CN106647274B - Operating condition stable state method of discrimination in a kind of continuous flow procedure - Google Patents
Operating condition stable state method of discrimination in a kind of continuous flow procedure Download PDFInfo
- Publication number
- CN106647274B CN106647274B CN201611240042.8A CN201611240042A CN106647274B CN 106647274 B CN106647274 B CN 106647274B CN 201611240042 A CN201611240042 A CN 201611240042A CN 106647274 B CN106647274 B CN 106647274B
- Authority
- CN
- China
- Prior art keywords
- stable state
- principal component
- window
- value
- mrow
- 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
-
- 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
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Complex Calculations (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides operating condition stable state method of discrimination in a kind of continuous flow procedure, including：S1 is based on operating parameter in production process, and principal component and corresponding characteristic value are obtained using Principal Component Analysis；S2 is based in the principal component continuous stable state segment in first principal component, determines polynomial filtering window；S3 is based on each described window, and stable state detection is carried out to each principal component in the principal component using polynomial filtering stable state diagnostic method；S4 is based on the corresponding characteristic value, assigns weights to the stable state testing result of each principal component, determines the window stable state testing result；According to the stable state testing result of the window, obtain operating condition stable state in production process and differentiate result.Production process adjustment more frequently, more than monitored parameters and when fluctuation is larger, method of the invention can be effectively prevented from the influence that single or several variable errors differentiate stable state, enhance adaptability in industrial processes by extracting the integrated informations of data.
Description
Technical field
The present invention relates to production process stable state detection field, more particularly, to running work in a kind of continuous flow procedure
Condition stable state method of discrimination.
Background technology
At present, with the development of automatic technology and monitoring technology, substantial amounts of equipment working condition and procedure parameter are accumulated,
These data messages of reasonable utilization, to ensureing that safety in production, Improving The Quality of Products are of great significance.Inside continuous flow procedure
Principle and complicated shows as between physical quantity there is stronger coupling, while shows extremely strong non-linear and time-varying
Property.It conducts a research for continuous flow procedure, stable state is most important and most common it is assumed that whether continuous flow procedure is in steady
State, modeling, control and optimization method are different used by being directly related to subsequently to flow.When flow is in unstable state, it is
The data characteristic of each variable of uniting changes violent, and numerically there are relatively large deviations with the input/output relation of real system.Only flow
Journey is under steady working condition, and parameters and variable just have stronger state consistency.Based on such situation, to continuous production
Evaluation, simulation and the optimization of process operation performance are required for premised on the operation stable state of acquisition continuous flow procedure.
For the research of stable state detection, mainly have at present based on Analysis on Mechanism, based on statistical theory and based on trend abstraction
Three classes steady state detecting method for use.Mechanism analysis method generally analyzes detection Properties of Objects in the case where idealizing supposed premise, and
Actual industrial process be difficult to meet this it is assumed that and object dependency it is stronger, without general applicability.Based on statistical theory
Method in, as CST methods and MTE methods assume measured value containing only random error, and error obedience is desired for 0 normal distribution,
The hypothesis is difficult to reach in practical applications, reduces the steady degree of method, once larger interference occur may result in erroneous judgement
It is disconnected.In method based on trend abstraction, such as wavelet analysis method, detection threshold value need to rely on the steady state data of history to be calculated, must
There must be reliable stable state historical data to rely on；Neutral net trend abstraction method, the precision of testing result is to sliding window value ratio
More sensitive, the value of detection process important parameter lacks effective theoretical foundation, causes the validity and reliability of method significantly
It reduces；Adaptive Polynomial filter method, less using parameter, filter window adaptive can should determine that, violent suitable for processing variation
Signal, there is stronger anti-noise ability and preferable robustness, but single argument stable state detection can only be carried out, and testing result
Easily influenced by error of measured data.As a result, in a kind of differentiation sides of atmospheric and vacuum distillation unit steady state condition of patent CN105389648A
In method, according to the standard deviation of the Monomial coefficient of data fit equation and fitness bias in current window, judge whether stable state；
In a kind of steady state condition method of discrimination optimized towards Atmospheric vacuum of patent CN104977847A, with the average in adjacent a period of time
With once fitting coefficient as stable state criterion.Both approaches are required for respectively detecting all variables, are more suitable for list
Variable or the relatively small number of process of variable, it is impossible to well solve in industrial process that monitored parameters are numerous and coupling is strong asks
Topic, while only all variable stable states just think Steady-state process, this easily causes mistake when technical process measurement data fluctuations are larger
Sentence.
