CN109188258A - The high-voltage circuitbreaker feature extraction and classification method being electrically coupled based on vibration - Google Patents
The high-voltage circuitbreaker feature extraction and classification method being electrically coupled based on vibration Download PDFInfo
- Publication number
- CN109188258A CN109188258A CN201810785672.6A CN201810785672A CN109188258A CN 109188258 A CN109188258 A CN 109188258A CN 201810785672 A CN201810785672 A CN 201810785672A CN 109188258 A CN109188258 A CN 109188258A
- Authority
- CN
- China
- Prior art keywords
- vibration
- signal
- time
- data
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G01R31/3271—Testing of circuit interrupters, switches or circuit-breakers of high voltage or medium voltage devices
- G01R31/3275—Fault detection or status indication
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Testing Electric Properties And Detecting Electric Faults (AREA)
Abstract
The invention discloses a kind of high-voltage circuitbreaker feature extractions and classification method being electrically coupled based on vibration, are related to the fault diagnosis of high-voltage circuitbreaker.Currently, it is more single individually to study electric current and vibration signal the high-voltage circuitbreaker fault type reflected to determine failure.The invention includes the following steps: 1) data acquire；2) characteristic vector pickup；It is extracted including the feature extraction of divide-shut brake coil current signal and vibration signal characteristics；3) pattern-recognition；Every kind of state m group data are extracted, as training sample；Remaining data is as test sample；Classification and Identification is carried out using SVM, by being trained to obtain training pattern to training set sample, test set data is inputted in training pattern and are calculated, classification results are obtained.The technical program in such a way that vibration signal combines, can accurately and comprehensively reflect the operation conditions of operating mechanism coil current curve, and fast and accurately judge the fault type of operating mechanism.
Description
Technical field
The present invention relates to the fault diagnosises of high-voltage circuitbreaker, more particularly to are mentioned based on the high-voltage circuitbreaker feature that vibration is electrically coupled
It takes and classification method.
Background technique
1High-voltage circuitbreaker as common electrical equipment, reliability of operation affect power equipment stability and
Reliability plays protective effect to power grid.Relevant investigation shows the failure for having nearly half in high-voltage circuitbreaker significant trouble
It is caused by being broken down due to operating mechanism, so being studied high-voltage circuit-breaker operating mechanism mechanical property for mentioning
High breaker operational reliability has a very important significance.
In breaker mechanic property curve, divide-shut brake coil current can reflect its secondary control loop, iron core and
The working condition of operating mechanism ontology, mechanical oscillation signal can reflect screw, transmission mechanism and buffer operating condition, and two
The characteristics of acquisition of kind signal all has non-intruding, so numerous domestic and international experts and scholars are special to high-voltage circuitbreaker electric current and vibration
Property is studied.Yang Lingxiao et al. is using probabilistic neural network to the opening coil electric current of high-voltage circuitbreaker and stroke curve
Feature is classified, and achieves certain effect.Zhang Yongkui et al. wavelet analysis method and dynamic time warping (DTW) are to electricity
Stream signal is analyzed, for determining fault state of circuit breaker.Li Chun cutting edge of a knife or a sword et al. leads 7 characteristic values of electric current respectively
Component Analysis dimensionality reduction, then classified using clustering method to identify circuit breaker failure.Mountain Sun Yin et al. be respectively adopted wavelet analysis with
Time domain asks extreme value point methods to be extracted time and electric current totally 7 extreme points of current curve, assesses high-voltage circuitbreaker with this
Operating status.Jiang Zhi is great et al. to combine Sakoe-Chiba window with conventional dynamic time wrapping algorithm, from divide-shut brake coil
Feature is extracted in electric current and carrys out the incipient fault in decision mechanism, and there is certain validity.About in the research of vibration signal, Miao
Uncommon benevolence et al. analyzes Circuit breaker vibration signal using wavelet decomposition method, and carries out using BP neural network to failure
Identification, works well.It is documented and is analyzed and researched using improved dynamic time warping algorithm to vibration signal, led to
The deviation more normally between test sample data is crossed, circuit breaker failure is diagnosed.Also have and utilize wavelet-packet energy
Entropy extracts vibration performance and has carried out fault diagnosis to breaker.It opens pendant, Zhao Shutao et al. and is believed by extracting vibration signal and sound wave
Number feature identification is classified to circuit breaker failure, work well.Program, Guan Yonggang et al. are excellent using factorial analysis dimensionality reduction
Change, support vector machines identifies failure, has carried out state recognition to vibration signal.The above method examines the failure of high-voltage circuitbreaker
Disconnected and status assessment has certain effect, but in these researchs, is individually studied electric current and vibration signal to sentence
Determine failure, the high-voltage circuitbreaker fault type reflected is more single.The reason of structure is complicated for high-voltage circuitbreaker, generates failure is more
Sample, mechanically and electrically failure is likely to occur.
Summary of the invention
The technical problem to be solved in the present invention and the technical assignment of proposition are to be improved and improved to prior art,
The high-voltage circuitbreaker feature extraction and classification method being electrically coupled based on vibration are provided, to improve the mechanical breakdown identification of high-voltage circuitbreaker
Validity purpose.For this purpose, the present invention takes following technical scheme.
