CN110309491A - A kind of transient phase division methods and system based on local Gaussian mixture model - Google Patents
A kind of transient phase division methods and system based on local Gaussian mixture model Download PDFInfo
- Publication number
- CN110309491A CN110309491A CN201910571289.5A CN201910571289A CN110309491A CN 110309491 A CN110309491 A CN 110309491A CN 201910571289 A CN201910571289 A CN 201910571289A CN 110309491 A CN110309491 A CN 110309491A
- Authority
- CN
- China
- Prior art keywords
- phase
- sample
- gauss
- model
- stable phase
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Algebra (AREA)
- Evolutionary Biology (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a kind of phase division methods and system based on local Gaussian mixture model comprising: S1, collecting sample simultaneously create historical training dataset；S2, first Gaussian distribution model is created to determine first stable phase from historical training dataset selected section sample；S3, based on the previous Gauss model having determined, gauss hybrid models of the creation comprising two gauss components are to determine next stable phase；S4, transient phase that may be present between two stable phases is determined based on fixed two adjacent stable phase models；S5, S3 and S4 is repeated to complete the phase division to whole sample datas.Present invention greatly reduces redundant computations, it improves computational efficiency and uses step-by-step movement more new strategy, according to sample time order, stable phase and transient phase are gradually determined, have many advantages, such as that phase division number does not need preassignment and division result does not need subsequent processing.
Description
Technical field
The present invention relates to transient state phases in batch process statistical modeling technical field more particularly to a kind of leggy batch process
Position division methods and system.
Background technique
Batch process be in modern industry a kind of very common production method be widely used in fine chemistry industry, pharmacy,
The industries such as metallurgy and semiconductor.With the development of technology with the diversification of demand, batch process also becomes to become increasingly complex, directly
Performance be in a batch process comprising multiple and different operational phases or multiple and different reaction/changes phases, in this way
Process be referred to as leggy batch process, each such stage is referred to as a phase.Such as penicillin fermentation process,
Assuming that the time of a penicillin fermentation process is 400h, preceding 45h is the preculture stage, and rear 355h belongs to feedback formula feeding stage,
To raw material is added in reaction kettle i.e. from 45h.For reaction mechanism, a typical penicillin fermentation process can divide again
For four phases (stage), including retardation stage, exponential growth phase, stabilization sub stage and self-dissolving stage.And penicillin fermentation mistake
Journey is a typical slow time-varying process, and the process of not instead of one mutation of transformation between out of phase, one slowly becomes
The process of change, therefore the transformation between out of phase is not so obvious, and will appear a kind of following situation, it is steady at two
Sample between state phase is the feature that part retains first stable phase, and includes the spy of next new stable phase
Sign, the phase for meeting such characteristic are referred to as transient phase.How the batch process of a leggy accurately reasonably to be divided
At out of phase, be conducive to reinforce the precision for further understanding and improving process model building to process mechanism.
Currently, had certain research achievement for leggy division, the leggy pivot including being similar to the method for exhaustion
Analytic approach carries out batch division using repetition factor index, however for the process that process does not have significant change inflection point, this
Kind method is no longer applicable in.The phase that the method for cluster is widely used in batch process divides, however, being removed based on K-means algorithm
It to preassign and divide outside classification number, the timing sexual intercourse between sample do not account for, and causes division result chaotic and needs
Further subsequent processing is wanted, the problems such as explaining is not easy to.Therefore in phase, what the sequential relationship between sample not can not ignore
One key factor, and multistable phase is not only wanted effectively to be divided, it also wants accurately divide transient phase.That is
Above two method does not account for the problem of the deficiencies of timing and transient phase division.
Summary of the invention
Based on this, for the deficiencies of timing and transient phase divide is not accounted in existing phase division methods, spy is mentioned
A kind of phase division methods based on local Gaussian mixture model are gone out.
A kind of phase division methods based on local Gaussian mixture model, include the following steps:
S1, collecting sample simultaneously create historical training dataset；
S2, first Gaussian Profile is created from the historical training dataset selected section sample according to sample time order
Model is to determine first stable phase；
S3, based on the previous Gauss model having determined, gauss hybrid models of the creation comprising two gauss components with
Determine next stable phase；
S4, based on fixed two adjacent stable phase models determine between two stable phases it is that may be present temporarily
State phase；
S5, S3 and S4 is repeated to complete the phase division to whole sample datas.