The content of the invention
In order to solve since continuous flow procedure monitored parameters are numerous and there is stronger coupling, measurement data fluctuation compared with
Greatly, while the error of single or several variables easily causes testing result very big influence, in a conventional method to continuous production mistake
Journey industrial data is difficult to realize the problem of good operating condition stable state detection, and the present invention provides in a kind of continuous flow procedure
Operating condition stable state method of discrimination.
Operating condition stable state method of discrimination in continuous flow procedure provided by the invention, including：
S1. based on operating parameter in production process, principal component and corresponding characteristic value are obtained using Principal Component Analysis；
S2. based on continuous stable state segment in first principal component in the principal component, polynomial filtering window is determined；
S3. based on each described window, using polynomial filtering stable state diagnostic method in the principal component it is each it is main into
Divide and carry out stable state detection；
S4. based on the corresponding characteristic value, weights is assigned to the stable state testing result of each principal component, determine institute
State window stable state testing result；According to the stable state testing result of the window, obtain operating condition stable state in production process and differentiate
As a result.
Operating condition stable state method of discrimination in continuous flow procedure proposed by the present invention is steady by univariate polynomial filtering
State detection expands to multivariable, and comprehensive multivariable information reduces the influence of single or several variable errors, while using smoothly
Treatment technology and improved stabilizing determination method reduce the influence of error in measurement data；Stable state difference is segmented by the present invention
It is completed in same window, improves the precision of stable state difference；In the industrial data verification of hydrocracking process, it was demonstrated that should
Method has certain engineering practical value.
Description of the drawings
Fig. 1 is the overall procedure schematic diagram of operating condition stable state method of discrimination in the continuous flow procedure according to the present invention；
Fig. 2 is the stream according to operating condition stable state method of discrimination in continuous flow procedure in a preferred embodiment of the invention
Journey schematic diagram；
Fig. 3 is according to standard deviation ratio before and after first principal component filtering during definite filter window in the embodiment of the present invention 1
It is worth variation diagram；
Fig. 4 is according to the comparison diagram before and after first principal component smoothing processing in the embodiment of the present invention 1；
Fig. 5 be according to be hydrocracked in the embodiment of the present invention 1 flow stable state testing result figure and monitoring Partial Variable figure.
Specific embodiment
With reference to the accompanying drawings and examples, the specific embodiment of the present invention is described in further detail.Implement below
Example is not limited to the scope of the present invention for illustrating the present invention.
Operating condition stable state method of discrimination in continuous flow procedure provided by the invention, as shown in Figure 1, including：
S1. based on operating parameter in production process, principal component and corresponding characteristic value are obtained using Principal Component Analysis；
S2. based on continuous stable state segment in first principal component in the principal component, polynomial filtering window is determined；
S3. based on each described window, using polynomial filtering stable state diagnostic method in the principal component it is each it is main into
Divide and carry out stable state detection；
S4. based on the corresponding characteristic value, weights is assigned to the stable state testing result of each principal component, determine institute
State window stable state testing result；According to the stable state testing result of the window, operating condition stable state during final production is obtained
Differentiate result.
The present invention using establishing principal component model to operating parameter initial data in production process, obtain principal component and
Corresponding characteristic value；The variation of standard deviation before and after foundation filtering determines the window size of polynomial filtering；In observation window more
It is preferred that each principal component smoothing processing, improved polynomial filtering stable state is recycled to differentiate, stable state inspection is carried out to each principal component
It surveys, different weights is assigned to each principal component stable state testing result according to the characteristic value of principal component, the window is determined so as to comprehensive
The stable state testing result of mouth；Then according to the stable state testing result of each window, it is steady to obtain final production process operating condition
State differentiates result.The method of the present invention is more frequent in production process adjustment, when monitored parameters are more and fluctuation is larger, passes through extraction
The integrated information of data can be effectively prevented from the influence that single or several variable errors differentiate stable state, enhance in industry
Adaptability in the process.
In the method for the invention, when Principal Component Analysis common in this field is used to divide operational parameter data
Analysis obtain principal component and corresponding characteristic value after, can by manually choosing continuous stable state segment in first principal component, according to
Typical ratio changes before and after glide filter, so that it is determined that the specific steps of polynomial filtering window size, i.e. S2.In this field
In, using Principal Component Analysis to transport supplemental characteristic analyzed to obtain the m principal component comprising former data Pe% information with
Corresponding characteristic value.Wherein, Pe% values are usually 80%-95%, which represents：Utilize the industrial number of screening during Principal Component Analysis
According to when, be typically chosen the principal component matrix of the information comprising former data 80%-95%.