Based on high-voltage circuitbreaker feature extraction and classification method that vibration is electrically coupled, comprising steps of
1) data acquire
Using Hall current sensor measure coil currents signal, using the vibration of acceleration transducer measurement operating mechanism
Signal；Hall current sensor, acceleration transducer are with set acquisition time, sample frequency to coil current signal and operation
The vibration signal of mechanism is acquired the abundant acquisition to guarantee vibration signal；
2) characteristic vector pickup
Then short-time energy value by calculating vibration signal is utilized with enhancing the shock characteristic of vibration event in signal
Double threshold method extracts time of origin and the end time of vibration event, and it is special to complete different faults state for makeup time parameter vector
Levy the construction of vector；
Characteristic vector pickup includes that the feature extraction of divide-shut brake coil current signal and vibration signal characteristics extract；Division brake cable
Loop current signal characteristic abstraction comprising steps of
201) to divide-shut brake current signal carry out smooth treatment, using ask the method for slope approximation derivation to Curve Maximization into
Row is sought；
202) processing is grouped to the extreme point detected, current signal is divided into three groups, seek three groups of extreme values respectively
Y value maximum value or minimum value in data seek maximum value for first group and determine T1, second group is minimized determination
T2, third group are maximized determining T3；
Vibration signal characteristics extract comprising steps of
203) signal is handled using short-time energy method, enhances the vibration performance of signal；
204) double threshold method detection displacement point is utilized；Short-time energy function value phase by preset threshold value, with signal
Compare, when the short-time energy function value of signal is more than given threshold, then judges this moment to conjugate point.During determining separating brake
The beginning of each event and finish time extract start and end times of maximum two vibration events totally 4 time parameters
T4, T5, T6, T7 set up feature vector；Wherein, the time of vibration that buffer spring generates is collided when T4 and T5 is closing operation,
T6 and T7 is the time of vibration that dynamic/static contact collision generates；N times are operated under every kind of operating status, are joined according to the respective time is extracted
Number completes the building of feature vector, and operating status is including normal, switching-in spring power is small, buffering elastic spring force is small, control loop voltage
It is low；
3) pattern-recognition
Every kind of state m group data are extracted, as training sample；Remaining data is as test sample；Classified using SVM
Test set data are inputted in training pattern and are calculated by being trained to obtain training pattern to training set sample by identification,
Obtain classification results.
First by the Analysis on Mechanism to high-voltage circuit-breaker switching on-off coil current signal and vibration signal, when proposing to utilize
Then intermediate node parameter extracts current signal time parameter using slope of curve method as feature vector, using based in short-term
The double threshold method of energy extracts the time parameter of vibration event, using the parameter of the two as the feature vector of pattern-recognition.Finally
Shown by SVM classification results: coil current curve is combined with vibration signal accurately can comprehensively reflect operating mechanism
Operation conditions, the fault type of operating mechanism can be fast and accurately judged using SVM, for breaker fault diagnosis and
Repair and maintenance has great importance.
The technical program operating status is classified, and the effective of the mechanical breakdown identification of high-voltage circuitbreaker is helped to improve
Property.
As optimization technique means: in step 1), setting acquisition time is 200ms, sample frequency 10K.
As optimization technique means: in step 201), it includes: a) to set i=1 that Curve Maximization, which seeks step, and xi and yi divide
It Wei not the corresponding coordinate value of i point；
B) judge whether i≤950, if entering in next step, otherwise terminate；
C) slope k i is calculated；
D) judge whether ki=0, if so, entering step h)；If it is not, entering in next step；
E) judge whether ki*k (i+1) < 0, if then entering in next step, if it is not, then entering step g)；
F) x (i+1) is exported, y (i+1)
G) i=i+1；And it is back to step b)；
H) (xi+x (i+1))/2 is exported, (yi+y (i+1))/2.
As optimization technique means: it in step 204), is operated 12 times under every kind of state, 4 kinds of states totally 48 groups of data,
Respective time parameter is extracted respectively, completes the building of feature vector.
As optimization technique means: in step 3), to 48 groups of sample datas, extracting 6 groups of data of every kind of state, 4 kinds of shapes
State totally 24 groups of data as training sample；Remaining 24 groups of data are as test sample；4 SVM are constructed, 4 groups " a pair three " are formed
Binary classifier, SVM classifier select default linear kernel function linear, by being trained to obtain to training set sample
Test set data are inputted in training pattern and are calculated, obtain classification results by training pattern.
The utility model has the advantages that the technical program will vibrate and current signal time parameter be used as feature vector, using support to
Amount machine (SVM) classifies to circuit breaker failure type.Circuit-breaker status can be more comprehensively reacted, SVM is for the electric signal that shakes
With preferable classifying quality, there is validity and practicability for the mechanical breakdown identification of high-voltage circuitbreaker.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Fig. 2 is exemplary currents curve graph.
Fig. 3 is the vibration signal figure of operating mechanism.
Fig. 4 is optimal classification surface figure.
Fig. 5 is extraction extreme point flow chart of the invention.
Fig. 6 is experiment current signal figure of the invention.
Fig. 7 is three kinds of Time Domain Analysis processing comparison diagram of the invention
Fig. 8 is that vibration signal temporal characteristics of the invention extract result figure.
Fig. 9 is svm classifier effect picture of the invention.
Specific embodiment
Technical solution of the present invention is described in further detail below in conjunction with Figure of description.