Optionally, described from historical training dataset selected section sample creation the in one of the embodiments,
One Gaussian distribution model includes: to determine first stable phase data
S21, successively from the historical training dataset choose before N1A sample simultaneously calculates its mean value and variance to obtain pair
The Gaussian distribution model p (x | 1) answered, wherein p (x | 1) indicates that the probability density function of first Gaussian distribution model, x indicate
The sample data of acquisition；
S22, sample drawn point carry out stable phase verifying, i.e., from N1/ 2 sample points start verifying to find continuous three
Serial number corresponding to the sample point for meeting first verification condition is simultaneously labeled as by a sample point for meeting the first verification condition
The verification condition is
Wherein, ρ is preassigned threshold value；
S23, judge N1Whether it is equal toIt is to indicate that i.e. first stable phase of result convergence has determined that and carry out next
Step；Otherwise it enablesAnd return step S21 is iterated until N1It is equal to
Optionally, described based on the previous Gauss model having determined in one of the embodiments, creation includes two
The gauss hybrid models of a gauss component include: to determine next stable phase
S31, based on the previous Gauss model having determined, mixed model of the creation comprising two gauss of distribution function is simultaneously
It is trained, without loss of generality, it is assumed that preceding c-1 stable phase is it has been determined that c is greater than the integer equal to 2, the mixed model
Corresponding formula is
p(x|θc)=αc-1p(x|c-1)+αcp(x|c)
Wherein, the probability density function p (x | c-1) of the c-1 Gauss model is it has been determined that note includes Nc-1The of a sample
C-1 stable phase is Xc-1, the probability density function p (x | c) of c-th of Gauss model is undetermined, it is assumed that includes NcThe of a sample
C stable phase is Xc, remember Xm={ Xc-1,XcGauss hybrid models thus training data, corresponding training parameter θc=
{αc-1,αc,μc,Σc, αc-1And αcIt is the c-1 combination coefficients with c-th of gauss component in mixed Gauss model, μ respectivelyc
And ΣcIt is the mean vector and variance matrix in c-th of Gaussian probability-density function p (x | c) respectively, utilizes EM algorithm
I.e. EM algorithm is trained above-mentioned mixed model；
S32, sample drawn point carry out stable phase verifying to mixed model,
I.e. from Nc-1/ 2 sample points start to verify to find the sample point of the second verification condition of continuous three satisfactions and incite somebody to action
Meet serial number corresponding to the sample point of second verification condition to be labeled asThe verification condition is
Wherein, ρ is preassigned threshold value；
S33, judgementWhether N is equal toc+Nc-1, it is to indicate that i.e. c-th of the stable phase of result convergence has determined that；Otherwise it enablesAnd return step S31 be iterated untilEqual to Nc+Nc-1。
Optionally, described in one of the embodiments, to be determined based on fixed two adjacent stable phase models
Transient phase that may be present includes: between two stable phases
Two stable phase X being had determined from twoc-1And Xc, examined since c-th of stable phase, first sample point
It tests and finds out and continuously meet p (xn| the c) sample point of < ρ is denoted as transient phase Xc-1,c。
In addition, to solve deficiency present in traditional technology, it is also proposed that a kind of phase based on local Gaussian mixture model
Position dividing system.
A kind of phase dividing system based on local Gaussian mixture model, comprising:
Acquisition unit is used for collecting sample and creates historical training dataset；
First Gaussian Profile creating unit is used for according to sample time order from the historical training dataset selector
Sample is divided to create first Gaussian distribution model to determine first stable phase data；
Gauss hybrid models creating unit is used for based on the previous Gauss model having determined, creation includes two
The gauss hybrid models of gauss component are to determine next stable phase and complete with the cooperation of transient phase acquiring unit to whole
The phase of sample data divides；
Transient phase acquiring unit is used to determine corresponding transient phase number based on two adjacent stable phase data
According to.
