CN110006645A - A kind of Mechanical Failure of HV Circuit Breaker diagnostic method of multi-source fusion - Google Patents
A kind of Mechanical Failure of HV Circuit Breaker diagnostic method of multi-source fusion Download PDFInfo
- Publication number
- CN110006645A CN110006645A CN201910387365.7A CN201910387365A CN110006645A CN 110006645 A CN110006645 A CN 110006645A CN 201910387365 A CN201910387365 A CN 201910387365A CN 110006645 A CN110006645 A CN 110006645A
- Authority
- CN
- China
- Prior art keywords
- sensor
- sample
- frequency
- measurement sample
- diagnosis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- 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
Abstract
The invention discloses a kind of Mechanical Failure of HV Circuit Breaker diagnostic methods of multi-source fusion, belong to multi-sensor information fusion diagnostic field.This method is primarily based on vibration measurement device while acquiring the vibration signal of the multiple positions of high-voltage circuitbreaker, calculates the wavelet energy entropy of vibration signal, constitutes the vibration performance vector of description mechanical state of high-voltage circuit breaker;Then, the vibration data of each sensor is grouped, form model training collection and assessment collection, Softmax regression diagnostics model is designed according to model training collection, and accuracy rate of diagnosis of each sensor under Softmax regression diagnostics model is calculated using assessment collection, form the confidence weight of multiple sensor mechanism fault diagnosises;Finally, the improvement D-S evidence fusion method based on each sensor diagnostic weight of proposition, realizes Mechanical Failure of HV Circuit Breaker identification.The unilateral influence of single-sensor diagnosis is effectively reduced in the present invention, greatly improves high-voltage circuitbreaker mechanical defect accuracy rate of diagnosis.
Description
Technical field
The invention belongs to multi-sensor information fusion diagnostic field, the high-voltage circuitbreaker of specifically a kind of multi-source fusion is mechanical
Method for diagnosing faults.
Background technique
High-voltage circuitbreaker is one of equipment important in electric system, carries crucial control and protective effect.With
The expansion of power grid scale, the utilization rate sustained, stable growth of high-voltage circuitbreaker.In order to cope with the complexity of electric system, power grid pair
Breaker performance requirement under different voltages grade is different, acts on irreplaceable.Circuit breaker failure in order to prevent needs replacing
Old or inoperative component maintains the operating status of breaker, at present to the maintenance mode of high-voltage circuitbreaker have subsequent maintenance,
Periodic inspection and repair based on condition of component.Subsequent maintenance cannot reduce the influence caused by power grid of high-voltage circuitbreaker failure, and endanger big;
There are blindness for periodic inspection, cause the cost of overhaul excessive, waste of manpower and material resources.Therefore, high-voltage circuitbreaker fortune is understood in time
Its fault type is predicted in row state, identification, and to improving, its operational reliability is particularly important.
Currently, detection and the condition diagnosing of effecting reaction mechanical characteristic of high-voltage circuit breaker, depend on high-voltage circuitbreaker
Moving contact stroke, opening/closing operation mechanism electric current and mechanical oscillation signal.Wherein contact travel and coil current detect skill
There are inconveniences in existing application test process for art, and some mechanical defect is difficult to recognize.And high-voltage circuitbreaker is mechanical
Vibration signal characteristics are extracted that there are still accuracy rate of diagnosis is low and poor universality etc. with Fault Classification by vibration detection.Base
It is numerous in the low reason of the Mechanical Failure of HV Circuit Breaker diagnostic accuracy of vibration information, as the big, feature of vibration signal dispersion property mentions
Insufficient, single position vibration information is taken to be difficult to the independent comprehensive description etc. obtained to various faults scene.
Therefore, efficient, the high precision breaker mechanical failure diagnostic method of research and development multiposition vibration information fusion, will be advantageous
Accuracy is recognized in improving, promotes the further development of mechanical state of high-voltage circuit breaker assessment.
Summary of the invention
The present invention mentions in view of the above-mentioned problems, propose a kind of Mechanical Failure of HV Circuit Breaker diagnostic method of multi-source fusion
The accuracy of high multisensor diagnosis fusion.
Specific step is as follows:
Step 1: the multisensor position vibration signal for obtaining different defects constitutes measurement sample, to each sensor position
When each measurement sample under setting carries out wavelet packet-frequency transformation, obtain it is each measure sample when-frequency energy matrix;
Firstly, acquiring high-voltage circuitbreaker simultaneously in the mechanical oscillation signal of n sensor position of different mechanical defects, often
The corresponding sample size of a sensor is m;For q-th of measurement sample X of i-th of sensori,qWhen progress wavelet packet-frequency
Transformation, when acquisition-frequency energy matrix Yi,q,t×f;Matrix Yi,q,t×fLine number be t, columns f.
