CN109538311A - Control performance method of real-time towards steam turbine in high-end power generating equipment - Google Patents
Control performance method of real-time towards steam turbine in high-end power generating equipment Download PDFInfo
- Publication number
- CN109538311A CN109538311A CN201811110373.9A CN201811110373A CN109538311A CN 109538311 A CN109538311 A CN 109538311A CN 201811110373 A CN201811110373 A CN 201811110373A CN 109538311 A CN109538311 A CN 109538311A
- Authority
- CN
- China
- Prior art keywords
- matrix
- new
- variable
- slow
- steam turbine
- 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
-
- F—MECHANICAL ENGINEERING; LIGHTING; HEATING; WEAPONS; BLASTING
- F01—MACHINES OR ENGINES IN GENERAL; ENGINE PLANTS IN GENERAL; STEAM ENGINES
- F01D—NON-POSITIVE DISPLACEMENT MACHINES OR ENGINES, e.g. STEAM TURBINES
- F01D21/00—Shutting-down of machines or engines, e.g. in emergency; Regulating, controlling, or safety means not otherwise provided for
- F01D21/003—Arrangements for testing or measuring
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- General Engineering & Computer Science (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a kind of control performance method of real-time towards steam turbine in high-end power generating equipment.There is complicated relationships between the load of steam turbine performance index and power plant units, operating parameter, the present invention problem difficult because of control performance monitoring caused by parameter is numerous, operating condition is changeable for steam turbine in high-end power generating equipment, the relevant information between extracting steam turbine control system variable is analyzed with canonical variable, slow feature analysis al is recycled, the multidate information in relevant information is extracted.Finally, in conjunction with the correlation and variation speed information structuring steam turbine control performance on-line monitoring model of variable.Numerous, control performance monitoring difficulty caused by operating condition variation problem that the method overcome large-size steam turbine dependent variables, substantially increase the accuracy of dynamic process control performance on-line monitoring, facilitate thermal power plant and steam turbine control system is effectively and timely monitored, it is significant to the safe and reliable operation for guaranteeing high-end power generating equipment.
Description
Technical field
The invention belongs to thermoelectricity Process Control System performance monitoring fields, more particularly to steamer in high-end power generating equipment
The on-line performance monitoring method of machine operation correlation information and multidate information.
Background technique
Control system occupies very important status, the quality of production, safe operation, object in the industrial process of modernization
The index that energy consumption etc. influences economic benefit is all direct or indirect related with the performance of control system.In actual production process
In, often performance is good at the initial stage of coming into operation for control system, but after operation a period of time, due to the abrasion of equipment, determines
Phase maintenance and the maintenance reasons such as not in time, may cause the performance decline of control system, control performance variation will have a direct impact on life
Yield and quality causes economic benefit to be lost, if therefore causing fault in production, can also be related to the life security even enterprise, society of people
The property safety of industry, brings great threat.Torrres et al. is to 2004-2005 Brazilian 12 factories (petrochemical industry, papermaking, water
Mud, steel, mining etc.), more than 700 control loops are tested, and the valve wear in 14% circuit is excessive as the result is shown,
15% valve is there are braking problems, and there are serious problem of tuning, 24% controller output has saturation in 16% circuit
Phenomenon, 41% circuit because the problem of problem of tuning, coupling, disturbance and actuator and there are oscillatory occurences.
In addition, a production process might have thousands of control loop collective effects in actual production,
Two rectifying production equipments in Eastman chemical company possess much 14000 control loops, in HVAC production process,
The quantity of its control loop can even reach 100,000.High-end power generating equipment complexity with higher, is embodied in rule
Mould is huge, equipment is numerous, parameter is diversified and influences each other etc..In addition, large-scale generating set, scene has height
The features such as temperature, high pressure and strong noise.
Control performance evaluation and monitoring technology are an emerging important technologies of process control field, it can utilize and set
Standby day-to-day operation data, the variation of real-time monitoring monitoring system control performance make EARLY RECOGNITION to the problem of control system
And optimization.For generating set, since the power load in electric system often changes, in order to maintain active power flat
Weighing apparatus keeps system frequency to stablize, and the department of generating electricity is needed to change the power output of generator accordingly to adapt to the variation of power load,
I.e. the operating condition of generating set is not stablized constant.But the evaluation of existing control performance and monitoring method such as principal component analysis,
Partial Least Squares, Fei Sheer discriminant analysis are all based on the ideal assuming lower progress of stable conditions, therefore, are used
In large-scale thermal power machine group steam turbine control system performance monitoring, good monitoring effect can not be obtained.
Summary of the invention
Steam turbine is steam continuously to flow as working medium, and the thermal energy of steam is converted to a kind of rotation of mechanical energy
Formula Thermal Motor.Generally it is made of jet pipe (or stator blade), movable vane piece, impeller, arbor bearing and cylinder etc..Steamer equipment
There are revolving speed height, the reliable, single-machine capacity that runs smoothly to can be made into very big (more than 1000MW) and convenient for directly coupling with generator
The advantages that, but whole structure is complicated, has very strict requirements to design, manufacture, installation, operation and service technique, and required
Configure the boiler of corresponding steam parameter and capacity.Except be widely used as in the modern times, large-size thermal power plant, nuclear power station and large size
Outside the major engine on naval vessel, it may also be used for the integrated complexs such as large-size chemical or steel.
