVC disaggregated model in step 5 is established.Specific steps are as follows: step 51 passes through nonlinear mapping functionData are reflected
It is mapped to higher dimensional space, establishes hyperplane g (x)=wTφ (x)+b=0.Optimal classification surface formula, that is, objective function is as follows:Constraint condition: di(wTφ (x)+b) >=1- ξ (ξ > 0), wherein i=1,
2 ..., p is the constant that user specifies, and C is to control the degree to error sample punishment, i.e. penalty factor；ξ is slack variable；
P be to learning data group number.Objective function is converted into method using Lagrange optimization by step 52:Constraint condition:
Wherein αiIt is Lagrange's multiplier, (x1,x2,...,xp) it is supporting vector,Step 53
ByIt can obtain SVC disaggregated model:
Step 54 selects kernel function K.
The embodiment of the present invention by comparing, has selected the highest kernel function of classification accuracy when parameter is defaulted, i.e.,
Radial Basis Function kernel function.The features of the present invention also characterized in that optimization kernel functional parameter g and punishment in step 6
The POS algorithm of factor C.Wherein population POS optimization algorithm realizes the optimal solution of complex space by individual cooperation and competition.
Specific steps are as follows:
Step 61 simulates flock of birds predation, initializes a group random particles, generates the spatial position of first generation population
And speed.2 particle of step 6 is constantly iterated, seek optimal solution.In each iteration, particle can all update two extreme values.One
A is particle optimal solution itself, and one is population total optimization solution.Particle optimal solution itself is referred to as individual extreme value.Population is whole most
Excellent solution is referred to as global extremum.Particle rapidity more new formula is as follows: V(t+1) id=W(t)×V(t) id+C1×rand()(pbest(t) id-
present(t) id)+C2×rdnd()×(gbes(t) id-present(t) id)
Wherein, d=1,2,3 ..., n, n are the n-dimensional space obtained in step 3, and i=1,2,3 ..., m, m is population rule
Mould, t are current evolutionary generation, V(t) idIt is that the i-th particle d in the t times iteration ties up speed, W(t)It is the t times iteration inertia power
Weight, present(t) idIt is that the i-th particle d in the t times iteration ties up position, pbest(t) idIt is the i-th particle in the t times iteration
D ties up individual extreme value, gbest(t) idWhen d ties up global extremum to the i-th particle in the t times iteration, rand () is between 0 to 1
Random number, C1C2It is the usual C of Studying factors1=C2=2.Particle position more new formula: present(t+1) id=present(t) id+V(t+1) id, wherein present(t+1) idIt is that the i-th particle d in the t+1 times iteration ties up position.
The invention feature also resides in, and optimizes and revises particle inertia weight, and when the number of iterations is small, dynamic increases inertia power
Weight, improves its global optimizing ability, avoids falling into local optimum.When the number of iterations increases, dynamic reduces inertia weight, improves
Convergence rate and precision adjust overall situation and partial situation's optimizing ability of PSO algorithm.Inertia weight more new formula:Wherein, TmaxFor maximum evolutionary generation, WendWhen to evolve to maximum algebra
Inertia weight value.WminFor initial maximum inertia weight value.Step 63 is constantly iterated, while judging whether to meet iteration
Default extreme value precision or iteration maximum times, if satisfied, then converging to optimal value.Feature of the present invention also resides in step 8 expert system
The exploitation of system, specific implementation step are as follows: step 81, using Visual Studio Developing Expert System, import characteristic parameter
With the diagnostic result based on particle group optimizing support vector machines, make it have live signal show, Image Rendering, condition diagnosing with
The functions such as classification, the inquiry of historical data have good human-computer interaction function.