As shown in Figure 1, the technical program comprising steps of
1) data acquire
Using Hall current sensor measure coil currents signal, using the vibration of acceleration transducer measurement operating mechanism
Signal；Hall current sensor, acceleration transducer are with set acquisition time, sample frequency to coil current signal and operation
The vibration signal of mechanism is acquired the abundant acquisition to guarantee vibration signal；
2) characteristic vector pickup
Then short-time energy value by calculating vibration signal is utilized with enhancing the shock characteristic of vibration event in signal
Double threshold method extracts time of origin and the end time of vibration event, and it is special to complete different faults state for makeup time parameter vector
Levy the construction of vector；
Characteristic vector pickup includes that the feature extraction of divide-shut brake coil current signal and vibration signal characteristics extract；Division brake cable
Loop current signal characteristic abstraction comprising steps of
201) to divide-shut brake current signal carry out smooth treatment, using ask the method for slope approximation derivation to Curve Maximization into
Row is sought；
202) processing is grouped to the extreme point detected, current signal is divided into three groups, seek three groups of extreme values respectively
Y value maximum value or minimum value in data seek maximum value for first group and determine T1, second group is minimized determination
T2, third group are maximized determining T3；
Vibration signal characteristics extract comprising steps of
203) signal is handled using short-time energy method, enhances the vibration performance of signal；
204) double threshold method detection displacement point is utilized；Short-time energy function value phase by preset threshold value, with signal
Compare, when the short-time energy function value of signal is more than given threshold, then judges this moment to conjugate point.During determining separating brake
The beginning of each event and finish time extract start and end times of maximum two vibration events totally 4 time parameters
T4, T5, T6, T7 set up feature vector；Wherein, the time of vibration that buffer spring generates is collided when T4 and T5 is closing operation,
T6 and T7 is the time of vibration that dynamic/static contact collision generates；N times are operated under every kind of operating status, are joined according to the respective time is extracted
Number completes the building of feature vector, and operating status is including normal, switching-in spring power is small, buffering elastic spring force is small, control loop voltage
It is low；
3) pattern-recognition
Every kind of state m group data are extracted, as training sample；Remaining data is as test sample；Classified using SVM
Test set data are inputted in training pattern and are calculated by being trained to obtain training pattern to training set sample by identification,
Obtain classification results.
It is described further below for the technical program.
1 electric current and Vibrations
1.1 divide-shut brake coil current tracing analysis
In the operational process of high-voltage circuitbreaker, the divide-shut brake coil current moment changes, and the variation of coil current is anti-
The variation of the motion state of iron core is reflected.It measures and analyzes by the feature to coil current, iron core can be fully understood
Working condition, while also being understood that the part operating status of spring operating mechanism, be high-voltage circuitbreaker failure prevention and
Maintenance plan provides an auxiliary judgment.As shown in Fig. 2, the figure is typical divide-shut brake coil current signal curve, by electric current
Beginning and ending time and T1, T2, T3, I1, I2, I3 can be divided into four-stage as characteristic parameter, the curve:
0-T1: in zero moment, coil is initially powered up, and electric current exponentially rises, in the process, the electromagnetic force that electric current generates
It gradually increases, reaches first wave crest to T1 moment electric current, the electromagnetic force that electric current generates at this time is greater than the resistance of the external world suffered by iron core
Power, iron core setting in motion.The stage can reflect coil voltage, loop resistance and iron core whether clamping stagnation situations such as.
After T1-T2:T1, iron core setting in motion, coil current reduces at this time, arrives the T2 moment, and electric current reaches minimum value,
Speed of dynamic iron core drastically reduces during this, arrives the T2 moment, and iron core hits lock/trip gear, then stop motion.This rank
Section can reflect the iron core movement failures such as bite situation and de- mouth failure.
After T2-T3:T2 moment iron core stop motion, the lock of transmission mechanism is opened, this stage current persistently rises, and moves
Contact starts to act, and tripping spring starts separating brake, arrives the T3 moment, and electric current reaches peak value.The stage can reflect transmission mechanism
Motion state.
T3- finish time: T3 moment dynamic/static contact is kept completely separate, and auxiliary switch disconnects, and is cut off coil power, is produced between contact
Raw electric arc, arc voltage sharply increase in a short time, this electric current directly resulted in strongly reduces, and is finally reduced to 0.The stage
Current curve can reflect auxiliary switch with the presence or absence of failure.
The present invention divide-shut brake coil current curve is carried out by using cubic spline interpolation method it is smooth, then using asking
The method of slope is extracted feature vector of the corresponding time parameter of three extreme points of electric current as current signal, accordingly to disconnected
The mechanical breakdown of road device is determined.
1.2 Vibrations
Mechanical oscillation are the effects by stress, cause motion morphology to change and a kind of vibration shape for showing
Formula.For breaker, the energy release of storage causes a series of connecting rod band moving contact to move, entire motion process
In, mechanical part is started, moved, buffered and hit according to certain logical order, this series of shock can be
Breaker casing and operating mechanism ontology generate shock wave, have stronger timing.The impact of mechanical part corresponds to vibration
Wave crest in time domain plethysmographic signal, the motion state of component is corresponding with overall waveform in entire motion process, as shown in Figure 3.
Vibration is the response of a variety of driving sources of circuit breaker internal, and specific to the breaker studied, driving source includes cam
The movement of the internal components such as mechanism, four-bar mechanism, spring mechanism, iron core lance.Breaker in actual motion is with operation
The aggravation of increase and the abrasion of number, the motion state of operating mechanism component include that the time interval of component movement and energy can be sent out
Changing causes collected vibration signal time domain waveform and normal condition difference occur, in this, as the spy for extracting breaker
Levy the theoretical foundation of parameter.Under different malfunctions, time, time interval and the vibration amplitude that vibration event occurs are not
Equally, the High Voltage Circuit Breaker Condition can be characterized effectively as the fault signature of breaker.
The present invention enhances the shock characteristic of vibration event in signal, so by the short-time energy value of calculating vibration signal
Time of origin and the end time of vibration event are extracted using double threshold method afterwards, makeup time parameter vector completes different faults
The construction of state feature vector.
2 theoretical basis
The 2.1 double threshold method principles based on short-time energy method
The processing technique of voice signal is used for reference, double threshold method is commonly used to carry out end-point detection to it.This method it is basic
Principle is that two threshold value Y1 and Y2 (Y1 < Y2) are arranged first, and short-time energy function value is compared with threshold value Y2, is higher than threshold value Y2
Be impact event certainly, start-stop point should be located at except time point corresponding to threshold value Y2 and short-time energy function value intersection point.
Then lower threshold value Y1 is chosen, from the joint with threshold value Y2 to two-sided search, finds two intersected with threshold value Y1 respectively
Point, this section be exactly double threshold method according in short-term can entropy than the impact event of judgement start-stop point.
