CN107701378B - A kind of wind-driven generator fault early warning method - Google Patents
A kind of wind-driven generator fault early warning method Download PDFInfo
- Publication number
- CN107701378B CN107701378B CN201710904941.1A CN201710904941A CN107701378B CN 107701378 B CN107701378 B CN 107701378B CN 201710904941 A CN201710904941 A CN 201710904941A CN 107701378 B CN107701378 B CN 107701378B
- Authority
- CN
- China
- Prior art keywords
- observed parameter
- wind
- class
- new class
- observed
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F03—MACHINES OR ENGINES FOR LIQUIDS; WIND, SPRING, OR WEIGHT MOTORS; PRODUCING MECHANICAL POWER OR A REACTIVE PROPULSIVE THRUST, NOT OTHERWISE PROVIDED FOR
- F03D—WIND MOTORS
- F03D17/00—Monitoring or testing of wind motors, e.g. diagnostics
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
- F05B—INDEXING SCHEME RELATING TO WIND, SPRING, WEIGHT, INERTIA OR LIKE MOTORS, TO MACHINES OR ENGINES FOR LIQUIDS COVERED BY SUBCLASSES F03B, F03D AND F03G
- F05B2260/00—Function
- F05B2260/80—Diagnostics
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
- F05B—INDEXING SCHEME RELATING TO WIND, SPRING, WEIGHT, INERTIA OR LIKE MOTORS, TO MACHINES OR ENGINES FOR LIQUIDS COVERED BY SUBCLASSES F03B, F03D AND F03G
- F05B2260/00—Function
- F05B2260/82—Forecasts
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F05—INDEXING SCHEMES RELATING TO ENGINES OR PUMPS IN VARIOUS SUBCLASSES OF CLASSES F01-F04
- F05B—INDEXING SCHEME RELATING TO WIND, SPRING, WEIGHT, INERTIA OR LIKE MOTORS, TO MACHINES OR ENGINES FOR LIQUIDS COVERED BY SUBCLASSES F03B, F03D AND F03G
- F05B2260/00—Function
- F05B2260/84—Modelling or simulation
Landscapes
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Sustainable Development (AREA)
- Sustainable Energy (AREA)
- Chemical & Material Sciences (AREA)
- Combustion & Propulsion (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Control Of Eletrric Generators (AREA)
- Wind Motors (AREA)
Abstract
The invention discloses a kind of wind-driven generator fault early warning methods, including extracting observed parameter, observed parameter clustering, classified using Ward system cluster analysis to the observed parameter, " mass center " extracts structural regime matrix, " mass center " of each classification is calculated according to shortest distance principle, the state matrix that characterization wind-driven generator operates normally situation is combined by " mass center " collection, state matrix described in similitude Modeling Calculation generates the estimated value of state, and residual analysis judges whether to trigger early warning.Wind-driven generator fault early warning method provided in an embodiment of the present invention can quickly and efficiently predict wind-driven generator failure.
Description
Technical field
The present embodiments relate to wind power generation field more particularly to a kind of wind-driven generator fault early warning methods.
Background technique
Wind power system running environment is severe, and the wind-powered electricity generation tertiary industry of domestic development sluggishness, causes Wind turbines equipment in addition
Failure high frequency occurs.The core equipment that doubly-fed wind turbine is realized as variable speed constant frequency wind power system, while being in wind field hair
The boundary of electric side and grid side, importance is self-evident, so no matter considering from economy or safety, to wind-power electricity generation
Machine carry out fault pre-alarming research it is necessory to.
The generation of wind-driven generator component failure is not accomplished in one move, and generates and development generally need to be via abnormal, scarce
Fall into, several processes such as failure and accident, such as vibrate the excessively high some failures of excessive, temperature be can be by status monitoring and pre-
Alert technology nips in the bud failure.When failure occurs, the variation of unit equipment observed parameter need to be via unobvious to aobvious
The progressive formation of work, if can just germinate in Wind turbines failure, the degree even slight stage identifies exception, compared to
Causing the subsequent maintenance of serious consequence has more great meaning.
