CN107016404A - Wind power generating set failure prediction method based on D S evidence fusions - Google Patents
Wind power generating set failure prediction method based on D S evidence fusions Download PDFInfo
- Publication number
- CN107016404A CN107016404A CN201710102612.5A CN201710102612A CN107016404A CN 107016404 A CN107016404 A CN 107016404A CN 201710102612 A CN201710102612 A CN 201710102612A CN 107016404 A CN107016404 A CN 107016404A
- Authority
- CN
- China
- Prior art keywords
- signal
- evidence
- fusions
- energy
- probability
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M99/00—Subject matter not provided for in other groups of this subclass
-
- 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
- G01R31/34—Testing dynamo-electric machines
- G01R31/343—Testing dynamo-electric machines in operation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
Based on the wind power generating set failure prediction method of D S evidence fusions, this method constructs two SVMs Jing Guo parameter optimization for two kinds of signals, and final prediction failure mode is provided after D S fusions as two evidences;Advantages of the present invention is as follows with good effect:(1)Traditional method for oscillating is simply analyzed vibration signal, according to vibrational energy characteristic vector constructing neural network, the machine learning algorithm model such as decision tree.But simply observation vibration signal can be by some malfunction misclassifications, such as bearing damage and rotor eccentricity can all cause vibration signal abnormal, now just can preferably distinguish two kinds of malfunctions by current signal,(2)The forecast model that this method is set up can be stored, it is not necessary to historical data extract repeatedly and trained, under the real-time estimate environment of wind field, can be quickly provide predicts the outcome.
Description
Technical field
The present invention relates to the failure prediction method of wind power generating set, and in particular to a kind of D-S evidence fusions vibrational energy
The method of feature and current energy feature, belongs to Research on Fan Fault Forecasting technical field.
Background technology
Being widely popularized and using with clean energy resource, wind generating technology has become the important of clean reproducible energy
Research object.But wind field is in uninhabited, the exceedingly odious environment of natural environment, the real-time monitoring to fan condition mostly
Just become more and more difficult, especially the large-scale promotion of offshore wind turbine, also allow Fault Diagnosis of Fan to become rich in challenge.So
And blower fan involves great expense, if can be found fan condition is not good early stage, timely expatriate personnel's spot repair, with regard to that can keep away
Exempt from that some failures are continuous worsening to ultimately cause great economic damage.Wind field has had many years in the foundation and operation of China
Time, a large amount of valuable data are have accumulated, how fan operation have been found out by machine learning algorithm and computer technology
Relation between parameter and fan trouble just becomes more and more valuable.
Because D-S evidence theory has the ability of very strong processing uncertain information, in recent years the weight as information fusion
Means are wanted, but how to construct basic probability assignment function (BPA) is an important topic for first having to solve in merging.Due to
Machine learning algorithm can provide a kind of method for calculating Basic Probability As-signment for D-S evidence theory, therefore many scholars taste
The algorithm of examination application machine learning obtains BPA, herein also exactly in view of this thinking, by merging blower fan electrically and mechanically
The operational factor of two aspects, constructs machine learning algorithm respectively, is then merged two kinds of machine learning algorithms, draws pair
The prediction of fan condition finally.
The content of the invention
Goal of the invention
For generator of wind generating set in the failure of mechanically and electrically aspect, wavelet packet is based on this paper presents one kind
Decompose and D-S evidence theory multi-model fusion method, wherein WAVELET PACKET DECOMPOSITION is used for extracting vibration signal and current signal
Every feature vectors are constructed a SVMs, then obtained finally by D-S evidence fusions by the feature parameter vectors respectively
A forecast model.
Technical scheme
Because D-S evidence theory has the ability of very strong processing uncertain information, in recent years the weight as information fusion
Means are wanted, but how to construct basic probability assignment function (BPA) is an important topic for first having to solve in merging.Due to
Machine learning algorithm can provide a kind of method for calculating Basic Probability As-signment for D-S evidence theory, therefore many scholars taste
The algorithm of examination application machine learning obtains BPA, herein also exactly in view of this thinking, because the different conditions of generator are with shaking
Dynamic signal and current signal have close relation, herein for two kinds of signals construct two supports Jing Guo parameter optimization to
Amount machine, final prediction failure mode is provided as two evidences after D-S fusions.Due to shaking for generator different conditions
Dynamic signal and current signal have very big discrimination in frequency domain, extract vibration and electricity using three layers of WAVELET PACKET DECOMPOSITION herein
The feature parameter vectors of stream calculate model after being merged by D-S eventually through the mode of cross validation as SVM input
Error rate and loss, and by set forth herein Fusion Model and the grader of attribute superimposed structure without weight make respectively
Contrast, method of the data display herein based on multi-model evidence fusion has higher performance.