It is an object of the invention to be directed to large-scale thermal power machine group steam turbine because parameter is numerous, structure is complicated, operating condition is more
The difficult problem of the monitoring of control performance caused by becoming extracts steamer with slow signature analysis blending algorithm with canonical variable analysis
Relevant information between machine control system variable and variation speed information, overcome that large-size steam turbine dependent variable is numerous, operating condition variation
Caused by the difficult problem of control performance monitoring.
The purpose of the present invention is be achieved through the following technical solutions: the control performance of steam turbine is real in high-end power generating equipment
When monitoring method, method includes the following steps:
(1) obtain training data: the control system for setting steam turbine has J measurand and performance variable, adopts each time
The observation vector y of available J × 1 of samplek, wherein subscript k is time index, samples the data obtained after n times and is expressed as
One two-dimensional observation matrixThe measurand is that steam turbine operation in the process can
Measured state parameter, including bearing metal temperature, bearing axial vibration, generator active power, steam turbine revolving speed, excitation
Three-phase temperature etc.；The performance variable include condensing water flow (three minutes average), feed pressure, feedwater flow, feeder to
Coal amount etc.；Training data should choose the sampled data of steam turbine under normal operating conditions.
(2) the timing relevant information of data is extracted using CVA algorithm, which is realized by following sub-step:
(2.1) timing expands building matrix and matrix in future in the past: in specific sampling instant k, by observation vector ykTo
P step is expanded before k generates observation vector in the pastF step is expanded after to k
Generate observation vector in futureAgain to yp,k, yf,kCarry out equalization processing:
Wherein: mean (yp,k) indicateMean value, mean (yf,k) indicateMean value.
Respectively with all past observation vectors and observation vector in future building past observing matrix YpSquare is observed with future
Battle array Yf:
Wherein, M=N-f-p+1, p, f are two Delay parameters, enable p=f, value can pass through sample autocorrelation function
To determine:
Wherein: autocorr (Yj, p) and representing matrix YpThe auto-correlation coefficient of j-th column vector and its time lag p；
(2.2) it constructs Hankel matrix: calculating the covariance matrix ∑ of matrix and matrix in future in the pastpp, ∑ffAnd he
Cross-covariance ∑fp, recycle covariance and Cross-covariance to construct Hankel matrix H:
(2.3) singular value decomposition: carrying out the available Jp group canonical variable of singular value decomposition to Hankel matrix and match,
With (ai TYp,bi TYf) indicate i-th group of canonical variable pairing, ai T、bi TIndicate the related coefficient between i-th group of canonical variable pairing:
H=UDVT (6)
U and V is respectively singular vector ui, viThe orthogonal matrix of composition, D are singular value matrix, and the singular vector in U, V is only
It is related in pairs, and correlation size is by i-th of singular value γ corresponding in DiCharacterization.Bigger (the γ of singular value1>γ2>…>
γJp), the correlation between canonical variable is bigger.
(2.4) it calculates transformation matrix and extracts canonical variable and residual error variable: interception matrixPreceding r
Column, the matrix after generating dimensionality reductionVrStill remain most of timing relevant information.Wherein, the size of r value can
To be determined by following criterion:
Cr expressiveness value, β are judgment threshold, β=0.5.
By VrCalculate canonical variable transition matrix C and residual error variable transition matrix L:
Recycle the available canonical variable space Z and residual error space E of transition matrix:
Column vector z in Z, Εk∈ r × 1, εkJp × 1 ∈ is illustrated respectively in the canonical variable of sampling instant k and residual error becomes
Amount；Row vector z in Z, Εt, εtSame variable is contained in the timing information of different moments.
(3) canonical variable space is extracted respectively using slow feature analysis al (Slow Feature Analysis, SFA)
Slow feature s in Z and residual error space ΕZ, sE.To extract slow feature s in the Z of canonical variable spaceZFor, this method key step
It is as follows:
(3.1) data normalization: canonical variable space Z is standardized by variable, calculation formula is as follows:
ztTime series vector of the same variable of table in different moments, mean (zt) indicate ztMean value, std (zt) indicate zt's
Standard deviation.
(3.2) output signal of the Z after projection is sZj, sZjIndicate sZJ-th of slow characteristic sequence.Consider linear conditions
Under,Indicate coefficient vector, this is equivalent to searching one and mentions from normalized input signal Z
Take slow characteristic signal sZ=[sZ1 T, sZ2 T..., sZr T]TTransition matrixI.e.
sZ=WZZ.Slow characteristic signal sZjThe objective function and constraint condition to be met are as follows:
Objective function:
Constraint condition are as follows:
Wherein:Indicate slow characteristic signal sZTiming Difference, operation<>is expressed ast1, t0Respectively indicate time bound.
(3.3) albefaction: singular value decomposition is utilized, to covariance matrix < ZZ of input dataT> carry out whitening processing can be with
The correlation in data is removed, the slow characteristic value extracted is made to carry different information:
Wherein: ΛZ -1/2BTFor whitening matrix, ΟZFor the input signal after corresponding albefaction.
(3.4) transition matrix W is calculatedZ: to input matrix OZIt does difference processing and obtains Timing Difference signalIt can demonstrate,prove
It is bright, it is rightCovariance matrixAfter carrying out singular value decomposition, a series of obtained singular value ωZjAs formula
(12) target function value described in
WZ=P ΛZ -1/2BT (17)
Slow feature s in the residual error space ΕEExtracting method and above-mentioned canonical variable space Z in slow feature sZMention
Take method identical.