However, traditional fault early warning method based on expert system, for such as DFIG machine-electric-thermal close coupling complexity
System, Knowledge Source are not enough to express and reflect that the feature of things, accuracy rate be not high；Tradition is based on Artificial Neural Network Modeling
Fault early warning method, modeling need to take a long time, and the selection of learning sample also lacks foundation, and model maintenance is difficult.
Summary of the invention
The embodiment of the present invention provides a kind of wind-driven generator fault early warning method, quickly and efficiently to predict wind-power electricity generation
Machine failure.
The embodiment of the invention provides a kind of wind-driven generator fault early warning methods, comprising:
Extract observed parameter；
Observed parameter clustering classifies to the observed parameter using Ward system cluster analysis；
" mass center " extracts structural regime matrix, " mass center " of each classification is calculated according to shortest distance principle, leads to
It crosses " mass center " collection and is combined into the state matrix that characterization wind-driven generator operates normally situation；
State matrix described in similitude Modeling Calculation generates the estimated value of state, and residual analysis judges whether to trigger early warning.
Further, after the extraction observed parameter, further includes:
Rough set attribute reduction is arranged by Attribute Significance of the Algorithm for Attribute Reduction to the observed parameter
Sequence reduces the observed parameter scale to reject redundant attributes item.
Further, after the extraction observed parameter, further includes:
Observed parameter pretreatment, rejects noise, non-operational data, invalid data and the unrelated number in the observed parameter
According to.
Further, whether the method for rejecting non-operational data is: being more than to cut wind speed and cut out with Wind observation parameter
On the basis of wind speed, reject lower than incision wind speed and all observed parameters obtained higher than the moment where cut-out wind speed.
It is further, described to be classified using Ward system cluster analysis to observed parameter, comprising:
Step 1, using each observed parameter column sample as a preliminary classification；
Step 2 will merge two-by-two apart from the smallest two preliminary classifications, and form new class；Between two preliminary classifications away from
It is calculated from using Euclidean distance formula；
All observed parameter samples are divided into C after above-mentioned merging two-by-two by step 31, C2... ..., Ck, k new classes, calculating
The class mean value of each new class, the class mean value are the mean vector of all observed parameters in corresponding new class, are calculated in each new class
Observed parameter to the square distance of its corresponding class mean value, obtain the sum of squares of deviations of each new class:
Wherein, CtFor t-th of new class；XitFor new class CtIn i-th of observed parameter；ntFor new class CtObserved parameter sum；For new class CtClass mean value；StFor new class CtSum of squares of deviations；
Merge sum of squares of deviations in class and increase the smallest two classification, i.e., forms new classification again；
Classify CpWith classification CqIt is merged into new class Cr, then the corresponding sum of squares of deviations S of threep、Sq、SrSatisfaction makes to increase
AmountMinimum, wherein new class CrSum of squares of deviations SrThe same S of calculation methodpAnd Sq；
Step 4, based on the new class after merging, it then follows two classes merged every time always increase sum of squares of deviations in class
Add minimum, circulation executes step 3, until meeting the Cluster Classification quantitative requirement of setting.
Further, the method up to meeting the Cluster Classification quantitative requirement of setting is: class after final merge
When number is 4~6 times of observation measuring point quantity, clustering terminates.
Further, described " mass center " extracts structural regime matrix, and the state matrix is as shown in following formula:
The state matrix XtIt is the matrix of (n+1) * m, state matrix XtThe first row X1~XmFor different observation
Parameter column serial number, state matrix XtRemaining every " mass center " to extract, i.e., specific observed parameter；
The state matrix XtRemaining every column vector represents the observed parameter of a certain moment difference observation measuring point, described
State matrix XtRemaining every row vector represents observed parameter of a certain observation measuring point within a period of time.
Further, the method that state matrix described in the similitude Modeling Calculation generates the estimated value of state is: with institute
Based on stating state matrix, predictive estimation is carried out to running status of wind generator according to similitude State Estimation Theory, is obtained
Estimated value.