Step 2: by each measurement sample when-frequency energy matrix distinguishes equal part with frequency axis along the time axis, constitute each
From it is several when-frequency block;
For when-frequency energy matrix Yi,q,t×f, divide n respectively etc. in time shaft t and the direction frequency axis f1And n2Part, constitute n1
×n2When a-frequency block;
n1It indicates along time orientation to matrix Yi,q,t×fEqual part number;n2It indicates along frequency direction to matrix Yi,q,t×f's
Equal part number;Each sub- period element number is t/n1, every sub- frequency band element number is f/n2。
Step 3: when calculating each-energy of frequency block, construct the vibration information feature description of each measurement sample;
Firstly, being directed to q-th of measurement sample of i-th of sensor, n is calculated1×n2When a-frequency block in (α, β) it is a
When-frequency block ENERGY Ei,q,α,β, formula is as follows:
Indicate (α, β) it is a when-frequency block in (S1, S2) a element;
Then, according to n1×n2When a-energy of frequency block, calculate the wavelet packet of q-th of measurement sample under i-th of sensor
Energy-Entropy, the vibration information feature for constituting the measurement sample describe WPEi,q={ WPEi,q,α,f,WPEi,q,t,β};
I=1,2 ..., n;Q=1,2 ..., m;α=1,2 ..., n1;β=1,2 ..., n2。
Step 4: being described according to the vibration information feature of each measurement sample, evaluate collection and training subset are divided, instruction is passed through
Practice Softmax regression diagnostics model, export the fault diagnosis accuracy rate of each sensor, and then calculates the confidence power of each sensor
Weight;
Firstly, m is measured the feature description composition set A of sample under i-th of sensori;
Set Ai={ Ai,1,Ai,2,...,Ai,q...,Ai,m};Ai,qIndicate q-th of measurement sample of i-th of sensor
Feature description;Ai,q=[xi,q,1,xi,q,2,...,xi,q,w]T;W indicates Characteristic Number, i.e. feature space dimension.
Then, characteristic descriptor set is closed into AiIt is divided into k subset, each subset is evaluate collection, each evaluate collection and sample
The difference set of this collection is corresponding training subset;
K evaluate collection is combined into { Bi,1,Bi,2...,Bi,k};Training subset is combined into { Ci,1,Ci,2...,Ci,k};Ci,k=Ai\
Bi,k;
According to k training subset training Softmax regression diagnostics model, inputs k evaluate collection and export corresponding examine
Disconnected accuracy rate calculates fault diagnosis Average Accuracy P of the average value as i-th of sensori;
The fault diagnosis Average Accuracy for similarly calculating n sensor, obtains the confidence weight of i-th of sensor;
The confidence weight of i-th of sensor are as follows:
Confidence weight vectors ω=[ω of corresponding n sensor1,ω2,...,ωn]。
Step 5: calculating the vibration letter of the sample A to be tested under i-th of sensor for some new sample A to be tested
Cease feature description vectors Oi;
Step 6: by vibration information feature description vectors OiInput diagnostic model Mi, obtaining the sample A to be tested may send out
The probability column vector Q of raw s class failurei;QiThe vector arranged for s row 1.
Diagnostic model MiFor the Softmax regression diagnostics model of i-th of sensor;The Softmax of each sensor is returned
Diagnostic model is respectively designated as corresponding diagnostic model;
Similarly calculate separately s class fault rate column vector set { Q of the sample A to be tested under n sensor1,
Q2,...Qi,...,Qn};
Step 7: using the confidence weight of n sensor and the s class fault rate column vector of each sensor, meter
Calculate the Mean Vector Q of s class fault rateλ;
Step 8: the sample A to be tested is directed to, by the s class fault rate column vector and the phase that calculate each sensor
Hope vector QλBetween Euclidean distance, further obtain the s class failure fusion probability column vector that the sample A to be tested may occur
mass;
Probability column vector QiWith Mean Vector QλBetween Euclidean distance diIt calculates as follows:
Using Euclidean distance define the s class failure that the sample A to be tested may occur for n sensor merge probability column to
Measure mass;
Step 9: after being merged column vector mass n-1 times using traditional D-S evidential reasoning, maximum probability in definition column
Place fault category is the corresponding fault type of sample A to be tested, completes final Mechanical Failure of HV Circuit Breaker diagnosis.
The excellent effect of the present invention is that:
1, the Mechanical Failure of HV Circuit Breaker diagnostic method of a kind of multi-source fusion is different from conventional high-tension circuit breaker failure and examines
Disconnected method, the present invention use for reference the relevant knowledge of Multi-source Information Fusion, using and improve D-S evidence theory, form more sensings
The diagnosis of device diagnostic message is merged, and is avoided single-sensor to the blindness of fault identification result, is improved fault diagnosis
Robustness.