(4) slow feature s is dividedZ: most slow feature corresponds to the smallest characteristic value, by the ascending arrangement of characteristic value, and according to
Preceding l feature is divided into s according to characteristic value sizeZThe slower feature of middle variation, uses sZ,dIt indicates；(r-l) a feature is drawn by after
It is divided into sZThe middle faster feature of variation, uses sZ,eIt indicates.The determination method of partitioning standards l is, first with slow characteristic value sZChange
Changing speed indicates process variableVariation speed:
Wherein: rjiFor matrix RZThe element that middle jth row i-th arranges, sZiIndicate that i-th of slow characteristic sequence, Δ () indicate meter
Calculate a kind of operation of the slow degree of sequence variation:
The also big feature of degree slower than input data will be slowly spent in the slow characteristic value extracted is divided into fast feature,
One shared MeA such fast feature:
Here card { } indicates element number in set { }.The M determined according to formula (19)eValue, it is corresponding by matrix ΩZ
It is also divided into two parts:
(5) calculate Monitoring Indexes: since first sample point in canonical variable space, each sample point can be with
Obtain one group of Monitoring Indexes (SZ,d 2, SZ,e 2)。
(6) it determines the control limit based on Monitoring Indexes: using the method for Density Estimator, first estimating dynamic and supervise
Survey index SZ,d 2Probability density function p (x), for give level of significance α, SZ,d 2Control limit SZ,d 2 UCLCalculation
Are as follows:
S can be calculated in the same wayZ,e 2Control limit SZ,e 2 UCL。
(7) step (6) the method is arrived according to step (3), extracts the slow feature s of residual error space ΕEAnd by sEIt is divided into
Two parts sE,d, sE,e, establish monitoring index SE,d 2, SE,e 2And calculate control limit SE,d 2 UCL, SE,e 2 UCL.With to canonical variable space
The processing mode of Z is identical, repeats no more.
(8) Online Monitoring Control performance: CVA-SFA model, step (5) based on step (2) to (4) foundation arrive step
(7) performance state of resulting four monitoring and statistics amounts on-line monitoring steam turbine control system, the step by following sub-step Lai
It realizes:
(8.1) new online data and new data pretreatment are obtained: collecting one section of new observation dataAfterwards, wherein following table new indicates new observation data, will first, in accordance with step (2)
YnewExpansion becomes the past matrix, and is standardized according to the mean value and standard deviation that obtain in step (2) to past matrix
Obtain Ypnew。
(8.2) canonical variable and residual error variable of new observation data are extracted: true using step (2) after standardization
Fixed transition matrix VrThe canonical variable space Z of new observation data is calculated with LnewWith residual error space Enew。
(8.3) the canonical variable space Z of new observation data is extractednewIn slow feature: first, in accordance in step (3.1) really
Fixed mean value and variance is to ZnewIt is standardized, utilizes the slow Feature Conversion matrix W determined in step (3.4) laterZ,
Extract standardization ZnewSlow feature sZnew, and according to division parameter before by sZnewIt is divided into sZ,d newAnd sZ,e new, together
Sample is according to WEAvailable Enew, further obtain sE,d newAnd sE,e new。
(8.4) new monitoring and statistics index is calculated: according to the calculating side determined in the model of foundation and step (5) (7)
Method calculates the monitoring and statistics index S under canonical variable spaceZ,d 2 new, SZ,e 2 newWith residual error space monitoring index SE,d 2 new,
SE,e 2 new:
(8.5) judge steam turbine control performance state online: relatively four monitoring indexes and its respective statistics are controlled in real time
System limit shows that control system works normally if four monitoring indexes are all located within Statisti-cal control limit；If there is one or more
Monitoring index is limited beyond normal control, shows that control system has unusual condition.
The beneficial effects of the present invention are: the present invention is for large-scale thermal power machine group steam turbine because parameter is numerous, structure
Problem complicated, the changeable caused control performance monitoring of operating condition is difficult, proposes one kind towards steam turbine in high-end power generating equipment
Control performance method of real-time, this method with canonical variable analysis extract steam turbine control system variable between correlation
Information recycles slow feature analysis al, extracts the multidate information in relevant information and residual information.Finally, in conjunction with variable
Correlation and variation speed information structuring steam turbine control performance on-line monitoring model.The method overcome large-size steam turbine because
The difficult problem of control performance monitoring, substantially increases dynamic process control performance and exists caused by variable is numerous, operating condition changes
The accuracy of line monitoring, facilitates thermal power plant and is effectively and timely monitored to steam turbine control system, help to ensure that large size
The safe and reliable operation of generating set, while reaching the production requirement for improving its productivity effect.
Detailed description of the invention
Fig. 1 is the flow chart of the invention towards the control performance method of real-time of steam turbine in high-end power generating equipment,
(a) it is off-line modeling process flow diagram flow chart, is (b) on-line monitoring process flow diagram flow chart；
Fig. 2 is the result figure that CVA-SFA method of the present invention is used for statistical process monitoring, and (a) is that monitoring is tied under normal circumstances
Fruit figure is (b) monitoring result figure under abnormal conditions.
Specific embodiment
With reference to the accompanying drawing and specific example, invention is further described in detail.
For the present invention by taking subordinate Jia Hua power plant, Zhe Neng group #5 power generator turbine as an example, the power of the unit is 600,000 thousand
Watt, it is large-scale thermal power machine group, including 60 process variables, including bearing metal temperature, bearing axial vibration, generator
Active power, excitation three-phase temperature, condensing water flow (three minutes average), feed pressure, feedwater flow, gives coal at steam turbine revolving speed
Machine coal-supplying amount etc. and some valve openings.