Further, the residual analysis judges whether that the method for triggering early warning is: according to the estimated value and actual value
Difference obtain residual values, the residual values are compared with pre-set overgauge threshold value and minus deviation threshold value, work as institute
Residual values are stated greater than the overgauge threshold value or when being less than the minus deviation threshold value, trigger early warning；Conversely, not triggering alarm.
Further, the extraction observed parameter, the attribute of the observed parameter include wind speed, frequency, environment temperature, machine
Cabin temperature, voltage, electric current, generator speed, generator active power, generator cooling air temperature, drive end bearing temperature
With non-driven bearing temperature.
In the embodiment of the present invention, by observed parameter clustering, using Ward system cluster analysis to observed parameter
Classify, extract " mass center " structural regime matrix, similitude Modeling Calculation state matrix generates the estimated value of state, residual error point
Analysis judges whether to trigger early warning, solves the problems, such as wind-driven generator fault pre-alarming, realize that prediction stability is high, adaptable and fault-tolerant
The high effect of rate.
Detailed description of the invention
Fig. 1 is the wind-driven generator fault early warning method flow chart that the embodiment of the present invention one provides；
Fig. 2 is wind-driven generator fault early warning method flow chart provided by Embodiment 2 of the present invention；
Fig. 3 is rejecting non-operational data display figure provided by Embodiment 2 of the present invention；
Fig. 4 is Ward system cluster analysis flow chart provided by Embodiment 2 of the present invention；
Fig. 5 is that similar state provided by Embodiment 2 of the present invention estimates output result figure.
Specific embodiment
To keep the technical problems solved, the adopted technical scheme and the technical effect achieved by the invention clearer, below
It will the technical scheme of the embodiment of the invention will be described in further detail in conjunction with attached drawing, it is clear that described embodiment is only
It is a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those skilled in the art exist
Every other embodiment obtained under the premise of creative work is not made, shall fall within the protection scope of the present invention.
Embodiment one
Fig. 1 is the wind-driven generator fault early warning method flow chart that the embodiment of the present invention one provides, and specifically includes following step
It is rapid:
Step S101: observed parameter is extracted；
Observed parameter is multiple observation measuring points with observation data in different time periods, extracts building shape by SCADA system
At initial observation parameter matrix, each column vector (referred to as column sample) in the initial observation parameter matrix is to adopt at some moment
The value of the wind-driven generator difference observation measuring point parameter collected；Each row vector in the initial observation parameter matrix (is referred to as gone
Sample) it is the value for observing measuring point different moments collected parameter same to wind-driven generator.
Step S102: the construction of observed parameter clustering, state matrix can be attributed to gather from the tradeoff of data mining angle
Class problem classifies to observed parameter with Ward system cluster analysis；
Ward system cluster analysis is also known as sum of squares of deviations method, is to integrate object according to the feature of many aspects
A kind of System Cluster Analysis of classification.Certainly, the embodiment of the present invention can also commonly use other systems clustering method, example
Such as single connection method is fully connected method, average connection method and organizes average connection method or common K mean cluster method etc..
Step S103: " mass center " extracts structural regime matrix, and " the matter of each classification is calculated according to shortest distance principle
The heart ", in each classification according to can summarize as far as possible this category feature to seek out each classification " mass center ", by
" mass center " is gathered to form the state matrix that can characterize double-fed generator accidental conditions.For example following formula of state matrix
It is shown:
Wherein, state matrix XtIt is one (n+1)*The matrix of m, the first row X of matrix1~XmFor different observed parameters
Column serial number, the number that observed parameter column serial number manually carries out do not have practical significance, state matrix XtIt is remaining every for extraction
" mass center ", i.e., specific observed parameter.State matrix XtIn remaining items, column vector is that a certain moment difference observes measuring point wind
Operational parameter value under power generator accidental conditions；Row vector is monitoring of a certain observation measuring point within a period of time
Value, for example, Xn(2) in observed parameter column serial number X2Ginseng is run under lower n-th of observation measuring point wind-driven generator accidental conditions
Numerical value, illustratively, m/n=4~6.
Step S104: similitude Modeling Calculation state matrix XtThe estimated value of generation state, residual analysis judge whether to touch
Send out early warning.
