CN106843172B - Complex industrial process On-line quality prediction method based on JY-KPLS - Google Patents
Complex industrial process On-line quality prediction method based on JY-KPLS Download PDFInfo
- Publication number
- CN106843172B CN106843172B CN201611243467.4A CN201611243467A CN106843172B CN 106843172 B CN106843172 B CN 106843172B CN 201611243467 A CN201611243467 A CN 201611243467A CN 106843172 B CN106843172 B CN 106843172B
- Authority
- CN
- China
- Prior art keywords
- new
- matrix
- prediction
- data
- variable
- 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
- 238000000034 method Methods 0.000 title claims abstract description 151
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 68
- 230000008569 process Effects 0.000 claims description 104
- 239000011159 matrix material Substances 0.000 claims description 69
- 238000013277 forecasting method Methods 0.000 claims description 12
- 102000008297 Nuclear Matrix-Associated Proteins Human genes 0.000 claims description 6
- 108010035916 Nuclear Matrix-Associated Proteins Proteins 0.000 claims description 6
- 210000000299 nuclear matrix Anatomy 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 6
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims description 3
- 238000002790 cross-validation Methods 0.000 claims description 3
- 238000009826 distribution Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 210000004940 nucleus Anatomy 0.000 claims description 3
- 238000005498 polishing Methods 0.000 claims description 3
- 238000009877 rendering Methods 0.000 claims description 3
- HAWJXYBZNNRMNO-UHFFFAOYSA-N furathiocarb Chemical compound CCCCOC(=O)N(C)SN(C)C(=O)OC1=CC=CC2=C1OC(C)(C)C2 HAWJXYBZNNRMNO-UHFFFAOYSA-N 0.000 claims 1
- 239000000047 product Substances 0.000 abstract description 7
- 238000005516 engineering process Methods 0.000 abstract description 6
- 238000005457 optimization Methods 0.000 abstract description 3
- 239000013589 supplement Substances 0.000 abstract description 3
- 230000003044 adaptive effect Effects 0.000 abstract description 2
- 238000011049 filling Methods 0.000 abstract description 2
- 229930182555 Penicillin Natural products 0.000 description 30
- JGSARLDLIJGVTE-MBNYWOFBSA-N Penicillin G Chemical compound N([C@H]1[C@H]2SC([C@@H](N2C1=O)C(O)=O)(C)C)C(=O)CC1=CC=CC=C1 JGSARLDLIJGVTE-MBNYWOFBSA-N 0.000 description 30
- 229940049954 penicillin Drugs 0.000 description 30
- 238000000855 fermentation Methods 0.000 description 14
- 230000004151 fermentation Effects 0.000 description 14
- 238000010923 batch production Methods 0.000 description 5
- 230000007246 mechanism Effects 0.000 description 5
- 241001269238 Data Species 0.000 description 3
- 230000008901 benefit Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 244000005700 microbiome Species 0.000 description 3
- 238000013508 migration Methods 0.000 description 3
- 230000005012 migration Effects 0.000 description 3
- 238000000611 regression analysis Methods 0.000 description 3
- IAZDPXIOMUYVGZ-UHFFFAOYSA-N Dimethylsulphoxide Chemical compound CS(C)=O IAZDPXIOMUYVGZ-UHFFFAOYSA-N 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000009776 industrial production Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000035479 physiological effects, processes and functions Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 235000001674 Agaricus brunnescens Nutrition 0.000 description 1
- 239000002028 Biomass Substances 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- WQZGKKKJIJFFOK-GASJEMHNSA-N Glucose Natural products OC[C@H]1OC(O)[C@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-GASJEMHNSA-N 0.000 description 1
- 235000010254 Jasminum officinale Nutrition 0.000 description 1
- 240000005385 Jasminum sambac Species 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000004113 cell culture Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000006052 feed supplement Substances 0.000 description 1
- 239000008103 glucose Substances 0.000 description 1
- 238000010921 in-depth analysis Methods 0.000 description 1
- 238000011534 incubation Methods 0.000 description 1
- 230000000813 microbial effect Effects 0.000 description 1
- 230000007269 microbial metabolism Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 229910052760 oxygen Inorganic materials 0.000 description 1
- 239000001301 oxygen Substances 0.000 description 1
- 238000011112 process operation Methods 0.000 description 1
- 230000035484 reaction time Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 229930000044 secondary metabolite Natural products 0.000 description 1
- 235000014347 soups Nutrition 0.000 description 1
- 239000000758 substrate Substances 0.000 description 1
- 238000003786 synthesis reaction Methods 0.000 description 1
- 238000009423 ventilation Methods 0.000 description 1
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
- G05B19/00—Programme-control systems
- G05B19/02—Programme-control systems electric
- G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
- G05B19/41885—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by modeling, simulation of the manufacturing system
-
- 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
- G05B2219/00—Program-control systems
- G05B2219/30—Nc systems
- G05B2219/32—Operator till task planning
- G05B2219/32339—Object oriented modeling, design, analysis, implementation, simulation language
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Engineering & Computer Science (AREA)
- Manufacturing & Machinery (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of complex industrial process On-line quality prediction method based on JY-KPLS, the On-line quality prediction of the offline quality prediction model of operating status, batch industrial process is established using the core offset minimum binary technology of joint Y variable, specifically prediction model is established offline first with the kernel partial least squares technology of joint Y, it is online to obtain all data of decision point, data supplement completely using mean value filling mode, are predicted using quality of the prediction model to product.The present invention can effectively overcome the problems, such as that new production process data is few and can not establish prediction model, it can accelerate modeling speed, model prediction accuracy can be improved, implementation model is adaptive, prediction result can be obtained according to real-time, site operation personnel can adjust production strategy in time, realize the real-time optimization of production process, improve the overall economic efficiency of factory.