It should be understood that the present invention is not limited to the thermoelectricity power generation process of examples detailed above, all technologies for being familiar with this field
Personnel can also make equivalent modifications or replacement without prejudice to the invention, these equivalent variation or replacement are wrapped
Containing within the scope defined in the claims of this application.
As shown in Figure 1, the present invention is a kind of control performance real-time monitoring side towards steam turbine in high-end power generating equipment
Method, comprising the following steps:
(1) obtain training data: the control system for setting steam turbine has J measurand and performance variable, adopts each time
The observation vector y of available J × 1 of samplek, wherein subscript k is time index, samples the data obtained after n times and is expressed as
One two-dimensional observation matrixIn this example, the sampling period is 10 minutes, totally 4566
A sample, 60 process variables, surveyed variable are the flow during steam turbine operation, vibration, temperature, pressure, valve opening
Deng；
(2) the timing relevant information of data is extracted using CVA algorithm, which is realized by following sub-step:
(2.1) timing expands building matrix and matrix in future in the past: in specific sampling instant k, by observation vector ykTo
P step is expanded before k generates observation vector in the pastF step is expanded after to k
Generate observation vector in futureAgain to yp,k, yf,kCarry out equalization processing:
Wherein: mean (yp,k) indicateMean value, mean (yf,k) indicateMean value.
Respectively with all past observation vectors and observation vector in future building past observing matrix YpSquare is observed with future
Battle array Yf:
Wherein, M=N-f-p+1, p, f are two Delay parameters, enable p=f, value can pass through sample autocorrelation function
To determine:
Wherein: autocorr (Yj, p) and representing matrix YpThe auto-correlation coefficient of j-th column vector and its time lag p；
(2.2) it constructs Hankel matrix: calculating the covariance matrix ∑ of matrix and matrix in future in the pastpp, ∑ffAnd he
Cross-covariance ∑fp, recycle covariance and Cross-covariance to construct Hankel matrix H:
(2.3) singular value decomposition: carrying out the available Jp group canonical variable of singular value decomposition to Hankel matrix and match,
With (ai TYp,bi TYf) indicate i-th group of canonical variable pairing, ai T、bi TIndicate the related coefficient between i-th group of canonical variable pairing:
H=UDVT (6)
U and V is respectively singular vector ui, viThe orthogonal matrix of composition, D are singular value matrix, and the singular vector in U, V is only
It is related in pairs, and correlation size is by i-th of singular value γ corresponding in DiCharacterization.Bigger (the γ of singular value1>γ2>…>
γJp), the correlation between canonical variable is bigger.
(2.4) it calculates transformation matrix and extracts canonical variable and residual error variable: interception matrixPreceding r
Column, the matrix after generating dimensionality reductionVrStill remain most of timing relevant information.Wherein, the size of r value can
To be determined by following criterion:
Cr expressiveness value, β are judgment threshold, β=0.5.
By VrCalculate canonical variable transition matrix C and residual error variable transition matrix L:
Recycle the available canonical variable space Z and residual error space E of transition matrix:
Column vector z in Z, Εk∈ r × 1, εkJp × 1 ∈ is illustrated respectively in the canonical variable of sampling instant k and residual error becomes
Amount；Row vector z in Z, Εt, εtSame variable is contained in the timing information of different moments.
(3) canonical variable space is extracted respectively using slow feature analysis al (Slow Feature Analysis, SFA)
Slow feature s in Z and residual error space ΕZ, sE.To extract slow feature s in the Z of canonical variable spaceZFor, this method key step
It is as follows:
(3.1) data normalization: canonical variable space Z is standardized by variable, calculation formula is as follows:
ztTime series vector of the same variable of table in different moments, mean (zt) indicate ztMean value, std (zt) indicate zt's
Standard deviation.
(3.2) output signal of the Z after projection is sZj, sZjIndicate sZJ-th of slow characteristic sequence.Consider linear conditions
Under,Indicate coefficient vector, this is equivalent to searching one and mentions from normalized input signal Z
Take slow characteristic signal sZ=[sZ1 T, sZ2 T..., sZr T]TTransition matrix
That is sZ=WZZ.Slow characteristic signal sZjThe objective function and constraint condition to be met are as follows:
Objective function:
Constraint condition are as follows:
Wherein:Indicate slow characteristic signal sZTiming Difference, operation<>is expressed ast1, t0Respectively indicate time bound.
(3.3) albefaction: singular value decomposition is utilized, to covariance matrix < ZZ of input dataT> carry out whitening processing can be with
The correlation in data is removed, the slow characteristic value extracted is made to carry different information:
Wherein: ΛZ -1/2BTFor whitening matrix, ΟZFor the input signal after corresponding albefaction.
(3.4) transition matrix W is calculatedZ: to input matrix OZIt does difference processing and obtains Timing Difference signalIt can demonstrate,prove
It is bright, it is rightCovariance matrixAfter carrying out singular value decomposition, a series of obtained singular value ωZjAs formula
(12) target function value described in
WZ=P ΛZ -1/2BT (17)
Slow feature s in the residual error space ΕEExtracting method and above-mentioned canonical variable space Z in slow feature sZMention
Take method identical.