The technical solution of the embodiment of the present invention, by observed parameter clustering, using Ward system cluster analysis pair
Observed parameter is classified, and " mass center " structural regime matrix is extracted, and similitude Modeling Calculation state matrix generates the estimation of state
Value, residual analysis judge whether to trigger early warning, solve the problems, such as wind-driven generator fault pre-alarming, realize that detection stability is high, adapt to
Property strong and high serious forgiveness effect.
Embodiment two
Fig. 2 is wind-driven generator fault early warning method flow chart provided by Embodiment 2 of the present invention.The embodiment of the present invention two
On the basis of embodiment one, the data processing after extraction observed parameter is illustrated.
Further, as shown in Fig. 2, it is coarse step S202 can also to be carried out after step S201 extracts observed parameter
Set attribute reduction and the pretreated operation of step S203 observed parameter.Step S202 rough set attribute reduction and step S203 observation
Parameter, which pre-processes on practical significance, to be arranged to history data, is reduced the scale of data processing, can also be mentioned
The accuracy of high fault pre-alarming.After handling history data, step S204 observed parameter cluster is then carried out
Structural regime matrix is extracted in analysis, step S205 " mass center " and step S206 similitude modeling state estimation judges whether that triggering is pre-
It is alert.
Wherein, step S202: rough set attribute reduction, by Algorithm for Attribute Reduction to the attribute weight of observed parameter
It spends and is ranked up, reject redundant attributes item, reduce the observed parameter scale, reduce state matrix row.
Table 1 is an an Attribute Significance provided in an embodiment of the present invention measuring and calculating, to acquisition from wind power plant SCADA system
Property parameters relevant to wind-driven generator carry out importance sorting analysis after, choose generator active power, bearing temperature
And the every 30 minutes observation data of ten measuring points such as revolving speed carry out modeling analysis to bearing of wind power generator.
The measuring and calculating of 1 Attribute Significance of table
| Serial number | Measuring point title | Different degree | Serial number | Measuring point title | Different degree |
| 1 | Generator speed 1 (tachometer disk) | 0.68 | 6 | Environment temperature | 0.24 |
| 2 | Generator speed 2 (PLC) | 0.67 | 7 | Drive end bearing temperature 1 | 0.98 |
| 3 | Wind speed | 0.71 | 8 | Drive end bearing temperature 2 | 0.98 |
| 4 | Generator active power | 0.41 | 9 | Non-driven-end bearing temperature | 0.92 |
| 5 | Cabin temperature | 0.37 | 10 | Generator cooling air temperature | 0.55 |
Step S203: noise, non-operational data, the invalid data in the observed parameter are rejected in observed parameter pretreatment
And extraneous data.
Specifically, reject invalid data, the presence of measuring device systematic error, such as sensor in transmission log data,
Or server is when handling data, in fact it could happen that the invalid situation of measurement data.For example, being acquired from wind power plant SCADA system
To data in, in the 11:00:00 on the 7th of August in 2013 the 475th column " generator drive end bearing temperature A1 " measuring point there is number
Value is 68742, this does not obviously meet convention, and 30 minute datas are 40 DEG C of magnitudes before and after the measuring point, can be concluded that and belong to
Invalid data, it is therefore desirable to reject.
Differential analysis removes extraneous data and is provided with " 1+1 " two sensings mainly in some important observation measuring point parts
It when device (1 is standby with 1) observes measuring point, is compared and analyzed using the collected data of the two sensors, removes extraneous data.
Wind-driven generator operates normally entire dynamic process: unit starting and stop phase, unit even running and
Acute variation.Since wind generator system design does not generate electricity when lower than incision wind speed and opens from grid disruption, blower is in certainly
By locking state, just start grid-connected work until wind speed is higher than incision wind speed and continues some short period length, when wind speed height
When cut-out wind speed, wind generator system can then cut out power grid.Therefore, the monitoring number of " wind speed " this observation measuring point can be chosen
Whether value is more than to reject on the basis of cutting wind speed and cut-out wind speed to non-operational data, rejects lower than incision wind speed and is higher than
All observed parameters that moment where cut-out wind speed obtains.Fig. 3 is that rejecting non-operational data provided by Embodiment 2 of the present invention is aobvious
Diagram carries out rejecting non-operational data processing.