Description
Technical field
The method of the present invention relates to a kind of batch process On-line quality prediction based on JY-KPLS, belongs to industrial production mistake
Journey prediction of quality field.
Background technique
The quality of enterprise product is its key that they depend on for existence in market competition forever.In science and technology rapid development
Accurate quality predictions can be obtained than whenever all can more cause the concern of people today.But in actual production
In the process, the final actual mass index of product is often merely able to the off-line measurement in production batch end of run and obtains, such as
Product design of biological fermentation process etc. can not direct-on-line acquisition.And the off-line measurement mode usual sampling period is longer, it is several
Even ten a few houres hour, measure serious lag, it is difficult to be directly used in Optimizing manufacture.However these are difficult to parameter measured directly
For guaranteeing that product quality, the operation characteristic for reflecting production process and operation conditions and adjustment production strategy etc. all have extremely
Important meaning.In order to solve this contradiction, prediction of quality technology is come into being.
Qualitative forecasting method mainly has two major classes at present, and one kind is the method based on modelling by mechanism；Another kind of is based on number
According to the method for driving, such as regression analysis and artificial neural network.Method based on modelling by mechanism is established in advance to work
It is often difficult when in face of having the industrial process of strong nonlinearity and model uncertainty on the basis of industry process is fully understood
To complete modelling by mechanism；And method based on data-driven mainly according to process can measured data, modeling process is not rely on
The mechanism of process itself.Passed through based on the qualitative forecasting method of regression analysis to the historical data accumulated in industrial processes
Regression analysis, the offline prediction model for establishing quality are carried out, and carries out On-line quality prediction, deflected secondary air therein
(PLS) it is most widely used.
Quality prediction model is established using the method for constituents extraction based on the qualitative forecasting method of offset minimum binary.Carry out
When Principle component extraction, while considering input data information and output data information, guarantees the institute from input data and output data
Correlation maximum between the principal component extracted carries out regression modeling with the principal component extracted.Partial least squares algorithm has
Less to amount of training data requirement, computational complexity is lower, explains the advantages that effect is preferable.
Currently, existing correlation scholar has done some researchs to batch process prediction of quality.Such as Zhao Chunhui proposes one kind
PLS modeling and qualitative forecasting method based on the period, suitable for solving the problems, such as the modeling and prediction of quality of multistage batch process.Cui
Long jasmine etc., proposes a kind of qualitative forecasting method based on multidirectional kernel partial least squares, supervises to current some statistic processes
Control and prediction of quality algorithm are improved.Qu Yaxin, Song Kai etc., the method proposed are confined to the data using single process
Information, and modeling process needs a large amount of training data.Dong Weiwei, Jia Runda, soup are strong etc., and the modeling method mentioned is based on core
Offset minimum binary (KPLS), being capable of strong nonlinearity problem in effective processing system.Sunlight etc. is opened, Kernel-Based Methods are carried out
Correlative study, and kernel method has been applied in the operation monitoring of penicillin fermentation process.
The above several method is all built upon possess adequate data sample on the basis of, just put into production for one new
Batch production process can not usually obtain sufficient process data since the time of production process operation is shorter.If using class
It is similar to above-mentioned modeling method, often precision of prediction is lower is unable to meet production demand for the prediction model established.And according to biography
System method must just redesign experiment with acquisition process data, this will expend a large amount of manpower financial capacity, and modeling efficiency is low, sternly
It has dragged slowly new process to realize the speed of production run optimization again, has been unfavorable for enterprise and adjusts production strategy and expanding production rule in real time
Mould.In addition, often there is strong nonlinearity in actual production process, linear deflected secondary air is caused to be difficult to fit
With.
In actual production process, the change of operating condition or the foundation of newborn producing line will all make the prediction originally established
Model failure, however being consistent due to the inherent mechanism of new and old process, certainly exist certain similitude between them.Such as
Fruit can using certain strategy, new process model building is helped using data information useful in similar process, will avoid into
Row is a large amount of to be repeated to test, and improves modeling efficiency.
Summary of the invention
In order to overcome existing prediction technique to be difficult to set up the deficiency of accurate predictive model because new production process data is insufficient,
The present invention provides a kind of complex industrial process On-line quality prediction method based on JY-KPLS, utilizes two similar process data
Combine and establish prediction model, Data Migration suitable in similar process is applied in new production process, can effectively be overcome
New production process data is few and the problem of prediction model can not be established, modeling speed can be accelerated, model prediction can be improved
Precision, implementation model update.