In one embodiment, step 2 in order to better understand further illustrates failure and extraction that the present invention simulates
Characteristic value:
Failure one: it is insufficient to simulate supply voltage as control loop supply voltage for 220V AC voltage；
Failure two: contactor mechanical movable componental movement is hindered using insulated ceramic plates, simulation is attracted clamping stagnation；
Failure three: being placed in iron core pole-face using insulation spacer, simulates iron core foreign matter；
Pass through the characteristic value for the different conditions that FIR low pass noise reduction and FFT butterfly frequency-domain transform obtain:
3 closed state electric current third week of 2 closed state electric current second round peak value of closed state electric current period 1 peak value
The 6 closed state electric current period 3 of 6 closed state electric current period 1 of phase peak value, 5 closed state electric current second round valley of valley
10 starting current maximum frequency 11 of peak value 8 steady state current period 1 of valley, valley 9 steady state current period 1 is steady
12 closed state dynamic iron core of constant current maximum frequency vibrates 13 stable state dynamic iron core of maximum frequency and vibrates maximum frequency.
In the step 3, normalization algorithm standardization characteristic value is utilized.Normalize formula:In the step 4, normalization data is found out to the covariance square of sample using PCA algorithm
The characteristic value of battle array and feature vector, that is, each Wesy are in the contribution rate for distinguishing total data.It is arranged according to contribution margin size.
Compositional model vector selects main Cheng Chengfen.PCA algorithm contribution rate calculation formula:
Cn×n=(ci,j,ci,j=cov (Dimi,Dimj)), in the step 5, carries out preliminary SVC disaggregated model and establish.It established
Journey: pass through nonlinear mapping functionData are mapped to higher dimensional space, establish hyperplane, release optimal classification surface formula into
And mapping function is found out to method by Lagrange optimization, optimal kernel function is obtained by experiment, finally obtains SVC disaggregated model.
Optimal classification formula is as follows:Constraint condition: di(wTφ(x)+b)
>=1- ξ (ξ > 0), objective function Equation is as follows:
Constraint condition:
Lagrange coefficient and hyperplane relation formula are as follows:Preliminary SVC disaggregated model is as follows:Optimal kernel function Radial Basis
Function expression formula is as follows: K (x, xi)=exp (- γ | | x-xi||2),γ>0。
Fig. 3 is the contactor diagnostic method according to an embodiment of the invention based on particle group optimizing support vector machines
Model foundation flow diagram in the step 6, seeks optimal SVC disaggregated model, process using population POS optimization algorithm
Figure is as shown in Figure 3.Specific implementation step the following steps are included:
Step 61: initialization a group random particles obtain spatial position and the speed of first generation population.
Step 62: particle is constantly iterated, and in each iteration, particle can all update two extreme values.One is particle sheet
Body optimal solution, one is population total optimization solution.Particle optimal solution itself is referred to as individual extreme value.Population total optimization solution is referred to as complete
Office's extreme value.
Particle rapidity more new formula is as follows:
V(t+1) id=W(t)×V(t) id+C1×rand()×(pbest(t) id-present(t) id)+C2×rand()×
(gbest(t) id-present(t) id)
Particle position more new formula is as follows:
present(t+1) id=present(t) id+V(t+1) id
Inertia weight more new formula is as follows:
Step 63: optimizing and revising particle inertia weight, and when the number of iterations is small, dynamic increases inertia weight.When iteration time
When number increases, dynamic reduces inertia weight, while constantly judging whether that meeting iteration presets extreme value precision or iteration maximum times.
If satisfied, then converging to optimal value, that is, optimize kernel functional parameter g and penalty factor, and then obtain optimal SVC disaggregated model.
Weight in speed formula in order to better understand is done further value to weight and illustrated: weight is to fly to be used to
Property, indicate particle to the constant quest in a part of space.If w is 0, algorithm is easily trapped into locally optimal solution.It is global that w increases particle
Search capability increases, therefore dynamic adjusts w to balance ability of searching optimum and local search.By constantly testing, w is changing every time
Gradually reducing from 1.4 to 0 during generation can achieve optimal effectiveness.