In a preferred embodiment, S2 is specially：
Manually choose continuous stable state segment Q in first principal componenttrain=[q1(1),q1(2),...,q1(nt),...,q1
(Nt)]T,Nt>=500,
Wherein, q1(nt) represent n-th in selected stable state segmenttA value, T are transposition symbol, calculate training data Qtrain
Standard deviation δt；
Equipped with filter window Hh=2h (h=1,2 .., nh), h is to filter half-window, nhTo be less than or equal to Nt/ 2 maximum
Integer, given threshold value αt∈ (0,1) is a smaller value, initializes half-window h=0, proceeds as follows：
S21.h=h+1 uses HhTo QtrainGlide filter is carried out, obtains filtered signal sequenceWherein,It represents n-th in filtered signal sequencetA value, T
For transposition symbol, calculateStandard deviationWith the standard deviation δ of training datatRatio
S22. if h≤nh, and vh-1-vh≥αt, then S21 is returned；Otherwise current h is required filtering half-window, that is, is filtered
Ripple window size Ht=2h.
In the present invention, in order to reduce the influence of error in measurement data, antialiasing and improved can be used
Stabilizing determination method, i.e. S3 can be：
Based on window each described, after each principal component carries out smooth place in the principal component, multinomial is recycled
It filters stable state diagnostic method and stable state detection is carried out to each principal component in the principal component.Wherein polynomial filtering stable state diagnostic method is
The polynomial filtering stable state diagnostic method after stable state criterion is improved, criterion is：The absolute value of Monomial coefficient is given respectively
Threshold value, the absolute value of Monomial coefficient and the poor threshold value both less than each set are set with standard deviation, are considered as just stable state.
In the present invention, the specific steps of S3 are more preferably：
S31. according to the size of each window, the principal component is split along sampling time direction；
S32. in each described window, each principal component in the principal component is smoothed, and keeps institute
The trend for stating principal component variation does not change；
S33. based on each principal component, the secondary multinomial table of each principal component is obtained using polynomial filtering
Up to formula, the Monomial coefficient absolute value of the secondary multi-term expression of each principal component and the mark of each principal component are utilized
It is accurate poor, stable state detection is carried out in the window to each principal component.
In a preferred embodiment, in order to be further reduced error, smoothing processing includes in S32：
Assuming that j-th of principal component of s-th of window to be processedWherein, qj(Ht(s-1)+st)
Represent the H of j-th of principal componentt(s-1)+stA value, T are transposition symbol, are calculatedAverage valueAnd standard deviationJust
Beginningization iterations a=0, proceeds as follows：
S321. countInherent domainNumber
If S322.A=a+1 returns to S321；Otherwise willInside it is less thanTax
It is worth and isIt willInside it is more thanBe assigned a value of
In order to make data more accurate, the improved stabilizing determination method more preferably used in the present invention is i.e. S33 tools
Body is preferably：
Assuming thatRepresent smooth
J-th of principal component of treated s-th of window, is denoted asStandard deviation is denoted asFiltered signal x (i) is expressed as the function of time, i.e.,：
X (i)=k0+k1i+k2i2+...,+kcic (1)
Wherein,Represent the H of j-th of principal component after smoothing processingt(s-1)+stA value, T are
Transposition symbol,Represent the h of j-th of principal component in s-th of window after smoothing processingtA value, c are model order；
Make θ=[k0,k1,..,kc]TFor model parameter vector, r (i)=[i0,i,..,ic]TFor regression variable, formula (1) letter
It is denoted as：
X (i)=θTr(i) (2)
The optimal estimation of parameter θ can be obtained using least square method：
Wherein,
C=2 is taken, k can be obtained according to formula (3)1Value, k1The speed of variation is represented, according to principal component in Analysis on Mechanism and window
Variation, sets suitable threshold valueWithIf represent j-th of principal component judge in the stable state of s-th window as a result,AndThen judge the principal component in the window for stable state,1 is denoted as, is not otherwise stable state,It is denoted as 0.