The technical solution used to solve the technical problems of the present invention is that:
A kind of complex industrial process On-line quality prediction method based on JY-KPLS,
It uses two sets of identical production equipments, and respective inner parameter setting is different, carries out a process and b respectively
Two production processes of process, wherein a process is completely new production process and data are few, and the production time of b process and long data
It is abundant；Assuming that process data matrix is X ∈ RN×J, N is sample number, and J is process variable number；Assuming that output data matrix is Y ∈
RN, include output process variable；Specific step is as follows for qualitative forecasting method:
Step 1: a process, the three-dimensional input data of b process are launched into two-dimensional matrix according to batch direction, respectively
Xa、Xb；
Step 2: to a process, b process input data matrix Xa, XbEach column carry out zero-mean and unit variance processing；Together
Reason, to output data matrix Ya, YbAlso it is standardized, and output data matrix YaWith YbThe number phase of middle quality variable
Together；
Step 3: by input data matrix Xa、XbThrough Nonlinear Mapping Φ: xi∈RN→Φ(xi) ∈ F project to higher-dimension spy
Space F is levied, and calculates nuclear matrix K in the space Fa、Kb: Ka=ΦTΦ, Kb=ΦTΦ；
Step 4: to nuclear matrix Ka、KbIt is standardized；
Step 5: running JY-KPLS algorithm to input nucleus matrix K and output data matrix Y:
Input data matrix becomes K at this timea、Kb, output data matrix becomes Ya, Yb, extracted from output data matrix Y
Convergent ui, work as i=1, KWi=KW, YWi=YW；
A1, Y is enabledWiIn any one column be equal to ui；
B1, K is calculatedaScore vector, t1i=KWiu1i, t1i←t1i/||t1i||；
c1、KbScore vector t2i=KWiu2i, t2i←t2i||t2i||；
D1, Y1 is calculatedWiScore vector u1i=Y1Wiqi, u1i←u1i||u1i||；F1, Y2 is calculatedWiScore vector u2i
=Y2Wiqi, u2i←u2i/||u2i||；
E1, judge u1iWith u2iWhether restrain, is transferred to step 6 if convergence, otherwise returns to a1；
Step 6: calculating KWiLoad matrix:
Step 7: extracting whole pivots, input data matrix K is calculatedWScore matrix T, input data matrix KWIt is negative
Carry matrix P, output data matrix YWScore matrix U and output data matrix YWLoad matrix Q, it is specific as follows:
A2, order
B2, i=i+1 is enabled, repeated Step 5: six until extracting A pivot, pivot number A can be true by cross-validation method
It is fixed；
c2、T1=[t1..., tA], T2=[t1..., tA], p=[p1..., pA], U2=[u1..., uA], Q=
[q1..., qA]；
If output data matrix Y is single output variable, the expression formula of JY-KPLS model is as follows:
Y=K*U2(T2*Knx*U2)-1*T'j*Yj+ F,
It is the joint of a the output of process variable and b the output of process variable,It is a process latent variable
With the joint of b process latent variable, that is, establish the most critical variable obtained required for quality prediction model, can be used for operating status from
The foundation of line mass prediction model；
As the new sample k of introducingnew, score matrix will obtain by following formula:
Wherein, knewIt is new samples xnewKernel function, can be calculated by following formula:
knew=Ф (X) Ф (xnew)=[k (x1, xnew) ... k (xn, xnew)]T,
To knewCarrying out equalization can obtain:
Wherein, 1t=1/n [11...1]T∈Rn；
Step 8: obtaining input data data x onlinenew, the part input data mean value polishing that can not be obtained, utilization
xnewIt carries out On-line quality prediction and obtains predicted valueAnd according to prediction result Instructing manufacture process；
Step 9: obtaining newest output data y at the end of current production batchnew, and calculate the pre- of newest batch
Survey error deltan, whereinOtherwise return step eight；
Step 10: model prediction accuracy is examined, rendering error curve；When process runs are greater than 2J times, J is process input
Total number of variable obtains prediction error delta all in addition to newest batchn-1；Sample δn-1Normal Distribution is found out as conspicuousness α
Confidence interval when=0.95When newest batch predicts error deltanFall in confidence interval
When interior, then 11 are entered step；Work as δnWhen falling in outside confidence interval, then 12 are entered step；
Step 11: the similarity degree during rejecting b with a process similarity degree the smallest data old process and new process
With similitude s (xi) indicate, s (x can be acquired with formula (10) and (11)i), formula is as follows:
Wherein, | | | | it is Euclidean distance,For the mean value of new process data, s (xi) value range be 0 to 1；
New X is formed Step 12: new data measured is added in a process initial dataa, Ya, and return step one,
Formula is as follows:
Compared with prior art, a kind of complex industrial process On-line quality prediction method based on JY-KPLS of the invention is
The offline quality prediction model of operating status is established using core offset minimum binary (abbreviation JY-KPLS) technology of joint Y variable, is criticized
The On-line quality prediction of secondary industrial process specifically establishes prediction mould first with the kernel partial least squares technology of joint Y offline
Type completes offline prediction model establishment process by step 1 to step 7, realize JYKPLS model modification, and online obtain is determined
All data of plan point supplement to data complete, using prediction model to product quality progress using mean value filling mode
Prediction.By by historical data existing in similar process migration be applied to it is new during, solve new process be difficult to set up it is pre-
The problem of surveying model；Compared with the common deflected secondary air directly modeled using new process data, this method can be accelerated
Modeling speed enables new production process quickly to put into production；It is identical that this method does not need two processes, for
The variable number and sample number of input data, this method do not do any restrictions, however due to needing the output data of similar process
Matrix is joined together, and the variables number of output data must be identical.Meanwhile for the nonlinear problem during solving, we
Method introduces Kernel-Based Methods, and kernel method passes through the input data that will acquire and is mapped in high-dimensional feature space, makes originally non-
The initial data of linearly inseparable becomes linear separability, so that the nonlinearity of data itself is reduced, it can be in nuclear space
Data are analyzed using partial least squares algorithm, introducing kernel method enables this method to have height non-thread suitable for processing
The actual industrial production process of property.