Step 8 in order to better understand, below with reference to Fig. 4 to the system based on particle group optimizing support vector machines do into
One step illustrates.SVC algorithm is write in matlab, utilizes the dynamic under the database and .NET environment of Access exploitation
Chained library develops the system based on particle group optimizing support vector machines.By importing characteristic parameter and being based on particle group optimizing branch
The diagnostic result for holding vector machine realizes the function of expert system, and user is facilitated to carry out human-computer interaction.
Fig. 5 is the structural schematic diagram of diagnostic system according to an embodiment of the invention, as shown in figure 5, a kind of implementation institute
The diagnostic system for stating the contactor diagnostic method based on particle group optimizing support vector machines includes
Data acquisition device 1 acquires the vibration signal of the contactor dynamic iron core under different working condition and the electricity of coil
Flow signal；
Data pre-processing unit 2 carries out FIR low pass noise reduction to vibration signal and current signal and FFT butterfly frequency domain becomes
Get characteristic parameter in return；
Normalizing unit 3 is normalized characteristic parameter and obtains normalization characteristic vector；
PCA processing unit 4, device is by normalization characteristic vector PCA processing and completes main at constituents extraction；
SVC classifier 5 establishes SVC sorter model；
Particle group optimizing unit 6 continues to optimize SVC sorter model using optimization particle swarm algorithm, obtains optimal SVC
Sorter model；
Training unit 7, training pattern and test data；
Display interface 8, by characteristic parameter import optimal SVC sorter model obtain live signal show, Image Rendering, shape
State diagnosis and classification and/or the inquiry of historical data.
In the preferred embodiment of diagnostic system of the present invention, diagnostic system is set on the server, and the server includes
Processor, hard disk, memory, bus and the communication port for being interacted with data acquisition device.
In the preferred embodiment of diagnostic system of the present invention, the hard disk of server include can fast reading and writing SDD hard disk
And the mobile read-write equipment of pluggable SD card.
Fig. 6 is the display interface schematic diagram of diagnostic system according to an embodiment of the invention, as shown in fig. 6, display circle
Face 8 by characteristic parameter import optimal SVC sorter model obtain live signal show, Image Rendering, condition diagnosing and classification and/
Or the inquiry of historical data.
Although embodiment of the present invention is described in conjunction with attached drawing above, the invention is not limited to above-mentioned
Specific embodiments and applications field, above-mentioned specific embodiment are only schematical, directiveness, rather than restricted
's.Those skilled in the art are under the enlightenment of this specification and in the range for not departing from the claims in the present invention and being protected
In the case where, a variety of forms can also be made, these belong to the column of protection of the invention.
Claims (10)
1. a kind of contactor diagnostic method based on particle group optimizing support vector machines, processing method include the following steps,
First step (S1) acquires the vibration signal of the contactor dynamic iron core under different working condition and the current signal of coil；
Second step (S2) carries out FIR low pass noise reduction to vibration signal and current signal and FFT butterfly frequency-domain transform obtains feature
Parameter；
Third step (S3), characteristic parameter, which is normalized, obtains normalization characteristic vector；
Four steps (S4), normalization characteristic vector PCA is main at constituents extraction；
5th step (S5), establishes SVC sorter model；
6th step (S6) continues to optimize SVC sorter model using optimization particle swarm algorithm, obtains optimized parameter C and g,
Obtain optimal SVC sorter model；
7th step (S7), is trained model and test data；
8th step (S8), by characteristic parameter import optimal SVC sorter model obtain live signal show, Image Rendering and/
Or condition diagnosing and classification.
2. the method according to claim 1, wherein further including following steps preferably:
Second step (S2), the FIR low pass filter for carrying out FIR filtering noise reduction use rectangular window function, and sample frequency is
70kHZ, cutoff frequency 15kHz, filter order 65, FFT butterfly frequency-domain transform choose 4096 points and carry out the change of butterfly FFT frequency domain
Get signal spectrum figure in return.
