CN108490908B - A kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation - Google Patents
A kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation Download PDFInfo
- Publication number
- CN108490908B CN108490908B CN201810141761.7A CN201810141761A CN108490908B CN 108490908 B CN108490908 B CN 108490908B CN 201810141761 A CN201810141761 A CN 201810141761A CN 108490908 B CN108490908 B CN 108490908B
- Authority
- CN
- China
- Prior art keywords
- monitoring
- matrix
- variable
- variables
- vector
- 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
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0224—Process history based detection method, e.g. whereby history implies the availability of large amounts of data
- G05B23/024—Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of dynamic distributed monitoring methods towards gigawatt extra-supercritical unit variable parameter operation.The identification that the present invention this typical extensive non-stationary process for gigawatt extra-supercritical unit, the sparse cointegrating analysis of R. concomitans and dynamic feature extraction method change for fault detection and operating condition.Large scale process variable automatically can be divided into different set of variables by this method, while extraction process static state and multidate information be distinguished in each set of variables, and the two is modeled and monitored respectively.It realizes depth mining process information, and effectively realizes the transformation and actual process failure of differentiation process nominal situation.This method effective solution process monitoring difficulty problem of this extensive non-stationary dynamical system of gigawatt extra-supercritical unit, substantially increase the performance of process monitoring, facilitate field engineer and accurately grasp operating states of the units, to guarantee the safety of gigawatt extra-supercritical unit and improve productivity effect.
Description
Technical field
The invention belongs to extensive non-stationary process statistical monitoring fields, super towards gigawatt especially for one kind
The dynamic distributed monitoring method of criticality benchmark variable parameter operation.
Background technique
Power industry is the emphasis of Chinese national economy most important basic energy resource industry and Strategy for economic development.China lacks
The few gas of oil, rich coal resources, coal electricity is always China's main power source for a long time.Coal fired generation process is increasingly sophisticated
Change, enlargement development.Large-scale coal fired generation process environment is complicated, and scene has the features such as high temperature, high pressure, strong noise.1001000
Watt extra-supercritical unit be in the world it is state-of-the-art efficiently, large capacity coal powerplant, there is apparent efficiency advantage, be me
The representative unit and main flow direction of state's electric power industry development.China's gigawatt extra-supercritical unit ownership the first in the world,
And development space is huge.
Gigawatt extra-supercritical unit has apparent extensive, non-stationary and dynamic characteristic.In order to guarantee million
The safety of kilowatt extra-supercritical unit simultaneously improves productivity effect, it is necessary to use effective process monitoring method.Process monitoring is just
It is monitoring process operation state, is sounded an alarm in time when occurring extremely.With the development of technology, data are obtained in industry spot
It becomes increasingly easy, a large amount of procedural information has been contained in process data, then, the process monitoring method based on data has become
For the hot spot of research.
Forefathers have done corresponding research to the process monitoring based on data.Principal component analysis (PCA), offset minimum binary
(PLS) etc. Multielement statistical analysis methods have been widely used in data procedures monitoring field.However, these methods assumed that
Journey is that smoothly, there is no the dynamic characteristics for considering process.Actually since process is disturbed in gigawatt extra-supercritical unit
Dynamic, the reasons such as ageing equipment process can have apparent non-stationary property.Simultaneously because frequently peak regulation process is in dynamic change
Among.Therefore, gigawatt extra-supercritical unit has apparent extensive, non-stationary and dynamic characteristic, this is supervised to process
Survey brings very big challenge.
For gigawatt extra-supercritical unit, this typical extensive, dynamic non-stationary process proposes a kind of knot to the present invention
Close the process monitoring method that sparse cointegrating analysis and behavioral characteristics extract.This method goes out non-stationary identification variables in the process
Come, establishes sparse co-integration model using these non-stationary variables and variable is divided into different set of variables using alternative manner.It mentions
The method of dynamic feature extraction is for multidate information and static information to be distinguished from out, and the two is modeled and monitored respectively.
Variable can be automatically divided into different set of variables by this method, while can effectively extract process dynamics and static letter
Breath, the effective change and actual process failure for distinguishing normal operating conditions, substantially increases the performance of process monitoring.To current
Until, there is not yet research related to the present invention is reported.