Step S204: observed parameter clustering.
Classified using Ward system cluster analysis to observed parameter.The minimum of Ward hierarchial-cluster analysis processing is single
Member is observed parameter column, each value for being classified as some moment collected wind-driven generator difference observation measuring point parameter.Wind-force
When generator is run, in certain different moments, wind-power electricity generation chance is under identical or very similar operating condition, Ward system
The purpose of system clustering flocks together at the time of being these under identical or very close operating condition and is classified as one
Class, this kind in can containing x different moments namely x column.
Fig. 4 is Ward system cluster analysis flow chart provided by Embodiment 2 of the present invention, as shown in Figure 4, comprising:
Step S401, using each observed parameter column sample as a preliminary classification；
Step S402 will merge two-by-two apart from the smallest two preliminary classifications, and form new class；Between two preliminary classifications
Distance is calculated using Euclidean distance formula；
All observed parameter samples are divided into C after above-mentioned merging two-by-two by step S4031, C2... ..., Ck, k new classes, meter
The class mean value of each new class is calculated, the class mean value is the mean vector of all observed parameters in corresponding new class, calculates each new class
In observed parameter to the square distance of its corresponding class mean value, obtain the sum of squares of deviations of each new class:
Wherein, CtFor t-th of new class；XitFor new class CtIn i-th of observed parameter；ntFor new class CtObserved parameter sum；For new class CtClass mean value；StFor new class CtSum of squares of deviations；
Merge sum of squares of deviations in class and increase the smallest two classification, i.e., forms new classification again；
Classify CpWith classification CqIt is merged into new class Cr, then the corresponding sum of squares of deviations S of threep、Sq、SrSatisfaction makes to increase
AmountMinimum, wherein new class CrSum of squares of deviations SrThe same S of calculation methodpAnd Sq；
Step S404, based on the new class after merging, it then follows two classes merged every time always make sum of squares of deviations in class
Increase minimum, circulation execution step S403, until meeting the Cluster Classification quantitative requirement of setting.
Step S205: " mass center " extracts structural regime matrix.
Specifically, it by running Ward hierarchial-cluster analysis Matlab simulated program, can obtain to wind-driven generator operating condition
What is extracted on the basis of data classification can most express " mass center " set for summarizing every category feature.Portion of program code is as follows:
Following state matrix XtThe state matrix as calculated:
Further, the class number after Cluster Classification finally merges is cluster point when observing 4~6 times of test point quantity
Analysis terminates.
Further, " mass center " extracts in structural regime matrix, and the column vector of state matrix represents a certain moment difference and surveys
The observed parameter of point, the row vector of state matrix represent observed parameter of a certain measuring point within a period of time.Such as state square
Battle array XtShown, row vector is 11, removes the sequence number of the acquisition of observed parameter representated by the first row, remaining 10 row vector generations
Observed parameter of a certain observation measuring point of table within a period of time, each column vector represent a certain moment difference observation measuring point
Observed parameter.Wherein the quantity of the present embodiment test point is total up to 10, and column vector is 48, and the class number after final merging is to see
4.8 times for surveying measuring point quantity, meeting the class number after Cluster Classification finally merges is 4~6 times for observing measuring point quantity.
Modeled using similitude and carry out the keys of fault pre-alarming success or failure and be the construction of state matrix, similitude modeling come into
Row fault pre-alarming is divided into multiple steps, but wherein the foundation of state matrix is a most key step.The section that state matrix is established
Degree, order of accuarcy all decide final early warning effect or early warning system success or not.
Step S206: similitude Modeling Calculation state matrix generates the estimated value of state, and residual analysis judges whether to trigger
Early warning.
Further, the method that similitude Modeling Calculation state matrix generates the estimated value of state is: being with state matrix
Basis carries out predictive estimation to running status of wind generator according to similitude State Estimation Theory, obtains estimated value.Work as state
After matrix determines, the model of characterization wind-driven generator accidental conditions is just determined, it can is estimated according to similitude state
Meter is theoretical to carry out predictive estimation to running status of wind generator.