3. according to the method described in claim 1, it is characterized by: third step (S3), by different dimension different dimensions data
Control is under same referential, formulaWherein: x is characterized parameter set；After X is normalization
Parameter sample；xminIt is the minimum value that total data is concentrated；xmaxIt is the maximum value that total data is concentrated.
4. according to the method described in claim 1, it is characterized by: four steps (S4), using PCA algorithm by normalization characteristic
Vector finds out the characteristic value and feature vector of the covariance matrix of sample, i.e., each Wesy in the contribution rate for distinguishing total data,
It carries out rearranging pattern vector according to contribution margin size, selects main Cheng Chengfen, wherein covariance formula is as followsCn×n=(ci,j,ci,j=cov (Dimi,Dimj)),
Middle xi,yiThe stochastic variable of two dimensions,For the average of variable of two dimensions, cov (x, y) indicate X and Y two with
The covariance of machine variable, Cn×nIndicate the covariance of n dimension data, Dim indicates that array dimension, n indicate that the array sample dimension, i are
I-th in n sample dimension.
5. according to the method described in claim 1, it is characterized by: the 5th step (S5), passes through nonlinear mapping function
Data are mapped to higher dimensional space, establish hyperplane, release optimal classification surface formula and then mapping is found out by Lagrange optimization
Function obtains optimal kernel function and SVC disaggregated model by experiment, wherein SVC disaggregated model is as follows:
Optimal kernel function expression
Formula is as follows: K (x, xi)=exp (- γ | | x-xi||2), γ > 0, wherein i=1,2 ..., p are the constants that user specifies；ξ is pine
Relaxation variable；P be to learning data group number, w is weight vectors；B is biasing；ξ is slack variable (ξ > 0), indicates number
According to the fault-tolerance of classification；C is penalty factor (C > 0), to control the degree to error sample punishment；α1,α2,…,αpRight and wrong
Negative Lagrange's multiplier, wherein αp>0；Sample (x1,x2,…,xp) it is supporting vector to determine decision boundary；Y (x) is classification
Classification.
6. according to the method described in claim 1, it is characterized by: the 6th step (S6), simulates flock of birds predation, initialization
A group random particles generate spatial position and the speed of first generation population；Particle is constantly iterated, and seeks optimal solution, often
In secondary iteration, particle can all update two extreme values, and one is particle optimal solution itself, the other is population total optimization solution, grain
Sub optimal solution itself is referred to as individual extreme value, and population total optimization solution is referred to as global extremum, optimizes and revises particle inertia weight, when repeatedly
Generation a few hours, dynamic increase inertia weight, and when the number of iterations increases, dynamic reduces inertia weight, while constantly judgement is
The no iteration that meets presets extreme value precision or iteration maximum times, if satisfied, then converge to optimal value, i.e. optimization kernel functional parameter g
And penalty factor, and then obtain optimal SVC disaggregated model.
7. according to the method described in claim 1, it is characterized by: second step (S2), to vibration signal and current signal into
Row FIR low pass noise reduction and FFT butterfly frequency-domain transform obtain the characteristic parameter of following different conditions: closed state electric current period 1
Peak value, closed state electric current second round peak value, closed state electric current period 3 peak value, closed state electric current period 1 paddy
Value, closed state electric current second round valley, closed state electric current period 3 valley, steady state current period 1 peak
Value, steady state current period 1 valley, starting current maximum frequency, stabling current maximum frequency, closed state dynamic iron core
It vibrates maximum frequency and stable state dynamic iron core vibrates maximum frequency.