Summary of the invention
It is an object of the invention to this typical extensive, dynamic non-stationary mistakes for gigawatt extra-supercritical unit
Journey provides a kind of fault detection method and diagnostic method towards large-scale Thermal generation unit non-stationary process.
The purpose of the present invention is what is be achieved through the following technical solutions: one kind is towards large-scale Thermal generation unit non-stationary mistake
The fault detection method of journey, comprising the following steps:
(1) normal processes data are obtained: being set comprising J process variable in a gigawatt extra-supercritical unit, every time
The vector of an available 1 × J is sampled, obtains the two-dimensional matrix X under a normal processes after sampling M timesn(M×J);
(2) it identifies non-stationary variable: identifying Two-Dimensional Moment using Augmented Dickey-Fuller (ADF) method of inspection
Battle array XnNon-stationary variable in (M × J) obtains non-stationary variable data matrix Xns(M × N), N indicate non-stationary variable number;
(3) non-stationary variable data matrix X obtained in (2) is utilizedns(M × N)=[x1,x2,…,xN], xt=(x1,
x2,…,xN)TSparse co-integration model is established, wherein xtFor data matrix XnsThe transposition of a row vector in (M × N) indicates
The sampled value of t moment, t=1,2 ..., M.It establishes sparse co-integration model and specifically includes following sub-step:
(3.1) to xtEstablish Vector Autoression Models
xt=Π1xt-1+…+Πpxt-p+c+μt (1)
Wherein, Π1~ΠpIt is the coefficient matrix of (N × N), μtFor (N × 1) matrix, white Gaussian noise, μ are indicatedt~N
(0, Ξ), c are (N × 1) matrix, indicate constant, p is model order;
(3.2) x is subtracted at formula (1) both endst-1Obtain error correction model
(3.3) Γ in step (3.2) is decomposed into matrix Γ=Α Β of two sequency spectrumsT, formula (2) becomes
Wherein, Α (N × R), Β (N × R);
(3.4) the whole vector matrix Β of association in formula (3) is estimated by Maximum Likelihood Estimation
Wherein, L (*) indicates maximum likelihood function, the mark of tr (*) representing matrix.X=(Δ xp+1,...,ΔxM)T, Y=
(ΔYp+1,...,ΔYM)T,L=p+1 ..., M, Ω=(Ω1,...,Ωp-1)T, Z=
(xt-1,...,xM-1)T, Θ=Ξ-1;
(3.5) characteristic equation solution procedure can be converted into the Maximum-likelihood estimation of formula (4)
Wherein,
Wherein parameter matrix ΘiAnd Φi, i=1,2 ..., p-1 can be acquired by least-squares algorithm;
(3.6) penalty is added to the objective function of formula (4) and obtains the sparse whole vector of association
Wherein, P1, P2, P3For parameter Β, the penalty of Ω, Θ, using a normal form;Wherein Β is to assist whole vector
Matrix, Ω, Θ are parameters to be estimated;Adjusting parameter λ1And λ2It is determined using crosscheck, adjusting parameter λ3Using Bayes
Information criterion determines.Pass through the available whole vector of sparse association of solution to formula (6).
(4) variable can be divided using the whole vector of sparse association obtained in formula (6), specifically includes following sub-step
It is rapid:
(4.1) the sparse whole vector matrix Β of association is obtained according to formula (6)s(N × K)=[βs,1,βs,2,...,βs,K], wherein
K indicates the number of the sparse whole vector of association;
(4.2) steady residual sequence is obtained according to the sparse whole vector of associationWherein k=1,2 ..., K, t=1,
2 ..., M.The consistent level for measuring residual sequence, and record residual difference sequence γ are examined using ADFk,tADF test statistics tk
(4.3) to obtained test statistics tkCarry out ascending sort, the corresponding sparse association of the smallest inspection statistics magnitude
Whole vector is retained.The corresponding variable of nonzero element in the sparse whole vector of association is assigned in subgroup, and is denoted as Xb。
(4.4) by XbIn variable from initial data concentrate remove, raw data set is denoted as X at this timeL。
(4.5) iteration step (4.1)-(4.4) are all assigned in different subgroups until all variables, original at this time
Variable is divided into Z different set of variables.