Detailed description of the invention
Present invention will be further explained below with reference to the attached drawings and examples.
Fig. 1 is the flow chart of qualitative forecasting method of the present invention.
Batch penicillin prediction of quality relative error when Fig. 2 is decision point difference.
Fig. 3 is that the concentration relative error of penicillin prediction of quality compares.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without creative efforts, belongs to protection scope of the present invention.
Complex industrial process based on JY-KPLS (abbreviation that JY-KPLS is the core offset minimum binary of joint Y variable) is online
Qualitative forecasting method,
It uses two sets of identical production equipments, and respective inner parameter setting is different, carries out a process and b respectively
Two production processes of process, wherein a process is completely new production process and data are few, and the production time of b process and long data
It is abundant；Assuming that process data matrix is X ∈ RN×J, N is sample number, and J is process variable number；Assuming that output data matrix is Y ∈
RN, include output process variable；Specific step is as follows for qualitative forecasting method:
Step 1: a process, the three-dimensional input data of b process are launched into two-dimensional matrix according to batch direction, respectively
Xa、Xb；
Step 2: to a process, b process input data matrix Xa, XbEach column carry out zero-mean and unit variance processing；Together
Reason, to output data matrix Ya, YbAlso it is standardized, and output data matrix YaWith YbThe number phase of middle quality variable
Together；
Step 3: by input data matrix Xa、XbThrough Nonlinear Mapping Φ: xi∈RN→Φ(xi) ∈ F project to higher-dimension spy
Space F is levied, and calculates nuclear matrix K in the space Fa、Kb: Ka=ΦTΦ, Kb=ΦTΦ；
Step 4: to nuclear matrix Ka、KbIt is standardized；
Step 5: running JY-KPLS algorithm to input nucleus matrix K and output data matrix Y:
Input data matrix becomes K at this timea、Kb, output data matrix becomes Ya, Yb, extracted from output data matrix Y
Convergent ui, work as i=1, KWi=KW, YWi=YW；
A1, Y is enabledWiIn any one column be equal to ui；
B1, K is calculatedaScore vector, t1i=KWiu1i, t1i←t1i/||t1i||；
c1、KbScore vector t2i=KWiu2i, t2i←t2i/||t2i||；
D1, Y1 is calculatedWiScore vector u1i=Y1Wiqi, u1i←u1i/||u1i||；F1, Y2 is calculatedWiScore vector
u2i=Y2Wiqi, u2i←u2i/||u2i||；
E1, judge u1iWith u2iWhether restrain, is transferred to step 6 if convergence, otherwise returns to a1；
Step 6: calculating KWiLoad matrix:
Step 7: extracting whole pivots, input data matrix K is calculatedWScore matrix T, input data matrix KWIt is negative
Carry matrix P, output data matrix YWScore matrix U and output data matrix YWLoad matrix Q, it is specific as follows:
A2, order
B2, i=i+1 is enabled, repeated Step 5: six until extracting A pivot, pivot number A can be true by cross-validation method
It is fixed；
c2、T1=[t1..., tA], T2=[t1..., tA], P=[p1..., pA], U2=[u1..., uA], Q=
[q1..., qA]；If output data matrix Y is single output variable, the expression formula of JY-KPLS model is as follows:
Y=K*U2(T2*Knx*U2)-1*T′j*Yj+,
It is the joint of a the output of process variable and b the output of process variable,It is a process latent variable
With the joint of b process latent variable, that is, establish the most critical variable obtained required for quality prediction model, can be used for operating status from
The foundation of line mass prediction model；
As the new sample k of introducingnew, score matrix will obtain by following formula:
Wherein, knewIt is new samples xnewKernel function, can be calculated by following formula:
knew=Φ (X) Φ (xnew)=[k (x1, xnew) ... k (xn, xnew)]T,
To knewCarrying out equalization can obtain:
Wherein, 1t=1/n [11...1]T∈Rn；
Step 8: obtaining input data data x onlinenew, the part input data mean value polishing that can not be obtained, utilization
xnewIt carries out On-line quality prediction and obtains predicted valueAnd according to prediction result Instructing manufacture process；
Step 9: obtaining newest output data y at the end of current production batchnew, and calculate the pre- of newest batch
Survey error deltan, whereinOtherwise return step eight；
Step 10: model prediction accuracy is examined, rendering error curve；When process runs are greater than 2J times, J is process input
Total number of variable obtains prediction error delta all in addition to newest batchn-1；Sample δn-1Normal Distribution is found out as conspicuousness α
Confidence interval when=0.95When newest batch predicts error deltanFall in confidence interval
When interior, then 11 are entered step；Work as δnWhen falling in outside confidence interval, then 12 are entered step；
Step 11: the similarity degree during rejecting b with a process similarity degree the smallest data old process and new process
With similitude s (xi) indicate, s (x can be acquired with formula (10) and (11)i), formula is as follows:
Wherein, | | | | it is Euclidean distance,For the mean value of new process data, s (xi) value range be 0 to 1；
New X is formed Step 12: new data measured is added in a process initial dataa, Ya, and return step one,
Formula is as follows:
Simulation example is provided below, more detailed explanation is made to technical solution of the present invention effect.