In the method for the invention, weights are assigned there are many method to the stable state testing result of each principal component, it is more excellent
Selection of land is：
The corresponding characteristic value of m principal component of selection is λ1,λ2,...,λm, then result is judged to the stable state of j principal component
Assigning weights is：
Stable state in calculation window judges result weighted sum：
Judge whether the production process is stable state according to the value of weighted sum.IfThen think continuous production mistake
Journey is stable state in the window, is not otherwise stable state.Wherein, pe% refers to flow and reaches minimum steady state data information needed for stable state
The percentage of total data information is accounted for, that is, thinks that when the former data message of pe% be stable state, then then assert at whole process
In stable state, Pe% values are included the percent value of former data by principal component, i.e., the data that the principal component of described selection is characterized are believed
Breath accounts for the percentage of total data information., i.e., described principal component includes former data Pe% information, Pe% values usually 80%-
95%, it is determined according to practical operation data；Wherein, usual pe% selects 50%-Pe%.
When pe% selects 50%, evenIt is stable state in the window then to think continuous flow procedure, otherwise not
It is stable state.Wherein, Pe% values are included the percent value of former data by principal component, i.e., described principal component includes former data Pe% letters
Breath, Pe% values are usually 80%-95%, are determined according to practical operation data.
It influences to eliminate initial data dimension, each operating parameter would generally be carried out before Principal Component Analysis is used
Pretreatment, it is more common to be：
Assuming that continuous flow procedure operational parameter data is：
(j=1,2 ..., the operating parameter that J is monitored by continuous flow procedure
Number), whereinFor the n-th of j-th of variable
A sampled value)；
Following standardization is carried out to each operating parameter, eliminates dimension impact：
Wherein
Wherein
Pretreated data X using principal component analysis is handled, obtains principal component and its corresponding characteristic value.
In the present invention, principal component and its corresponding characteristic value are obtained using principal component analysis technology (PCA), PCA is actual
It is a kind of multivariate statistical method that multiple correlated variables of research object are turned to a few incoherent variable, and to the greatest extent may be used
The information of former data can be retained.
Assuming that pretreated data be a sample × quantitative variable tables of data X (N × J) (wherein, N be data sample
This number, J are the number of process variable), i.e., each corresponding observational variable of row corresponds to a sample per a line.Matrix X
(N × J) can be decomposed into：
Wherein, pj∈RJFor load vector；qj∈RNFor score vector, that is, the principal component to be extracted, correspondingly, P=
[p1,p2,...,pJ] for matrix of loadings, Q=[q1,q2,...,qJ] it is score matrix, represent projections of the X in load direction；
Divide vector sum load vector all mutually orthogonal, and load vector is unit vector.Then p is multiplied by the right side in (8) formula both sides simultaneouslyjIt can turn
It turns to：
qj=XpjOr Q=XP (9)
According to mathematical conclusions, to the variance matrix of matrix XDo singular value decomposition, can obtain matrix of loadings P and
Eigenvalue matrix D.I.e.：
S=PDPT (10)
Wherein,The descending arrangement of element on its diagonal, λj(j=1,2 ..., J)
As characteristic value.
The selection of characteristic value and principal component：MeetMinimum m values for selected characteristic value number, that is, choose
Preceding m eigenvalue λj(j=1,2 .., m), then corresponding matrix of loadings PmIt is arranged for the preceding m of matrix of loadings P；It can using formula (9)
In the hope of corresponding score matrix Qm=XPm, i.e., selected principal component matrix is Q after dimensionality reductionm。
The present invention is described in further detail by taking the continuous flow procedure being hydrocracked as an example.
Embodiment 1
The embodiment provide continuous flow procedure in operating condition stable state method of discrimination, using be hydrocracked flow as pair
As, for being hydrocracked more than flow monitoring parameters and the problem of coupling is larger, principal component being extracted using PCA, realize dimensionality reduction and
The effect of uncoupling realizes that stable state detects to principal component after smoothing processing using polynomial filtering, passes through the characteristic value of principal component
Certain weights are assigned to single principal component stable state testing result, finally, are hydrocracked whether flow is located according to weighted sum judgement
In stable state.Detailed process is as follows：
The monitored parameters that are hydrocracked of this example selection have general export feedstock oil flow-rate adjustment, 1 bed of hydrofining reactor
Temperature (on), 2 bed temperature of hydrofining reactor (on), 3 bed temperature of hydrofining reactor (on), hydrocracking reaction
1 bed temperature of device (on), 2 bed temperature of hydrocracking reactor (on), boat coal goes out device flow, diesel oil goes out device flow, tail
Oil goes out 22 primary variables such as device flow.Sampling interval is 5 minutes, the whole month data in 2 months in 2016；2 months in 2016
It is that 2,400,000 tons of devices of certain factory just come into operation, for parameter by repeatedly adjusting, data have diversity.