(5) local static, dynamic monitoring statistic are established in each set of variables:
γs,b,tFor the steady residual sequence in each set of variables, Bf,bFor the whole moment of a vector of association calculated in each set of variables
Battle array, Λs,b=(XbBf,b)T(XbBf,b)/(M-1) (b=1 ..., Z) be covariance matrix, M be sampled point number.Z is variable point
The number of group indicates that original variable is divided into Z set of variables, xb,tFor XbThe transposition of middle row vector indicates the sampling in t moment
Point.
(5.2) local dynamic station monitoring and statistics amount is established
B indicates b-th of set of variables, and p indicates time lag, Δ xp+1,bIndicate the Difference Terms, that is, Δ x at p+1 momentp+1,b=
xp+1,b-xp,b, xp,bFor XbThe transposition of middle row vector indicates the sampled point at the p moment, ΘbAnd Φb(b=1 ..., Z) it is wait estimate
Parameter is counted, can be acquired by least square.Be,bAnd Bf,bFor EbAnd FbCalculate the weight matrix obtained after canonical correlation analysis.It is dynamic
State monitoring and statistics amount is
Wherein, te,bFrom matrix Te,b, it is Te,bMiddle column vector.
(6) global monitoring and statistics amount is established
(6.1) global static monitoring techniques statistic is established:
Wherein, Z indicates the number of set of variables; PS(xb)=P
(xb|Ns)P(Ns)+P(xb|Fs)P(Fs);P(Ns) it is confidence interval, i.e. P (Ns)=α, P (Fs)=1- α; For sample xb'sMonitoring and statistics
Amount,For monitoring and statistics amountControl limit;
(6.2) global dynamic monitoring statistic is established:
Wherein, Z indicates the number of set of variables; Pe(xb)=P (xb
|Ne)P(Ne)+P(xb|Fe)P(Fe);P(Ne) it is confidence interval, i.e. P (Ne)=α, P (Fe)=1- α; For sample xb'sMonitoring and statistics
Amount,For monitoring and statistics amountControl limit;
(7) online process monitoring
(7.1) to new collected non-stationary variable sample xnew,t(N × 1) carries out variable grouping, which is divided into Z
Set of variables, that is, xnew,b,t, b=1,2 ..., Z.
(7.2) method recorded according to step 5 calculates local static monitoring and statistics amount
xnewi,tWith xnew,t(N × 1) what relationship, xnewi,tWhat is indicated
(7.3) local dynamic station monitoring and statistics amount is calculated
(7.4) global static, dynamic monitoring statistic is calculated
IfWithIt is limited beyond control, it is meant that static, dynamic long-run equilibrium relationship is broken, process hair
Raw failure;IfWhile alarm,Beyond normal range (NR) is returned to after control limit, meaned at this time
The operating point of journey changes;IfWithIt is all limited without departing from control, means process work in new work at this time
Static state, dynamic long-run equilibrium relationship under condition, but between variable are not broken or the operating condition is included in modeling data.
If local static, dynamic monitoring statistic are influenced by failure, BICsOr BICeAlso it beyond control limit, anticipates at this time
Taste the failure not only influence the local state of process, while global state is also affected, if BICsOr BICeDo not exceed
Control limit, indicates the local state of the failure influence process.
The invention has the benefit that extensive non-stationary process can be divided into different set of variables by the present invention to be reduced
The complexity of process, while by monitoring static and dynamic change in different set of variables, the identification of operating condition variation may be implemented
With the detection to failure, provides local monitoring information while providing global monitoring information.
Detailed description of the invention
Fig. 1 is the flow chart of the dynamic distributed monitoring the present invention is based on sparse cointegrating analysis;
Fig. 2 is area monitoring's result of this method;
Fig. 3 is the global monitoring result of this method;
Fig. 4 is the monitoring result of traditional cointegrating analysis.
Fig. 5 is the monitoring result of traditional cointegrating analysis.
Specific embodiment
With reference to the accompanying drawing and specific example, invention is further described in detail.