The step of a kind of method for carrying out Research on Fan Fault Forecasting based on D-S evidence fusions is included is as follows:
The first step, carries out data scrubbing to the historical data that wind field is gathered, removes some parameters unrelated with fan condition,
Because historical data is got off according to sequence of event, blower fan is all in good running status, institute in most cases
Retained with the data of several hours before for the data balancing before fault modeling, emphasis that blower fan breaks down
Come, and the failure mode of blower fan is indicated in behind data set, be convenient for the input of machine learning algorithm;
Second step, wind-driven generator is under different working condition, and the vibration signal and current signal of generator can all be showed
To be extremely strong non-stationary.Wavelet packet analysis can be simultaneously analyzed the time domain and frequency domain information of signal, can be effectively right
Non-stationary signal carries out feature extraction.Vibrating sensor gathers dynamo bearing x shaking to, y to, 45 degree, 135 degree four directions
Dynamic signal, current signal includes A phase, B phase, C phase three-phase electricity flow valuve, and sample frequency 50Hz is divided by three layers of WAVELET PACKET DECOMPOSITION
Not Chong Gou vibration energy signal and stator current frequency band energy distribution, finally respectively obtain one 32 dimension vibrational energy feature to
The current energy characteristic vector of amount and one 24 dimension;
3rd step, with vibrational energy characteristic signal and using current energy characteristic signal as input, constructs two warps respectively
Cross the SVMs of parameter optimization.Because the input parameter of D-S evidence fusions is the elementary probability on all classifying spaces
Assignment, so the hard output of traditional grader is converted into soft output, i.e., after the tag along sort output of grader is changed to
Test the prediction probability of probability output, i.e. grader in different classifications;
4th step, the output of two SVMs passes through D-S evidences as the Basic Probability As-signment of D-S evidence fusions
Fusion formula calculates predicted state of the final disaggregated model to blower fan.Analysis of experimental results meter by the way of five foldings intersection
The error rate and loss of context of methods are calculated, while being compared to be merged with D-S, tests and is not dividing proper subspace
In the case of, all properties participate in SVM training, and the weight of all properties is equal, and experimental result shows D-S evidence fusions
Classification accuracy and recall rate it is higher.
The step of wavelet packet extracts characteristic vector in second step is as follows:
(1) three layers of WAVELET PACKET DECOMPOSITION are carried out to vibration energy signal first, so as to obtain third layer 8 from low to high
The WAVELET PACKET DECOMPOSITION coefficient of sub-band
(2) coefficient of wavelet decomposition is reconstructed, extracts the signal of a sub- frequency band rangeThen resultant signal can be represented
For
(3) each sub-band signal energy is calculated.WithThe reconstruction signal of the 3rd layer of each node is represented,
Corresponding energy isBy SijThe energy of correspondence frequency band is designated as Eij, then have
X in formulajk(j=0 ... 7, k=1,2 ... n) represents reconstruction signalThe vibration amplitude of discrete point;
(4) energy of each sub-band is combined into a vector T, be expressed as
This characteristic vector is normalized
Normalization frequency band energy distribution after the WAVELET PACKET DECOMPOSITION of table 1
Normalization frequency band energy distribution after the WAVELET PACKET DECOMPOSITION of table 2
By vibration signal and current signal by three layers of WAVELET PACKET DECOMPOSITION and to being obtained after energy normalized in 3rd step
Characteristic vector as SVMs input, in order to ensure that the base classifier performance selected before fusion preferably, is melting
Tuning first is carried out to parameter before conjunction, D-S fusions are carried out from two minimum graders of test errors rate;SVM regulation ginseng
Number includes penalty factor, and gaussian kernel function parameter σ, initially set up standard SVM models, used in SVM parameters are selected
The mode of the folding cross validation of grid 5 determines optimized parameter, and wherein C hunting zone is set to [2-5,210], σ hunting zone is set
It is set to [2-10,25], both step-size in search changes are set to Cn+1=2*Cn,σn+1=2* σn+1, it is final to determine that two feature are empty
Between in SVM parameters it is as shown in table 3;
SVM parameters under 3 two proper subspaces of table
In 4th step, in D-S evidence theory, by mutual exclusive elementary sentence (it is assumed that) perfect set that constitutes is collectively referred to as
Identification framework Θ:
(1) basic probability assignment
BPA on identification framework Θ is one 2ΘThe function m of → [0,1], referred to as mass functions.And meetAndWherein so that m (A)>0 A is referred to as burnt first (Focal elements).Wherein Θ is also referred to as
Assuming that space, corresponds to [normal, bearing fault, turn-to-turn short circuit, rotor eccentricity] in this experiment, A represents the variable in Θ.
(2) Dempster composition rules
Dempster composition rules are also referred to as combining evidences formula[8], it is defined as follows:ForTwo on Θ
Mass functions m1,m2Dempster composition rules be:
Wherein, A represents to assume a kind of fault mode under space, m1(B) represent evidence 1 to failure A prediction probability, m2
(C) evidence 2 is represented to failure A prediction probability, and K is normaliztion constant
(3) mass functions are calculated, by taking bearing fault as an example
Wherein m1 represents the posterior probability that SVM1 is predicted bearing fault, and m2 represents the posteriority that SVM2 is predicted bearing fault
Probability, after being merged by above-mentioned formula, draws prediction probability of the Fusion Model to bearing fault, can similarly calculate fusion
Model is to the prediction probability of other three kinds of states, the final predicted state of last probability highest state representation Fusion Model.
The error rate and loss of context of methods are calculated by the way of five foldings intersection, while in order to merge work with D-S
Contrast, is tested in the case where not dividing proper subspace, and all properties participate in SVM training, and the weight of all properties is equal
It is equal.