Optionally, the first Gaussian Profile creating unit includes: in one of the embodiments,
First data acquisition module is used to successively choose preceding N from the historical training dataset1A sample simultaneously calculates it
Mean value and variance are to obtain corresponding gauss of distribution function p (x | 1), wherein and p (x | 1) indicate first Gaussian distribution model
Probability density function, x indicate the sample data of acquisition；
First stable phase authentication module is used for sample drawn point and carries out stable phase verifying, i.e., from N1/ 2 samples
This point starts to verify to find the sample point of the first verification condition of continuous three satisfactions and will meet the sample of first verification condition
The corresponding serial number of this point is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value；
First stable phase determining module, is used to judge N1Whether it is equal toIt is to indicate result convergence i.e. first
A stable phase has determined that and carries out in next step；Otherwise it enablesAnd iteration is re-started by the first stable phase authentication module
Until N1It is equal to
Optionally, the gauss hybrid models creating unit includes: in one of the embodiments,
Second data acquisition module is used for based on the previous Gauss model having determined, creation includes two Gausses
The gauss hybrid models of ingredient are simultaneously trained, it is assumed that preceding c-1 stable phase it has been determined that c is greater than the integer equal to 2,
Formula corresponding to the mixed model is
p(x|θc)=αc-1p(x|c-1)+αcp(x|c)
Wherein, the probability density function p (x | c-1) of the c-1 Gauss model is it has been determined that note includes Nc-1The of a sample
C-1 stable phase is Xc-1, the probability density function p (x | c) of c-th of Gauss model is undetermined, it is assumed that includes NcThe of a sample
C stable phase is Xc, remember Xm={ Xc-1,XcGauss hybrid models thus training data, corresponding training parameter θc=
{αc-1,αc,μc,Σc, αc-1And αcIt is the c-1 combination coefficients with c-th of gauss component in mixed Gauss model, μ respectivelyc
And ΣcIt is the mean vector and variance matrix in c-th of Gaussian probability-density function p (x | c) respectively, utilizes EM algorithm
I.e. EM algorithm is trained above-mentioned mixed model；
Above-mentioned mixed model is trained using EM algorithm, that is, EM algorithm, wherein note Xm={ Xc-1,Xc, institute
Corresponding trained difference θc={ αc-1,αc,μc,Σc}；
Second stable phase authentication module is used for sample drawn point and carries out stable phase verifying to mixed model, i.e., certainly
Nc-1/ 2 sample points start to verify second to be tested with finding the sample point of the second verification condition of continuous three satisfactions and meeting this
Serial number corresponding to the sample point of card condition is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value；
Second stable phase determining module, is used to judgeWhether N is equal toc+Nc-1, it is to indicate result convergence i.e.
C-th of stable phase has determined that；Otherwise it enablesAnd by the second stable phase authentication module re-start iteration untilEqual to Nc+Nc-1。
Optionally, in one of the embodiments, the treatment process of the transient phase acquiring unit include: from two
Through two determining stable phase Xc-1And Xc, examined since c-th of stable phase, first sample point and find out continuous satisfaction
p(xn| the c) sample point of < ρ is denoted as transient phase Xc-1,c。
In addition, to solve deficiency existing for traditional technology, it is also proposed that a kind of computer readable storage medium, including calculate
Machine instruction, when the computer instruction is run on computers, so that computer executes the method.
Implement the embodiment of the present invention, does not account for timing and transient phase in addition to solving in existing phase division methods
The deficiencies of division, the present invention also has following the utility model has the advantages that i.e. (1) present invention is independent with one from angle of Gaussian Profile
Gaussian Profile describes a stable phase, transient phase is described with the mixed model of two adjacent Gaussian Profiles, so that phase
Division methods not only can effectively mark off stable phase, but also can determine transient phase simultaneously；(2) present invention every time only adopt by iteration
Modeling verifying is carried out with the data of part, redundant computation is greatly reduced, improves computational efficiency；(3) present invention is using step
Stable phase and transient phase are gradually determined according to sample time order into formula more new strategy, and there is phase to divide number not
The advantages that needing preassignment and division result not to need subsequent processing.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Wherein:
Fig. 1 a is that initial model updates schematic diagram in one embodiment；
Fig. 1 b is that mixed model updates schematic diagram in one embodiment；
Fig. 2 is that local Gaussian mixture model phase divides schematic diagram in one embodiment；
Fig. 3 is penicillin fermentation process schematic diagram in one embodiment；
Fig. 4 is core flow chart of steps in one embodiment.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Unless otherwise defined, all technical and scientific terms used herein and belong to technical field of the invention
The normally understood meaning of technical staff is identical.Term as used herein in the specification of the present invention is intended merely to description tool
The purpose of the embodiment of body, it is not intended that the limitation present invention.It is appreciated that term " first " used in the present invention, " second "
Etc. can be used to describe various elements herein, but these elements should not be limited by these terms.These terms are only used to by first
A element and another element are distinguished.For example, in the case where not departing from scope of the present application, first element can be claimed
It can be first element by second element for second element, and similarly.First element and second element both element, but
It is not identity element.