(4) slow feature s is dividedZ: most slow feature corresponds to the smallest characteristic value, by the ascending arrangement of characteristic value, and according to
Preceding l feature is divided into s according to characteristic value sizeZThe slower feature of middle variation, uses sZ,dIt indicates；(r-l) a feature is drawn by after
It is divided into sZThe middle faster feature of variation, uses sZ,eIt indicates.The determination method of partitioning standards l is, first with slow characteristic value sZChange
Changing speed indicates process variableVariation speed:
Wherein: rjiFor matrix RZThe element that middle jth row i-th arranges, sZiIndicate that i-th of slow characteristic sequence, Δ () indicate meter
Calculate a kind of operation of the slow degree of sequence variation:
The also big feature of degree slower than input data will be slowly spent in the slow characteristic value extracted is divided into fast feature,
One shared MeA such fast feature:
Here card { } indicates element number in set { }.The M determined according to formula (19)eValue, it is corresponding by matrix ΩZ
It is also divided into two parts:
(5) calculate Monitoring Indexes: since first sample point in canonical variable space, each sample point can be with
Obtain one group of Monitoring Indexes (SZ,d 2, SZ,e 2)。
(6) it determines the control limit based on Monitoring Indexes: using the method for Density Estimator, first estimating dynamic and supervise
Survey index SZ,d 2Probability density function p (x), for give level of significance α, SZ,d 2Control limit SZ,d 2 UCLCalculation
Are as follows:
S can be calculated in the same wayZ,e 2Control limit SZ,e 2 UCL。
(7) step (6) the method is arrived according to step (3), extracts the slow feature s of residual error space ΕEAnd by sEIt is divided into
Two parts sE,d, sE,e, establish monitoring index SE,d 2, SE,e 2And calculate control limit SE,d 2 UCL, SE,e 2 UCL.With to canonical variable space
The processing mode of Z is identical, repeats no more.
(8) Online Monitoring Control performance: CVA-SFA model, step (5) based on step (2) to (4) foundation arrive step
(7) resulting four monitoring and statistics amounts can monitor the performance state of steam turbine control system on-line.The step is by following sub-step
It is rapid to realize:
(8.1) new online data and new data pretreatment are obtained: collecting one section of new observation dataAfterwards, wherein following table new indicates new observation data, will first, in accordance with step (2)
YnewExpansion becomes the past matrix, and is standardized according to the mean value and standard deviation that obtain in step (2) to past matrix
Obtain Ypnew.In this example, new data shares two parts, and data one are the data acquired under nominal situation, and the sampling period is 10 points
Clock, totally 2271 samples, 60 process variables, data two are to be abnormal the data recorded under operating condition, and the sampling period is 10 points
Clock, totally 2563 samples, 60 process variables, surveyed variable are flow during steam turbine operation, vibration, temperature, turn
Speed, electric current etc.；
(8.2) canonical variable and residual error variable of new observation data are extracted: true using step (2) after standardization
Fixed transition matrix VrThe canonical variable space Z of new observation data is calculated with LnewWith residual error space Enew。
(8.3) the canonical variable space Z of new observation data is extractednewIn slow feature: first, in accordance in step (3.1) really
Fixed mean value and variance is to ZnewIt is standardized, utilizes the slow Feature Conversion matrix W determined in step (3.4) laterZ,
Extract standardization ZnewSlow feature sZnew, and according to division parameter before by sZnewIt is divided into sZ,d newAnd sZ,e new, together
Sample is according to WEAvailable Enew, further obtain sE,d newAnd sE,e new。
(8.4) new monitoring and statistics index is calculated: according to the calculating side determined in the model of foundation and step (5) (7)
Method calculates the monitoring and statistics index S under canonical variable spaceZ,d 2 new, SZ,e 2 newWith residual error space monitoring index SE,d 2 new,
SE,e 2 new:
(8.5) judge steam turbine control performance state online:
Relatively four monitoring indexes and its respective Statisti-cal control limit in real time, if four monitoring indexes are all located at statistics control
Within system limit, show that control system works normally；
If there is one or more monitoring index to limit beyond normal control, show that control system has unusual condition.
In Fig. 2 (a), in four groups of statistics and corresponding control line, the statistic only put individually has been more than control line, is being set
Under conditions of believing horizontal α=0.05, it is believed that new floor data be it is normal, i.e., control system is acted normally；Fig. 2 (b)
In, four groups of statistic SZ,d 2, SZ,e 2, SE,d 2, SE,e 2In the 1330th-1430,1808-1825,1943-1972,2093-2128
Four sections obviously exceed threshold line, statistic QkThe shape that transfinites is maintained always after the 326th sampled point or so obviously transfinites for the first time
State, may determine that has abnormal generation in these period controlling systems accordingly, at this moment can use fault diagnosis side appropriate
Method, such as contribution drawing method analysis isolate possible failure variable.
The present invention extracts the relevant information between steam turbine control system variable with canonical variable analysis, recycles slow special
Parser is levied, the behavioral characteristics in relevant information are extracted, can be made by the feature that this method is extracted with the adjusting of response controller
With.Finally, in conjunction with the correlation and variation speed information structuring steam turbine control performance on-line monitoring model of variable, this method
It overcomes that large-size steam turbine dependent variable is numerous, the difficult problem of control performance monitoring caused by operating condition variation, substantially increases
The accuracy of dynamic process control performance on-line monitoring facilitates thermal power plant and is effectively and timely monitored to turbine system,
It helps to ensure that the safe and reliable operation of large-scale thermal power machine group, while reaching the production requirement for improving its productivity effect.