In short-time analysis, short-time energy analysis, which is equivalent to, converts the advanced row index of signal, then with segmentation or framing
The method of superposition is pocessed.Vibration signal is divided into multistage multiplied by a time-limited window function by collected vibration signal,
Its characteristic parameter is analyzed respectively, wherein each section is known as one " frame ", frame length generally takes 10~30ms.Body vibration is believed in this way
For number, each frame characteristic parameter constitutes characteristic parameter time series.
If signal sequence is x (i), i=0 ..., N-1, then short-time energy function S (n) is defined as
In formula: w (n) is sliding window function, n=0 ..., M-1；S (n) represents signal in the local energy of moment n.
Short-time energy analysis is actually that signal square passes through the linear filter that a unit function res is w (n)
Output.So its performance depends on the selection of window function, in practice, there are commonly rectangular windows and hamming window, it is contemplated that hamming
The bandwidth of window is larger, and the outer rate of decay is very fast, causes the distortion of its input signal smaller, so the present invention is using hamming window to vibration
Dynamic signal is handled.The selection long for window, the small then high resolution of length, but cannot show in short-term can high excellent of signal-to-noise ratio
Gesture comprehensively considers, and combined floodgate vibration signal sample frequency of the invention is 10K, and window is long to choose 90.
2.2SVM principle
Support vector machines (Support Vector Machine, SVM) is to realize one kind of structural risk minimization
General learning algorithm, sample is mapped to high-dimensional feature space using kernel function by it, then the structural classification interval in this space
Maximum linear classification hyperplane, thus support vector machines is relatively more suitable for the classification of Small Sample Database.Its basic thought is shown in figure
4。
Figure orbicular spot and side's point respectively indicate two class training samples, and two class samples are by errorless point completely of optimal separating hyper plane
Open, by sample nearest from hyperplane in sample and be parallel to the straight line of hyperplane, between them between be divided into class interval,
Sample point above is exactly supporting vector.The sum of two class supporting vectors and optimal hyperlane spacing are 2 | | w | |, therefore, construction is most
Excellent hyperplane problem is converted into optimization problem:
In formula: the normal vector of w- optimal hyperlane；B- threshold value.
Extend to linearly inseparable situation, it is contemplated that some samples cannot correctly be classified by hyperplane, introduce relaxation vector
εi>=0 loosens condition, then hyperplane constraint condition is yi(w·x+b)+εi>=1, while punishment parameter C is added to introduce to ε i most
The target of smallization.Objective function are as follows:
The above problem is solved using method of Lagrange multipliers, obtains optimization object function:
Corresponding constraint condition are as follows:0≤αi≤C。αiFor Lagrange multiplier.Nonlinear problem is expanded to,
It can use mapping phi (x) sample in lower dimensional space is mapped as in higher dimensional space, so that data sample is in higher dimensional space
Linear separability.Define kernel function K (xi,xj)=φ (xi)Tφ(xj), optimization object function is at this time
3 experimental applications and analysis
The experiment of 3.1 high-voltage circuitbreakers
The present invention is using a 35kV outdoor high-voltage SF6 breaker as experimental subjects.Hall current sensor is selected to measure line
Loop current signal selects the vibration signal of the 1A102E type acceleration transducer measurement operating mechanism of Jiangsu east magnificent test company.
The transducer range is 0~5000m/s2, frequency response: 0.5~10000Hz.In view of the breaker closing time be 80 ±
15ms, setting acquisition time are 200ms, sample frequency 10K, ensure that the abundant acquisition of vibration signal.
3.2 characteristic vector pickup
3.2.1 divide-shut brake coil current signal extracts
Collected current signal includes many spikes and trough in experiment, affects the extraction of signal characteristic, so this
Invention is smooth to the progress of divide-shut brake current signal using the method for cubic spline interpolation, utilizes the method pair for asking slope approximation derivation
Curve Maximization is sought, and it is as shown in Figure 5 to seek process.
Because the sampling interval is sufficiently small, the derivative value of the point is replaced with the slope approximation of adjacent two o'clock.Wherein, xi with
Yi is respectively the corresponding coordinate value of i point.If i point slope is 0, the point is judged for stationary point, and taking the midpoint coordinates of i and i+1 is pole
Value point coordinate；If i point slope is not 0, the symbol of adjacent two o'clock slope is investigated, contrary sign then determines that i+1 point is extreme point.Its
In, the value of i depends on closing time, it is known that closing time is up to 95ms, so i takes 950 points.
For most of signal, three groups of extreme values can be obtained, but still there is some small spikes and trough in curve unavoidably, are had
When the case where will appear more than three extreme point, as shown in Figure 6.
So to be grouped processing to the extreme point detected, it is grouped true according to division brake current specific experiment data
It is fixed.According to switching current curve waveform it is found that Wave crest and wave trough appearance sequence be first wave crest again trough finally again be wave crest, so
Current signal is divided into three groups, seeks y value maximum value or minimum value in three groups of extreme value data respectively, i.e., for first group
It seeks maximum value and determines T1, second group is minimized determining T2, and third group is maximized determining T3, and table 1 is Fig. 6 current curve
Extreme point group result.The part-time parameter that divide-shut brake coil current signal extracts is shown in Table 2 after normalized.
1 extreme point group result of table
3.2.2 vibration signal time parameter extracts
By Fig. 7, it is apparent that short-time energy method matches well with vibration signal characteristics, relative to vibration signal wave
Shape, short-time energy method are more sensitive for shock characteristic.Compared to other two kinds of time domain approach, short-time energy method is for compared with favourable opposition
What the temporal characteristics hit embodied becomes apparent from, and facilitates the subsequent extraction to temporal characteristics.So present invention application short-time energy method pair
Signal is handled, and the vibration performance of signal is enhanced.
After handling signal, feature extraction is carried out with double threshold method, process is as follows: utilizing the detection displacement of double threshold method
Point.By preset threshold value, compared with the short-time energy function value of signal, when the short-time energy function value of signal is more than to set
When determining threshold value, then this moment is judged to conjugate point.By screen suitable threshold value can accurately judge it is each during separating brake
The beginning of a event and finish time, as a result as shown in Figure 8.