Advantage and effect
Advantages of the present invention is as follows with good effect:
(1) traditional method for oscillating is simply analyzed vibration signal, and nerve is constructed according to vibrational energy characteristic vector
Network, the machine learning algorithm model such as decision tree.But simply observation vibration signal can be by some malfunction misclassifications, such as
Bearing damage and rotor eccentricity can all cause vibration signal abnormal, now just can preferably distinguish two kinds of failures by current signal
State
(2) forecast model that this method is set up can be stored, it is not necessary to historical data extract repeatedly and trained,
Under the real-time estimate environment of wind field, can be quickly provide predicts the outcome.
Brief description of the drawings
Fig. 1 is the flow chart of whole prediction process, i.e. the Research on Fan Fault Forecasting flow chart based on D-S evidence fusions.
Fig. 2 is the schematic diagram of SVMs.
Embodiment:
The present invention is described in further detail below in conjunction with the accompanying drawings.
As shown in Figure 1, the D-S evidence fusions algorithm that the present invention is designed is used for the failure predication of wind power generating set, it
The step of it is as follows:
The first step, carries out data scrubbing to the historical data that wind field is gathered, removes some parameters unrelated with fan condition,
Because historical data is got off according to sequence of event, blower fan is all in good running status, institute in most cases
Retained with the data of several hours before for the data balancing before fault modeling, emphasis that blower fan breaks down
Come, and the failure mode of blower fan is indicated in behind data set, be convenient for the input of machine learning algorithm;
Second step, wind-driven generator is under different working condition, and the vibration signal and current signal of generator can all be showed
To be extremely strong non-stationary.Wavelet packet analysis can be simultaneously analyzed the time domain and frequency domain information of signal, can be effectively right
Non-stationary signal carries out feature extraction.Vibrating sensor gathers dynamo bearing x shaking to, y to, 45 degree, 135 degree four directions
Dynamic signal, current signal includes A phase, B phase, C phase three-phase electricity flow valuve, and sample frequency 50Hz is divided by three layers of WAVELET PACKET DECOMPOSITION
Not Chong Gou vibration energy signal and stator current frequency band energy distribution, finally respectively obtain one 32 dimension vibrational energy feature to
The current energy characteristic vector of amount and one 24 dimension.Experiment is very stable in the range of 30m/s ± 5 from wind speed interval in data set
Data set, now blower fan be in normal power generating state.Bear vibration acceleration signal is decomposed with wavelet packet, carried
Each frequency band Wavelet Packet Energy Spectrum is taken, and energy normalized processing is carried out to it, characteristic vector is used as.Wavelet packet extracts characteristic vector
The step of it is as follows:
(1) three layers of WAVELET PACKET DECOMPOSITION are carried out to vibration energy signal first, so as to obtain third layer 8 from low to high
The WAVELET PACKET DECOMPOSITION coefficient of sub-band
(2) coefficient of wavelet decomposition is reconstructed, extracts the signal of a sub- frequency band rangeThen resultant signal can be represented
For
(3) each sub-band signal energy is calculated.WithThe reconstruction signal of the 3rd layer of each node is represented,
Corresponding energy isBy SijThe energy of correspondence frequency band is designated as Eij, then have
X in formulajk(j=0 ... 7, k=1,2 ... n) represents reconstruction signalThe vibration amplitude of discrete point;
(4) energy of each sub-band is combined into a vector T, be expressed as
This characteristic vector is normalized
Normalization frequency band energy distribution after the WAVELET PACKET DECOMPOSITION of table 1
Normalization frequency band energy distribution after the WAVELET PACKET DECOMPOSITION of table 2
3rd step, with vibrational energy characteristic signal and using current energy characteristic signal as input, constructs two warps respectively
Cross the SVMs of parameter optimization.Because the input parameter of D-S evidence fusions is the elementary probability on all classifying spaces
Assignment, so the hard output of traditional grader is converted into soft output, i.e., after the tag along sort output of grader is changed to
Test the prediction probability of probability output, i.e. grader in different classifications.By vibration signal and current signal by three layers of wavelet packet
Decompose and to the input of the characteristic vector that is obtained after energy normalized as SVMs, in order to ensure to be selected before fusion
Base classifier performance preferably, first tuning is carried out before fusion to parameter, from two classification that test errors rate is minimum
Device carries out D-S fusions.SVM regulation parameter include penalty factor, and gaussian kernel function parameter σ, initially set up standard
SVM models[14], optimized parameter, wherein C search model are determined by the way of the folding cross validation of grid 5 in the selection of SVM parameters
Enclose and be set to [2-5,210], σ hunting zone is set to [2-10,25], both step-size in search changes are set to Cn+1=2*Cn,
σn+1=2* σn+1, finally determine that SVM parameters are as shown in table 3 in two proper subspaces.