Referring to Fig. 2, Fig. 2 is operating condition stable state method of discrimination in the continuous flow procedure proposed in the present invention.
Step 1：To being hydrocracked the pretreatment of process flow operation supplemental characteristic, it is assumed that be hydrocracked process flow operation supplemental characteristic
For：
(j=1,2 ..., J；J=22 is to be hydrocracked the operation ginseng that flow is monitored
Several numbers), wherein For j-th variable
N-th of sampled value)；
First, following standardization is carried out to each operating parameter, eliminates dimension impact：
Wherein
Wherein
Obtain pretreated data matrix X.
Step 2：Pretreated data X using principal component analysis (PCA) is handled, obtains principal component and its corresponding
Characteristic value.
First, the variance matrix of data matrix X is asked forTo variance matrix S singular value decompositions, load is asked for
Matrix P and eigenvalue matrix D.
Secondly, Pe%=95% is made, metMinimum value be that selected characteristic value number m is 5, it is corresponding
Characteristic value is as shown in table 1.
Table 1 chooses the characteristic value corresponding to principal component
Finally, the preceding 5 row P of matrix of loadings P is takenm, then the principal component matrix chosen is Qm=XPm。
Step 3：Manually choose continuous stable state segment (i.e. training data) Q in first principal componenttrain=[q1(1),q1
(2),...,q1(Nt)]T(Nt≥500).Calculate the standard deviation δ of training datat。
Equipped with filter window Hh=2h (h=1,2 .., nh), h is to filter half-window, nhTo be less than or equal to Nt/ 2 maximum is whole
Number, given threshold value αt∈ (0,1) is a smaller value.Half-window h=0 is initialized, is proceeded as follows：
1) h=h+1 uses HhTo QtrainGlide filter is carried out, obtains filtered signal sequenceIt calculatesStandard deviationWith the standard deviation δ of training datatRatio
If 2) h≤nt, and vh-1-vh≥αt, then return 1)；Otherwise current h is required filtering half-window, i.e. spectral window
Mouth size Ht=2h.
Take threshold alphat=5.5 × 10-5, standard deviation ratio before and after corresponding glide filter, trend chart as shown in figure 3,
Filter window can be obtained as Ht=250.
Step 4：First according to the size of window, the principal component matrix of selection is split along sampling time direction.
In each observation window, each principal component of selection is smoothed, i.e.,：
Assuming that principal component to be processedRepresent s
J-th of principal component of a window calculatesAverage valueAnd standard deviationIterations a=0 is initialized, is grasped as follows
Make：
1) countInherent domainNumber
If 2)1) a=a+1 is returned；Otherwise willInside it is less thanBe assigned a value ofIt willInside it is more thanBe assigned a value of
Data after smoothing processing can eliminate the random error in initial data measurement, and keep becoming for principal component variation
Gesture does not change, if Fig. 4 is comparison diagram before and after first principal component smoothing processing.
Principal component after the smoothing processing of selection calculates standard deviation noteAnd calculating matrix value (RTR)-1RT。
WhereinThe Monomial coefficient absolute value of the quadratic polynomial wave filter of corresponding principal component
It is as shown in table 2 with the standard deviation threshold method selection of principal component.
The standard deviation threshold method of 2 Monomial coefficient absolute value of table and principal component
Using polynomial filtering, the quadratic polynomial wave filter of principal component is obtained, it is assumed that use
It represents to need to carry out the principal component after the smoothing processing of stable state differentiation using polynomial filtering, thenSecond
Item is k1Value.If represent j-th of principal component judge in the stable state of s-th window as a result,AndThen judge
The principal component is stable state in the window,1 is denoted as, is not otherwise stable state,It is denoted as 0.
Step 5：The different power shown in table 3 are assigned to principal component stable state testing result according to the corresponding characteristic value of principal component
Value.
3 principal component stable state weights of table
Stable state in calculation window judges result weighted sum：
Pe%=50% is taken, ifThen think that it is stable state to be hydrocracked flow in the window, otherwise
It is not stable state.
With reference to the steady result of all observation windows, so as to fulfill the differentiation of flow stable state is hydrocracked.Stable state differentiates knot
Fruit is as shown in Figure 5.
Finally, the present processes are only preferable embodiment, are not intended to limit the scope of the present invention.It is all
Within the spirit and principles in the present invention, any modifications, equivalent replacements and improvements are made should be included in the protection of the present invention
Within the scope of.