Gigawatt extra-supercritical unit is typical non-stationary process, and a portion variable has apparent non-stationary special
Property, such as condenser circulating water pressure, high-pressure heater initial steam pressure, oxygen-eliminating device incoming condensing water amount.The present invention is to praise Hua Ran
For the unit of power plant 8, it is gigawatt extra-supercritical unit, including 159 that the power of the unit, which is 10000MW,
Process variable, these variables are related to pressure, temperature, water level, flow velocity etc..
As shown in Figure 1, the present invention is a kind of towards the dynamic distributed of gigawatt extra-supercritical unit variable parameter operation
Monitoring method, comprising the following steps:
Step 1: acquisition process data: setting in a gigawatt extra-supercritical unit comprising J process variable, adopt every time
The vector of the available 1 × J of sample, the data obtained after sampling K times can be described as a two-dimensional matrix X (K × J).It obtains
Take normal data Xn(M × J), wherein subscript n indicates normal data.In this example, the sampling period is 1 minute, for normal number
According to 740 samples of acquisition, 159 process variables.So normal data sample is Xn(740×159)。
Step 2: identification non-stationary variable: normal using the identification of Augmented Dickey-Fuller (ADF) method of inspection
Data XnNon-stationary variable in (740 × 159), is examined by ADF, wherein 51 variables are non-stationary variable, is obtained non-flat
Steady variable data Xns(740 × 51), wherein subscript n s indicates Non-stationary Data.Below for convenience of expression, by sampled point number 740
It is indicated with M, non-stationary variables number 51 is indicated with N.
Step 3: utilizing non-stationary variable data matrix X obtained in (2)ns(M × N)=[x1,x2,…,xN], xt=
(x1,x2,…,xN)TSparse co-integration model is established, wherein xtFor data matrix XnsThe transposition of a row vector in (M × N), table
Show the sampled value in t moment, t=1,2 ..., M.It establishes sparse co-integration model and specifically includes following sub-step:
(3.1) to xtEstablish Vector Autoression Models
xt=Π1xt-1+…+Πpxt-p+c+μt (1)
Wherein, Π1~ΠpIt is the coefficient matrix of (N × N), μtFor (N × 1) matrix, white Gaussian noise, μ are indicatedt~N
(0, Ξ), c are (N × 1) matrix, indicate constant, p is model order;
(3.2) x is subtracted at formula (1) both endst-1Obtain error correction model
(3.3) Γ in step (3.2) is decomposed into matrix Γ=Α Β of two sequency spectrumsT, formula (2) becomes
Wherein, Α (N × R), Β (N × R);
(3.4) the whole vector matrix Β of association in formula (3) is estimated by Maximum Likelihood Estimation
Wherein, L (*) indicates maximum likelihood function, the mark of tr (*) representing matrix.X=(Δ xp+1,...,ΔxM)T, Y=
(ΔYp+1,...,ΔYM)T,L=p+1 ..., M, Ω=(Ω1,...,Ωp-1)T, Z=
(xt-1,...,xM-1)T, Θ=Ξ-1;
(3.5) characteristic equation solution procedure can be converted into the Maximum-likelihood estimation of formula (4)
Wherein,
Wherein parameter matrix ΘiAnd Φi, i=1,2 ..., p-1 can be acquired by least-squares algorithm;
(3.6) penalty is added to the objective function of formula (4) and obtains the sparse whole vector of association
Wherein, P1, P2, P3For parameter Β, the penalty of Ω, Θ, using a normal form;Wherein Β is to assist whole vector
Matrix, Ω, Θ are parameters to be estimated;Adjusting parameter λ1And λ2It is determined using crosscheck, adjusting parameter λ3Using Bayes
Information criterion determines.Pass through the available whole vector of sparse association of solution to formula (6).
(4) variable can be divided using the whole vector of sparse association obtained in formula (6), specifically includes following sub-step
It is rapid:
(4.1) the sparse whole vector matrix Β of association is obtained according to formula (6)s(N × K)=[βs,1,βs,2,...,βs,K], wherein
K indicates the number of the sparse whole vector of association;
(4.2) steady residual sequence is obtained according to the sparse whole vector of associationWherein k=1,2 ..., K, t=
1,2 ..., M.The consistent level for measuring residual sequence, and record residual difference sequence γ are examined using ADFk,tADF test statistics
tk
(4.3) to obtained test statistics tkCarry out ascending sort, the corresponding sparse association of the smallest inspection statistics magnitude
Whole vector is retained.The corresponding variable of nonzero element in the sparse whole vector of association is assigned in subgroup, and is denoted as Xb。
(4.4) by XbIn variable from initial data concentrate remove, raw data set is denoted as X at this timeL。
(4.5) iteration step (4.1)-(4.4), all variables are all assigned in 5 different subgroups.