Claims (1)
1. a kind of control performance method of real-time towards steam turbine in high-end power generating equipment, which is characterized in that this method packet
Include following steps:
(1) obtain training data: the control system for setting steam turbine has J measurand and performance variable, and sampling can each time
To obtain the observation vector y of J × 1k, wherein subscript k is time index, samples the data obtained after n times and is expressed as one
Two-dimensional observation matrixThe measurand is that steam turbine operation can be measured in the process
State parameter, including bearing metal temperature, bearing axial vibration, generator active power, steam turbine revolving speed, excitation three-phase temperature
Deng；The performance variable includes condensing water flow (three minutes average), feed pressure, feedwater flow, feeder coal-supplying amount etc.；Instruction
The sampled data of steam turbine under normal operating conditions should be chosen by practicing data.
(2) the timing relevant information of data is extracted using CVA algorithm, which is realized by following sub-step:
(2.1) timing expands building matrix and matrix in future in the past: in specific sampling instant k, by observation vector ykTo before k
It expands p step and generates observation vector in the pastF step generation is expanded after to k will
Carry out observation vectorAgain to yp,k, yf,kCarry out equalization processing:
Wherein: mean (yp,k) indicateMean value, mean (yf,k) indicate
Mean value.
Respectively with all past observation vectors and observation vector in future building past observing matrix YpWith observing matrix Y in futuref:
Wherein, M=N-f-p+1, p, f are two Delay parameters, enable p=f, and value can be by sample autocorrelation function come really
It is fixed:
Wherein: autocorr (Yj, p) and representing matrix YpThe auto-correlation coefficient of j-th column vector and its time lag p；
(2.2) it constructs Hankel matrix: calculating the covariance matrix ∑ of matrix and matrix in future in the pastpp, ∑ffAnd theirs is mutual
Covariance matrix ∑fp, recycle covariance and Cross-covariance to construct Hankel matrix H:
(2.3) singular value decomposition: the available Jp group canonical variable of singular value decomposition is carried out to Hankel matrix and is matched, is used
(ai TYp,bi TYf) indicate i-th group of canonical variable pairing, ai T、bi TIndicate the related coefficient between i-th group of canonical variable pairing:
H=UDVT (6)
U and V is respectively singular vector ui, viThe orthogonal matrix of composition, D are singular value matrix, and the singular vector in U, V is only pairs of
Correlation, and correlation size is by i-th of singular value γ corresponding in DiCharacterization.Bigger (the γ of singular value1>γ2>…>γJp), it is typical
Correlation between variable is bigger.
(2.4) it calculates transformation matrix and extracts canonical variable and residual error variable: interception matrixPreceding r column, it is raw
At the matrix after dimensionality reductionVrStill remain most of timing relevant information.Wherein, the size of r value can pass through
Following criterion determines:
Cr expressiveness value, β are judgment threshold, β=0.5.
By VrCalculate canonical variable transition matrix C and residual error variable transition matrix L:
Recycle the available canonical variable space Z and residual error space E of transition matrix:
Column vector z in Z, Εk∈ r × 1, εkJp × 1 ∈ is illustrated respectively in the canonical variable and residual error variable of sampling instant k；Z,
Row vector z in Εt, εtSame variable is contained in the timing information of different moments.
(3) canonical variable space Z and residual is extracted respectively using slow feature analysis al (Slow Feature Analysis, SFA)
Slow feature s in difference space ΕZ, sE.Slow feature s in the Z of canonical variable spaceZExtracting method it is as follows:
(3.1) data normalization: canonical variable space Z is standardized by variable, calculation formula is as follows:
ztTime series vector of the same variable of table in different moments, mean (zt) indicate ztMean value, std (zt) indicate ztStandard
Difference.
(3.2) output signal of the Z after projection is sZj, sZjIndicate j-th of sZ slow characteristic sequence.Consider under linear conditions, Indicate coefficient vector, this is equivalent to find the slow spy of extraction from normalized input signal Z
Reference sZ=[sZ1 T, sZ2 T..., sZr T]TTransition matrixThat is sZ=
WZZ.Slow characteristic signal sZjThe objective function and constraint condition to be met are as follows:
Objective function:
Constraint condition are as follows:
Wherein:Indicate slow characteristic signal sZTiming Difference, operation<>is expressed ast1, t0Respectively indicate time bound.
(3.3) albefaction: singular value decomposition is utilized, to covariance matrix < ZZ of input dataT> progress whitening processing can remove divisor
Correlation in makes the slow characteristic value extracted carry different information:
Wherein: ΛZ -1/2BTFor whitening matrix, ΟZFor the input signal after corresponding albefaction.
(3.4) transition matrix W is calculatedZ: to input matrix OZIt does difference processing and obtains Timing Difference signalIt can be proved that rightCovariance matrixAfter carrying out singular value decomposition, a series of obtained singular value ωZjAs formula (12) institute
The target function value stated
WZ=P ΛZ -1/2BT (17)
Slow feature s in the residual error space ΕEExtracting method and above-mentioned canonical variable space Z in slow feature sZExtraction side
Method is identical.