2, the Mechanical Failure of HV Circuit Breaker diagnostic method of a kind of multi-source fusion, by the training dataset of single-sensor point
For several subsets, and permutation and combination formed multiple training subsets and corresponding assessment collection (training dataset and training subset
Difference set), Softmax regression model is designed using training subset, according to the diagnostic result of assessment collection, defines the diagnosis of this sensor
Accuracy rate.The accuracy rate of diagnosis for normalizing multiple sensors obtains the confidence of each sensor, improves multisensor diagnosis and melts
The accuracy of conjunction.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the Mechanical Failure of HV Circuit Breaker diagnostic method of multi-source fusion of the present invention;
Fig. 2 is that multisensor confidence of the present invention calculates schematic diagram;
Fig. 3 is the circuit breaker failure taxonomic structure figure the present invention is based on Softmax regression model;
Fig. 4 is the multi-sensor information fusion diagnostic flow chart that the present invention improves D-S evidential reasoning;
Fig. 5 is certain model high-voltage circuitbreaker vibration information acquisition lab diagram of the invention.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description.
The present invention is based on the fault diagnosis model of the multiple position vibration informations of high-voltage circuitbreaker, a kind of multi-source fusion is formed
Mechanical Failure of HV Circuit Breaker diagnostic method, firstly, obtaining multiposition vibration data and original wavelet Energy-Entropy feature sky simultaneously
Between construct;Then, training subset and evaluate collection and evaluation sensor fault diagnosis result confidence are constructed;Finally, with each sensing
The Softmax model of device and the diagnostic result of the diagnosis confidence fusion multiple Softmax of decision.
Specifically: it is primarily based on vibration measurement device while acquiring high-voltage circuitbreaker mechanical defect under multiple measurement points
Vibration data sample constructs traditional wavelet-packet energy entropy to it, forms vibration information feature space;Then, by each sensing
For the vibration data of device averagely, without intersection to be grouped, the vibration performance sample set under each measurement point (sensor) is evaluate collection
And the training subset of data difference set is acquired with original accordingly;Then, Softmax regression diagnostics are carried out to each training subset to build
Mould calculates accuracy rate of diagnosis of each sensor under Softmax regression diagnostics model and each using corresponding evaluate collection
The Average Accuracy of measuring point, and define with this confidence weight of each sensor;Finally, according to the confidence of each sensor and examining
Disconnected probability calculation diagnosis probability vector expectation, is based on each sensor diagnostic probability vector and the desired distance of gained probability vector,
Define all kinds of fault rate vectors of the test sample under the fusion of multiple sensors;Based on improve D-S evidence fusion method,
With the Softmax model of each sensor and the diagnostic result of the diagnosis confidence fusion multiple Softmax of decision, repeatedly merge multiple
All kinds of fault rate vectors under sensor fusion, defining the corresponding fault type of maximum probability, test sample occurs thus
Failure, complete the last diagnostic of high-voltage circuitbreaker mechanical defect.
Analysis of experimental results shows that the unilateral influence of single-sensor diagnosis can be effectively reduced in the present invention, is greatly improved
High-voltage circuitbreaker mechanical defect accuracy rate of diagnosis.
As shown in Figure 1, the specific steps are as follows:
Step 1: the multisensor position vibration signal for obtaining different defects constitutes measurement sample, to each sensor position
When each measurement sample under setting carries out wavelet packet-frequency transformation, obtain it is each measure sample when-frequency energy matrix;
Firstly, n vibration for being acquired high-voltage circuitbreaker simultaneously using vibration information measuring device in different mechanical defects is added
The mechanical oscillation signal of velocity sensor position, the corresponding sample size of each sensor are m;For i-th sensor
Q-th of measurement sample Xi,qWhen progress wavelet packet-frequency transformation, wherein i=1,2 ... n, i indicate sensor number;Q=1,2 ...,
M indicates measurement sample number, when acquisition-frequency energy matrix Yi,q,t×f;Matrix Yi,q,t×fLine number be t, columns f.