(5) local static, dynamic monitoring statistic are established in each set of variables:
Wherein,
γs,b,tFor the steady residual sequence in each set of variables, Bf,bFor the whole moment of a vector of association calculated in each set of variables
Battle array, Λs,b=(XbBf,b)T(XbBf,b)/(M-1) (b=1 ..., 5) be covariance matrix, M be sampled point number. xb,tFor Xb
The transposition of middle row vector indicates the sampled point in t moment.
(5.2) local dynamic station monitoring and statistics amount is established
Wherein,
Indicate b-th of set of variables, p indicates time lag, Δ xp+1,bIndicate the Difference Terms, that is, Δ x at p+1 momentp+1,b=xp+1,b-xp,b,
xp,bFor XbThe transposition of middle row vector indicates the sampled point at the p moment, ΘbAnd Φb(b=1 ..., Z) is parameter to be estimated, can
It is acquired by least square.Be,bAnd Bf,bFor EbAnd FbCalculate the weight matrix obtained after canonical correlation analysis.Dynamic monitoring statistics
Amount is
Wherein, te,bFrom matrix Te,b, it is Te,bMiddle column vector.
(6) global monitoring and statistics amount is established
(6.1) global static monitoring techniques statistic is established:
Wherein,PS(xb)=P (xb|Ns)P(Ns)+P(xb|Fs)P(Fs);P(Ns) be
Confidence interval, i.e. P (Ns)=α, P (Fs)=1- α; For sample xb'sMonitoring and statistics amount,For monitoring and statistics amountControl limit;
(6.2) global dynamic monitoring statistic is established:
Wherein,Pe(xb)=P (xb|Ne)P(Ne)+P(xb|Fe)P(Fe);P(Ne) be
Confidence interval, i.e. P (Ne)=α, P (Fe)=1- α; For sample xb'sMonitoring and statistics amount,For monitoring and statistics amountControl limit;
Step 7: online process monitoring
(7.1) to new collected non-stationary variable sample xnew,t(N × 1) carries out variable grouping, which is divided into Z
Set of variables, that is, xnew,b,t, b=1,2 ..., 5.
(7.2) method recorded according to step 5 calculates local static monitoring and statistics amount
(7.3) local dynamic station monitoring and statistics amount is calculated
(7.4) global static, dynamic monitoring statistic is calculated
IfWithMean that static, dynamic long-run equilibrium relationship is broken beyond control limit, at this time process
It breaks down;IfWhile alarm,Beyond normal range (NR) is returned to after control limit, mean at this time
The operating point of process changes;IfWithIt is all limited without departing from control, means that process may work at this time
Static state, dynamic long-run equilibrium relationship under new operating condition, but between variable are not broken or the operating condition is included in modeling data
In.
If local static, dynamic monitoring statistic are influenced by failure, BICsOr BICeAlso it beyond control limit, anticipates at this time
Taste the failure not only influence the local state of process, while global state is also affected, if BICsOr BICeDo not exceed
Control limit indicates the local state of the failure influence process.
Table 1 is set of variables dividing condition, and non-stationary variable is divided into 5 different set of variables.It can be seen from Fig. 2
The load of gigawatt extra-supercritical unit often changes, it means that the operating condition of unit does not stop to change.Fig. 3 is the prison of this method
Survey result, it can be seen that the static monitoring techniques statistic of set of variables 1-4 is beyond control limit after operating point changes, it means that
Static long-run equilibrium relationship is broken.However dynamic monitoring statistic connects only after operating point changes beyond control limit
Get off to return to normal range (NR).The static state and dynamic monitoring statistic of set of variables 5 are all limited without departing from control, this indicates the change of work
Change does not influence the variable in set of variables 5.Fig. 4 is global monitoring result, it can be seen that global monitoring result and area monitoring
As a result consistent.Fig. 5 illustrates the monitoring result of traditional cointegrating analysis, it can be seen that monitoring and statistics amount after operating point changes
It is constantly in alarm condition, this can impact field engineer, so that engineer is mistakenly considered process and be in abnormality.Pass through
This method can help field engineer effectively to distinguish the influence of the change and failure of process operating point to process, ensure that hundred
The efficient operation of ten thousand kilowatts of extra-supercritical units.