8. the contactor diagnostic method based on particle group optimizing support vector machines described in a kind of any one of implementation claim 1-7
Diagnostic system, it is characterised in that: the diagnostic system includes:
Data acquisition device (1) acquires the vibration signal of the contactor dynamic iron core under different working condition and the electric current of coil
Signal；
Data pre-processing unit (2) carries out FIR low pass noise reduction and FFT butterfly frequency-domain transform to vibration signal and current signal
Obtain characteristic parameter；
Normalizing unit (3) is normalized characteristic parameter and obtains normalization characteristic vector；
PCA processing unit (4), device is by normalization characteristic vector PCA processing and completes main at constituents extraction；
SVC classifier (5), establishes SVC sorter model；
Particle group optimizing unit (6) continues to optimize SVC sorter model using optimization particle swarm algorithm, obtains optimal SVC points
Class device model；
Training unit (7), training pattern and test data；
Display interface (8), by characteristic parameter import optimal SVC sorter model obtain live signal show, Image Rendering, state
Diagnosis and classification and/or the inquiry of historical data.
9. Cubicle Gas-Insulated Switchgear according to claim 7, it is characterised in that: diagnostic system is located at service
On device, the server includes processor, hard disk, memory, bus and the communication port for interacting with data acquisition device.
10. Cubicle Gas-Insulated Switchgear according to claim 9, it is characterised in that: the hard disk packet of server
Including can the SDD hard disk of fast reading and writing and the mobile read-write equipment of pluggable SD card.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810090128.XA CN110096927A (en) | 2018-01-30 | 2018-01-30 | Contactor diagnostic method and diagnostic system based on particle group optimizing support vector machines |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810090128.XA CN110096927A (en) | 2018-01-30 | 2018-01-30 | Contactor diagnostic method and diagnostic system based on particle group optimizing support vector machines |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110096927A true CN110096927A (en) | 2019-08-06 |
Family
ID=67441910
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810090128.XA Pending CN110096927A (en) | 2018-01-30 | 2018-01-30 | Contactor diagnostic method and diagnostic system based on particle group optimizing support vector machines |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110096927A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120233071A1 (en) * | 2003-08-29 | 2012-09-13 | Starbucks Corporation D/B/A Starbucks Coffee Company | Method and apparatus for automatically reloading a stored value card |
| US20120330743A1 (en) * | 2011-06-27 | 2012-12-27 | Felix Schul | Method and a system for providing a service transaction for a mobile device |
| EP2541480A1 (en) * | 2011-06-27 | 2013-01-02 | Coupies GmbH | A method and system for providing a service transaction for a mobile device |
| CN105445656A (en) * | 2015-11-16 | 2016-03-30 | 北京机械设备研究所 | Fault detection method for large-power power vehicle contactor |
| CN105866664A (en) * | 2016-06-20 | 2016-08-17 | 金陵科技学院 | Intelligent fault diagnosis method for analog circuit based on amplitude frequency features |
| CN106124212A (en) * | 2016-06-16 | 2016-11-16 | 燕山大学 | Based on sparse coding device and the Fault Diagnosis of Roller Bearings of support vector machine |
| CN106529124A (en) * | 2016-10-14 | 2017-03-22 | 云南电网有限责任公司昆明供电局 | A transformer insulation state assessment method based on principal component analysis and support vector machines |
| CN106682682A (en) * | 2016-10-20 | 2017-05-17 | 北京工业大学 | Method for optimizing support vector machine based on Particle Swarm Optimization |
| CN107390118A (en) * | 2017-06-13 | 2017-11-24 | 西安交通大学 | A kind of A.C. contactor on-line monitoring system and its on-line monitoring method |
-
2018
- 2018-01-30 CN CN201810090128.XA patent/CN110096927A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120233071A1 (en) * | 2003-08-29 | 2012-09-13 | Starbucks Corporation D/B/A Starbucks Coffee Company | Method and apparatus for automatically reloading a stored value card |