Step 2: by each measurement sample when-frequency energy matrix distinguishes equal part with frequency axis along the time axis, constitute each
From it is several when-frequency block;
For when-frequency energy matrix Yi,q,t×f, divide n respectively etc. in time shaft t and the direction frequency axis f1And n2Part, constitute n1
×n2When a-frequency block;
n1It indicates along time orientation (line direction) to matrix Yi,q,t×fEqual part number;n2It indicates along frequency direction (column side
To) to matrix Yi,q,t×fEqual part number;Each sub- period element number is t/n1, every sub- frequency band element number is f/
n2。
Step 3: when calculating each-energy of frequency block, construct the vibration information feature description of each measurement sample;
Show as impact free damping signal based on high voltage circuit breaker closing process, i.e., signal be mutation, frequency at any time
Between Variation Features, therefore utilize wavelet packet Time-Frequency Analysis, calculate wavelet-packet energy entropy, measurement vibration information feature description;
Firstly, being directed to q-th of measurement sample of i-th of acceleration transducer of high-voltage circuitbreaker, n is calculated1×n2When a-frequency
When (α, β) in block is a-ENERGY E of frequency blocki,q,α,β, formula is as follows:
Wherein Ei,q,α,βWhen indicating q-th of measurement sample wavelet packet under i-th of acceleration transducer-frequency matrix Yi,q,t×f
Carry out n1×n2When (α, β) is a after piecemeal-energy of frequency block,It indicates under i-th of acceleration transducer
When (α, β) is a in q-th of measurement sample-frequency block in (S1, S2) a element;
Then, according to n1×n2When a-energy of frequency block, calculate the wavelet packet of q-th of measurement sample under i-th of sensor
Energy-Entropy, as shown in formula (2) and (3), the vibration information feature for constituting the measurement sample describes WPEi,q={ WPEi,q,α,f,
WPEi,q,t,β};
I=1,2 ..., n;Q=1,2 ..., m;α=1,2 ..., n1;β=1,2 ..., n2。
Step 4: being described according to the vibration information feature of each measurement sample, evaluate collection and training subset are divided, instruction is passed through
Practice Softmax regression diagnostics model, export the fault diagnosis accuracy rate of each sensor, and then calculates the confidence power of each sensor
Weight;
As shown in Fig. 2, firstly, m is measured the feature description composition set A of sample under i-th of sensori;
The feature of q-th of measurement sample under i-th of sensor is described into WPEi,q={ WPEi,q,α,f,WPEi,q,t,β, definition
ForWherein Ai,qIndicate the spy of the wavelet-packet energy entropy of q-th of measurement sample under i-th of sensor
Collection, w indicate Characteristic Number, i.e. feature space dimension;The vibration wavelet packet Energy-Entropy feature sample of sensor confidence is not calculated
This set Ai={ Ai,1,Ai,2,...,Ai,q...,Ai,m};
Then, characteristic descriptor set is closed into AiIt is divided into k subset, the intersection between each subset is empty set, union is all
Sample set, defining this subset is evaluate collection, and each evaluate collection is corresponding training subset with the difference set of sample set;
K evaluate collection is combined into { Bi,1,Bi,2...,Bi,k, meet
Training subset is combined into { Ci,1,Ci,2...,Ci,k};Ci,k=Ai\Bi,k;
Softmax regression diagnostics model is constructed according to k training subset, as shown in figure 3, and optimizing using gradient descent method
It is C based on training subseti,kSoftmax regression diagnostics MODEL C Mi,k.Assuming that sharing s fault type, returned in Softmax
Shown in middle system equation such as formula (4).
Wherein p (yq=s | xi,q;θ) indicate s class failure of q-th of measurement sample under parameter θ under i-th of sensor
The probability of generation, θ indicate the matrix of a s × w,Calculate gradientSoftmax regression diagnostics MODEL C Mi,kIt calculates under i-th of sensor
K-th of evaluate collection Bi,kAccuracy rate Pi,k, input k evaluate collection and export corresponding accuracy rate of diagnosis, after completing k evaluation
Calculate fault diagnosis Average Accuracy P of the average value as i-th of sensori;
The fault diagnosis Average Accuracy for similarly calculating n sensor, according to the Average Accuracy of each sensor, definition is each
The confidence weight of sensor;
The confidence weight of i-th of sensor are as follows:
Confidence weight vectors ω=[ω of corresponding n sensor1,ω2,...,ωn]。
Step 5: calculating the vibration letter of the sample A to be tested under i-th of sensor for some new sample A to be tested
Cease feature description vectors Oi;
Oi=[xi,1,xi,2,...,xi,w]T
Step 6: by vibration information feature description vectors OiInput diagnostic model Mi, obtaining the sample A to be tested may send out
The probability column vector Q of raw s class failurei;
QiThe vector arranged for s row 1.