Penicillin is the secondary metabolite of microorganism, and the main approach by microbial fermentation produces, high cost and height
Energy consumption is the characteristic of penicillin fermentation process.Therefore, optimize penicillin fermentation production process, for reducing penicillin production cost
It is of great significance with penicillin yield is improved.In order to grasp the quality of product in time, penicillin fermentation process is realized
Real-time optimization, can will be in penicillin quality on-line prediction that this method applies to.
In current penicillin industrial processes, the fermentation method of fed-batch is occupied an leading position.The production of penicillin
Process is typical a non-linear, dynamic, multistage batch production process.The fermentation process pH value and temperature of penicillin use
Closed-loop control, and feed supplement is controlled using open loop value can be with by the temperature in control reaction process in pH value and fermentation reactor
Run reaction at optimum conditions.The entire production cycle contains four physiology phases: response lag phase, thallus mushroom out
Phase, penicillin synthesize phase, thallus death phase；Two physics sub-periods: the cell culture stage (corresponding the first two physiology phase, about
Continue 45h) with penicillin fed-batch fermentation stage (in latter two corresponding period, last about 355h greatly).As secondary microbial metabolism
Process, the fermentation process common practice are to carry out the culture of microorganism under certain condition first, this is the initial incubation stage,
Then promote the synthesis of penicillin, this meaning penicillin fermentation stage by constantly supplement glucose.
Pass through the in-depth analysis to penicillin production process and consider the actual production process at scene, 6 processes can be chosen
Variable and 1 output variable are used for the on-line prediction of penicillin quality.This 6 process variables are respectively: ventilation rate, substrate feeding
The concentration of temperature, power of agitator, culture volume, CO2, pH value；1 output variable is: penicillin concn.Each input and output become
Amount is as shown in table 1:
1 input/output variable table of table
1) foundation of concentration prediction model
Professor Cinar of Illinois technical college develops the simulation software Pensim2.0 of penicillin fermentation process, side
Penicillin data acquisition.The software can be to the microorganism concn of penicillin production process under the conditions of different operation, dense
Degree, CO2Value, pH value, penicillin concn, concentration of carbon, oxygen concentration and heat of generation etc. are emulated.Required setting it is initial
Changing parameter includes: reaction time, sampling time, biomass, yeasting, temperature control parameter, pH control parameter.It is soft using this
Part obtains the data that a production process obtains 46 batches, wherein the data of 5 batches are used to model, the data of 41 batches
For examining.B production process obtains the data of 40 batches, the foundation for penicillin concn prediction model.It was produced using a
5 batches of journey and 40 lot datas of b process are modeled according to above-mentioned method, then utilize a process remaining 41
A lot data is detected.
2) determination of predicted time
During on-line prediction, need enough input datas that could obtain more satisfied prediction result, unknown is defeated
Entering data can be replaced with mean value.Therefore the selection of predicted time is extremely important, and predicted time can early make very much penicillin concn
Prediction error it is too big, the time will affect the validity of on-line prediction again too late.In order to acquire optimum prediction time point, select here
The data for selecting first batch are tested, and carry out a prediction of quality respectively every 10h, obtained data are plotted in Fig. 2.
As seen from Figure 2, when predicted time is greater than 17 (170h), prediction error can incite somebody to action very close to minimum value
Predicted time is scheduled on 170 between 240h, and predicted time is scheduled at 200h by this emulation.