(4) slow feature s is dividedZ: most slow feature corresponds to the smallest characteristic value, by the ascending arrangement of characteristic value, and according to feature
Preceding l feature is divided into s by value sizeZThe slower feature of middle variation, uses sZ,dIt indicates；(r-l) a feature is divided into s by afterZIn
Change faster feature, uses sZ,eIt indicates.The determination method of partitioning standards l is, first with slow characteristic value sZVariation speed table
Show process variableVariation speed:
Wherein: rjiFor matrix RZThe element that middle jth row i-th arranges, sZiIndicate that i-th of slow characteristic sequence, Δ () indicate to calculate sequence
Column change a kind of operation of slow degree:
The also big feature of degree slower than input data will be slowly spent in the slow characteristic value extracted and is divided into fast feature, and one is shared
MeA such fast feature:
Here card { } indicates element number in set { }.The M determined according to formula (19)eValue, it is corresponding by matrix ΩZAlso it draws
It is divided into two parts:
(5) Monitoring Indexes are calculated: since first sample point in canonical variable space, each sample point available one
Group Monitoring Indexes (SZ,d 2, SZ,e 2)。
(6) it determines the control limit based on Monitoring Indexes: using the method for Density Estimator, first estimating Monitoring Indexes
SZ,d 2Probability density function p (x), for give level of significance α, SZ,d 2Control limitCalculation are as follows:
S can be calculated in the same wayZ,e 2Control limit
(7) step (6) the method is arrived according to step (3), extracts the slow feature s of residual error space ΕEAnd by sEIt is divided into two parts
sE,d, sE,e, establish monitoring index SE,d 2, SE,e 2And calculate control limit
(8) Online Monitoring Control performance: CVA-SFA model, step (5) based on step (2) to (4) foundation arrive step (7) institute
The performance state of the four monitoring and statistics amounts on-line monitoring steam turbine control system obtained, the step are realized by following sub-step:
(8.1) new online data and new data pretreatment are obtained: collecting one section of new observation dataAfterwards, wherein following table new indicates new observation data, will first, in accordance with step (2)
YnewExpansion becomes the past matrix, and is standardized according to the mean value and standard deviation that obtain in step (2) to past matrix
To Ypnew。
(8.2) canonical variable and residual error variable of new observation data are extracted: after standardization, being determined using step (2)
Transition matrix VrThe canonical variable space Z of new observation data is calculated with LnewWith residual error space Enew。
(8.3) the canonical variable space Z of new observation data is extractednewIn slow feature: first, in accordance in step (3.1) determine
Mean value and variance are to ZnewIt is standardized, utilizes the slow Feature Conversion matrix W determined in step (3.4) laterZ, extract
Z is standardized outnewSlow feature sZnew, and according to division parameter before by sZnewIt is divided into sZ,d newAnd sZ,e new, same root
According to WEAvailable Enew, further obtain sE,d newAnd sE,e new。
(8.4) it calculates new monitoring and statistics index: according to the calculation method determined in the model of foundation and step (5) (7), calculating
Monitoring and statistics index under canonical variable spaceWith residual error space monitoring index
(8.5) judge steam turbine control performance state online: relatively four monitoring indexes and its respective Statisti-cal control limit in real time,
If four monitoring indexes are all located within Statisti-cal control limit, show that control system works normally；If there is one or more monitoring to refer to
Mark is limited beyond normal control, shows that control system has unusual condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811110373.9A CN109538311B (en) | 2018-09-21 | 2018-09-21 | Real-time monitoring method for control performance of steam turbine in high-end power generation equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811110373.9A CN109538311B (en) | 2018-09-21 | 2018-09-21 | Real-time monitoring method for control performance of steam turbine in high-end power generation equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109538311A true CN109538311A (en) | 2019-03-29 |
| CN109538311B CN109538311B (en) | 2020-08-04 |
Family
ID=65843543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811110373.9A Active CN109538311B (en) | 2018-09-21 | 2018-09-21 | Real-time monitoring method for control performance of steam turbine in high-end power generation equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109538311B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111199246A (en) * | 2019-12-24 | 2020-05-26 | 泉州装备制造研究所 | Working condition classification method |
| CN111474920A (en) * | 2020-04-29 | 2020-07-31 | 电子科技大学 | ICA and ACVA-based fault detection method |
| CN113065583A (en) * | 2021-03-13 | 2021-07-02 | 宁波大学科学技术学院 | Rectification process abnormity monitoring method based on online nonlinear discriminant feature analysis |
| CN114396317A (en) * | 2021-12-01 | 2022-04-26 | 上海发电设备成套设计研究院有限责任公司 | Multi-target multi-dimensional online combined monitoring method and system for nuclear turbine |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111188761B (en) * | 2019-12-31 | 2021-09-10 | 杭州哲达科技股份有限公司 | Monitoring method for pump equipment based on Fourier-CVA model |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1494020A1 (en) * | 2003-06-30 | 2005-01-05 | Siemens Westinghouse Power Corporation | Method and apparatus for measuring on line failure of turbine thermal barrier coatings |
| US20160033580A1 (en) * | 2012-05-29 | 2016-02-04 | Board Of Regents Of The University Of Nebraska | Detecting Faults in Turbine Generators |
| CN106097151A (en) * | 2016-06-24 | 2016-11-09 | 清华大学 | A kind of method reducing power plant's data uncertainty based on data harmonization |