According to double threshold method testing principle, each vibration event time started is marked with red solid line in Fig. 8, the end time
It is marked with red dotted line.By being found to analysis of vibration signal under different working condition, vibration event under different faults state
Time of origin, time interval and end time are different, so at the beginning and end of extracting maximum two vibration events
Between totally 4 time parameters (T4, T5, T6, T7) set up feature vector.Wherein, buffer spring is collided when T4 and T5 is closing operation
The vibration of generation, T6 and T7 are the vibration that dynamic/static contact collision generates.It is operated 12 times under every kind of state, 4 kinds of states totally 48 groups of numbers
According to extracting respective time parameter respectively, complete the building of feature vector, the part-time parameter attribute vector extracted is shown in
Table 2.
2 part-time parameter attribute vector of table
3.3 pattern-recognition
Since breaker cannot be operated continually, cause collected data in experiment limited, most of pattern-recognition side
Method needs a large amount of test sample to guarantee the accuracy of result, and SVM is solving small sample, the knowledge of non-linear and high dimensional pattern
There is distinctive advantage in not, be suitable for the fault identification and classification of breaker.For above-mentioned 48 groups of sample datas, every kind is extracted
6 groups of data of state, 4 kinds of states totally 24 groups of data as training sample；Remaining 24 groups of data are as test sample.SVM essence is
A two classifier here using the classification policy of " a pair of remaining ", constructs 4 SVM, forms 4 groups for four kinds of working conditions
The binary classifier of " a pair three ", SVM classifier select the linear kernel function linear of default, by carrying out to training set sample
Training obtains training pattern, and test set data are inputted in training pattern and are calculated, classification results are obtained.Test set in total 24
Group data, 6 groups of data of every kind of state, are as shown in Figure 9 to the classifying quality of vibration electric signal.
In order to verify the validity of SVM method, recombination time parameter attribute is carried out with BP neural network and FCM respectively
Classification.Wherein, BP neural network sets 7 input nodes, 1 output node and 8 concealed nodes, and smoothing factor is set as m in FCM
=2,3 kinds of algorithm for pattern recognition classification results are as shown in table 3.
3 different mode recognizer classification results of table
From table 3 it can be seen that the classifying quality of SVM is substantially better than BP neural network and FCM for same signal, for
The recognition accuracy of working state of circuit breaker is higher, and diagnosis effect is good, and SVM is suitable for the identification of small sample training, thus have compared with
High fault recognition rate.
4 different faults tagsort result of table
It is available from table 4, the identification of signal time parameter is electrically coupled for single signal and vibration, SVM is for vibration electricity
Time parameter classifying quality is more preferable, and extracted fault signature becomes apparent from, and classification results are more accurate, can be accurate and comprehensively anti-
Reflect the operating status of high-voltage circuit-breaker operating mechanism.
The present invention is using coil current and operating mechanism vibration signal as research object, using the double threshold based on short-time energy
Method carries out feature extraction to high-voltage circuitbreaker signal, and is classified using SVM to four kinds of states of breaker, has following spy
Point:
(1) using divide-shut brake coil current curve and fault feature vector of the vibration signal time parameter as breaker, phase
For single signal, extracted fault signature is become apparent from, and classification results are more accurate, accurately and comprehensively can reflect that height is broken
The operating status of road device operating mechanism；
(2) svm classifier recognition result is substantially better than BP neural network and the classification recognition result of FCM, and this method is for disconnected
The recognition accuracy of road device working condition is high, has stronger practicability；
(3) the double threshold method based on short-time energy method can fast and effeciently extract the feature vector of breaker time parameter,
SVM can accurately carry out state recognition and classification, provide and a kind of determined according to time parameter for circuit breaker failure diagnosis
Method.
It based on the high-voltage circuitbreaker feature extraction that is electrically coupled of vibration and classification method is of the invention specific shown in figure 1 above
Embodiment has embodied substantive distinguishing features of the present invention and progress, needs can be used according to actual, in enlightenment of the invention
Under, the equivalent modifications of shape, structure etc., the column in the protection scope of this programme are carried out to it.
Claims (5)
1. the high-voltage circuitbreaker feature extraction and classification method that are electrically coupled based on vibration, it is characterised in that comprising steps of
1) data acquire
Using Hall current sensor measure coil currents signal, the vibration using acceleration transducer measurement operating mechanism is believed
Number；Hall current sensor, acceleration transducer are with set acquisition time, sample frequency to coil current signal and operation machine
The vibration signal of structure is acquired the abundant acquisition to guarantee vibration signal；
2) characteristic vector pickup
By calculating the short-time energy value of vibration signal, to enhance the shock characteristic of vibration event in signal, then using two-door
Limit method extract vibration event time of origin and the end time, makeup time parameter vector, complete different faults state feature to
The construction of amount；
Characteristic vector pickup includes that the feature extraction of divide-shut brake coil current signal and vibration signal characteristics extract；Divide-shut brake coil electricity
Flow signal characteristic abstraction comprising steps of
201) smooth treatment is carried out to divide-shut brake current signal, Curve Maximization is asked using the method for asking slope approximation derivation
It takes；
202) processing is grouped to the extreme point detected, current signal is divided into three groups, seek three groups of extreme value data respectively
In y value maximum value or minimum value, i.e., seek maximum value for first group and determine T1, second group is minimized determining T2,
Third group is maximized determining T3；
Vibration signal characteristics extract comprising steps of
203) signal is handled using short-time energy method, enhances the vibration performance of signal；
204) double threshold method detection displacement point is utilized；By preset threshold value, compared with the short-time energy function value of signal,
When the short-time energy function value of signal is more than given threshold, then this moment is judged to conjugate point；It determines each during separating brake
The beginning of event and finish time, extract start and end times of maximum two vibration events totally 4 time parameter T4,
T5, T6, T7 set up feature vector；Wherein, T4 and T5 be closing operation when collide buffer spring generate time of vibration, T6 and
T7 is the time of vibration that dynamic/static contact collision generates；N times are operated under every kind of operating status, according to extracting respective time parameter,
The building of feature vector is completed, operating status is including normal, switching-in spring power is small, buffering elastic spring force is small, control loop voltage is low；
3) pattern-recognition
Every kind of state m group data are extracted, as training sample；Remaining data is as test sample；Classification knowledge is carried out using SVM
Not, by being trained to obtain training pattern to training set sample, test set data is inputted in training pattern and are calculated, are obtained
To classification results.