3) the penicillin quality on-line prediction based on JY-KPLS
In order to test to the provided qualitative forecasting method based on JY-KPLS, used here as identical batch number
It is compared according to the qualitative forecasting method based on KPLS, as a result as shown in Figure 3.From before Fig. 3 in 12 batches as can be seen that
The initial stage of production process is significantly less than KPLS using the prediction error of JY-KPLS, therefore the prediction model of JY-KPLS can have
The foundation of prediction model is accelerated on effect ground, obtains satisfactory control performance.It is emphasized that fewer pair of new process data
JY-KPLS is more advantageous, more unfavorable to KPLS, therefore the data at new process operation initial stage is selected to compare here, to embody this
The advantage of method.
4) data of legacy data are rejected
With the increase of production batch number, new process data sample constantly expands, the precision of prediction of KPLS method also because
This is continuously improved.The precision of prediction of JY-KPLS method is equally improved with the increase of batch number, but due to wherein similar process number
According to presence, improve speed be obviously not so good as KPLS method.Therefore in order to further increase precision of prediction, it is necessary to be obtained with new
Creation data legacy data is rejected.
It can be the 18th batch in the hope of rejecting point according to the method in specification, therefore according to the method described above from the 18th
Batch starts to reject legacy data, as a result as shown in Figure 3.From figure 3, it can be seen that similitude s ought be weeded out every time
(xi) minimum legacy data when, whole precision of prediction has obtained further raising, and the prediction technique based on JY-KPLS is another
The secondary gap widened between KPLS method.
It can be seen that the present invention by the above simulation example and establish prediction model using two similar process data aggregates,
Data Migration suitable in similar process is applied in new production process, new penicillin production process can be effectively overcome
Data are few and can not establish this problem of prediction model.Kernel-Based Methods are very suitable to non-linear, the time-varying that processing has height
Probabilistic penicillin fermentation process of property and model.Utilize the batch industrial process prediction of quality migrated based on latent variable
Method (JY-KPLS) establishes quality prediction model offline, can accelerate modeling speed, predicts the ultimate density of penicillin
With important practical usage.This method using each batch after the new data that generates model is updated, can not
Model prediction accuracy is enough improved, implementation model is adaptive.After obtaining enough new process datas, to similitude during old compared with
Low data are rejected, to eliminate influence caused by legacy data itself deviation.By means of prediction as a result, execute-in-place people
Member can adjust and improve in time production strategy, improve production efficiency.
The above is only presently preferred embodiments of the present invention, not does limitation in any form to the present invention, it is all according to
According to technical spirit of the invention, any simple modification and same variation, each fall within guarantor of the invention to made by above embodiments
Within the scope of shield.
Claims (1)
1. a kind of complex industrial process On-line quality prediction method based on JY-KPLS, uses two sets of identical productions
Equipment, respective inner parameter setting is different, carries out two production processes of a process and b process respectively, wherein a process is completely new
Production process and data it is few, and the production time of b process and long data rich；Assuming that process data matrix is X ∈ RN×J, N is
Sample number, J are process variable number；Assuming that output data matrix is Y ∈ RN, include output process variable；Qualitative forecasting method tool
Steps are as follows for body:
Step 1: a process, the three-dimensional input data of b process are launched into two-dimensional matrix, respectively X according to batch directiona、Xb；
Step 2: to a process, b process input data matrix Xa, XbEach column carry out zero-mean and unit variance processing；Similarly,
To output data matrix Ya, YbAlso it is standardized, and output data matrix YaWith YbThe number of middle quality variable is identical；
Step 3: by input data matrix Xa、XbThrough Nonlinear Mapping Φ: xi∈RN→Φ(xi) ∈ F project to high dimensional feature sky
Between F, and in the space F calculate nuclear matrix Ka、Kb: Ka=ΦTΦ, Kb=ΦTΦ；
Step 4: to nuclear matrix Ka、KbIt is standardized；
Step 5: running JY-KPLS algorithm to input nucleus matrix K and output data matrix Y:
Input data matrix becomes K at this timea、Kb, output data matrix becomes Ya, Yb, extracted from output data matrix Y convergent
ui, work as i=1, KWi=KW, YWi=YW；
A1, Y is enabledWiIn any one column be equal to ui；
B1, K is calculatedaScore vector, t1i=KWiu1i, t1i←t1i/||t1i||；
c1、KbScore vector t2i=KWiu2i, t2i←t2i/||t2i||；qi=Yj T[t1；t2]；
D1, Y1 is calculatedWiScore vector u1i=Y1Wiqi, u1i←u1i/||u1i||；F1, Y2 is calculatedWiScore vector u2i=
Y2Wiqi, u2i←u2i/||u2i||；
E1, judge u1iWith u2iWhether restrain, is transferred to step 6 if convergence, otherwise returns to a1；