Further, it carries out residual analysis to judge whether to trigger early warning, according to the difference of the estimated value of prediction and actual value
Residual values are obtained, residual values are compared with pre-set overgauge threshold value and minus deviation threshold value, when residual values are greater than just
Deviation threshold or be less than minus deviation threshold value when, trigger early warning；Conversely, when residual values be located at overgauge threshold value and minus deviation threshold value it
Between (including two endpoints of overgauge threshold value and minus deviation threshold value) when, alarm is not triggered.Here overgauge threshold value and minus deviation
Threshold value can as the case may be or requirement is flexibly set and adjustment, wherein minus deviation threshold value is less than overgauge threshold value.
Specifically, Fig. 5 is that similar state provided by Embodiment 2 of the present invention estimates output result figure.With drive end bearing temperature
Failure example is spent, the embodiment of Fig. 5 sets 15, when negative bias difference is set as 6 for positively biased difference, system state estimation output.
Its middle line 2 is drive end bearing temperature actual value, and line 3 is state of temperature discreet value, line 1 × when indicating that deviation is more than threshold value
Warning note.Since No. 21 21:30:00 of September, early warning system monitors that drive end bearing temperature shows fluctuating and increases state
Gesture, No. 22 temperature values of September reach 70 DEG C of magnitudes, and wind field service personnel generator drive end temperature is reminded in the triggering of temperature rise early warning at this time
Rising excessively high failure, there are possibility occurrences.By consulting running log, 2 unit of wind power plant, which truly has, within this period quotes wind
Power generator exception record, further demonstrate proposition of the embodiment of the present invention diagnoses wind-force based on similitude state estimation early warning
The feasibility that generator failure method is implemented.
Further, observed parameter is extracted, as shown in table 1, the attribute of observed parameter may include wind speed, frequency, environment
Temperature, cabin temperature, voltage, electric current, generator speed, generator active power, generator cooling air temperature, drive end axle
Hold temperature and non-driven bearing temperature etc..Actual conditions are not limited to above-mentioned listed attribute, can according to equipment supplier or
Owner requires to make corresponding change.
Note that the above is only a better embodiment of the present invention and the applied technical principle.It will be appreciated by those skilled in the art that
The invention is not limited to the specific embodiments described herein, be able to carry out for a person skilled in the art it is various it is apparent variation,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above embodiments to the present invention
It is described in further detail, but the present invention is not limited to the above embodiments only, without departing from the inventive concept, also
It may include more other equivalent embodiments, and the scope of the invention is determined by the scope of the appended claims.
Claims (10)
1. a kind of wind-driven generator fault early warning method characterized by comprising
Extract observed parameter；
Observed parameter clustering classifies to the observed parameter using Ward system cluster analysis；
" mass center " extracts structural regime matrix, " mass center " of each classification is calculated according to shortest distance principle, passes through institute
It states " mass center " collection and is combined into the state matrix that characterization wind-driven generator operates normally situation；
State matrix described in similitude Modeling Calculation generates the estimated value of state, and residual analysis judges whether to trigger early warning；
Wherein, described " mass center " is specific observed parameter.
2. the method according to claim 1, wherein after the extraction observed parameter, further includes:
Rough set attribute reduction is ranked up by Attribute Significance of the Algorithm for Attribute Reduction to the observed parameter,
To reject redundant attributes item, reduce the observed parameter scale.
3. method according to claim 1 or 2, which is characterized in that after the extraction observed parameter, further includes:
Observed parameter pretreatment, rejects noise, non-operational data, invalid data and the extraneous data in the observed parameter.
4. according to the method described in claim 3, it is characterized in that, the method for rejecting non-operational data is: being joined with Wind observation
Whether number is more than rejecting moment acquisition lower than incision wind speed and where being higher than cut-out wind speed on the basis of incision wind speed and cut-out wind speed
All observed parameters.