SVM parameters under 3 two proper subspaces of table
4th step, the output of two SVMs passes through D-S evidences as the Basic Probability As-signment of D-S evidence fusions
Fusion formula calculates predicted state of the final disaggregated model to blower fan.In D-S evidence theory, by mutual exclusive basic
Proposition (it is assumed that) composition perfect set be collectively referred to as identification framework Θ, several basic conceptions are first explained below:
(1) basic probability assignment
BPA on identification framework Θ is one 2ΘThe function m of → [0,1], referred to as mass functions.And meetAndWherein so that m (A)>0 A is referred to as burnt first (Focal elements).Wherein Θ is also referred to as
Assuming that space, corresponds to [normal, bearing fault, turn-to-turn short circuit, rotor eccentricity] in this experiment, A represents the variable in Θ.
(2) Dempster composition rules
Dempster composition rules are also referred to as combining evidences formula[8], it is defined as follows:ForTwo on Θ
Mass functions m1,m2Dempster composition rules be:
Wherein, A represents to assume a kind of fault mode under space, m1(B) represent evidence 1 to failure A prediction probability, m2
(C) evidence 1 is represented to failure A prediction probability, and K is normaliztion constant
(3) mass functions are calculated, by taking bearing fault as an example
Wherein m1 represents the posterior probability that SVM1 is predicted bearing fault, and m2 represents the posteriority that SVM2 is predicted bearing fault
Probability, after being merged by above-mentioned formula, draws prediction probability of the Fusion Model to bearing fault, can similarly calculate fusion
Model is to the prediction probability of other three kinds of states, the final predicted state of last probability highest state representation Fusion Model.
The error rate and loss of context of methods are calculated by the way of five foldings intersection, while in order to merge work with D-S
Contrast, is tested in the case where not dividing proper subspace, and all properties participate in SVM training, and the weight of all properties is equal
Equal, experimental result shows that the classification accuracy and recall rate of D-S evidence fusions are higher, and experimental result is as shown in table 4.
The all properties of table 4 are compared (mean value ± variance) (%) without Weighted Fusion with context of methods
Multiple experimental result shows, D-S Fusion Models the predicting the outcome in error rate to failure based on multi-categorizer
With the forecast model for being below not passing through evidence fusion in loss.
Claims (6)
1. the wind power generating set failure prediction method based on D-S evidence fusions, it is characterised in that:This method is for two kinds of signals
Two SVMs Jing Guo parameter optimization are constructed, final prediction is provided after D-S fusions as two evidences
Failure mode;
This method constructs machine learning algorithm, so respectively by merging operational factor of the blower fan at electrically and mechanically two aspects
Two kinds of machine learning algorithms are merged afterwards, the prediction final to fan condition is drawn;Extracted using three layers of WAVELET PACKET DECOMPOSITION
The feature parameter vectors of vibration and electric current are calculated eventually through the mode of cross validation and merged by D-S as SVM input
The error rate and loss of model afterwards, and by set forth herein Fusion Model and the classification of the attribute superimposed structure without weight
Device is compared respectively.
2. the wind power generating set failure prediction method according to claim 1 based on D-S evidence fusions, its feature exists
In:
The step of a kind of method for carrying out Research on Fan Fault Forecasting based on D-S evidence fusions is included is as follows:
The first step, carries out data scrubbing to the historical data that wind field is gathered, removes some parameters unrelated with fan condition, because
Historical data is got off according to sequence of event, and blower fan is all in good running status in most cases, so being
Data balancing before fault modeling, emphasis blower fan is broken down before the data of several hours remain,
And the failure mode of blower fan is indicated in behind data set, it is convenient for the input of machine learning algorithm;
Second step, wind-driven generator is under different working condition, and the vibration signal and current signal of generator can all show as pole
Strong is non-stationary;Wavelet packet analysis is analyzed the time domain and frequency domain information of signal simultaneously, and effectively non-stationary can be believed
Number carry out feature extraction;Vibrating sensor gathers vibration signals of the dynamo bearing x to, y to, 45 degree, 135 degree four directions,
Current signal includes A phase, B phase, C phase three-phase electricity flow valuve, and sample frequency 50Hz is shaken by three layers of WAVELET PACKET DECOMPOSITION to reconstruct respectively
Energy signal and the distribution of stator current frequency band energy, finally respectively obtain the one 32 vibrational energy characteristic vector tieed up and one
The current energy characteristic vector of 24 dimensions;
3rd step, with vibrational energy characteristic signal and using current energy characteristic signal as input, constructs two by ginseng respectively
The SVMs of number optimization;Because the input parameter of D-S evidence fusions is the Basic Probability As-signment on all classifying spaces,
So the hard output of traditional grader is converted into soft output, i.e., the tag along sort output of grader is changed to posterior probability
Output, i.e. prediction probability of the grader in different classifications;
4th step, the output of two SVMs passes through D-S evidence fusions as the Basic Probability As-signment of D-S evidence fusions
Formula calculates predicted state of the final disaggregated model to blower fan;Analysis of experimental results is calculated by the way of five foldings intersection
The error rate and loss of context of methods, while being compared to be merged with D-S, test and are not dividing the feelings of proper subspace
Under condition, all properties participate in SVM training, and the weight of all properties is equal, and experimental result shows point of D-S evidence fusions
Class accuracy rate and recall rate are higher.