Step 6: calculating KWiLoad matrix:
Step 7: extracting whole pivots, input data matrix K is calculatedWScore matrix T, input data matrix KWThe moment of load
Battle array P, output data matrix YWScore matrix U and output data matrix YWLoad matrix Q, it is specific as follows:
A2, order
B2, i=i+1 is enabled, repeated Step 5: six until extracting A pivot, pivot number A can be determined by cross-validation method；
c2、T1=[t1..., tA], T2=[t1..., tA], P=[p1..., pA], U2=[u1..., uA], Q=[q1...,
qA]；
If output data matrix Y is single output variable, the expression formula of JY-KPLS model is as follows:
Y=K*U2(T2*Knx*U2)-1*T′j*Yj+ F,
It is the joint of a the output of process variable and b the output of process variable,It is a process latent variable and b mistake
The joint of Cheng Qian's variable establishes the most critical variable obtained required for quality prediction model, can be used for the offline matter of operating status
Measure the foundation of prediction model；
As the new sample k of introducingnew, score matrix will obtain by following formula:
Wherein, knewIt is new samples xnewKernel function, can be calculated by following formula:
knew=Φ (X) Φ (xnew)=[k (x1, xnew) ... k (xn, xnew)]T,
To knewCarrying out equalization can obtain:
Wherein, 1t=1/n [11...1]T∈Rn；
Step 8: obtaining input data data x onlinenew, the part input data mean value polishing that can not be obtained utilizes xnewInto
Row On-line quality prediction obtains predicted valueAnd according to prediction result Instructing manufacture process；
Step 9: obtaining newest output data y at the end of current production batchnew, and the prediction for calculating newest batch misses
Poor δn, whereinOtherwise return step eight；
Step 10: model prediction accuracy is examined, rendering error curve；When process runs are greater than 2J times, J is process input variable
Sum obtains prediction error delta all in addition to newest batchn-1；Sample δn-1Normal Distribution, find out when conspicuousness α=
Confidence interval when 0.95When newest batch predicts error deltanIt falls in confidence interval
When, then enter step 11；Work as δnWhen falling in outside confidence interval, then 12 are entered step；
Step 11: the similarity degree phase during rejecting b with the old process of the smallest data of a process similarity degree and new process
Like property s (xi) indicate, s (x can be acquired with formula (10) and (11)i), formula is as follows:
Wherein, | | | | it is Euclidean distance,For the mean value of new process data, s (xi) value range be 0 to 1；
New X is formed Step 12: new data measured is added in a process initial dataa, Ya, and return step one, formula
It is as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611243467.4A CN106843172B (en) | 2016-12-29 | 2016-12-29 | Complex industrial process On-line quality prediction method based on JY-KPLS |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611243467.4A CN106843172B (en) | 2016-12-29 | 2016-12-29 | Complex industrial process On-line quality prediction method based on JY-KPLS |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106843172A CN106843172A (en) | 2017-06-13 |
| CN106843172B true CN106843172B (en) | 2019-04-09 |
Family
ID=59113928
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611243467.4A Active CN106843172B (en) | 2016-12-29 | 2016-12-29 | Complex industrial process On-line quality prediction method based on JY-KPLS |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106843172B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107463093B (en) * | 2017-07-13 | 2019-05-21 | 东北大学 | A kind of blast-melted quality monitoring method based on KPLS robust reconstructed error |
| WO2019094787A1 (en) * | 2017-11-09 | 2019-05-16 | Zymergen Inc. | Optimization of organisms for performance in larger-scale conditions based on performance in smaller-scale conditions |
| CN108388218B (en) * | 2018-02-08 | 2020-06-19 | 中国矿业大学 | Correction self-adaptive batch process optimization method based on latent variable process migration model |
| CN108664002B (en) * | 2018-04-27 | 2021-06-29 | 中国石油大学(华东) | Quality-oriented nonlinear dynamic process monitoring method |
| CN109242224A (en) * | 2018-11-29 | 2019-01-18 | 杭州电子科技大学 | A kind of novel chemical procedure quality prediction technique |
| CN110426956B (en) * | 2019-07-30 | 2021-07-23 | 中国矿业大学 | Intermittent process optimal compensation control strategy based on process migration model |
| CN110703714B (en) * | 2019-11-06 | 2020-09-15 | 点睛数据科技(杭州)有限责任公司 | Batch process model building method |
| CN110794782A (en) * | 2019-11-08 | 2020-02-14 | 中国矿业大学 | Batch industrial process online quality prediction method based on JY-MKPLS |
| CN111487950B (en) * | 2020-04-24 | 2021-11-16 | 西安交通大学 | 'prediction-verification-feedback-optimization' closed-loop system for online early warning and offline diagnosis |
| CN111679643B (en) * | 2020-06-28 | 2021-10-15 | 中国矿业大学 | Intermittent process second-order correction self-adaptive optimization method based on process migration model |
| CN112859790A (en) * | 2021-02-01 | 2021-05-28 | 刘金涛 | Production process quality prediction and control method based on KEPLS |
| CN115576293B (en) * | 2022-12-08 | 2023-03-14 | 台昌树脂(佛山)有限公司 | Pressure-sensitive adhesive on-line production analysis method and system based on data monitoring |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101446831A (en) * | 2008-12-30 | 2009-06-03 | 东北大学 | Decentralized process monitoring method |