Diagnostic model MiFor the Softmax regression diagnostics model of i-th of sensor;The Softmax of each sensor is returned
Diagnostic model is respectively designated as corresponding diagnostic model;
S class failure of the sample A to be tested under n sensor Softmax diagnostic model is similarly calculated separately to occur generally
Rate column vector set { Q1,Q2,...Qi,...,Qn};
Step 7: using the confidence weight of n sensor and the s class fault rate column vector of each sensor, meter
Calculate the Mean Vector Q of s class fault rateλ;
Step 8: the sample A to be tested is directed to, by the s class fault rate column vector and the phase that calculate each sensor
Hope vector QλBetween Euclidean distance, further obtain the s class failure fusion probability column vector that the sample A to be tested may occur
mass;
Probability column vector QiWith Mean Vector QλBetween Euclidean distance diIt calculates as follows:
Using Euclidean distance define the s class failure that the sample A to be tested may occur for n sensor merge probability column to
Measure mass;
The probability that s class failure may be occurred to test sample by forming n sensor is assigned;
Step 9: using traditional D-S evidential reasoning by column vector massiAfter fusion n-1 times, maximum probability in definition column
Place fault category is the corresponding fault type of sample A to be tested, completes final Mechanical Failure of HV Circuit Breaker diagnosis.
It is as follows according to traditional D-S Evidential reasoning algorithm:
Define mass1And mass2It is the basic probability assignment of framework of identification F, framework of identification F={ F1,F2,...,FsBe by
The s finite aggregates that mutual exclusion element is constituted two-by-two;FsIt indicates that s class failure occurs, while defining massiIndicate i-th of sensor
To for SoftMax model MiOutput, i.e. the basic probability assignment for s class failure, then multisensor massiFusion process
As shown in formula (5).
Wherein massc(Fs) indicate fusion after s class failure probability, as i=1, massc(Fs)=mass1(Fs);
massi+1(Fsj) indicating that i+1 sensor test sample occurs the probability value of sj class failure, this illustrates n sensor only
It needs to merge n-1 times;kcIndicate that conflict coefficient is equal to
It defines n sensor of high-voltage circuitbreaker and probability column vector mass is merged to the s class failure that test sample may occur
For the basic probability assignment of framework of identification F, s class failure is framework of identification F, will using formula (5) with D-S Evidential reasoning algorithm
After mass Vector Fusion n-1 times, the maximum probability place fault category corresponding fault type of test sample thus is defined, is completed most
The Mechanical Failure of HV Circuit Breaker diagnosis of D-S evidential reasoning is improved eventually.As shown in figure 4, to improve more sensings of D-S evidential reasoning
Device information merges process.
Example is as shown in figure 5, certain model high-voltage circuitbreaker is that experiment porch is arranged normal condition, tripping spring fatigue, closes
Lock spring fatigue, oil bumper oil leak, transmission shaft damping increases, the abrasion of transmission pivot pin and foundation bolt loosen 7 kinds of operating conditions,
Done each 350 times of combined floodgate experiment altogether, every kind operating condition 50 times, the vibration of experiment 3 measurement positions of acquisition simultaneously is believed every time
Breath, selects 70% data of every kind of operating condition at random, i.e. 35 data are trained, and remaining data carries out surveying experiment card.Instruction
Practice sample number m=245 and divides n along time shaft etc. in wavelet packet time-frequency matrix1=12, divide n along frequency axis etc.2=13, then it is divided into
For 12 × 13 energy blocks, the original feature space dimension w that wavelet-packet energy entropy is constituted is that 12+13 is equal to 25, based on every class event
The wavelet-packet energy entropy feature for hindering 15 test samples is utilized respectively what Softmax model was diagnosed in three test positions
Accuracy rate is 84.76%, 76.19% and 73.33%, and the accuracy rate of diagnosis using traditional D-S evidential reasoning fusion method is
95.24%, and after the method for the invention makes inferences fusion, high-voltage circuitbreaker mechanical defect accuracy rate of diagnosis is 97.14%,
Show that this method significantly reduces single-sensor and breaks to the height based on vibration information by the diagnostic result of test data
Road device fault diagnosis error probability establishes the improvement D-S multi-source fusion method of multiple sensor confidence weights, can be further
Improve the accuracy rate of failure modes.
Finally it should be noted that: described embodiment is only some embodiments of the present application, rather than whole realities
Apply example.Based on the embodiment in the application, those of ordinary skill in the art are obtained without making creative work
Every other embodiment, shall fall in the protection scope of this application.