5. the method according to claim 1, wherein described use Ward system cluster analysis to observed parameter
Classify, comprising:
Step 1, using each observed parameter column sample as a preliminary classification；
Step 2 will merge two-by-two apart from the smallest two preliminary classifications, and form new class；The distance between two preliminary classifications are adopted
It is calculated with Euclidean distance formula；
All observed parameter samples are divided into C after above-mentioned merging two-by-two by step 31, C2... ..., Ck, k new classes, calculating is each
The class mean value of new class, the class mean value are the mean vector of all observed parameters in corresponding new class, calculate the sight in each new class
Parameter is surveyed to the square distance of its corresponding class mean value, obtains the sum of squares of deviations of each new class:
Wherein, CtFor t-th of new class；XitFor new class CtIn i-th of observed parameter；ntFor new class CtObserved parameter sum；For
New class CtClass mean value；StFor new class CtSum of squares of deviations；
Merge sum of squares of deviations in class and increase the smallest two classification, i.e., forms new classification again:
Classify CpWith classification CqIt is merged into new class Cr, then the corresponding sum of squares of deviations S of threep、Sq、SrSatisfaction makes incrementssMinimum, wherein new class CrSum of squares of deviations SrThe same S of calculation methodpAnd Sq；
Step 4, based on the new class after merging, it then follows two classes merged every time always make in class sum of squares of deviations increase most
Small, circulation executes step 3, until meeting the Cluster Classification quantitative requirement of setting.
6. according to the method described in claim 5, it is characterized in that, described until meeting the Cluster Classification quantitative requirement of setting
Method is: the class number after final merge is when observing 4~6 times of measuring point quantity, and clustering terminates.
7. the method according to claim 1, wherein " mass center " extracts structural regime matrix, the state
Matrix is as shown in following formula:
The state matrix XtIt is the matrix of (n+1) * m, state matrix XtThe first row X1~XmFor different observed parameters
Column serial number, state matrix XtRemaining every " mass center " to extract, i.e., specific observed parameter；
The state matrix XtRemaining every column vector represents the observed parameter of a certain moment difference observation measuring point, the state
Matrix XtRemaining every row vector represents observed parameter of a certain observation measuring point within a period of time.
8. method according to claim 1 or claim 7, which is characterized in that state matrix described in the similitude Modeling Calculation produces
The method of the estimated value of raw state is: based on the state matrix, according to similitude State Estimation Theory to wind-power electricity generation
Machine operating status carries out predictive estimation, obtains estimated value.
9. according to the method described in claim 8, it is characterized in that, the residual analysis judges whether the method for triggering early warning
It is: residual values is obtained according to the difference of the estimated value and actual value, by the residual values and pre-set overgauge threshold value
It is compared with minus deviation threshold value, when the residual values are greater than the overgauge threshold value or are less than the minus deviation threshold value, touching
Send out early warning；Conversely, not triggering alarm.
10. the method according to claim 1, wherein the extraction observed parameter, the attribute of the observed parameter
It is cold including wind speed, frequency, environment temperature, cabin temperature, voltage, electric current, generator speed, generator active power, generator
But air themperature, drive end bearing temperature and non-driven bearing temperature.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710904941.1A CN107701378B (en) | 2017-09-29 | 2017-09-29 | A kind of wind-driven generator fault early warning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710904941.1A CN107701378B (en) | 2017-09-29 | 2017-09-29 | A kind of wind-driven generator fault early warning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107701378A CN107701378A (en) | 2018-02-16 |
| CN107701378B true CN107701378B (en) | 2019-09-27 |
Family
ID=61174648