2. the high-voltage circuitbreaker feature extraction according to claim 1 being electrically coupled based on vibration and classification method, feature are existed
In: in step 1), setting acquisition time is 200ms, sample frequency 10K.
3. the high-voltage circuitbreaker feature extraction according to claim 2 being electrically coupled based on vibration and classification method, feature are existed
In: in step 201), Curve Maximization seeks step and includes:
A) setting i=1, xi and yi is respectively the corresponding coordinate value of i point；
B) judge whether i≤950, if entering in next step, otherwise terminate；
C) slope k i is calculated；
D) judge whether ki=0, if so, entering step h)；If it is not, entering in next step；
E) judge whether ki*k (i+1) < 0, if then entering in next step, if it is not, then entering step g)；
F) x (i+1) is exported, y (i+1)
G) i=i+1；And it is back to step b)；
H) (xi+x (i+1))/2 is exported, (yi+y (i+1))/2.
4. the high-voltage circuitbreaker feature extraction according to claim 3 being electrically coupled based on vibration and classification method, feature are existed
In: in step 204), operated 12 times under every kind of state, 4 kinds of states totally 48 groups of data extract respective time parameter respectively,
Complete the building of feature vector.
5. the high-voltage circuitbreaker feature extraction according to claim 4 being electrically coupled based on vibration and classification method, feature are existed
In: in step 3), to 48 groups of sample datas, 6 groups of data of every kind of state are extracted, totally 24 groups of data are used as training sample to 4 kinds of states
This；Remaining 24 groups of data are as test sample；4 SVM are constructed, the binary classifier of 4 groups " a pair three ", SVM classifier are formed
The linear kernel function linear for selecting default, by being trained to obtain training pattern to training set sample, by test set data
It is calculated in input training pattern, obtains classification results.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810785672.6A CN109188258A (en) | 2018-07-17 | 2018-07-17 | The high-voltage circuitbreaker feature extraction and classification method being electrically coupled based on vibration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810785672.6A CN109188258A (en) | 2018-07-17 | 2018-07-17 | The high-voltage circuitbreaker feature extraction and classification method being electrically coupled based on vibration |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109188258A true CN109188258A (en) | 2019-01-11 |
Family
ID=64936859
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810785672.6A Pending CN109188258A (en) | 2018-07-17 | 2018-07-17 | The high-voltage circuitbreaker feature extraction and classification method being electrically coupled based on vibration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109188258A (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110426192A (en) * | 2019-08-14 | 2019-11-08 | 南方电网科学研究院有限责任公司 | A kind of the acoustic fingerprints detection system and method for breaker |
| CN110553830A (en) * | 2019-09-26 | 2019-12-10 | 国网四川省电力公司泸州供电公司 | online detection method for mechanical characteristics of metal-enclosed switchgear |
| CN110674725A (en) * | 2019-09-20 | 2020-01-10 | 电子科技大学 | Equipment signal type identification method based on multi-dimensional feature vector combination of detection signals |
| CN111157853A (en) * | 2019-12-31 | 2020-05-15 | 广西诚新慧创科技有限公司 | Method and system for identifying discharge state of power transmission line |
| CN111474471A (en) * | 2020-05-29 | 2020-07-31 | 国网安徽省电力有限公司电力科学研究院 | Method for extracting current characteristic parameters of opening and closing coil of high-voltage alternating-current circuit breaker |
| CN111948528A (en) * | 2019-05-16 | 2020-11-17 | 株式会社日立制作所 | Diagnostic method and device for opening and closing device |
| CN112085328A (en) * | 2020-08-03 | 2020-12-15 | 北京贝壳时代网络科技有限公司 | Risk assessment method, system, electronic device and storage medium |
| CN112729381A (en) * | 2020-12-11 | 2021-04-30 | 广州致新电力科技有限公司 | Fault diagnosis method of high-voltage circuit breaker based on neural network |
| CN112816859A (en) * | 2021-01-19 | 2021-05-18 | 国网宁夏电力有限公司培训中心 | On-line monitoring and intelligent evaluation system for on-off coil state of circuit breaker |
| CN112924861A (en) * | 2021-01-28 | 2021-06-08 | 国网宁夏电力有限公司培训中心 | Capacitor switching circuit breaker state monitoring method |
| CN113219328A (en) * | 2021-01-27 | 2021-08-06 | 中国国家铁路集团有限公司 | Intelligent fault diagnosis method for circuit breaker operating mechanism based on information fusion |
| CN113376515A (en) * | 2020-03-09 | 2021-09-10 | 西门子股份公司 | Method and device for determining closing time of circuit breaker and computer readable medium |
| CN113466679A (en) * | 2021-05-17 | 2021-10-01 | 浙江工业大学 | Method for estimating service life of circuit breaker |
| CN113945835A (en) * | 2020-07-16 | 2022-01-18 | 上海汽车集团股份有限公司 | Relay health state online prediction method and device and electronic equipment |
| CN114662548A (en) * | 2022-04-12 | 2022-06-24 | 安徽中安昊源电力科技有限公司 | Breaker diagnosis method and system based on action abnormity |