Claims (9)
1. a kind of operating condition stable state method of discrimination in continuous flow procedure, which is characterized in that including：
S1. based on operating parameter in production process, principal component and corresponding characteristic value are obtained using Principal Component Analysis；
S2. based on continuous stable state segment in first principal component in the principal component, polynomial filtering window is determined；
S3. based on each described window, using polynomial filtering stable state diagnostic method to each principal component in the principal component into
Row stable state detects；
S4. based on the corresponding characteristic value, weights is assigned to the stable state testing result of each principal component, determine the window
Mouth stable state testing result；According to the stable state testing result of the window, obtain operating condition stable state in production process and differentiate result；
Wherein, S3 is specially：
S31. according to the size of each window, the principal component is split along sampling time direction；
S32. in each described window, each principal component in the principal component is smoothed, and keeps the master
The trend of composition transfer does not change；
S33. based on each principal component, the secondary multinomial expression of each principal component is obtained using polynomial filtering
Formula utilizes the Monomial coefficient absolute value of the secondary multi-term expression of each principal component and the standard of each principal component
Difference carries out stable state detection to each principal component in the window.
2. stable state method of discrimination according to claim 1, which is characterized in that S2 is specially：
Continuous stable state segment in first principal component is manually chosen, is changed according to standard deviation ratio before and after glide filter, determined more
Item formula filter window size.
3. stable state method of discrimination according to claim 2, which is characterized in that S2 is specially：
Manually choose continuous stable state segment Q in first principal componenttrain=[q1(1),q1(2),...,q1(nt),...,q1(Nt)
]T,Nt>=500,
Wherein, q1(nt) represent n-th in selected stable state segmenttA value, T are transposition symbol, calculate training data QtrainMark
Quasi- difference δt；
Equipped with filter window Hh=2h, h=1,2 .., nh, h is to filter half-window, nhTo be less than or equal to Nt/ 2 maximum integer, gives
Determine threshold alphat∈ (0,1) initializes half-window h=0, proceeds as follows：
S21.h=h+1 uses HhTo QtrainGlide filter is carried out, obtains filtered signal sequenceWherein,It represents n-th in filtered signal sequencetA value, T
For transposition symbol, calculateStandard deviationWith the standard deviation δ of training datatRatio
S22. if h≤nt, vh-1-vh≥αt, then S21 is returned；Otherwise current h is required filtering half-window, i.e. filter window
Size Ht=2h.
4. stable state method of discrimination according to claim 1 or 2, which is characterized in that S3 is specially：
Based on window each described, after each principal component carries out smooth place in the principal component, polynomial filtering is recycled
Stable state diagnostic method carries out stable state detection to each principal component in the principal component.
5. stable state method of discrimination according to claim 3, which is characterized in that smoothing processing includes in S32：
Assuming that j-th of principal component of s-th of window to be processedWherein, qj(Ht(s-1)+st)
Represent the H of j-th of principal componentt(s-1)+stA value, T are transposition symbol, are calculatedAverage valueAnd standard deviationJust
Beginningization iterations a=0, proceeds as follows：
S321. countInherent domainNumber
If S322.A=a+1 returns to S321；Otherwise willInside it is less thanBe assigned a value ofIt willInside it is more thanBe assigned a value of
6. stable state method of discrimination according to claim 1, which is characterized in that S33 is specially：
Assuming thatRepresent smoothing processing
J-th of principal component of s-th of window afterwards, is denoted asStandard deviation is denoted asFilter
Signal x (i) after ripple is expressed as the function of time, i.e.,：
X (i)=k0+k1i+k2i2+...,+kcic (1)
Wherein,Represent the H of j-th of principal component after smoothing processingt(s-1)+stA value, T are transposition
Symbol,Represent the h of j-th of principal component in s-th of window after smoothing processingtA value, c are model order；
Make θ=[k0,k1,..,kc]TFor model parameter vector, r (i)=[i0,i,..,ic]TFor regression variable, formula (1) is abbreviated as：
X (i)=θTr(i) (2)
The optimal estimation of parameter θ can be obtained using least square method：
<mrow>
<mover>
<mi>&theta;</mi>
<mo>^</mo>
</mover>
<mo>=</mo>
<msup>
<mrow>
<mo>(</mo>
<msup>
<mi>R</mi>
<mi>T</mi>
</msup>
<mi>R</mi>
<mo>)</mo>
</mrow>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
<msup>
<mi>R</mi>
<mi>T</mi>
</msup>
<mover>
<mi>x</mi>
<mo>~</mo>
</mover>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>3</mn>
<mo>)</mo>
</mrow>
</mrow>
Wherein,
C=2 is taken, k can be obtained according to formula (3)1Value, k1The speed of variation is represented, according to the change of principal component in Analysis on Mechanism and window
Change, set suitable threshold valueWith If represent j-th of principal component judge in the stable state of s-th window as a result,
AndThen judge the principal component in the window for stable state,1 is denoted as, is not otherwise stable state,It is denoted as 0.