3. the wind power generating set failure prediction method according to claim 2 based on D-S evidence fusions, its feature exists
In:The step of wavelet packet extracts characteristic vector in second step is as follows:
(1) three layers of WAVELET PACKET DECOMPOSITION are carried out to vibration energy signal first, so as to obtain third layer 8 son frequencies from low to high
The WAVELET PACKET DECOMPOSITION coefficient of band
(2) coefficient of wavelet decomposition is reconstructed, extracts the signal of a sub- frequency band rangeThen resultant signal is expressed as
(3) each sub-band signal energy is calculated;WithThe reconstruction signal of the 3rd layer of each node is represented, correspondence
Energy beBy SijThe energy of correspondence frequency band is designated as Eij, then have
X in formulajk(j=0 ... 7, k=1,2 ... n) represents reconstruction signalThe vibration amplitude of discrete point;
(4) energy of each sub-band is combined into a vector T, be expressed as
This characteristic vector is normalized
4. the wind power generating set failure prediction method according to claim 3 based on D-S evidence fusions, its feature exists
In:
Normalization frequency band energy distribution after the WAVELET PACKET DECOMPOSITION of table 1
Normalization frequency band energy distribution after the WAVELET PACKET DECOMPOSITION of table 2
5. the wind power generating set failure prediction method according to claim 2 based on D-S evidence fusions, its feature exists
In:Feature in 3rd step by vibration signal and current signal by three layers of WAVELET PACKET DECOMPOSITION and to being obtained after energy normalized
The vectorial input as SVMs, in order to ensure the base classifier performance selected before fusion preferably, before fusion
Tuning first is carried out to parameter, D-S fusions are carried out from two minimum graders of test errors rate;SVM regulation parameter includes
Penalty factor, and gaussian kernel function parameter σ, initially set up standard SVM models, SVM parameters selection in use grid 5
The mode of folding cross validation determines optimized parameter, and wherein C hunting zone is set to [2-5,210], σ hunting zone is set to
[2-10,25], both step-size in search changes are set to Cn+1=2*Cn,σn+1=2* σn+1, finally determine in two proper subspaces
SVM parameters are as shown in table 3;
SVM parameters under 3 two proper subspaces of table
6. the wind power generating set failure prediction method according to claim 2 based on D-S evidence fusions, its feature exists
In:In 4th step, in D-S evidence theory, by mutual exclusive elementary sentence (it is assumed that) perfect set that constitutes is collectively referred to as identification
Framework Θ:
(1) basic probability assignment
BPA on identification framework Θ is one 2ΘThe function m of → [0,1], referred to as mass functions;And meet
AndWherein so that m (A)>0 A is referred to as burnt first (Focal elements);Wherein Θ is also referred to as assuming empty
Between, [normal, bearing fault, turn-to-turn short circuit, rotor eccentricity] in this experiment is corresponded to, A represents the variable in Θ;
(2) Dempster composition rules
Dempster composition rules are also referred to as combining evidences formula[8], it is defined as follows:ForTwo mass letters on Θ
Number m1,m2Dempster composition rules be:
Wherein, A represents to assume a kind of fault mode under space, m1(B) represent evidence 1 to failure A prediction probability, m2(C) table
Show prediction probability of the evidence 2 to failure A, K is normaliztion constant
(3) mass functions are calculated, by taking bearing fault as an example
Wherein m1Represent the posterior probability that SVM1 is predicted bearing fault, m2The posterior probability that SVM2 is predicted bearing fault is represented,
After being merged by above-mentioned formula, prediction probability of the Fusion Model to bearing fault is drawn, Fusion Model is similarly calculated to it
The prediction probability of his three kinds of states, the final predicted state of last probability highest state representation Fusion Model;
The error rate and loss of context of methods are calculated by the way of five foldings intersection, while compared to be merged with D-S,
Test in the case where not dividing proper subspace, all properties participate in SVM training, the weight of all properties is equal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710102612.5A CN107016404A (en) | 2017-02-24 | 2017-02-24 | Wind power generating set failure prediction method based on D S evidence fusions |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710102612.5A CN107016404A (en) | 2017-02-24 | 2017-02-24 | Wind power generating set failure prediction method based on D S evidence fusions |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107016404A true CN107016404A (en) | 2017-08-04 |
Family
ID=59440530