| CN115144742A (en) * | 2022-07-06 | 2022-10-04 | 云南电网有限责任公司电力科学研究院 | Method and device for distinguishing mechanical state of circuit breaker energy storage system |
| CN115184787A (en) * | 2022-06-29 | 2022-10-14 | 云南电网有限责任公司电力科学研究院 | Online measuring method, device and equipment for ablation degree of circuit breaker |
| WO2023026728A1 (en) * | 2021-08-27 | 2023-03-02 | Imv株式会社 | Vibration test device |
| CN117033950A (en) * | 2023-10-08 | 2023-11-10 | 华中科技大学 | GIS isolating switch mechanical fault on-line diagnosis method and device |
| CN117872123A (en) * | 2024-01-24 | 2024-04-12 | 广东电网有限责任公司江门供电局 | High-voltage circuit breaker fault diagnosis method based on mechanical vibration signals |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000304834A (en) * | 1999-04-22 | 2000-11-02 | Meidensha Corp | Test device for circuit-breaker |
| CN202041618U (en) * | 2010-12-13 | 2011-11-16 | 云南电力试验研究院(集团)有限公司 | Online monitoring device for high-voltage breaker |
| CN103323770A (en) * | 2013-05-23 | 2013-09-25 | 国家电网公司 | Device for detection of mechanical characteristics and diagnosis of faults of high-voltage circuit breaker |
| CN103487749A (en) * | 2013-09-18 | 2014-01-01 | 国家电网公司 | On-line monitoring and diagnosing system and method for mechanical state of high-voltage circuit breaker |
| CN105759160A (en) * | 2016-05-11 | 2016-07-13 | 郑州瑞能电气有限公司 | Online monitoring method of non-invasive line breaker or transformer |
| WO2016206056A1 (en) * | 2015-06-25 | 2016-12-29 | 国家电网公司 | Circuit breaker detection method, device and system |
-
2018
- 2018-07-17 CN CN201810785672.6A patent/CN109188258A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000304834A (en) * | 1999-04-22 | 2000-11-02 | Meidensha Corp | Test device for circuit-breaker |
| CN202041618U (en) * | 2010-12-13 | 2011-11-16 | 云南电力试验研究院(集团)有限公司 | Online monitoring device for high-voltage breaker |
| CN103323770A (en) * | 2013-05-23 | 2013-09-25 | 国家电网公司 | Device for detection of mechanical characteristics and diagnosis of faults of high-voltage circuit breaker |
| CN103487749A (en) * | 2013-09-18 | 2014-01-01 | 国家电网公司 | On-line monitoring and diagnosing system and method for mechanical state of high-voltage circuit breaker |
| WO2016206056A1 (en) * | 2015-06-25 | 2016-12-29 | 国家电网公司 | Circuit breaker detection method, device and system |
| CN105759160A (en) * | 2016-05-11 | 2016-07-13 | 郑州瑞能电气有限公司 | Online monitoring method of non-invasive line breaker or transformer |
Non-Patent Citations (5)
| Title |
|---|
| 孙银山等: ""高压断路器分合闸线圈电流信号特征提取与故障判别方法研究"", 《高压电器》 * |
| 杨为等: ""基于样条插值和线性拟合分析的高压断路器弹簧机构状态检测"", 《高压电器》 * |
| 王飞: ""基于线圈电流和振动信号分析的断路器机械状态评估方法研究"", 《中国优秀硕士学位论文全文数据库（工程科技Ⅱ辑）》 * |
| 费宇泉等: ""语音端点检测算法研究"", 《仪器仪表与检测技术》 * |
| 赵科等: ""基于信号特征融合与优化的高压断路器机械状态评估"", 《高压电器》 * |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111948528A (en) * | 2019-05-16 | 2020-11-17 | 株式会社日立制作所 | Diagnostic method and device for opening and closing device |
| CN111948528B (en) * | 2019-05-16 | 2023-04-18 | 株式会社日立制作所 | Diagnostic method and device for opening and closing device |
| CN110426192A (en) * | 2019-08-14 | 2019-11-08 | 南方电网科学研究院有限责任公司 | A kind of the acoustic fingerprints detection system and method for breaker |
| CN110674725A (en) * | 2019-09-20 | 2020-01-10 | 电子科技大学 | Equipment signal type identification method based on multi-dimensional feature vector combination of detection signals |
| CN110674725B (en) * | 2019-09-20 | 2022-06-03 | 电子科技大学 | Equipment signal type identification method based on multi-dimensional feature vector combination of detection signals |
| CN110553830B (en) * | 2019-09-26 | 2021-07-20 | 国网四川省电力公司泸州供电公司 | Online detection method for mechanical characteristics of metal-enclosed switchgear |
| CN110553830A (en) * | 2019-09-26 | 2019-12-10 | 国网四川省电力公司泸州供电公司 | online detection method for mechanical characteristics of metal-enclosed switchgear |
| CN111157853A (en) * | 2019-12-31 | 2020-05-15 | 广西诚新慧创科技有限公司 | Method and system for identifying discharge state of power transmission line |
| CN113376515A (en) * | 2020-03-09 | 2021-09-10 | 西门子股份公司 | Method and device for determining closing time of circuit breaker and computer readable medium |
| CN111474471A (en) * | 2020-05-29 | 2020-07-31 | 国网安徽省电力有限公司电力科学研究院 | Method for extracting current characteristic parameters of opening and closing coil of high-voltage alternating-current circuit breaker |
| CN113945835B (en) * | 2020-07-16 | 2024-04-05 | 上海汽车集团股份有限公司 | Relay health state online prediction method and device and electronic equipment |
| CN113945835A (en) * | 2020-07-16 | 2022-01-18 | 上海汽车集团股份有限公司 | Relay health state online prediction method and device and electronic equipment |
| CN112085328B (en) * | 2020-08-03 | 2024-05-24 | 北京贝壳时代网络科技有限公司 | Risk assessment method, system, electronic equipment and storage medium |
| CN112085328A (en) * | 2020-08-03 | 2020-12-15 | 北京贝壳时代网络科技有限公司 | Risk assessment method, system, electronic device and storage medium |
| CN112729381A (en) * | 2020-12-11 | 2021-04-30 | 广州致新电力科技有限公司 | Fault diagnosis method of high-voltage circuit breaker based on neural network |
| CN112729381B (en) * | 2020-12-11 | 2023-05-02 | 广州致新电力科技有限公司 | Fault diagnosis method of high-voltage circuit breaker based on neural network |