| US20120330743A1 (en) * | 2011-06-27 | 2012-12-27 | Felix Schul | Method and a system for providing a service transaction for a mobile device |
| EP2541480A1 (en) * | 2011-06-27 | 2013-01-02 | Coupies GmbH | A method and system for providing a service transaction for a mobile device |
| CN105445656A (en) * | 2015-11-16 | 2016-03-30 | 北京机械设备研究所 | Fault detection method for large-power power vehicle contactor |
| CN106124212A (en) * | 2016-06-16 | 2016-11-16 | 燕山大学 | Based on sparse coding device and the Fault Diagnosis of Roller Bearings of support vector machine |
| CN105866664A (en) * | 2016-06-20 | 2016-08-17 | 金陵科技学院 | Intelligent fault diagnosis method for analog circuit based on amplitude frequency features |
| CN106529124A (en) * | 2016-10-14 | 2017-03-22 | 云南电网有限责任公司昆明供电局 | A transformer insulation state assessment method based on principal component analysis and support vector machines |
| CN106682682A (en) * | 2016-10-20 | 2017-05-17 | 北京工业大学 | Method for optimizing support vector machine based on Particle Swarm Optimization |
| CN107390118A (en) * | 2017-06-13 | 2017-11-24 | 西安交通大学 | A kind of A.C. contactor on-line monitoring system and its on-line monitoring method |
Non-Patent Citations (4)
| Title |
|---|
| 李人厚等: "《智能控制理论和方法 第2版》", 28 February 2013 * |
| 郁磊等: "《MATLAB智能算法30个案例分析》", 31 August 2015 * |
| 韦鹏程等: "《基于R语言数据挖掘的统计与分析》", 31 December 2017 * |
| 韩世军等: "基于粒子群优化支持向量机的变压器故障诊断", 《电测与仪表》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109271975A (en) | A kind of electrical energy power quality disturbance recognition methods based on big data multi-feature extraction synergetic classification | |
| CN110347547A (en) | Log method for detecting abnormality, device, terminal and medium based on deep learning | |
| CN109284672A (en) | A kind of Mechanical Failure of HV Circuit Breaker diagnostic method based on PSO-Kmeans algorithm | |
| CN102651088B (en) | Classification method for malicious code based on A_Kohonen neural network | |
| CN109165604A (en) | The recognition methods of non-intrusion type load and its test macro based on coorinated training | |
| CN108616120A (en) | A kind of non-intrusive electrical load decomposition method based on RBF neural | |
| Zhang et al. | A survey on algorithm adaptation in evolutionary computation | |
| CN109359693A (en) | A kind of Power Quality Disturbance Classification Method | |
| Davies et al. | Deep neural networks for appliance transient classification | |
| CN112766140A (en) | Transformer fault identification method based on kernel function extreme learning machine | |
| Vasilic et al. | An improved neural network algorithm for classifying the transmission line faults | |
| CN110502989A (en) | A kind of small sample EO-1 hyperion face identification method and system | |
| CN110738232A (en) | grid voltage out-of-limit cause diagnosis method based on data mining technology | |
| Su | Analysis of optimisation method for online education data mining based on big data assessment technology | |
| CN115600088A (en) | Distribution transformer fault diagnosis method based on vibration signals | |
| CN108830304A (en) | A kind of image identification system based on support vector machines | |
| CN109873425A (en) | Electric system based on deep learning and user behavior adjusts trend method and system | |
| Reif et al. | Dataset generation for meta-learning | |
| CN113869451A (en) | Rolling bearing fault diagnosis method under variable working conditions based on improved JGSA algorithm | |
| CN110096927A (en) | Contactor diagnostic method and diagnostic system based on particle group optimizing support vector machines | |
| Chakraborty et al. | A dual‐tree complex wavelet transform‐based approach for recognition of power system transients | |
| Srivastava et al. | PSO & neural-network based signature recognition for harmonic source identification | |
| CN116298725A (en) | Fault arc detection method, system and storage medium | |
| CN110287759A (en) | A kind of eye strain detection method based on simplified input convolutional neural networks O-CNN | |
| CN115840116A (en) | Switch cabinet partial discharge fault mode identification method based on GAN-CNN |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190806 |