7. stable state method of discrimination according to claim 6, which is characterized in that S4 is specially：
The corresponding characteristic value of m principal component of selection is λ1,λ2,...,λm, then result, which assigns, to be judged to the stable state of j principal component
Weights are：
<mrow>
<msub>
<mi>&omega;</mi>
<mi>j</mi>
</msub>
<mo>=</mo>
<mfrac>
<msub>
<mi>&lambda;</mi>
<mi>j</mi>
</msub>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>m</mi>
</munderover>
<msub>
<mi>&lambda;</mi>
<mi>j</mi>
</msub>
</mrow>
</mfrac>
</mrow>
Stable state in calculation window judges result weighted sum：
IfIt is stable state in the window then to think continuous flow procedure, is not otherwise stable state；Wherein, pe% refers to stream
The minimum steady state data information that journey reaches needed for stable state accounts for the percentage of total data information, that is, thinks to believe when the former data of pe%
Breath is stable state, then then assert that whole process is in stable state, Pe% values are included the percent value of former data by principal component, i.e.,
The data message that the principal component of the selection is characterized accounts for the percentage of total data information.
8. stable state method of discrimination according to claim 1, which is characterized in that be additionally included in S1 using Principal Component Analysis
It is preceding that the operating parameter is pre-processed.
9. stable state method of discrimination according to claim 1, which is characterized in that the continuous flow procedure is hydrocracked
Continuous flow procedure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611240042.8A CN106647274B (en) | 2016-12-28 | 2016-12-28 | Operating condition stable state method of discrimination in a kind of continuous flow procedure |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611240042.8A CN106647274B (en) | 2016-12-28 | 2016-12-28 | Operating condition stable state method of discrimination in a kind of continuous flow procedure |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106647274A CN106647274A (en) | 2017-05-10 |
| CN106647274B true CN106647274B (en) | 2018-05-18 |
Family
ID=58836397
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611240042.8A Active CN106647274B (en) | 2016-12-28 | 2016-12-28 | Operating condition stable state method of discrimination in a kind of continuous flow procedure |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106647274B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107239653A (en) * | 2017-05-18 | 2017-10-10 | 华北电力大学 | A kind of power station unit steady state condition determination methods based on multivariable |
| CN108664000A (en) * | 2018-03-26 | 2018-10-16 | 中南大学 | A kind of alumina producing evaporation process steady state detecting method for use and system |
| CN111338310B (en) * | 2020-03-30 | 2023-02-07 | 南京富岛信息工程有限公司 | Industrial process steady-state working condition identification and classification method |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1849599A (en) * | 2003-09-12 | 2006-10-18 | 东京毅力科创株式会社 | Method and system of diagnosing a processing system using adaptive multivariate analysis |
| WO2012040916A1 (en) * | 2010-09-29 | 2012-04-05 | 东北大学 | Fault monitoring method of continuous annealing process based on recursive kernel principal component analysis |
| CN102662321A (en) * | 2012-03-23 | 2012-09-12 | 清华大学 | Online updating method of principal component analysis monitoring model |
| CN103323861A (en) * | 2013-06-19 | 2013-09-25 | 电子科技大学 | Method for improving steady-state performance of adaptive algorithm |
| CN103488091A (en) * | 2013-09-27 | 2014-01-01 | 上海交通大学 | Data-driving control process monitoring method based on dynamic component analysis |
| CN103941254A (en) * | 2014-03-03 | 2014-07-23 | 中国神华能源股份有限公司 | Soil physical property classification recognition method and device based on geological radar |
| US8930001B2 (en) * | 2011-09-19 | 2015-01-06 | Yokogawa Electric Corporation | Method of model identification for a process with unknown initial conditions in an industrial plant |
| CN104698837A (en) * | 2014-12-11 | 2015-06-10 | 华侨大学 | Method and device for identifying operating modal parameters of linear time-varying structure and application of the device |
| CN104932263A (en) * | 2015-06-03 | 2015-09-23 | 辽宁石油化工大学 | Minimum operation time control method of multistage intermittent process |
| CN105373110A (en) * | 2015-12-16 | 2016-03-02 | 浙江中烟工业有限责任公司 | Cigarette superspeed film packaging machine multi-loading-condition production process on-line monitoring and fault diagnosis method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6757569B2 (en) * | 2000-04-19 | 2004-06-29 | American Gnc Corporation | Filtering process for stable and accurate estimation |