Claims (2)
1. a kind of Mechanical Failure of HV Circuit Breaker diagnostic method of multi-source fusion, which is characterized in that specific step is as follows:
Step 1: the multisensor position vibration signal for obtaining different defects constitutes measurement sample, under each sensor position
Each measurement sample carry out wavelet packet when-frequency transformation, obtain it is each measure sample when-frequency energy matrix;
Firstly, acquiring high-voltage circuitbreaker simultaneously in the mechanical oscillation signal of n sensor position of different mechanical defects, Mei Gechuan
The corresponding sample size of sensor is m;For q-th of measurement sample X of i-th of sensori,qWhen progress wavelet packet-frequency transformation,
When acquisition-frequency energy matrix Yi,q,t×f;Matrix Yi,q,t×fLine number be t, columns f;
Step 2: by it is each measurement sample when-frequency energy matrix along the time axis with frequency axis distinguish equal part, constitute it is respective
When several-frequency block;
Step 3: when calculating each-energy of frequency block, construct the vibration information feature description of each measurement sample;
Firstly, being directed to q-th of measurement sample of i-th of sensor, n is calculated1×n2When a-frequency block in (α, β) it is a when-frequency
The ENERGY E of blocki,q,α,β, formula is as follows:
Indicate (α, β) it is a when-frequency block in (S1, S2) a element;
Then, according to n1×n2When a-energy of frequency block, calculate the wavelet-packet energy of q-th of measurement sample under i-th of sensor
Entropy, the vibration information feature for constituting the measurement sample describe WPEi,q={ WPEi,q,α,f,WPEi,q,t,β};
I=1,2 ..., n;Q=1,2 ..., m;α=1,2 ..., n1;β=1,2 ..., n2;
Step 4: being described according to the vibration information feature of each measurement sample, evaluate collection and training subset are divided, training is passed through
Softmax regression diagnostics model exports the fault diagnosis accuracy rate of each sensor, and then calculates the confidence power of each sensor
Weight;
Step 5: the vibration information for calculating the sample A to be tested under i-th of sensor is special for some new sample A to be tested
Levy description vectors Oi;
Step 6: by vibration information feature description vectors OiInput diagnostic model Mi, obtain the s that the sample A to be tested may occur
The probability column vector Q of class failurei;QiThe vector arranged for s row 1;
Diagnostic model MiFor the Softmax regression diagnostics model of i-th of sensor;By the Softmax regression diagnostics mould of each sensor
Type is respectively designated as corresponding diagnostic model;
Similarly calculate separately s class fault rate column vector set { Q of the sample A to be tested under n sensor1,
Q2,...Qi,...,Qn};
Step 7: calculating s class using the confidence weight of n sensor and the s class fault rate column vector of each sensor
The Mean Vector Q of fault rateλ;
Step 8: be directed to the sample A to be tested, by calculate the s class fault rate column vector of each sensor and it is expected to
Measure QλBetween Euclidean distance, further obtain the s class failure fusion probability column vector that the sample A to be tested may occur
mass;
Probability column vector QiWith Mean Vector QλBetween Euclidean distance diIt calculates as follows:
N sensor is defined using Euclidean distance, and probability column vector is merged to the s class failure that the sample A to be tested may occur
mass;
Step 9: after being merged column vector mass n-1 times using traditional D-S evidential reasoning, in definition column where maximum probability
Fault category is the corresponding fault type of sample A to be tested, completes final Mechanical Failure of HV Circuit Breaker diagnosis.
2. a kind of Mechanical Failure of HV Circuit Breaker diagnostic method of multi-source fusion as described in claim 1, which is characterized in that institute
The step of stating four specifically:
Firstly, m is measured the feature description composition set A of sample under i-th of sensori;
Set Ai={ Ai,1,Ai,2,...,Ai,q...,Ai,m};Ai,qIndicate that the feature of q-th of measurement sample of i-th of sensor is retouched
It states;Ai,q=[xi,q,1,xi,q,2,...,xi,q,w]T;W indicates Characteristic Number, i.e. feature space dimension;
Then, characteristic descriptor set is closed into AiIt is divided into k subset, each subset is evaluate collection, each evaluate collection and sample set
Difference set is corresponding training subset;
K evaluate collection is combined into { Bi,1,Bi,2...,Bi,k};Training subset is combined into { Ci,1,Ci,2...,Ci,k};Ci,k=Ai\Bi,k;
According to k training subset training Softmax regression diagnostics model, inputs k evaluate collection and export corresponding diagnosis standard
True rate calculates fault diagnosis Average Accuracy P of the average value as i-th of sensori;