| CN102601881A (en) * | 2012-03-20 | 2012-07-25 | 天津大学 | Method for monitoring on-line quality and updating prediction model of rubber hardness |
| CN104133991A (en) * | 2014-07-15 | 2014-11-05 | 东北大学 | Penicillin fermentation process fault diagnosis method based on kernel partial least squares reconstitution |
| CN105278520A (en) * | 2015-06-29 | 2016-01-27 | 中国矿业大学 | Complex industrial process running state evaluation method and application based on T-KPRM |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9110452B2 (en) * | 2011-09-19 | 2015-08-18 | Fisher-Rosemount Systems, Inc. | Inferential process modeling, quality prediction and fault detection using multi-stage data segregation |
-
2016
- 2016-12-29 CN CN201611243467.4A patent/CN106843172B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101446831A (en) * | 2008-12-30 | 2009-06-03 | 东北大学 | Decentralized process monitoring method |
| CN102601881A (en) * | 2012-03-20 | 2012-07-25 | 天津大学 | Method for monitoring on-line quality and updating prediction model of rubber hardness |
| CN104133991A (en) * | 2014-07-15 | 2014-11-05 | 东北大学 | Penicillin fermentation process fault diagnosis method based on kernel partial least squares reconstitution |
| CN105278520A (en) * | 2015-06-29 | 2016-01-27 | 中国矿业大学 | Complex industrial process running state evaluation method and application based on T-KPRM |
Non-Patent Citations (4)
| Title |
|---|
| Combining fundamental knowledge and latent variable techniques to transfer process monitoring models between plants;Emanuele Tomba, etc.;《Chemometrics and Intelligent Laboratory Systems》;20120507;第67-77页 * |
| Product transfer between sites using Joint-Y PLS;Salvador G. M. etc.;《Chemometrics and Intelligent Laboratory Systems》;20050803;第101-114页 * |
| Scale-up of a Pharmaceutical Roller Compaction Process Using a Joint-Y Partial Least Squares Model;Zheng Liu, etc.;《Industrial & Engineering Chemistry Reasearch》;20110727;第10696–10706页 * |
| 基于KPLS模型的间歇过程产品质量控制;贾润达, 等;《化工学报》;20130430;第64卷(第4期);第1332-1339页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106843172A (en) | 2017-06-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106843172B (en) | Complex industrial process On-line quality prediction method based on JY-KPLS | |
| CN102231057B (en) | Method for carrying out soft-sensing on lysine fermenting process on basis of chaos particle swarm optimization (CPSO) | |
| CN101630376A (en) | Soft-sensing modeling method and soft meter of multi-model neural network in biological fermentation process | |
| Tian et al. | TS fuzzy neural network predictive control for burning zone temperature in rotary kiln with improved hierarchical genetic algorithm | |
| Li et al. | Dynamic time features expanding and extracting method for prediction model of sintering process quality index | |
| Chu et al. | Final quality prediction method for new batch processes based on improved JYKPLS process transfer model | |
| He et al. | Soft sensor development for the key variables of complex chemical processes using a novel robust bagging nonlinear model integrating improved extreme learning machine with partial least square | |
| CN104462850A (en) | Multi-stage batch process soft measurement method based on fuzzy gauss hybrid model | |
| CN105335619A (en) | Collaborative optimization method applicable to parameter back analysis of high calculation cost numerical calculation model | |
| CN102004444A (en) | Multi-model predictive control method for component content in process of extracting rare earth | |
| Yuan et al. | Modelling and pathway identification involving the transport mechanism of a complex metabolic system in batch culture | |
| Yuan et al. | Robust optimal control problem with multiple characteristic time points in the objective for a batch nonlinear time-varying process using parallel global optimization | |
| CN106022532A (en) | Glutamic acid product concentration online estimation method and device, and monitoring system | |
| CN102129242A (en) | Product quality control method during batch processing production process based on two-layer hybrid intelligent optimization | |
| CN101916394A (en) | Knowledge fusion-based online soft measurement method | |
| CN103279030B (en) | Dynamic soft measuring modeling method and device based on Bayesian frame | |
| Wang et al. | Soft-sensing modeling based on ABC-MLSSVM inversion for marine low-temperature alkaline protease MP fermentation process | |
| Julien et al. | Bioreactor monitoring, modeling, and simulation | |
| CN103617316A (en) | Multi-kernel support vector machine based dynamic fermentation process predicating method | |
| Jianlin et al. | Run-to-run optimization for fed-batch fermentation process with swarm energy conservation particle swarm optimization algorithm | |
| CN103439964A (en) | On-line updating system and method for OCSVM monitoring model | |
| Teng et al. | Explicit distributed model predictive control design for chemical processes under constraints and uncertainty | |
| Liu et al. | Thermal efficiency prediction model of cement clinker production based on fuzzy c-means monitoring clustering | |
| Wang et al. | IUP: an intelligent utility prediction scheme for solid-state fermentation in 5G IoT | |
| Zhang et al. | A combustion control strategy of hot blast stove based on kernel fuzzy c-means (FCM) |
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 |