Not the deficiencies of timing and transient phase division are not accounted in being directed to existing phase division methods, in the present embodiment
In, spy proposes a kind of phase division methods based on local Gaussian mixture model, by mixed using improved local Gaussian
Model method is closed, using step-type probabilistic Modeling, phase division is carried out to leggy batch process；Pass through foundation office every time
The gauss hybrid models in portion, the i.e. initial model of only one gauss component and the mixed model comprising two gauss components, benefit
With the mode of iteration can stable phase and transient phase in determination process simultaneously, specifically, as shown in figure 4, this method includes
Following steps:
S1, collecting sample simultaneously create historical training dataset；In some specific embodiments, by acquiring batch process
DataAnd it is launched intoObtain historical training dataset；
S2, first Gaussian Profile is created from the historical training dataset selected section sample according to sample time order
Model is the introductory die established only comprising a gauss component to determine first stable phase data, the purpose of this step
Type, initial model update shown in schematic diagram such as Fig. 1 (a)；In some specific embodiments, selected from the historical training dataset
Part sample first Gaussian distribution model of creation, which is selected, to determine first stable phase data includes:
S21, successively from the historical training dataset choose before N1A sample simultaneously calculates its mean value and variance to obtain pair
The Gaussian distribution model p (x | 1) answered, wherein p (x | 1) indicates that the probability density function of first Gaussian distribution model, x indicate
The sample data of acquisition；
S22, sample drawn point carry out stable phase verifying, i.e., from N1/ 2 sample points start verifying to find continuous three
Serial number corresponding to the sample point for meeting first verification condition is simultaneously labeled as by a sample point for meeting the first verification conditionThe verification condition is
Wherein, ρ is preassigned threshold value, such as ρ=0.001, if N1/ 2 be non-integer, then forward be rounded or to
After be rounded, even N1/ 2 be 16.5, then can use 16 or 17；
S23, judge N1Whether it is equal toIt is to indicate that i.e. first stable phase of result convergence has determined that and carry out next
Step；Otherwise it enablesAnd return step S21 is iterated until N1It is equal to
S3, based on the previous Gauss model having determined, gauss hybrid models of the creation comprising two gauss components with
Determine next stable phase, the purpose of this step is to establish the mixed model comprising two gauss components, without loss of generality, by
In preceding c-1 stable phase it has been determined that now determining that out c-th of stable phase, mixed model updates schematic diagram such as Fig. 1
(b) shown in；In some specific embodiments, the creation gauss hybrid models are to determine next stable phase data packet
It includes:
The mixed model of S31, creation comprising two gauss of distribution function is simultaneously trained, public affairs corresponding to the mixed model
Formula is
p(x|θc)=αc-1p(x|c-1)+αcp(x|c)
In which it is assumed that preceding c-1 stable phase is it has been determined that note includes Nc-1The c-1 stable phase of a sample be
Xc-1, include NcC-th of stable phase of a sample is Xc, c is greater than the integer equal to 2；
Above-mentioned mixed model is trained using EM algorithm, that is, EM algorithm, wherein note Xm={ Xc-1,Xc, institute
Corresponding trained difference θc={ αc-1,αc,μc,Σc}；Due to preceding c-1 stable phase it has been determined that such as first gauss component
Mean value and variance determine that then this step only calculates trained difference θ via S2c={ αc-1,αc,μc,Σc?；
S32, sample drawn point carry out stable phase verifying to mixed model, i.e., from Nc-1/ 2 sample points start to verify
To find the sample point of the second verification condition of continuous three satisfactions and will meet corresponding to the sample point of second verification condition
Serial number is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value, such as ρ=0.001；
S33, judgementWhether N is equal toc+Nc-1, it is to indicate that i.e. c-th of the stable phase of result convergence has determined that；Otherwise it enablesAnd return step S31 be iterated untilEqual to Nc+Nc-1。
It is described based on the previous Gauss model having determined, gauss hybrid models of the creation comprising two gauss components with
Determine that next stable phase includes:
S31, based on the previous Gauss model having determined, mixed model of the creation comprising two gauss of distribution function is simultaneously
It is trained, without loss of generality, it is assumed that preceding c-1 stable phase is it has been determined that c is greater than the integer equal to 2, the mixed model
Corresponding formula is
p(x|θc)=αc-1p(x|c-1)+αcp(x|c)
Wherein, the probability density function p (x | c-1) of the c-1 Gauss model is it has been determined that note includes Nc-1The of a sample
C-1 stable phase is Xc-1, the probability density function p (x | c) of c-th of Gauss model is undetermined, it is assumed that includes NcThe of a sample
C stable phase is Xc, remember Xm={ Xc-1,XcGauss hybrid models thus training data, corresponding training parameter θc=
{αc-1,αc,μc,Σc, αc-1And αcIt is the c-1 combination coefficients with c-th of gauss component in mixed Gauss model, μ respectivelyc
And ΣcIt is the mean vector and variance matrix in c-th of Gaussian probability-density function p (x | c) respectively, utilizes EM algorithm
I.e. EM algorithm is trained above-mentioned mixed model；
S32, sample drawn point carry out stable phase verifying to mixed model,
I.e. from Nc-1/ 2 sample points start to verify to find the sample point of the second verification condition of continuous three satisfactions and incite somebody to action
Meet serial number corresponding to the sample point of second verification condition to be labeled asThe verification condition is
Wherein, ρ is preassigned threshold value；
S33, judgementWhether N is equal toc+Nc-1, it is to indicate that i.e. c-th of the stable phase of result convergence has determined that；Otherwise it enablesAnd return step S31 be iterated untilEqual to Nc+Nc-1。
S4, it is described based on fixed two adjacent stable phase models determine between two stable phases there may be
Transient phase, it is as shown in Figure 2 that phase of the present invention divides schematic diagram.It is described based in some specific embodiments
The adjacent stable phase model of determining two determines that transient phase that may be present includes: from two between two stable phases
Two stable phase X having determinedc-1And Xc, examine and find out continuous full since c-th of stable phase, first sample point
Sufficient p (xn| the c) sample point of < ρ is denoted as transient phase Xc-1,c。
S5, S3 and S4 is repeated to complete the phase division to whole sample datas.