The fault diagnosis Average Accuracy for similarly calculating n sensor, obtains the confidence weight of i-th of sensor;
The confidence weight of i-th of sensor are as follows:
Confidence weight vectors ω=[ω of corresponding n sensor1,ω2,...,ωn]。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910387365.7A CN110006645B (en) | 2019-05-10 | 2019-05-10 | Multi-source fusion high-voltage circuit breaker mechanical fault diagnosis method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910387365.7A CN110006645B (en) | 2019-05-10 | 2019-05-10 | Multi-source fusion high-voltage circuit breaker mechanical fault diagnosis method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110006645A true CN110006645A (en) | 2019-07-12 |
CN110006645B CN110006645B (en) | 2020-07-03 |
Family
ID=67176410
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910387365.7A Active CN110006645B (en) | 2019-05-10 | 2019-05-10 | Multi-source fusion high-voltage circuit breaker mechanical fault diagnosis method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110006645B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110879373A (en) * | 2019-12-12 | 2020-03-13 | 国网电力科学研究院武汉南瑞有限责任公司 | Oil-immersed transformer fault diagnosis method with neural network and decision fusion |
CN111272405A (en) * | 2020-03-26 | 2020-06-12 | 广西电网有限责任公司电力科学研究院 | High-voltage circuit breaker mechanical fault diagnosis method and system |
CN111460701A (en) * | 2020-03-09 | 2020-07-28 | 中海油田服务股份有限公司 | Fault diagnosis model training method and device |
CN111504627A (en) * | 2020-05-07 | 2020-08-07 | 山东泰开高压开关有限公司 | Method for detecting defective parts of circuit breaker |
CN111734669A (en) * | 2020-07-02 | 2020-10-02 | 重庆大学 | Multi-source information layered fusion centrifugal blower fault diagnosis method |
CN112083292A (en) * | 2020-10-26 | 2020-12-15 | 积成电子股份有限公司 | Power distribution network active fault studying and judging method based on multi-source non-sound information fusion |
CN112733951A (en) * | 2021-01-19 | 2021-04-30 | 中国矿业大学(北京) | Multi-information decision weight distribution and fusion method for mechanical defect diagnosis of circuit breaker |
CN112748331A (en) * | 2020-12-24 | 2021-05-04 | 国网江苏省电力有限公司电力科学研究院 | Circuit breaker mechanical fault identification method and device based on DS evidence fusion |
CN112906472A (en) * | 2021-01-19 | 2021-06-04 | 中国矿业大学(北京) | Circuit breaker defect identification method based on self-service sampling method |
CN117077505A (en) * | 2023-06-14 | 2023-11-17 | 中国人民解放军海军航空大学 | Design method of equipment built-in fault diagnostic device based on testability evaluation |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011185632A (en) * | 2010-03-05 | 2011-09-22 | Ntn Corp | Device and method for detecting faulure of bearing |
CN106897739A (en) * | 2017-02-15 | 2017-06-27 | 国网江苏省电力公司电力科学研究院 | A kind of grid equipment sorting technique based on convolutional neural networks |
CN107015145A (en) * | 2017-06-01 | 2017-08-04 | 湖州知维技术服务有限公司 | Primary cut-out is monitored on-line and state evaluation system |
CN107449994A (en) * | 2017-07-04 | 2017-12-08 | 国网江苏省电力公司电力科学研究院 | Partial discharge method for diagnosing faults based on CNN DBN networks |
CN107796602A (en) * | 2016-08-31 | 2018-03-13 | 华北电力大学(保定) | A kind of circuit breaker failure diagnostic method of sound and vibration signal fused processing |
CN109145886A (en) * | 2018-10-12 | 2019-01-04 | 西安交通大学 | A kind of asynchronous machine method for diagnosing faults of Multi-source Information Fusion |
CN109472097A (en) * | 2018-11-19 | 2019-03-15 | 国网湖北省电力有限公司黄石供电公司 | A kind of transmission line of electricity on-line monitoring equipment method for diagnosing faults |
-
2019
- 2019-05-10 CN CN201910387365.7A patent/CN110006645B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2011185632A (en) * | 2010-03-05 | 2011-09-22 | Ntn Corp | Device and method for detecting faulure of bearing |
CN107796602A (en) * | 2016-08-31 | 2018-03-13 | 华北电力大学(保定) | A kind of circuit breaker failure diagnostic method of sound and vibration signal fused processing |
CN106897739A (en) * | 2017-02-15 | 2017-06-27 | 国网江苏省电力公司电力科学研究院 | A kind of grid equipment sorting technique based on convolutional neural networks |
CN107015145A (en) * | 2017-06-01 | 2017-08-04 | 湖州知维技术服务有限公司 | Primary cut-out is monitored on-line and state evaluation system |
CN107449994A (en) * | 2017-07-04 | 2017-12-08 | 国网江苏省电力公司电力科学研究院 | Partial discharge method for diagnosing faults based on CNN DBN networks |
CN109145886A (en) * | 2018-10-12 | 2019-01-04 | 西安交通大学 | A kind of asynchronous machine method for diagnosing faults of Multi-source Information Fusion |
CN109472097A (en) * | 2018-11-19 | 2019-03-15 | 国网湖北省电力有限公司黄石供电公司 | A kind of transmission line of electricity on-line monitoring equipment method for diagnosing faults |
Non-Patent Citations (1)
Title |
---|