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710102612.5A Pending CN107016404A (en) | 2017-02-24 | 2017-02-24 | Wind power generating set failure prediction method based on D S evidence fusions |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107016404A (en) |
Cited By (35)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107644231A (en) * | 2017-09-19 | 2018-01-30 | 广东工业大学 | A kind of generator amature method for diagnosing faults and device |
CN107831438A (en) * | 2017-10-25 | 2018-03-23 | 上海交通大学 | The Forecasting Methodology and forecasting system of a kind of electrical fault |
CN107908598A (en) * | 2017-11-24 | 2018-04-13 | 东北电力大学 | OPLC Fault Locating Methods based on DS evidence theories |
CN108444710A (en) * | 2018-05-02 | 2018-08-24 | 中车永济电机有限公司 | A kind of train traction motor bearing life testing machine |
CN108490350A (en) * | 2018-03-14 | 2018-09-04 | 山东科技大学 | A kind of high-power permanent magnet motor fault diagnosis system and diagnostic method |
CN108664010A (en) * | 2018-05-07 | 2018-10-16 | 广东省电信规划设计院有限公司 | Generating set fault data prediction technique, device and computer equipment |
CN108760302A (en) * | 2018-05-08 | 2018-11-06 | 南京风电科技有限公司 | A kind of on-line monitoring and fault diagnosis system of wind power generating set bearing |
CN109040742A (en) * | 2018-07-04 | 2018-12-18 | 广东惠禾科技发展有限公司 | Camera intelligent fault is sought because of method, apparatus and camera |
CN109100608A (en) * | 2018-07-23 | 2018-12-28 | 华北电力大学(保定) | A kind of diagnostic method of the double-fed fan stator shorted-turn fault based on D-S information fusion algorithm |
CN109255333A (en) * | 2018-09-25 | 2019-01-22 | 内蒙古工业大学 | A kind of large-scale wind electricity unit rolling bearing fault Hybrid approaches of diagnosis |
CN109342882A (en) * | 2018-11-06 | 2019-02-15 | 长沙理工大学 | Cable fault on-line monitoring method based on multi-source Eigenvalues analysis |
CN109404226A (en) * | 2017-08-16 | 2019-03-01 | 通用电气公司 | For detecting the system and method for the damage in rotary machine |
CN109916627A (en) * | 2019-03-27 | 2019-06-21 | 西南石油大学 | Bearing fault detection and diagnosis based on Active Learning |
CN109959411A (en) * | 2017-12-26 | 2019-07-02 | 北京金风科创风电设备有限公司 | Method and equipment for identifying abnormity of wind generating set |
CN109995736A (en) * | 2017-12-31 | 2019-07-09 | 中国移动通信集团四川有限公司 | Detection threatens method, apparatus, equipment and the storage medium of attack |
CN110426634A (en) * | 2019-09-10 | 2019-11-08 | 上海大制科技有限公司 | A kind of method and apparatus of the predicting abnormality for drive system |
CN110441060A (en) * | 2019-08-07 | 2019-11-12 | 贵州电网有限责任公司 | One kind being based on mechanical failure diagnostic method combined of multi-sensor information |
CN110531266A (en) * | 2019-10-14 | 2019-12-03 | 国网山东省电力公司电力科学研究院 | A kind of synchronous machinery excitation winding interturn short-circuit fault early warning method |
CN110968069A (en) * | 2018-09-28 | 2020-04-07 | 新疆金风科技股份有限公司 | Fault prediction method of wind generating set, corresponding device and electronic equipment |
CN110988534A (en) * | 2019-12-03 | 2020-04-10 | 北京特种机械研究所 | Performance test method for universal servo system |
CN111079705A (en) * | 2019-12-31 | 2020-04-28 | 北京理工大学 | Vibration signal classification method |
CN111175651A (en) * | 2020-01-10 | 2020-05-19 | 河北大学 | Big data-based wind turbine generator fault early warning and diagnosis system |
CN111175596A (en) * | 2018-11-13 | 2020-05-19 | 发那科株式会社 | State determination device and state determination method |
CN111521937A (en) * | 2020-04-29 | 2020-08-11 | 粤电集团贵州有限公司 | Method and device for diagnosing starting fault of gas generator |
CN111539381A (en) * | 2020-05-18 | 2020-08-14 | 中车永济电机有限公司 | Construction method of wind turbine bearing fault classification diagnosis model |
CN111798650A (en) * | 2020-06-22 | 2020-10-20 | 河北新天科创新能源技术有限公司 | Wind turbine generator fault early warning method based on multi-model combination |
CN112492630A (en) * | 2019-09-11 | 2021-03-12 | 中国电信股份有限公司 | Fault prediction method and device of base station equipment and base station |
CN112668626A (en) * | 2020-12-23 | 2021-04-16 | 中国人民解放军海军航空大学 | Reliability prediction method and system for radar seeker |
CN112850408A (en) * | 2021-02-05 | 2021-05-28 | 浙江新再灵科技股份有限公司 | Elevator emergency stop trapped person fault detection method based on multi-model fusion |
CN113325314A (en) * | 2021-07-20 | 2021-08-31 | 安徽巨森电器有限公司 | Motor fault diagnosis method |
CN113673442A (en) * | 2021-08-24 | 2021-11-19 | 燕山大学 | Variable working condition fault detection method based on semi-supervised single classification network |
CN113887807A (en) * | 2021-10-08 | 2022-01-04 | 天津大学 | Robot game tactical prediction method based on machine learning and evidence theory |
CN114509158A (en) * | 2022-01-04 | 2022-05-17 | 东南大学 | Acoustic-vibration fused blade crack fault detection method and application |
CN115293286A (en) * | 2022-08-20 | 2022-11-04 | 华南理工大学 | Offshore wind turbine monitoring data processing method based on big data feature extraction |