In addition, to solve deficiency present in traditional technology, it is also proposed that a kind of phase based on local Gaussian mixture model
Position dividing system comprising:
Acquisition unit is used for collecting sample and creates historical training dataset；In some specific embodiments, pass through
Acquire batch process dataAnd it is launched intoObtain historical training dataset；
First Gaussian Profile creating unit is used for according to sample time order from the historical training dataset selector
Sample is divided to create first Gaussian distribution model to determine first stable phase data；In some specific embodiments,
The first Gaussian Profile creating unit includes:
First data acquisition module is used to successively choose preceding N from the historical training dataset1A sample simultaneously calculates it
Mean value and variance are to obtain corresponding Gaussian distribution model p (x | 1), wherein and p (x | 1) indicate first Gaussian distribution model
Probability density function, x indicate the sample data of acquisition；
First stable phase authentication module is used for sample drawn point and carries out stable phase verifying, i.e., from N1/ 2 samples
This point starts to verify to find the sample point of the first verification condition of continuous three satisfactions and will meet the sample of first verification condition
The corresponding serial number of this point is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value, such as ρ=0.001；
First stable phase determining module, is used to judge N1Whether it is equal toIt is to indicate result convergence i.e. first
A stable phase has determined that and carries out in next step；Otherwise it enablesAnd iteration is re-started by the first stable phase authentication module
Until N1It is equal to
Gauss hybrid models creating unit is used for based on the previous Gauss model having determined, creation includes two
The gauss hybrid models of gauss component are to determine next stable phase and complete with the cooperation of transient phase acquiring unit to whole
The phase of sample data, which divides, determines stable phase by gauss hybrid models creating unit, true by transient phase acquiring unit
Make transient phase；In some specific embodiments, the gauss hybrid models creating unit includes:
Second data acquisition module is used for based on the previous Gauss model having determined, creation includes two Gausses
The mixed model of distribution function is simultaneously trained, without loss of generality, it is assumed that preceding c-1 stable phase is it has been determined that c is greater than
In 2 integer, formula corresponding to the mixed model is
p(x|θc)=αc-1p(x|c-1)+αcp(x|c)
Wherein, the probability density function p (x | c-1) of the c-1 Gauss model is it has been determined that note includes Nc-1The of a sample
C-1 stable phase is Xc-1, the probability density function p (x | c) of c-th of Gauss model is undetermined, it is assumed that includes NcThe of a sample
C stable phase is Xc, remember Xm={ Xc-1,XcGauss hybrid models thus training data, corresponding training parameter θc=
{αc-1,αc,μc,Σc, αc-1And αcIt is the c-1 combination coefficients with c-th of gauss component in mixed Gauss model, μ respectivelyc
And ΣcIt is the mean vector and variance matrix in c-th of Gaussian probability-density function p (x | c) respectively, utilizes EM algorithm
I.e. EM algorithm is trained above-mentioned mixed model；
Second stable phase authentication module is used for sample drawn point and carries out stable phase verifying to mixed model, i.e., certainly
Nc-1/ 2 sample points start to verify second to be tested with finding the sample point of the second verification condition of continuous three satisfactions and meeting this
Serial number corresponding to the sample point of card condition is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value, such as ρ=0.001；
Second stable phase determining module, is used to judgeWhether N is equal toc+Nc-1, it is to indicate result convergence i.e.
C-th of stable phase has determined that；Otherwise it enablesAnd by the second stable phase authentication module re-start iteration untilEqual to Nc+Nc-1。
Transient phase acquiring unit is used to determine corresponding transient phase number based on two adjacent stable phase data
It completes to divide the phase of whole sample datas accordingly.In some specific embodiments, the transient phase acquiring unit
Treatment process includes: two stable phase X having determined from twoc-1And Xc, from c-th of stable phase, first sample point
Start to examine and find out continuously to meet p (xn| the c) sample point of < ρ is denoted as transient phase Xc-1,c。
Based on identical inventive concept, the invention also provides a kind of computer readable storage mediums, including computer to refer to
It enables, when the computer instruction is run on computers, so that computer executes the method.