-
2016
- 2016-12-28 CN CN201611240042.8A patent/CN106647274B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1849599A (en) * | 2003-09-12 | 2006-10-18 | 东京毅力科创株式会社 | Method and system of diagnosing a processing system using adaptive multivariate analysis |
| WO2012040916A1 (en) * | 2010-09-29 | 2012-04-05 | 东北大学 | Fault monitoring method of continuous annealing process based on recursive kernel principal component analysis |
| US8930001B2 (en) * | 2011-09-19 | 2015-01-06 | Yokogawa Electric Corporation | Method of model identification for a process with unknown initial conditions in an industrial plant |
| CN102662321A (en) * | 2012-03-23 | 2012-09-12 | 清华大学 | Online updating method of principal component analysis monitoring model |
| CN103323861A (en) * | 2013-06-19 | 2013-09-25 | 电子科技大学 | Method for improving steady-state performance of adaptive algorithm |
| CN103488091A (en) * | 2013-09-27 | 2014-01-01 | 上海交通大学 | Data-driving control process monitoring method based on dynamic component analysis |
| CN103941254A (en) * | 2014-03-03 | 2014-07-23 | 中国神华能源股份有限公司 | Soil physical property classification recognition method and device based on geological radar |
| CN104698837A (en) * | 2014-12-11 | 2015-06-10 | 华侨大学 | Method and device for identifying operating modal parameters of linear time-varying structure and application of the device |
| CN104932263A (en) * | 2015-06-03 | 2015-09-23 | 辽宁石油化工大学 | Minimum operation time control method of multistage intermittent process |
| CN105373110A (en) * | 2015-12-16 | 2016-03-02 | 浙江中烟工业有限责任公司 | Cigarette superspeed film packaging machine multi-loading-condition production process on-line monitoring and fault diagnosis method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106647274A (en) | 2017-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yang et al. | A control chart pattern recognition system using a statistical correlation coefficient method | |
| Das et al. | Process monitoring and fault detection strategies: a review | |
| CN106647274B (en) | Operating condition stable state method of discrimination in a kind of continuous flow procedure | |
| CN112232447A (en) | Construction method of complete sample set of power equipment state monitoring data | |
| CN106773693A (en) | A kind of sparse causality analysis method of Industry Control multi-loop oscillation behavior | |
| CN105278520A (en) | Complex industrial process running state evaluation method and application based on T-KPRM | |
| CN107272667A (en) | A kind of industrial process fault detection method based on parallel PLS | |
| WO2022171788A1 (en) | Prediction model for predicting product quality parameter values | |
| Rasay et al. | An integrated model of statistical process control and maintenance planning for a two-stage dependent process under general deterioration | |
| Ghute et al. | A multivariate synthetic control chart for process dispersion | |
| CN111414943B (en) | Anomaly detection method based on mixed hidden naive Bayes model | |
| CN103995985B (en) | Fault detection method based on Daubechies wavelet transform and elastic network | |
| Domański et al. | Robust and asymmetric assessment of the benefits from improved control–industrial validation | |
| CN114200914A (en) | MW-OCCA-based quality-related early fault detection method | |
| CN110135281B (en) | Intelligent online identification method for low-frequency oscillation of power system | |
| CN108388232B (en) | Method for monitoring operation mode fault in crude oil desalting process | |
| Sheu et al. | Monitoring autocorrelated process mean and variance using a GWMA chart based on residuals | |
| Modarresi et al. | Performance of hybrid exponentially weighted moving average control chart in the presence of measurement errors | |
| CN112199830B (en) | Variable structure system level health state evaluation method of flow program system | |
| ABOUEI et al. | An analytic variable limit np control chart | |
| Wang et al. | Study on pump fault diagnosis based on rough sets theory | |
| Badcock et al. | Two projection methods for use in the analysis of multivariate process data with an illustration in petrochemical production | |
| Aminnayeri et al. | An Analytic Variable Limit np Control Chart | |
| Ma et al. | Fault detection of chemical process based on functional kernel entropy component analysis | |
| Shamsuzzamana et al. | Design of multiattribute EWMA chart |
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 |