| CN106680012A (en) * | 2017-01-25 | 2017-05-17 | 浙江大学 | Fault detection method and fault diagnosis method for the non-stationary process of large coal-fired generator set |
| CN107766457A (en) * | 2017-09-27 | 2018-03-06 | 华能国际电力股份有限公司上海石洞口第二电厂 | A kind of thermal power plant's operation management system and its task executing method |
| US20180171774A1 (en) * | 2016-12-21 | 2018-06-21 | Schlumberger Technology Corporation | Drillstring sticking management framework |
| CN108334674A (en) * | 2018-01-17 | 2018-07-27 | 浙江大学 | A kind of steam turbine high-pressure cylinder method for monitoring operation states based on parameter association intellectual analysis |
-
2018
- 2018-09-21 CN CN201811110373.9A patent/CN109538311B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1494020A1 (en) * | 2003-06-30 | 2005-01-05 | Siemens Westinghouse Power Corporation | Method and apparatus for measuring on line failure of turbine thermal barrier coatings |
| US20160033580A1 (en) * | 2012-05-29 | 2016-02-04 | Board Of Regents Of The University Of Nebraska | Detecting Faults in Turbine Generators |
| CN106097151A (en) * | 2016-06-24 | 2016-11-09 | 清华大学 | A kind of method reducing power plant's data uncertainty based on data harmonization |
| US20180171774A1 (en) * | 2016-12-21 | 2018-06-21 | Schlumberger Technology Corporation | Drillstring sticking management framework |
| CN106680012A (en) * | 2017-01-25 | 2017-05-17 | 浙江大学 | Fault detection method and fault diagnosis method for the non-stationary process of large coal-fired generator set |
| CN107766457A (en) * | 2017-09-27 | 2018-03-06 | 华能国际电力股份有限公司上海石洞口第二电厂 | A kind of thermal power plant's operation management system and its task executing method |
| CN108334674A (en) * | 2018-01-17 | 2018-07-27 | 浙江大学 | A kind of steam turbine high-pressure cylinder method for monitoring operation states based on parameter association intellectual analysis |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111199246A (en) * | 2019-12-24 | 2020-05-26 | 泉州装备制造研究所 | Working condition classification method |
| CN111199246B (en) * | 2019-12-24 | 2024-04-05 | 泉州装备制造研究所 | Working condition classification method |
| CN111474920A (en) * | 2020-04-29 | 2020-07-31 | 电子科技大学 | ICA and ACVA-based fault detection method |
| CN111474920B (en) * | 2020-04-29 | 2021-04-06 | 电子科技大学 | ICA and ACVA-based fault detection method |
| CN113065583A (en) * | 2021-03-13 | 2021-07-02 | 宁波大学科学技术学院 | Rectification process abnormity monitoring method based on online nonlinear discriminant feature analysis |
| CN113065583B (en) * | 2021-03-13 | 2023-11-14 | 宁波大学科学技术学院 | Rectification process abnormity monitoring method based on online nonlinear discriminant feature analysis |
| CN114396317A (en) * | 2021-12-01 | 2022-04-26 | 上海发电设备成套设计研究院有限责任公司 | Multi-target multi-dimensional online combined monitoring method and system for nuclear turbine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109538311B (en) | 2020-08-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109538311A (en) | Control performance method of real-time towards steam turbine in high-end power generating equipment | |
| CN109491358A (en) | A kind of control performance monitoring method towards gigawatt Boiler of Ultra-supercritical Unit multidate information | |
| Hsu et al. | An adaptive forecast-based chart for non-Gaussian processes monitoring: with application to equipment malfunctions detection in a thermal power plant | |
| CN110262450B (en) | Fault prediction method for cooperative analysis of multiple fault characteristics of steam turbine | |
| CN109471420A (en) | Intelligent power plant's large size Thermal generation unit air preheater control performance monitoring method based on CVA-SFA | |
| CN109143995B (en) | Quality-related slow characteristic decomposition closed-loop system fine operation state monitoring method | |
| CN111582392A (en) | Multi-working-condition health state online monitoring method for key components of wind turbine generator | |
| CN108490908B (en) | A kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation | |
| CN109184821B (en) | Online monitoring method for closed-loop information analysis of steam turbine of large-scale generator set | |
| CN109238760B (en) | Online monitoring method of coal mill of intelligent power plant coal-fired generator set based on typical correlation analysis and slow characteristic analysis | |
| CN108334674A (en) | A kind of steam turbine high-pressure cylinder method for monitoring operation states based on parameter association intellectual analysis | |
| CN111610021B (en) | Early fault detection method for gearbox | |
| CN118194026A (en) | Gas power generation data analysis system | |
| Fantoni | Experiences and applications of PEANO for online monitoring in power plants | |
| CN114326486A (en) | Process monitoring method based on probability slow feature analysis and elastic weight consolidation | |
| Haghani et al. | Data-driven multimode fault detection for wind energy conversion systems | |
| CN110441081B (en) | Intelligent diagnosis method and intelligent diagnosis system for faults of rotating machinery | |
| Kruger et al. | A data-driven approach for sensor fault diagnosis in gearbox of wind energy conversion system | |
| CN116305733B (en) | Quality-related fault detection method based on global and local feature extraction | |
| CN109270917B (en) | Intelligent power plant steam turbine bearing-oriented closed-loop control system fault degradation state prediction method | |
| Chu | Anomaly detection of hydropower generating set using operation condition and adaptive PCA | |
| Rosli et al. | Application of Principle Component Analysis in Resolving Influential Factor Subject to Industrial Motor Failure | |
| Ma et al. | Process monitoring of the pneumatic control valve using canonical variate analysis | |
| Tang et al. | Distributed Process Monitoring Based on multi-block KGLPP | |
| CN117034157B (en) | Hydropower equipment fault identification method and system combining multimodal operation 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 |