Based on the above-mentioned technical proposal, it carries out verifying by taking specific experiment example-penicillin fermentation process as an example below of the invention
The schematic diagram of validity, penicillin fermentation process is as shown in Figure 3.
It is specific:
It in collecting sample and creates stage of historical training dataset: generating 20 batches of normal datas altogether here and be used as phase
It divides, and every batch data all joined the white noise that size is N (0,0.04)；The reaction time for setting each batch is
400h, every 1h sampling is primary, therefore each batch includes 400 sample points, and each sample point includes 11 variables, referring to table 1
In.
Table 1
First stable phase data and next stable phase data are being determined according to sample time order and to whole
The phase of sample data divides the stage: the sample point number of first phase is set as the three times of variable number as initial modeling sample
This point, i.e. N1=33.Division result as ρ=0.001 is as shown in table 2, it can be seen that whole process is classified approximately into 10 in table
A stable phase and three transient phases.In first and the 5th, there are three the phases of very little between stable phase.Real attenuation is anti-
During answering, the incipient stage is that the preculture stage is more stable, corresponds to first stable phase.Subsequently into vigorous reaction rank
Section, corresponds to three small stable phases next.Then process enters feedback formula feeding stage, then undergoes process after one section of conversion
It is finally the self-dissolving stage into fermentation stage is stablized.It can be seen that result and real processes stage that the method divides can be well
And it corresponds to.
Table 2
Implement the embodiment of the present invention, will have the following beneficial effects:
In addition to solving the deficiencies of not accounting for timing and transient phase division in existing phase division methods, the present invention
Also with following the utility model has the advantages that i.e. (1) present invention describes one surely with an independent Gaussian Profile from the angle of Gaussian Profile
State phase describes transient phase with the mixed model of two adjacent Gaussian Profiles, so that phase division methods both can be effective
Mark off stable phase, and can determine transient phase simultaneously；(2) present invention every time built only with the data of part by iteration
Mould verifying, greatly reduces redundant computation, improves computational efficiency；(3) present invention uses step-by-step movement more new strategy, according to adopting
Sample time sequencing, gradually determines stable phase and transient phase, and there is phase to divide number and do not need preassignment and divide knot
Fruit does not need the advantages that subsequent processing.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
The limitation to the application the scope of the patents therefore cannot be interpreted as.It should be pointed out that for those of ordinary skill in the art
For, without departing from the concept of this application, various modifications and improvements can be made, these belong to the guarantor of the application
Protect range.Therefore, the scope of protection shall be subject to the appended claims for the application patent.
Claims (8)
1. a kind of phase division methods based on local Gaussian mixture model, include the following steps:
S1, collecting sample simultaneously create historical training dataset；
S2, first Gaussian distribution model is created from the historical training dataset selected section sample according to sample time order
To determine first stable phase；
S3, based on the previous Gauss model having determined, gauss hybrid models of the creation comprising two gauss components are with determination
Next stable phase out；
S4, transient state phase that may be present between two stable phases is determined based on fixed two adjacent stable phase models
Position；
S5, S3 and S4 is repeated to complete the phase division to whole sample datas.
2. phase division methods according to claim 1, which is characterized in that from the historical training dataset selected section
Sample creates first Gaussian distribution model to determine first stable phase data
S21, successively from the historical training dataset choose before N1A sample simultaneously calculates its mean value and variance to obtain corresponding height
This distributed model p (x | 1), wherein p (x | 1) indicates that the probability density function of first Gaussian distribution model, x indicate acquisition
Sample data；
S22, sample drawn point carry out stable phase verifying, i.e., from N1It is full to find continuous three that/2 sample points start verifying
Serial number corresponding to the sample point for meeting first verification condition is simultaneously labeled as by the sample point of the first verification condition of footIt is described
Verification condition is
Wherein, ρ is preassigned threshold value；
S23, judge N1Whether it is equal toIt is to indicate that i.e. first stable phase of result convergence has determined that and carry out in next step；
Otherwise it enablesAnd return step S21 is iterated until N1It is equal to
3. phase division methods according to claim 2, which is characterized in that
It is described based on the previous Gauss model having determined, gauss hybrid models of the creation comprising two gauss components are with determination
Next stable phase includes: out