| CN112816859A (en) * | 2021-01-19 | 2021-05-18 | 国网宁夏电力有限公司培训中心 | On-line monitoring and intelligent evaluation system for on-off coil state of circuit breaker |
| CN113219328A (en) * | 2021-01-27 | 2021-08-06 | 中国国家铁路集团有限公司 | Intelligent fault diagnosis method for circuit breaker operating mechanism based on information fusion |
| CN113219328B (en) * | 2021-01-27 | 2023-02-17 | 中国国家铁路集团有限公司 | Intelligent fault diagnosis method for circuit breaker operating mechanism based on information fusion |
| CN112924861A (en) * | 2021-01-28 | 2021-06-08 | 国网宁夏电力有限公司培训中心 | Capacitor switching circuit breaker state monitoring method |
| CN113466679B (en) * | 2021-05-17 | 2024-01-05 | 浙江工业大学 | Circuit breaker service life estimation method |
| CN113466679A (en) * | 2021-05-17 | 2021-10-01 | 浙江工业大学 | Method for estimating service life of circuit breaker |
| JP2023032978A (en) * | 2021-08-27 | 2023-03-09 | Imv株式会社 | Vibration testing device |
| WO2023026728A1 (en) * | 2021-08-27 | 2023-03-02 | Imv株式会社 | Vibration test device |
| JP7261270B2 (en) | 2021-08-27 | 2023-04-19 | Imv株式会社 | Vibration test equipment |
| CN114662548A (en) * | 2022-04-12 | 2022-06-24 | 安徽中安昊源电力科技有限公司 | Breaker diagnosis method and system based on action abnormity |
| CN115184787A (en) * | 2022-06-29 | 2022-10-14 | 云南电网有限责任公司电力科学研究院 | Online measuring method, device and equipment for ablation degree of circuit breaker |
| CN115144742A (en) * | 2022-07-06 | 2022-10-04 | 云南电网有限责任公司电力科学研究院 | Method and device for distinguishing mechanical state of circuit breaker energy storage system |
| CN117033950A (en) * | 2023-10-08 | 2023-11-10 | 华中科技大学 | GIS isolating switch mechanical fault on-line diagnosis method and device |
| CN117033950B (en) * | 2023-10-08 | 2023-12-26 | 华中科技大学 | GIS isolating switch mechanical fault on-line diagnosis method and device |
| CN117872123A (en) * | 2024-01-24 | 2024-04-12 | 广东电网有限责任公司江门供电局 | High-voltage circuit breaker fault diagnosis method based on mechanical vibration signals |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109188258A (en) | The high-voltage circuitbreaker feature extraction and classification method being electrically coupled based on vibration | |
| CN106199412B (en) | A kind of permanent magnet mechanism high-pressure vacuum breaker method of fault pattern recognition | |
| CN106646096B (en) | Transformer fault classification and recognition methods based on vibration analysis method | |
| WO2021103496A1 (en) | Abnormal vibration-based gas insulated combined switchgear mechanical failure diagnosis method | |
| Ma et al. | A PMU-based data-driven approach for classifying power system events considering cyberattacks | |
| CN109948597A (en) | A kind of Mechanical Failure of HV Circuit Breaker diagnostic method | |
| CN109784310B (en) | Power distribution switch mechanical fault feature extraction method based on CEEMDAN and weighted time-frequency entropy | |
| CN107367687B (en) | A kind of high-voltage breaker spring fault degree detection method and device | |
| CN108919104B (en) | Breaker fault diagnosis method based on Fisher discriminant classification method | |
| CN114757110B (en) | Circuit breaker fault diagnosis method based on sliding window detection and current extraction signals | |
| CN112924861A (en) | Capacitor switching circuit breaker state monitoring method | |
| CN109800740A (en) | A kind of OLTC mechanical failure diagnostic method based on Sample Entropy and SVM | |
| CN1232834C (en) | Online detection method for vacuum circuit breaker contact on-off time based on vibration analysis | |
| CN112557966A (en) | Transformer winding looseness identification method based on local mean decomposition and support vector machine | |
| CN112730964A (en) | Lightning overvoltage identification method based on overvoltage waveform characteristics | |
| CN105241643A (en) | High-voltage circuit breaker mechanical state monitoring method employing HS transformation and single-type support vector machine | |
| CN113297786B (en) | Low-voltage fault arc sensing method based on semi-supervised machine learning | |
| CN112284707B (en) | Method for processing vibration signal of circuit breaker | |
| CN113901862A (en) | Electromagnetic repulsion mechanism fault monitoring method, system, equipment and readable storage medium | |
| CN110568300B (en) | Power distribution network single-phase earth fault identification method based on multi-source information | |
| CN115113038B (en) | Circuit breaker fault diagnosis method based on current signal phase space reconstruction | |
| Li et al. | High-voltage circuit breaker fault diagnosis model based on coil current and KNN | |
| Aljohani | Centralized fault detection and classification for motor power distribution centers utilizing MLP-NN and stockwell transform | |
| CN110988597A (en) | Resonance type detection method based on neural network | |
| CN110244219B (en) | Breaker fault identification method based on closing coil current time domain statistical characteristics |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190111 |
|
| WD01 | Invention patent application deemed withdrawn after publication |