1. set of variables division result of table
Set of variables | Variable |
1 | 4,7,15,16,20,22,26,27,29,30,33,34,37,39,43,47,49 |
2 | 5,6,13,18,24,31,35,40,41,42,45,50,51 |
3 | 28,44,46,48 |
4 | 8,9,11,12,17,19,21,23,25,36,38, |
5 | 1,2,3,10,14,32, |
Claims (1)
1. a kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation, which is characterized in that
The following steps are included:
(1) normal processes data are obtained: setting in a gigawatt extra-supercritical unit comprising J process variable, samples every time
The vector of an available 1 × J obtains the two-dimensional matrix X under a normal processes after sampling M timesn(M×J);
(2) it identifies non-stationary variable: identifying two-dimensional matrix X using Augmented Dickey-Fuller (ADF) method of inspectionn(M
× J) in non-stationary variable, obtain non-stationary variable data matrix Xns(M × N), wherein N is the number of non-stationary variable;
(3) non-stationary variable data matrix X obtained in (2) is utilizedns(M × N)=[x1,x2,…,xN], xt=(x1,x2,…,
xN)TEstablish sparse co-integration model;It establishes sparse co-integration model and specifically includes following sub-step:
(3.1) to xtEstablish Vector Autoression Models:
xt=Π1xt-1+…+Πpxt-p+c+μt (1)
Wherein, Π1~ΠpIt is the coefficient matrix of (N × N), μtFor (N × 1) matrix, white Gaussian noise, μ are indicatedt~N (0, Ξ),
C is (N × 1) matrix, indicates constant, xtIndicate data matrix XnsThe transposition of a row vector in (M × N) is indicated in t moment
Sampled value, t=1,2 ..., M, p is model order;
(3.2) x is subtracted at formula (1) both endst-1Obtain error correction model:
(3.3) Γ in step (3.2) is decomposed into matrix Γ=Α Β of two sequency spectrumsT, formula (2) becomes:
Wherein, Α (N × R), Β (N × R);
(3.4) the whole vector matrix Β of association in formula (3) is estimated by Maximum Likelihood Estimation:
Wherein, L (*) indicates maximum likelihood function, the mark of tr (*) representing matrix;X=(Δ xp+1,...,ΔxM)T, Y=(Δ
Yp+1,...,ΔYM)T,Ω=(Ω1,...,Ωp-1)T, Z=
(xt-1,...,xM-1)T, Θ=Ξ-1;
(3.5) characteristic equation solution procedure can be converted into the Maximum-likelihood estimation of formula (4):
Wherein,
Wherein parameter matrix ΘiAnd Φi, i=1,2 ..., p-1 can be acquired by least-squares algorithm;
(3.6) penalty is added to the objective function of formula (4) and obtains the sparse whole vector of association:
Wherein, P1, P2, P3For parameter Β, the penalty of Ω, Θ, using a normal form;Wherein Β is to assist whole vector matrix,
Ω, Θ are parameters to be estimated;Adjusting parameter λ1And λ2It is determined using crosscheck, adjusting parameter λ3Using Bayesian Information
Criterion determines;Pass through the available whole vector of sparse association of solution to formula (6);
(4) variable can be divided using the whole vector of sparse association obtained in formula (6), specifically includes following sub-step:
(4.1) the sparse whole vector matrix Β of association is obtained according to formula (6)s(N × K)=[βs,1,βs,2,...,βs,K], wherein K is indicated
The number of the sparse whole vector of association;
(4.2) steady residual sequence is obtained according to the sparse whole vector of associationWherein k=1,2 ..., K, t=1,2 ...,
M;The consistent level for measuring residual sequence, and record residual difference sequence γ are examined using ADFk,tADF test statistics tk;
(4.3) to obtained test statistics tkCarry out ascending sort, the corresponding sparse whole vector of association of the smallest inspection statistics magnitude
It is retained;The corresponding variable of nonzero element in the sparse whole vector of association is assigned in subgroup, and is denoted as Xb;
(4.4) by XbIn variable from initial data concentrate remove, raw data set is denoted as X at this timeL;
(4.5) iteration step (4.1)-(4.4) are all assigned in different subgroups until all variables, at this time original variable
It is divided into Z different set of variables;
(5) local static, dynamic monitoring statistic are established in each set of variables:
γs,b,tFor the steady residual sequence in each set of variables, Bf,bFor the whole vector matrix of association calculated in each set of variables,