S31, based on the previous Gauss model having determined, mixed model of the creation comprising two gauss of distribution function and progress
Training, it is assumed that it has been determined that c is greater than the integer equal to 2, formula corresponding to the mixed model is preceding c-1 stable phase
p(x|θc)=αc-1p(x|c-1)+αcp(x|c)
Wherein, the probability density function p (x | c-1) of the c-1 Gauss model is it has been determined that note includes Nc-1C-1 of a sample
Stable phase is Xc-1, the probability density function p (x | c) of c-th of Gauss model is undetermined, it is assumed that includes NcC-th of a sample is steady
State phase is Xc, remember Xm={ Xc-1,XcGauss hybrid models thus training data, corresponding training parameter θc={ αc-1,
αc,μc,Σc, αc-1And αcIt is the c-1 combination coefficients with c-th of gauss component in mixed Gauss model, μ respectivelycAnd Σc
It is the mean vector and variance matrix in c-th of Gaussian probability-density function p (x | c) respectively, utilizes EM algorithm, that is, EM
Algorithm is trained above-mentioned mixed model；
S32, sample drawn point carry out stable phase verifying to mixed model,
I.e. from Nc-1/ 2 sample points start to verify to find the sample point of the second verification condition of continuous three satisfactions and will meet
Serial number corresponding to the sample point of second verification condition is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value；
S33, judgementWhether N is equal toc+Nc-1, it is to indicate that i.e. c-th of the stable phase of result convergence has determined that；Otherwise it enablesAnd return step S31 be iterated untilEqual to Nc+Nc-1。
4. phase division methods according to claim 3, which is characterized in that described to be based on fixed two adjacent stable states
Phase model determines that transient phase that may be present includes: two stable states having determined from two between two stable phases
Phase Xc-1And Xc, examine and find out since c-th of stable phase, first sample point and continuously meet p (xn| the c) sample of < ρ
Point is denoted as transient phase Xc-1,c。
5. a kind of phase dividing system based on local Gaussian mixture model characterized by comprising
Acquisition unit is used for collecting sample and creates historical training dataset；
First Gaussian Profile creating unit is used for according to sample time order from the historical training dataset selected section sample
First Gaussian distribution model of this creation is to determine first stable phase data；
Gauss hybrid models creating unit is used for based on the previous Gauss model having determined, creation includes two Gausses
The gauss hybrid models of ingredient are to determine next stable phase and complete with the cooperation of transient phase acquiring unit to whole samples
The phase of data divides；
Transient phase acquiring unit is used to determine two stable phases based on fixed two adjacent stable phase models
Between transient phase that may be present.
6. system according to claim 5, which is characterized in that the first Gaussian Profile creating unit includes:
First data acquisition module is used to successively choose preceding N from the historical training dataset1A sample simultaneously calculates its mean value
With variance to obtain corresponding gauss of distribution function p (x | 1), wherein p (x | 1) indicates the probability of first Gaussian distribution model
Density function, x indicate the sample data of acquisition；
First stable phase authentication module is used for sample drawn point and carries out stable phase verifying, i.e., from N1/ 2 sample points are opened
Verifying begin to find the sample point of the first verification condition of continuous three satisfactions and will meet the sample point institute of first verification condition
Corresponding serial number is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value；
First stable phase determining module, is used to judge N1Whether it is equal toIt is to indicate that result restrains i.e. first surely
State phase has determined that and carries out in next step；Otherwise it enablesAnd by the first stable phase authentication module re-start iteration until
N1It is equal to
7. system according to claim 6, which is characterized in that the gauss hybrid models creating unit includes:
Second data acquisition module is used for based on the previous Gauss model having determined, creation includes two gauss components
Gauss hybrid models and be trained, it is assumed that for preceding c-1 stable phase it has been determined that c is greater than the integer equal to 2, this is mixed
Formula corresponding to molding type is
p(x|θc)=αc-1p(x|c-1)+αcp(x|c)
Wherein, the probability density function p (x | c-1) of the c-1 Gauss model is it has been determined that note includes Nc-1C-1 of a sample
Stable phase is Xc-1, the probability density function p (x | c) of c-th of Gauss model is undetermined, it is assumed that includes NcC-th of a sample is steady
State phase is Xc, remember Xm={ Xc-1,XcGauss hybrid models thus training data, corresponding training parameter θc={ αc-1,
αc,μc,Σc, αc-1And αcIt is the c-1 combination coefficients with c-th of gauss component in mixed Gauss model, μ respectivelycAnd Σc
It is the mean vector and variance matrix in c-th of Gaussian probability-density function p (x | c) respectively, utilizes EM algorithm, that is, EM
Algorithm is trained above-mentioned mixed model；
Second stable phase authentication module is used for sample drawn point and carries out stable phase verifying to mixed model, i.e., from the
Nc-1/ 2 sample points start to verify to find the sample point of the second verification condition of continuous three satisfactions and will meet second verifying