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710904941.1A Active CN107701378B (en) | 2017-09-29 | 2017-09-29 | A kind of wind-driven generator fault early warning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107701378B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108460207A (en) * | 2018-02-28 | 2018-08-28 | 上海华电电力发展有限公司 | A kind of fault early warning method of the generating set based on operation data model |
| CN108506171B (en) * | 2018-03-20 | 2019-08-16 | 明阳智慧能源集团股份公司 | A kind of large-scale half direct-drive unit cooling system for gear box fault early warning method |
| CN108872852A (en) * | 2018-05-04 | 2018-11-23 | 上海交通大学 | A kind of wind-driven generator fault diagnosis system and method |
| WO2019223350A1 (en) * | 2018-05-25 | 2019-11-28 | 北京金风科创风电设备有限公司 | Fault handling method and apparatus for wind power generator set, and computer readable storage medium |
| CN110805534B (en) * | 2019-11-18 | 2021-02-12 | 长沙理工大学 | Fault detection method, device and equipment of wind driven generator |
| CN112834224B (en) * | 2021-01-05 | 2023-05-23 | 广东核电合营有限公司 | Nuclear power steam turbine generator health state assessment method and system |
| CN112906473B (en) * | 2021-01-19 | 2023-06-20 | 杭州安脉盛智能技术有限公司 | Fault diagnosis method for rotary equipment |
| CN114088389A (en) * | 2021-12-10 | 2022-02-25 | 华润电力技术研究院有限公司 | Data processing method and related device for gearbox |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102009007938A1 (en) * | 2009-02-06 | 2010-08-19 | Baumer Innotec Ag | Measuring device for measuring deformations of elastically deformable objects |
| US8924162B2 (en) * | 2010-05-13 | 2014-12-30 | University Of Cincinnati | Turbine-to-turbine prognostics technique for wind farms |
| CN105464912B (en) * | 2016-01-27 | 2019-02-19 | 国电联合动力技术有限公司 | A kind of method and apparatus of wind generator set blade icing detection |
| CN106779096B (en) * | 2016-11-10 | 2018-07-10 | 国网山东省电力公司济南供电公司 | Power distribution network reports situation active forewarning system for repairment |
-
2017
- 2017-09-29 CN CN201710904941.1A patent/CN107701378B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN107701378A (en) | 2018-02-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107701378B (en) | A kind of wind-driven generator fault early warning method | |
| CN111539553B (en) | Wind turbine generator fault early warning method based on SVR algorithm and off-peak degree | |
| Liu et al. | A hybrid forecasting method for wind power ramp based on orthogonal test and support vector machine (OT-SVM) | |
| CN105403811B (en) | Wind power plant power grid fault diagnosis method and device | |
| CN111412579B (en) | Air conditioning unit fault type diagnosis method and system based on big data | |
| CN108376298A (en) | A kind of Wind turbines generator-temperature detection fault pre-alarming diagnostic method | |
| CN102588210B (en) | Filtering method for preprocessing fitting data of power curve | |
| CN109543210A (en) | A kind of Wind turbines failure prediction system based on machine learning algorithm platform | |
| CN106054104A (en) | Intelligent ammeter fault real time prediction method based on decision-making tree | |
| CN108072524B (en) | Wind turbine generator gearbox bearing fault early warning method | |
| CN103150473A (en) | Method and device for monitoring and diagnosing generating efficiency of wind turbine generator in real-time manner | |
| CN107728059B (en) | Pitch system state evaluation method | |
| Verstraeten et al. | Fleetwide data-enabled reliability improvement of wind turbines | |
| Butler et al. | A feasibility study into prognostics for the main bearing of a wind turbine | |
| CN111878320B (en) | Monitoring method and system of wind generating set and computer readable storage medium | |
| US20170370368A1 (en) | Predicting a Surge Event in a Compressor of a Turbomachine | |
| CN109492790A (en) | Wind turbines health control method based on neural network and data mining | |
| Hu et al. | Hierarchical fault diagnosis for power systems based on equivalent-input-disturbance approach | |
| CN109595130A (en) | A kind of fan blade icing failure prediction method and system | |
| Zhou et al. | Structural health monitoring of offshore wind power structures based on genetic algorithm optimization and uncertain analytic hierarchy process | |
| CN112001511A (en) | Equipment reliability and dynamic risk evaluation method, system and equipment based on data mining | |
| Deng et al. | Rule-based fault diagnosis expert system for wind turbine | |
| CN102096770A (en) | Network-based performance degradation assessment method of rotary machine | |
| CN107218180B (en) | A kind of wind power generating set driving unit fault alarm method based on vibration acceleration measurement | |
| Zhang et al. | An unsupervised anomaly detection approach based on industrial big 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 |