CN117648565A (en) * | 2023-10-27 | 2024-03-05 | 甘肃倚莱克特电力科技有限公司 | Generator fault recognition algorithm |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104833534A (en) * | 2015-04-21 | 2015-08-12 | 广州市地下铁道总公司 | Train running fault diagnosis device based on multi-source information fusion, and method |
-
2017
- 2017-02-24 CN CN201710102612.5A patent/CN107016404A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104833534A (en) * | 2015-04-21 | 2015-08-12 | 广州市地下铁道总公司 | Train running fault diagnosis device based on multi-source information fusion, and method |
Non-Patent Citations (2)
Title |
---|
曹斌: "《风电机组振动监测与故障诊断系统研究》", 《中国优秀硕士学位论文全文数据库 工程科技||辑》 * |
焦露琴等: "《基于SVM与D-S证据理论的异步电动机转子段条故障诊断方法》", 《工矿自动化》 * |
Cited By (47)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109404226A (en) * | 2017-08-16 | 2019-03-01 | 通用电气公司 | For detecting the system and method for the damage in rotary machine |
CN107644231A (en) * | 2017-09-19 | 2018-01-30 | 广东工业大学 | A kind of generator amature method for diagnosing faults and device |
CN107831438A (en) * | 2017-10-25 | 2018-03-23 | 上海交通大学 | The Forecasting Methodology and forecasting system of a kind of electrical fault |
CN107908598A (en) * | 2017-11-24 | 2018-04-13 | 东北电力大学 | OPLC Fault Locating Methods based on DS evidence theories |
CN107908598B (en) * | 2017-11-24 | 2021-04-30 | 东北电力大学 | OPLC fault positioning method based on DS evidence theory |
CN109959411A (en) * | 2017-12-26 | 2019-07-02 | 北京金风科创风电设备有限公司 | Method and equipment for identifying abnormity of wind generating set |
CN109995736A (en) * | 2017-12-31 | 2019-07-09 | 中国移动通信集团四川有限公司 | Detection threatens method, apparatus, equipment and the storage medium of attack |
CN108490350B (en) * | 2018-03-14 | 2020-07-31 | 山东科技大学 | Fault diagnosis system and method for high-power permanent magnet motor |
CN108490350A (en) * | 2018-03-14 | 2018-09-04 | 山东科技大学 | A kind of high-power permanent magnet motor fault diagnosis system and diagnostic method |
CN108444710A (en) * | 2018-05-02 | 2018-08-24 | 中车永济电机有限公司 | A kind of train traction motor bearing life testing machine |
CN108664010A (en) * | 2018-05-07 | 2018-10-16 | 广东省电信规划设计院有限公司 | Generating set fault data prediction technique, device and computer equipment |
CN108760302A (en) * | 2018-05-08 | 2018-11-06 | 南京风电科技有限公司 | A kind of on-line monitoring and fault diagnosis system of wind power generating set bearing |
CN109040742A (en) * | 2018-07-04 | 2018-12-18 | 广东惠禾科技发展有限公司 | Camera intelligent fault is sought because of method, apparatus and camera |
CN109100608A (en) * | 2018-07-23 | 2018-12-28 | 华北电力大学(保定) | A kind of diagnostic method of the double-fed fan stator shorted-turn fault based on D-S information fusion algorithm |
CN109255333A (en) * | 2018-09-25 | 2019-01-22 | 内蒙古工业大学 | A kind of large-scale wind electricity unit rolling bearing fault Hybrid approaches of diagnosis |
CN110968069B (en) * | 2018-09-28 | 2022-01-25 | 新疆金风科技股份有限公司 | Fault prediction method of wind generating set, corresponding device and electronic equipment |
CN110968069A (en) * | 2018-09-28 | 2020-04-07 | 新疆金风科技股份有限公司 | Fault prediction method of wind generating set, corresponding device and electronic equipment |
CN109342882A (en) * | 2018-11-06 | 2019-02-15 | 长沙理工大学 | Cable fault on-line monitoring method based on multi-source Eigenvalues analysis |
CN111175596A (en) * | 2018-11-13 | 2020-05-19 | 发那科株式会社 | State determination device and state determination method |
CN111175596B (en) * | 2018-11-13 | 2024-05-31 | 发那科株式会社 | State determination device and state determination method |
CN109916627A (en) * | 2019-03-27 | 2019-06-21 | 西南石油大学 | Bearing fault detection and diagnosis based on Active Learning |
CN110441060A (en) * | 2019-08-07 | 2019-11-12 | 贵州电网有限责任公司 | One kind being based on mechanical failure diagnostic method combined of multi-sensor information |
CN110426634B (en) * | 2019-09-10 | 2020-08-18 | 大制(苏州)科技有限公司 | Method and equipment for predicting abnormity of driving system |
CN110426634A (en) * | 2019-09-10 | 2019-11-08 | 上海大制科技有限公司 | A kind of method and apparatus of the predicting abnormality for drive system |
CN112492630A (en) * | 2019-09-11 | 2021-03-12 | 中国电信股份有限公司 | Fault prediction method and device of base station equipment and base station |
CN112492630B (en) * | 2019-09-11 | 2023-02-03 | 中国电信股份有限公司 | Fault prediction method and device of base station equipment and base station |
CN110531266A (en) * | 2019-10-14 | 2019-12-03 | 国网山东省电力公司电力科学研究院 | A kind of synchronous machinery excitation winding interturn short-circuit fault early warning method |
CN110531266B (en) * | 2019-10-14 | 2021-08-24 | 国网山东省电力公司电力科学研究院 | Synchronous motor excitation winding turn-to-turn short circuit fault early warning method |
CN110988534B (en) * | 2019-12-03 | 2022-09-16 | 北京特种机械研究所 | Performance test method for universal servo system |
CN110988534A (en) * | 2019-12-03 | 2020-04-10 | 北京特种机械研究所 | Performance test method for universal servo system |
CN111079705B (en) * | 2019-12-31 | 2023-07-25 | 北京理工大学 | Vibration signal classification method |