Λs,b=(XbBf,b)T(XbBf,b)/(M-1) (b=1 ..., Z) it is covariance matrix;Z is the number of variable grouping, is indicated original
Variable is divided into Z set of variables, xb,tFor XbThe transposition of middle row vector indicates the sampled point in t moment;
(5.2) local dynamic station monitoring and statistics amount is established:
Wherein,
B indicates b-th of set of variables, and p indicates time lag, Δ xp+1,bIndicate the Difference Terms, that is, Δ x at p+1 momentp+1,b=
xp+1,b-xp,b, xp,bFor XbThe transposition of middle row vector indicates the sampled point at the p moment, ΘbAnd Φb(b=1 ..., Z) it is wait estimate
Parameter is counted, can be acquired by least square;Be,bAnd Bf,bFor EbAnd FbCalculate the whole moment of a vector of association obtained after canonical correlation analysis
Battle array;Dynamic monitoring statistic is
Wherein, te,bFrom matrix Te,b, it is Te,bMiddle column vector;
(6) global monitoring and statistics amount is established:
(6.1) global static monitoring techniques statistic is established:
Wherein, Z indicates the number of set of variables;PS(xb)=P (xb|
Ns)P(Ns)+P(xb|Fs)P(Fs);P(Ns) it is confidence interval, i.e. P (Ns)=α, P (Fs)=1- α; For sample xb'sMonitoring and statistics
Amount,For monitoring and statistics amountControl limit;
(6.2) global dynamic monitoring statistic is established:
Wherein, Z indicates the number of set of variables;Pe(xb)=P (xb|Ne)P
(Ne)+P(xb|Fe)P(Fe);P(Ne) it is confidence interval, i.e. P (Ne)=α, P (Fe)=1- α; For sample xb'sMonitoring and statistics
Amount,For monitoring and statistics amountControl limit;
(7) online process monitoring:
(7.1) to new collected non-stationary variable sample xnew,t(N × 1) carries out variable grouping, which is divided into Z variable
Group is xnew,b,t, b=1,2 ..., Z;
(7.2) method recorded according to step 5 calculates local static monitoring and statistics amount
(7.3) local dynamic station monitoring and statistics amount is calculated:
(7.4) global static, dynamic monitoring statistic is calculated:
IfWithIt is limited beyond control, it is meant that static, dynamic long-run equilibrium relationship is broken, and event occurs for process
Barrier;IfWhile alarm,Beyond normal range (NR) is returned to after control limit, mean process at this time
Operating point changes;IfWithIt is all limited without departing from control, means process work in new operating condition at this time
Under, but the static state, dynamic long-run equilibrium relationship between variable are not broken or the operating condition is included in modeling data;
If local static, dynamic monitoring statistic are influenced by failure, BICsOr BICeAlso beyond control limit, mean at this time
The failure not only influences the local state of process, while also affecting global state, if BICsOr BICeNot beyond control
Limit, indicates the local state of the failure influence process.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810141761.7A CN108490908B (en) | 2018-02-11 | 2018-02-11 | A kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810141761.7A CN108490908B (en) | 2018-02-11 | 2018-02-11 | A kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108490908A CN108490908A (en) | 2018-09-04 |
CN108490908B true CN108490908B (en) | 2019-10-15 |
Family
ID=63340143
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810141761.7A Active CN108490908B (en) | 2018-02-11 | 2018-02-11 | A kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108490908B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109188905B (en) * | 2018-09-07 | 2019-10-15 | 浙江大学 | A kind of on-line monitoring method of the sound feature Cooperative Analysis towards gigawatt extra-supercritical unit |
CN109491358B (en) * | 2018-09-21 | 2020-08-14 | 浙江大学 | Control performance monitoring method for boiler dynamic information of million-kilowatt ultra-supercritical unit |
CN111880090B (en) * | 2019-06-28 | 2021-07-06 | 浙江大学 | Online fault detection method for million-kilowatt ultra-supercritical unit |
CN112598030B (en) * | 2020-12-08 | 2023-03-24 | 山东科技大学 | Non-stationary process monitoring method based on recursive covariance analysis and elastic weight consolidation |