Serial number corresponding to the sample point of condition is labeled asThe verification condition is
Wherein, ρ is preassigned threshold value；
Second stable phase determining module, is used to judgeWhether N is equal toc+Nc-1, it is to indicate that result restrains i.e. c-th
Stable phase has determined that；Otherwise it enablesAnd by the second stable phase authentication module re-start iteration untilDeng
In Nc+Nc-1。
8. system according to claim 7, which is characterized in that the treatment process of the transient phase acquiring unit includes:
Two stable phase X being had determined from twoc-1And Xc, examine and find out since c-th of stable phase, first sample point
Continuously meet p (xn| the c) sample point of < ρ is denoted as transient phase Xc-1,c。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910571289.5A CN110309491B (en) | 2019-06-26 | 2019-06-26 | Transient phase partitioning method and system based on local Gaussian mixture model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910571289.5A CN110309491B (en) | 2019-06-26 | 2019-06-26 | Transient phase partitioning method and system based on local Gaussian mixture model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110309491A true CN110309491A (en) | 2019-10-08 |
| CN110309491B CN110309491B (en) | 2022-10-14 |
Family
ID=68077805
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910571289.5A Active CN110309491B (en) | 2019-06-26 | 2019-06-26 | Transient phase partitioning method and system based on local Gaussian mixture model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110309491B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130124438A1 (en) * | 2011-11-11 | 2013-05-16 | Sung-young Lee | Method of recognizing patterns based on markov chain hidden conditional random field model |
| CN107451101A (en) * | 2017-07-21 | 2017-12-08 | 江南大学 | It is a kind of to be layered integrated Gaussian process recurrence soft-measuring modeling method |
| CN108804784A (en) * | 2018-05-25 | 2018-11-13 | 江南大学 | A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models |
-
2019
- 2019-06-26 CN CN201910571289.5A patent/CN110309491B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130124438A1 (en) * | 2011-11-11 | 2013-05-16 | Sung-young Lee | Method of recognizing patterns based on markov chain hidden conditional random field model |
| CN107451101A (en) * | 2017-07-21 | 2017-12-08 | 江南大学 | It is a kind of to be layered integrated Gaussian process recurrence soft-measuring modeling method |
| CN108804784A (en) * | 2018-05-25 | 2018-11-13 | 江南大学 | A kind of instant learning soft-measuring modeling method based on Bayes's gauss hybrid models |
Non-Patent Citations (2)
| Title |
|---|
| 王仲等: "基于数据驱动的全工况下燃气轮机基准值确定", 《自动化仪表》 * |
| 王凯南等: "基于高斯混合模型的EM算法改进与优化", 《工业控制计算机》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110309491B (en) | 2022-10-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116503676B (en) | Picture classification method and system based on knowledge distillation small sample increment learning | |
| CN110689136B (en) | Deep learning model obtaining method, device, equipment and storage medium | |
| CN114356578B (en) | Parallel computing method, device, equipment and medium for natural language processing model | |
| Nityasuddhi et al. | Asymptotic properties of the EM algorithm estimate for normal mixture models with component specific variances | |
| CN117290721A (en) | Digital twin modeling method, device, equipment and medium | |
| CN114970830A (en) | Flexible communication method for accelerating data parallel distributed deep learning training | |
| CN103605493A (en) | Parallel sorting learning method and system based on graphics processing unit | |
| CN110070120B (en) | Depth measurement learning method and system based on discrimination sampling strategy | |
| CN110309491A (en) | A kind of transient phase division methods and system based on local Gaussian mixture model | |
| CN108427773B (en) | Distributed knowledge graph embedding method | |
| CN110889251A (en) | Variable fidelity proxy model establishing method based on two-step self-adaptive dotting strategy | |
| CN115795355A (en) | Classification model training method, device and equipment | |
| CN108021985A (en) | A kind of model parameter training method and device | |
| AU2020103440A4 (en) | A method for optimizing the convergence performance of data learning with minimal computational steps | |
| CN116994015A (en) | Self-distillation classification method based on progressive knowledge transfer | |
| Hu et al. | Pwsnas: powering weight sharing nas with general search space shrinking framework | |
| CN111767981B (en) | Approximate calculation method of Mish activation function | |
| CN113837293A (en) | mRNA subcellular localization model training method, mRNA subcellular localization model localization method and readable storage medium | |
| CN108206024B (en) | Voice data processing method based on variational Gaussian regression process | |
| CN112738724A (en) | Method, device, equipment and medium for accurately identifying regional target crowd | |
| CN114580578B (en) | Method and device for training distributed random optimization model with constraints and terminal | |
| CN115130592B (en) | Sample generates chip | |
| CN112750074B (en) | Small sample image feature enhancement method and system and image classification method and system | |
| CN113296964B (en) | Data processing method and device | |
| CN112529081B (en) | Real-time semantic segmentation method based on efficient attention calibration |
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 |