CN111079705A (en) * | 2019-12-31 | 2020-04-28 | 北京理工大学 | Vibration signal classification method |
CN111175651A (en) * | 2020-01-10 | 2020-05-19 | 河北大学 | Big data-based wind turbine generator fault early warning and diagnosis system |
CN111521937A (en) * | 2020-04-29 | 2020-08-11 | 粤电集团贵州有限公司 | Method and device for diagnosing starting fault of gas generator |
CN111539381B (en) * | 2020-05-18 | 2022-04-05 | 中车永济电机有限公司 | Construction method of wind turbine bearing fault classification diagnosis model |
CN111539381A (en) * | 2020-05-18 | 2020-08-14 | 中车永济电机有限公司 | Construction method of wind turbine bearing fault classification diagnosis model |
CN111798650A (en) * | 2020-06-22 | 2020-10-20 | 河北新天科创新能源技术有限公司 | Wind turbine generator fault early warning method based on multi-model combination |
CN112668626A (en) * | 2020-12-23 | 2021-04-16 | 中国人民解放军海军航空大学 | Reliability prediction method and system for radar seeker |
CN112850408A (en) * | 2021-02-05 | 2021-05-28 | 浙江新再灵科技股份有限公司 | Elevator emergency stop trapped person fault detection method based on multi-model fusion |
CN113325314A (en) * | 2021-07-20 | 2021-08-31 | 安徽巨森电器有限公司 | Motor fault diagnosis method |
CN113673442B (en) * | 2021-08-24 | 2024-05-24 | 燕山大学 | Variable working condition fault detection method based on semi-supervised single classification network |
CN113673442A (en) * | 2021-08-24 | 2021-11-19 | 燕山大学 | Variable working condition fault detection method based on semi-supervised single classification network |
CN113887807A (en) * | 2021-10-08 | 2022-01-04 | 天津大学 | Robot game tactical prediction method based on machine learning and evidence theory |
CN113887807B (en) * | 2021-10-08 | 2024-04-30 | 天津大学 | Robot game tactics prediction method based on machine learning and evidence theory |
CN114509158A (en) * | 2022-01-04 | 2022-05-17 | 东南大学 | Acoustic-vibration fused blade crack fault detection method and application |
CN115293286A (en) * | 2022-08-20 | 2022-11-04 | 华南理工大学 | Offshore wind turbine monitoring data processing method based on big data feature extraction |
CN117648565A (en) * | 2023-10-27 | 2024-03-05 | 甘肃倚莱克特电力科技有限公司 | Generator fault recognition algorithm |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107016404A (en) | Wind power generating set failure prediction method based on D S evidence fusions | |
Zhang et al. | The multiclass fault diagnosis of wind turbine bearing based on multisource signal fusion and deep learning generative model | |
Lahouar et al. | Hour-ahead wind power forecast based on random forests | |
Tan et al. | A deep imbalanced learning framework for transient stability assessment of power system | |
Liu et al. | Data-driven event identification in the US power systems based on 2D-OLPP and RUSBoosted trees | |
Kamwa et al. | Development of rule-based classifiers for rapid stability assessment of wide-area post-disturbance records | |
Li et al. | Enhanced generative adversarial networks for fault diagnosis of rotating machinery with imbalanced data | |
Li et al. | Transient stability assessment of power system based on XGBoost and factorization machine | |
CN109325553B (en) | Wind power gear box fault detection method, system, equipment and medium | |
CN111553495A (en) | Small circuit breaker fault analysis method based on probabilistic neural network | |
CN106897945A (en) | The clustering method and equipment of wind power generating set | |
Koochi et al. | A synchrophasor-based decision tree approach for identification of most coherent generating units | |
CN112200694A (en) | Dominant instability mode identification model construction and application method based on graph neural network | |
Razavi-Far et al. | Incremental design of a decision system for residual evaluation: A wind turbine application | |
CN109141884A (en) | Method for Bearing Fault Diagnosis based on EEMD-AR model and DBN | |
Ren et al. | Generative adversarial networks driven by multi-domain information for improving the quality of generated samples in fault diagnosis | |
CN104182803A (en) | Wind power data preprocessing method, wind power forecast method and system | |
Vitor et al. | Induction motor short circuit diagnosis and interpretation under voltage unbalance and load variation conditions | |
CN117419010A (en) | Deep learning-based fan blade health monitoring method and fault detection method thereof | |
Tun et al. | Power quality event classification in distribution grids using machine learning | |
Chen et al. | A dual-view network for fault diagnosis in rotating machinery using unbalanced data | |
Hassani et al. | Design of a cost-effective deep convolutional neural network–based scheme for diagnosing faults in smart grids | |
CN108536877A (en) | The appraisal procedure and system of vibration of wind generating set degree | |
Ma et al. | Data generation for rare transient events: A generative adversarial network approach | |
CN115853722A (en) | Wind turbine generator fault diagnosis method based on feature selection and stacking model fusion |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170804 |