CN113050606A (en) * | 2021-03-29 | 2021-06-29 | 中国人民解放军火箭军工程大学 | Quality-related fault detection method for complex non-stationary process |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104794457A (en) * | 2015-05-05 | 2015-07-22 | 哈尔滨工程大学 | Hyperspectral image target detection method based on sparse error matrix |
CN107544477A (en) * | 2017-10-23 | 2018-01-05 | 中国石油大学(华东) | Nonlinear industrial processes fault detection method based on core pivot element analysis |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106680012B (en) * | 2017-01-25 | 2019-01-08 | 浙江大学 | A kind of fault detection method and diagnostic method towards large-scale Thermal generation unit non-stationary process |
-
2018
- 2018-02-11 CN CN201810141761.7A patent/CN108490908B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104794457A (en) * | 2015-05-05 | 2015-07-22 | 哈尔滨工程大学 | Hyperspectral image target detection method based on sparse error matrix |
CN107544477A (en) * | 2017-10-23 | 2018-01-05 | 中国石油大学(华东) | Nonlinear industrial processes fault detection method based on core pivot element analysis |
Also Published As
Publication number | Publication date |
---|---|
CN108490908A (en) | 2018-09-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108490908B (en) | A kind of dynamic distributed monitoring method towards gigawatt extra-supercritical unit variable parameter operation | |
Chen et al. | Fault detection for non-Gaussian processes using generalized canonical correlation analysis and randomized algorithms | |
CN111582392B (en) | Multi-working-condition health state online monitoring method for key components of wind turbine generator | |
CN106680012B (en) | A kind of fault detection method and diagnostic method towards large-scale Thermal generation unit non-stationary process | |
CN109211548A (en) | A kind of mechanical failure diagnostic method | |
CN105469138A (en) | Control system performer fault diagnosis method based on particle swarms and support vector machine | |
CN112257530B (en) | Rolling bearing fault diagnosis method based on blind signal separation and support vector machine | |
CN105004498A (en) | Vibration fault diagnosis method of hydroelectric generating set | |
CN109471420B (en) | CVA-SFA-based method for monitoring control performance of air preheater of large coal-fired power generator set of intelligent power plant | |
CN106483847B (en) | A kind of water cooler fault detection method based on adaptive ICA | |
CN111881159A (en) | Fault detection method and device based on cost-sensitive extreme random forest | |
CN113298133B (en) | Method for diagnosing explosion tube fault of supercritical unit boiler | |
Wang et al. | Research on anomaly detection and positioning of marine nuclear power steam turbine unit based on isolated forest | |
CN113640607B (en) | Early fault diagnosis method for inverter circuit and motor of high-speed train | |
Xin et al. | Fault diagnosis of nuclear power plant based on simplified signed directed graph with principal component analysis and support vector machine | |
CN110441081B (en) | Intelligent diagnosis method and intelligent diagnosis system for faults of rotating machinery | |
CN113886765A (en) | Method and device for detecting error data injection attack | |
Liu et al. | A novel process monitoring method based on dynamic related ReliefF-SFA method | |
CN113885371B (en) | Mixed variable process monitoring method based on health state data | |
Niu et al. | A reformative PCA-based fault detection method suitable for power plant process | |
Taoslavi et al. | Clustering approach for change point estimation in multivariate normal processes | |
CN108508866A (en) | A kind of gigawatt extra-supercritical unit failure identification variables method based on sparse opposite discriminant analysis | |
Marseguerra et al. | A soft-computing based classification procedure for the identification of transients in the steam generator of a pressurized water reactor | |
Eklund | Using synthetic data to train an accurate real-world fault detection system | |
Zhang et al. | Filter-Based Fault Diagnosis of Heat Exchangers |
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 |