胡洪武: "基于振动信号处理的高压断路器故障诊断系统研究", 《万方数据》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110879373A (en) * | 2019-12-12 | 2020-03-13 | 国网电力科学研究院武汉南瑞有限责任公司 | Oil-immersed transformer fault diagnosis method with neural network and decision fusion |
CN111460701A (en) * | 2020-03-09 | 2020-07-28 | 中海油田服务股份有限公司 | Fault diagnosis model training method and device |
CN111272405A (en) * | 2020-03-26 | 2020-06-12 | 广西电网有限责任公司电力科学研究院 | High-voltage circuit breaker mechanical fault diagnosis method and system |
CN111504627B (en) * | 2020-05-07 | 2022-04-29 | 山东泰开高压开关有限公司 | Method for detecting defective parts of circuit breaker |
CN111504627A (en) * | 2020-05-07 | 2020-08-07 | 山东泰开高压开关有限公司 | Method for detecting defective parts of circuit breaker |
CN111734669A (en) * | 2020-07-02 | 2020-10-02 | 重庆大学 | Multi-source information layered fusion centrifugal blower fault diagnosis method |
CN112083292A (en) * | 2020-10-26 | 2020-12-15 | 积成电子股份有限公司 | Power distribution network active fault studying and judging method based on multi-source non-sound information fusion |
CN112083292B (en) * | 2020-10-26 | 2023-09-05 | 积成电子股份有限公司 | Active fault studying and judging method for power distribution network based on multisource non-robust information fusion |
CN112748331A (en) * | 2020-12-24 | 2021-05-04 | 国网江苏省电力有限公司电力科学研究院 | Circuit breaker mechanical fault identification method and device based on DS evidence fusion |
CN112733951A (en) * | 2021-01-19 | 2021-04-30 | 中国矿业大学(北京) | Multi-information decision weight distribution and fusion method for mechanical defect diagnosis of circuit breaker |
CN112733951B (en) * | 2021-01-19 | 2021-09-28 | 中国矿业大学(北京) | Multi-information decision weight distribution and fusion method for mechanical defect diagnosis of circuit breaker |
CN112906472A (en) * | 2021-01-19 | 2021-06-04 | 中国矿业大学(北京) | Circuit breaker defect identification method based on self-service sampling method |
CN117077505A (en) * | 2023-06-14 | 2023-11-17 | 中国人民解放军海军航空大学 | Design method of equipment built-in fault diagnostic device based on testability evaluation |
Also Published As
Publication number | Publication date |
---|---|
CN110006645B (en) | 2020-07-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110006645A (en) | A kind of Mechanical Failure of HV Circuit Breaker diagnostic method of multi-source fusion | |
CN110376522B (en) | Motor fault diagnosis method of data fusion deep learning network | |
Ranjan et al. | K-nearest neighbors and grid search cv based real time fault monitoring system for industries | |
CN110108431B (en) | Mechanical equipment fault diagnosis method based on machine learning classification algorithm | |
CN110704801B (en) | Bridge cluster structure operation safety intelligent monitoring and rapid detection complete method | |
CN108304661A (en) | Diagnosis prediction method based on TDP models | |
CN110361176A (en) | A kind of intelligent failure diagnosis method for sharing neural network based on multitask feature | |
CN105868770B (en) | A kind of Fault Diagnosis for HV Circuit Breakers method based on unsupervised learning model | |
CN105466693B (en) | The pre- diagnostic method of Fault of Diesel Fuel System based on gray model | |
CN105629958B (en) | A kind of batch process method for diagnosing faults based on sub-period MPCA SVM | |
CN109948194B (en) | High-voltage circuit breaker mechanical defect integrated learning diagnosis method | |
CN109583520A (en) | A kind of state evaluating method of cloud model and genetic algorithm optimization support vector machines | |
CN114993669B (en) | Multi-sensor information fusion transmission system fault diagnosis method and system | |
CN115542099B (en) | Online GIS partial discharge detection method and device | |
CN112069930A (en) | Vibration signal processing method and device for improving GIS equipment fault diagnosis accuracy | |
CN102608519B (en) | Circuit failure diagnosis method based on node information | |
CN112949402A (en) | Fault diagnosis method for planetary gear box under minimum fault sample size | |
CN115238785A (en) | Rotary machine fault diagnosis method and system based on image fusion and integrated network | |
CN111260823B (en) | Fault diagnosis method based on I/O (input/output) measuring point fault dependency matrix | |
CN106644436B (en) | A kind of assessment method of breaker mechanic property | |
CN106054859B (en) | The double-deck integrated form industrial process fault detection method based on amendment type independent component analysis | |
CN110057588B (en) | Bearing early fault detection and diagnosis method and system based on fusion of singular value and graph theory characteristics | |
CN115659249A (en) | Intelligent station-finding control system anomaly detection method | |
Niu et al. | A hybrid bearing prognostic method with fault diagnosis and model fusion | |
CN108053093A (en) | A kind of k- neighbour's method for diagnosing faults based on the conversion of average influence Value Data |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |