CN107341359A - Stalk fermentation produces the flexible measurement method of ethanol process key parameters - Google Patents

Stalk fermentation produces the flexible measurement method of ethanol process key parameters Download PDF

Info

Publication number
CN107341359A
CN107341359A CN201710685210.2A CN201710685210A CN107341359A CN 107341359 A CN107341359 A CN 107341359A CN 201710685210 A CN201710685210 A CN 201710685210A CN 107341359 A CN107341359 A CN 107341359A
Authority
CN
China
Prior art keywords
degree
sample
variable
data
stalk fermentation
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.)
Pending
Application number
CN201710685210.2A
Other languages
Chinese (zh)
Inventor
姜哲宇
陈永琪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Marine Growth Engineering Equipment Co Ltd Of Jiangsu Section
Jiangsu University
Original Assignee
Marine Growth Engineering Equipment Co Ltd Of Jiangsu Section
Jiangsu University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Marine Growth Engineering Equipment Co Ltd Of Jiangsu Section, Jiangsu University filed Critical Marine Growth Engineering Equipment Co Ltd Of Jiangsu Section
Priority to CN201710685210.2A priority Critical patent/CN107341359A/en
Publication of CN107341359A publication Critical patent/CN107341359A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Z—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS, NOT OTHERWISE PROVIDED FOR
    • G16Z99/00—Subject matter not provided for in other main groups of this subclass

Landscapes

  • Preparation Of Compounds By Using Micro-Organisms (AREA)
  • Feedback Control In General (AREA)

Abstract

The flexible measurement method of stalk fermentation production ethanol process key parameters is a kind of method for the On-line Estimation for being used to solving the problems, such as to be difficult to the critical biochemical amount with physical sensors On-line sampling system during stalk fermentation.This method passes through the analysis to stalk fermentation process mechanism first, choose suitable auxiliary variable and training sample database is established according to history tank batch data, sample data is divided using core Fuzzy c-means Clustering and obtains the degree of membership of each cluster, DE LSSVM soft-sensing models are carried out to the data of each classification, finally realized to concentration of alcohol y1, total sugar content y2, cell concentration y3Online real-time soft measuring.This method relies on hardware platform and measuring instrumentss, the computer system and intelligent controller for carrying out software calculating, and the software obtains real-time process data by measuring instrumentss and carries out hard measurement.The present invention realizes stalk fermentation process key state variable and predicted in real time online, the optimal control and optimization for realizing stalk fermentation process is run significant.

Description

Stalk fermentation produces the flexible measurement method of ethanol process key parameters
Technical field
The present invention is a kind of for solving to be difficult to be surveyed in real time online with physical sensors in stalk fermentation production ethanol process The concentration of alcohol of amount, total sugar concentration, the On-line Estimation problem of cell concentration these three critical biochemical variables, belong to hard measurement and soft The technical field of instrument constitution.
Background technology
In many Industry Control occasions, variable as a major class be present：They are closely related with product quality, it is necessary to tight Lattice control, but technology or it is economical due to, be still difficult at present or can not directly be detected by physical sensors. Typical example has the product component concentration of rectifying column, the reactant concentration of chemical reactor and product distribution, and biology hair Fermentation tank mesophytization variable etc..In order to solve the problems, such as the measurement of this class variable, soft-measuring technique arises at the historic moment.So-called hard measurement is just It is according to certain criterion, selects one group not only there are close ties easily to measure again with being estimated variable (be measured or leading variable) Direct measurable variable (i.e. auxiliary variable), by constructing certain functional relation, with computer software realize to be measured Estimation.The flexible measurement method used at present, majority is based on lineary system theory, for non-linear as chemical industry, biochemical process The serious complex process of characteristic, this method can only be in the working regions of very little effectively, it is impossible to solve whole working region The hard measurement problem of measured variable.
Least square method supporting vector machine (LS-SVM) modeling method uses structural risk minimization and nuclear technology, due to It is applied to finite sample, nonlinear problem, so its application in hard measurement field formed based on least square support The flexible measurement method of vector machine, for the solution of the hard measurement problem of biochemical, chemical process critical biochemical variable, there is provided Qiang You The means of power.Practice have shown that radial direction sound stage width degree and penalty coefficient are very big on performance of modeling influence in LS-SVM model process.Simply Experiment to join method surely time-consuming.Differential evolution (DE) algorithm possesses that ability of searching optimum is strong, control parameter is few, easy to use, is excellent Change a kind of suitable method of LS-SVM model parameters.
During stalk fermentation is handled, with the progress of fermentation process, fermentation parameter changes therewith, single regression model No longer adapt to new operating mode.Therefore, it is necessary to seek new method, this method can be according to the different stage of fermenting to hard measurement mould Type is corrected.
Therefore the present invention is classified the fermentation sample data of collection using core Fuzzy c-Means Clustering Algorithm, with DE- LSSVM is combined, and establishes a kind of multi-model (KFCM-DE-LSSVM) stalk fermentation soft-sensing model, and design matched Soft instrument constructs.
The content of the invention
Technical problem：It is an object of the invention to provide a kind of hard measurement side of stalk fermentation production ethanol process key parameters Method.I.e. a kind of stalk fermentation process is extremely important it can be difficult to measuring with physical sensors On-line sampling system or in real time cost The On-line Estimation method of very high biochemical variable (such as cell concentration, total sugar concentration and concentration of alcohol) and corresponding soft instrument Building method.
Technical scheme：A kind of flexible measurement method specific steps of stalk fermentation production ethanol process key parameters of the present invention It is as follows:
Step 1, choosing auxiliary variables：Energy direct measurement and the external variable closely related with process are chosen, with consistent phase The degree of association of pass degree method analysis external variable and key stato variable, takes degree of association rij>=0.7 external variable is as hard measurement The auxiliary variable of model；
Step 2, establish tranining database：Gather the auxiliary variable of some history tank batches and crucial shape under same process State variable data, the set of construction input and output vector pair, generate muscle-setting exercise sample database；Wherein input vector is auxiliary Variable, output vector are key stato variable；
Step 3, training sample is obtained using step 2, carrying out clustering using core Fuzzy c-means Clustering chooses cluster Number k=3, calculates degree of membership of each sample in nuclear space；
Step 4, the sample of each cluster is entered with the least square method supporting vector machine DE-LSSVM based on differential evolution Row modeling, obtains stalk fermentation key parameters hard measurement submodelUpdated by subordinated-degree matrix Formula U：Obtain the degree of membership u of each submodelij, finally combine degree of membership and submodel obtain To multi-model stalk fermentation soft-sensing modelWherein, x is input variable, aiIt is Lagrange multiplier, K (x,xi) it is positive definite kernel function, b is threshold value, and i represents i-th of data, uijFor data xiTo the degree of membership of j-th of cluster centre Value, vjFor cluster centre, m is weighted number, and k is to cluster number, K (xi,vj) it is gaussian kernel function；
Step 5, key stato variable prediction：Using the soft-sensing model trained, according to current tank to be predicted batch Vector is newly entered, after soft-sensing model is established, is realized using embedded type C Programming with Pascal Language, and be embedded into intelligent controller In, when the input vector of tank to be predicted batch, after measuring instrumentss read in intelligent controller, intelligent controller utilizes hard measurement program The predicted value of key stato variable is calculated, and prediction result is sent on host computer through data channel and shown.
Wherein：
It is described with the closely related external variable of process be fermentation jar temperature t, fermentation tank pressure p, motor speed of agitator r, Fermentating liquid volume v, air mass flow q, CO2Release rate u, glucose feeding speed ρ, ammonia aqua stream rate of acceleration η, dissolved oxygen DO, fermentation Liquid acidity-basicity ph, key stato variable are concentration of alcohol X, total sugar content P and cell concentration S.
The degree of association of the analysis external variable and key stato variable, it is specially：The fermentation jar temperature degree of association= 0.927, the fermentation pressure tank degree of association=0.348, the motor speed of agitator degree of association=0.143, the fermentating liquid volume degree of association= 0.8475, air mass flow degree of association=0.563, CO2The release rate degree of association=0.946, glucose feeding Rate relating extent= 0.725, the dissolved oxygen degree of association=0.296, the zymotic fluid acid-base value degree of association=0.787.
Input vector described in step 5 is fermentating liquid volume v, air mass flow q, CO2Release rate u, dissolved oxygen and fermentation Liquid acidity-basicity ph；Output vector is concentration of alcohol X, total sugar content P and cell concentration S.
Clustering is carried out using core Fuzzy c-means Clustering, calculates degree of membership of each sample in nuclear space by following Step is carried out：
If stalk fermentation collecting sample collection：X={ x1,x2,…,xn, create k cluster, in data with sample set X It is similar for one group；It is dissimilar that data are classified by seeking the minimum value J of object function as far as possible not at one group,It is constrained in：
Wherein k is to cluster number, xiFor i-th of input data, vjFor cluster centre, uijFor data xiTo in j-th of cluster The heart is subordinate to angle value, and m is weighted number, and U is subordinated-degree matrix more new formula, and V is cluster centre calculation formula.
The degree of membership of each submodel of basis described in step 4 obtains final multi-model soft-sensing model, according to following Process is realized:
By described object function, the classification of data sample is obtained, carries out DE-LSSVM data modeling：
1) original sample is normalized：By the sample [x ' after normalizing1,y′1],…,[x′n,y′n] be divided into it is N number of Training sample and M test sample,
2) DE algorithm parameters are initialized：Current algebraically G=0 is made, according to penalty coefficient C and the upper and lower limit of radial direction base width cs Produce one group of random [C1,σ1]；
3) by N number of training sample [x '1,…,x′N] input as LS-SVM, current [C, σ] be used as parameter, training LS- SVM obtains corresponding output [y "1,…,y″N]；
4) output [y " after training1,…,y″N] and reality output [y '1,…,y′N] square-error and result as DE Whether the object function of algorithm, error in judgement meet whether requirement or G are equal to Gmax；If meeting one of which, step is gone to 7), otherwise into 5)；
5) G=G+1；
6) row variation, intersection, selection operation are entered to [C, σ], produces new [C, σ], return to step 3)；
7) parameter for obtaining optimal penalty coefficient C and radial direction base width cs as LSSVM carries out soft sensor modeling；
Obtaining every a kind of output is：
U in formulaijRepresent jth class fuzzy membership, f corresponding to i-th of samplei(x) i-th of submodel is represented.
Beneficial effect：The online process data that the present invention is provided using computer system and conventional instrumentation, only Lead to too small amount of artificial sample, realize the soft sensor modeling based on multi-model of the key stato variable of stalk fermentation process. Solves the problem for not having state variable on-line checking instrument to be difficult to on-line checking in fermentation process.Phase is chemically examined with manual sampling Than reducing the workload of site operation personnel, reducing the uncertainty of the measurement that manual operation introduces in fermentation process, carry High measurement it is ageing, reduce and sample the problem of data brought lag offline.The present invention by core Fuzzy c-means Clustering with It is combined using the least square method supporting vector machine of differential evolution, this hair compared with traditional SVMs flexible measurement method Overcome in bright middle flexible measurement method SVMs to parameter selection and the most suitable parameter of stalk fermentation with fermentation time change and The problem of change.In addition, the present invention considers the factor for influenceing stalk fermentation process key state variable comprehensively, it is a large amount of using existing Some conventional detection signals realize the on-line prediction of key stato variable, using it is simple, easy, cost is relatively low, hard measurement knot Fruit is also more accurate.This method contributes to the real optimal control in stalk fermentation process and optimization to run.
Brief description of the drawings
Stalk fermentation key variables soft-sensing model Establishing process figures of the Fig. 1 based on multi-model；
Fig. 2 a, Fig. 2 b, Fig. 2 c are the comparison diagrams of key stato variable predicted value and actual value；Wherein, Fig. 2 a are that ethanol is dense Spend hard measurement value to contrast with actual value, Fig. 2 b are that total sugar concentration hard measurement value contrasts with actual value, and Fig. 2 c are the soft surveys of cell concentration Value contrasts with actual value,
Fig. 3 a, Fig. 3 b, Fig. 3 c are key stato variable Relative Error curve maps.Wherein, Fig. 3 a are that ethanol content is soft Curve of the Measuring Error, Fig. 3 b are total sugar content hard measurement error curves, and Fig. 3 c are cell concentration hard measurement error curves.
Embodiment
The system includes biological fermentation tank, steam generator, air compressor, air cleaner, water system, intelligence control Device and host computer processed, the input of biological fermentation tank pass through pipeline and steam generator, air cleaner and water system Connect, air compressor connects with air cleaner, water system, the dissolved oxygen electrode of device, pressure difference sensing on biological fermentation tank Device, CO2Gas sensing electrode, air flow sensor and pH electrodes are used to gather stalk fermentation process data and send intelligent control to Device, intelligent controller is used to establishing tranining database according to fermentation process data, data divisions carried out according to degree of membership, it is soft to establish Measurement model, and vector is newly entered according to current tank to be predicted batch using the more soft-sensing models trained, closed The predicted value of strong state variable, is finally sent to host computer through data channel by prediction result and shows.
The implementing procedure figure shown in examples of implementation and Fig. 1 predicted below in conjunction with stalk fermentation process key state variable, Of the embodiment of the present invention is described in detail:
Fermentation jar temperature-t, machine speed of agitator-r, fermentating liquid volume-v
Air mass flow-q, CO2Release rate-u, glucose feeding speed-ρ,
Ammonia aqua stream rate of acceleration-η, dissolved oxygen DO, zymotic fluid acid-base value-pH.
Mechanical agitating fermentation tank specification is nominal volume 50L, tank body Φ 400x750cm, coefficient town 70%；Design pressure Power ability 0.35MPa, can bear 0.11MPa negative pressure, and steam generator electric thermal power is 11KW, steam production 12Kg/h, air Filter uses highly effective air sterilizing filter, 2 grades of filter efficiencies 99.99%.
With automatically controlling for intelligent controller (single-chip microcomputer) optimized integration control loop, and according to model need to filter, count Calculation obtains fermentating liquid volume v, air mass flow q, CO2Release rate u, dissolved oxygen DO and zymotic fluid acidity-basicity ph.
Above-mentioned process data is read, monitor is realized with the visual c++ software of Microsoft in computer is monitored Machine interface.
Soft-sensing model is to use C language programming realization, and in data storage DB modules, model data is provided in monitoring system Interface is changed, model parameter is changed for off-line analysis.Soft-sensor software is run on intelligent controller, is effectively guaranteed mould Type exports ageing, facilitates the process monitoring of system.
Tank temperature control system is in 120~180r/min, throughput in 36 DEG C ± 1 DEG C, the control of mixer rotating speed in fermentation process 0.2L/min, pH control are 6.0 ± 0.2, and the voltage-controlled system of fermentation tank is in 0.2MPa ± 0.01MPa.
Zymotic fluid is centrifuged offline inspection after device 14 separates and obtains concentration of alcohol, total sugar concentration and cell concentration.
Consistent Controlling UEP
For obtained external variable data (fermentation jar temperature t, fermentation tank pressure p, motor speed of agitator r, fermented liquid Product v, air mass flow q, CO2Release rate u, glucose feeding speed ρ, ammonia aqua stream rate of acceleration η, dissolved oxygen DO, zymotic fluid acid-base value PH) its associating with key stato variable (concentration of alcohol X, total sugar concentration S and cell concentration P) is analyzed with consistent degree of correlation method Degree, takes auxiliary variable of the larger external variable of the degree of association as soft-sensing model.
With fermentation jar temperature t, exemplified by cell concentration X, specific algorithm is as follows:
Wherein vij(k) it is rate of change related system, rijFor the degree of association, β is influence of the data variation rate to the degree of association, ξij For variant correlation coefficient.For fermentation jar temperature t and concentration of alcohol X, provided with m1Individual trend identical pointm2It is individual become The point of gesture onrelevantm3The opposite point of individual trendSubstituting into above formula can obtain:
Wherein Pij, Zij, Nij.Positive association degree, zero degree of association and negative customers degree are represented respectively.When | Pij|≥|Nij| when, hair Fermentation tank temperature t and concentration of alcohol X are based on positive correlation, and their variation tendency is similar, and degree of correlation is by rij,PijTwo factors Size determines.Work as rij=ZijWhen=0, fermentation jar temperature t and concentration of alcohol are unrelated；|Pij|≤|Nij| fermentation jar temperature t and second Based on determining alcohol correlation, i.e., they variation tendency it is opposite, degree of correlation is by by rij, PijTwo factor sizes determine.External variable It is as shown in table 1 with concentration of alcohol X calculation of relationship degree result
The calculation of relationship degree value of the external variable of table 1
External variable The degree of association
Fermentation jar temperature t 0.927
Fermentation tank pressure p 0.348
Motor speed of agitator r 0.143
Fermentating liquid volume v 0.875
Air mass flow q 0.563
CO2Release rate u 0.946
Glucose feeding speed ρ 0.725
Dissolved oxygen DO 0.296
Zymotic fluid acidity-basicity ph 0.787
From upper result of calculation, angle value r is associated with the setting of fermentation process experience by correlation analysisij>=0.7 bar Under part, external variable can be surveyed -- fermentation jar temperature t, fermentating liquid volume v, CO2Release rate u, glucose feeding speed ρ, zymotic fluid Acidity-basicity ph and the concentration of alcohol X during stalk fermentation are mostly concerned, select above-mentioned five variables as soft-sensing model Auxiliary variable.
2. establish training sample database:
Stalk fermentation process forms sample according to following structure, and collects some history tank batch training under same process Sample data.Sample is expressed as { xI,yI, wherein xIFor the input of sample, that is, the auxiliary vector chosen-fermentation jar temperature t, hair Zymotic fluid volume v, CO2Release rate u, glucose feeding speed ρ, zymotic fluid acidity-basicity ph.The output of sample is dominated for be predicted Variable-concentration of alcohol X, total sugar content S, cell concentration P.Training sample acquisition and recording structure such as table 2, the time is fermentation process In the middle sampling period, to reduce the offline Error and Assay of leading variable, entered according to same sample leading variable using result of laboratory test three times Row sample is accepted or rejected, and is finally averaged：
The sample data structure of table 2
Should be representative in view of sample data, and coverage is wider as far as possible, should at least include fermentation Process normal range of operation, manually regulation and control fermentation pressure tank, fermentation jar temperature and motor speed of agitator, permit in production technology Perhaps the operating point of fermentation process, sample examination after each operating condition changes system steadily are changed in the range of as far as possible.
3. data sample carries out clustering
The data application core Fuzzy c-means Clustering obtained from step 2 is classified, and calculates respective degree of membership. Comprise the following steps that：
If stalk fermentation collecting sample collection：X={ x1,x2,…,xn, create cluster number k, in data with sample set X is similar for one group；It is dissimilar as far as possible not at one group.Data are classified by seeking the minimum value J of object function.
It is constrained in：
Wherein k is to cluster number, vjFor cluster centre, uijFor data xiAngle value is subordinate to j-th cluster centre, m is Weighted number.
Introduce Nonlinear Mapping φ：X → φ (x), the sample distance definition of feature space are:
||φ(xi)-φ(xj)||2=K (xi,xj)+K(vj,vj)-2K(xi,vj) (5)
Wherein K is kernel function.The present invention uses a kind of Definite core：
K (x, y)=- (| | x-y | |2+b2)1/2, b ∈ R (6)
B=1 is made, positive definite kernel is substituted into formula (3), can obtain new object function is：
Lagrangian is constructed, local derviation is asked to v, u respectively, obtained cluster centre vjWith subordinated-degree matrix U renewal Formula is as follows：
4. establish soft-sensing model
The data divided by fuzzy clustering obtained with step 3 establish DE-LSSVM soft-sensing models.The present invention is to use LSSVM establishes stalk fermentation soft-sensing model, recycles differential evolution algorithm to the punishment parameter C and radial direction sound stage width in LS-SVM Degree σ is optimized.
In given one group of sample set { (xi,yi) | i=1,2,3....l }, xi∈RnSample input, y are tieed up for ni∈ R are sample Output.By approaching sample data, Function Fitting problem can be described as optimization problem:
In formula, C is penalty coefficient, and e is allowable error.Lagrangian is constructed, and local derviation is asked simultaneously to wherein each variable Arrangement obtains system of linear equations：
In formula:
Q=[1 ..., 1]T, a=[a1,a2,…,al]T, y=[y1,y2,…,yl]T,
K is kernel matrix.Can obtain kernel function according to Mercer conditions is：
The a and b in formula (9) are tried to achieve with least square method, the output that can to sum up obtain least square method supporting vector machine is：
Kernel function is used as using RBF (RBF) herein：
Wherein σ is radial direction sound stage width degree.
The flow of differential evolution algorithm is from randomly generating initial populationStart, NPPopulation Size is represented, i is population number, G For current algebraically.
(1) in mutation operation, to randomly generating initial populationMutation operation, which is carried out, by formula (13) obtains new individual
Wherein F is zoom factor.h1,h2,h3∈(1,2,…,NP) it is different and different from i random number.
(2) it is right in crossover operationWithCrossover operation generation experimental subjects are carried out according to formula (14)
Wherein, CR is crossover operator of the scope between [0,1], random numbers of the rand (j) between [0,1].
(3) in selection operation, for minimization problem, of the low individual of selection target functional value as new population Body, i.e., such as formula (15).
Wherein, f () is object function.
The soft-sensing model is derived according to procedure below：
(1) original sample is normalized.By the sample [x ' after normalizing1,y′1],…,[x′n,y′n] it is divided into N Individual training sample and M test sample.
(2) population N is initializedP, mutation operator F, evolution maximum iteration Gmax, crossing-over rate CR, terminate threshold value, punishment Coefficient C and radial direction base width cs upper and lower value.G=0 is made, one group of random [C is produced according to C and σ upper and lower limit1,σ1]。
(3) by N number of training sample [x '1,…,x′N] input as LS-SVM, current [C, σ] be used as parameter, training LS- SVM obtains corresponding output [y "1,…,y″N]。
(4) output [y " after training1,…,y″N] and reality output [y '1,…,y′N] square-error and result conduct Whether the object function of DE algorithms, error in judgement meet whether requirement or G are equal to Gmax.If meeting one of which, step is gone to Suddenly (7), otherwise (5) are entered.
(5) G=G+1.
(6) row variation, intersection, selection operation are entered to [C, σ].Produce new [C, σ], return to step (3).
(7) parameter for obtaining optimal penalty coefficient C and radial direction base width cs as LSSVM carries out soft sensor modeling.
Multi-model DE-LSSVM soft sensor modeling thoughts are：, will using effective classification of core Fuzzy c-Means Clustering Algorithm Sample data set X is divided into { Xi| i=1,2 ..., k }, k cluster.To each XiDE-LSSVM is respectively adopted to be trained, obtains It is to every a kind of output：
X in formulai, for the input of sample, that is, input vector fermentation jar temperature t, fermentating liquid volume v, the CO chosen2Release rate U, glucose feeding speed ρ, zymotic fluid acidity-basicity ph；F (x) be leading variable to be predicted-mycelial concentration X, matrix depth S, Production concentration P；
To m obtained output function, last result is integrated using fuzzy membership：
U in formulaikRepresent the i-th class fuzzy membership, f corresponding to k-th of samplei(x) i-th of submodel is represented.
5. predict key stato variable
After soft-sensing model is established, realized, and be embedded into intelligent controller 15 using embedded type C Programming with Pascal Language, As the input vector x of tank to be predicted batchi+1, after measuring instrumentss read in intelligent controller 15, intelligent controller 15 utilizes hard measurement The predicted value of key stato variable is calculated in program, and prediction result is sent on host computer through data channel and shown Show.
The present invention be the stalk fermentation soft-sensing model process based on multi-model during normal operation, can be according to fermenting Journey is newly entered vector forecasting key stato variable.Fig. 2 gives key stato variable (mycelial concentration, total sugar concentration, product Concentration) predicted value is to the tracking effects of true laboratory values.Fig. 3 gives Relative Error curve, it can be seen that second Determining alcohol X maximum relative error is 2.23%, and total sugar concentration P maximum relative error is 3.13%, cell concentration S maximum Relative error is 1.24%, and its variation tendency has approached truth well.This shows the fuzzy support that the present invention is proposed Vector machine modeling method is effective, reliable, can improve key stato variable (mycelial concentration, total reducing sugar during stalk fermentation Concentration, production concentration) hard measurement precision, reached the set goal satisfiedly, it is online to solve key stato variable The problem of precision of hard measurement is not high, laid a solid foundation to implement optimal control.

Claims (6)

  1. A kind of 1. flexible measurement method of stalk fermentation production ethanol process key parameters, it is characterised in that, comprise the following steps that:
    Step 1, choosing auxiliary variables：Energy direct measurement and the external variable closely related with process are chosen, with the consistent degree of correlation Method analyzes the degree of association of external variable and key stato variable, takes degree of association rij>=0.7 external variable is as soft-sensing model Auxiliary variable；
    Step 2, establish tranining database：The auxiliary variable of some history tank batches and key state under same process is gathered to become Data are measured, the set of construction input and output vector pair, generate muscle-setting exercise sample database；Wherein input vector is that auxiliary becomes Amount, output vector is key stato variable；
    Step 3, training sample is obtained using step 2, carrying out clustering using core Fuzzy c-means Clustering chooses cluster numbers k =3, calculate degree of membership of each sample in nuclear space；
    Step 4, the sample of each cluster is built with the least square method supporting vector machine DE-LSSVM based on differential evolution Mould, obtain stalk fermentation key parameters hard measurement submodelPass through subordinated-degree matrix more new formula U：Obtain the degree of membership u of each submodelij, finally obtained with reference to degree of membership and submodel more Model stalk fermentation soft-sensing modelWherein, x is input variable, aiIt is Lagrange multiplier, K (x, xi) It is positive definite kernel function, b is threshold value, and i represents i-th of data, uijFor data xiAngle value, v are subordinate to j-th cluster centrejFor Cluster centre, m are weighted number, and k is to cluster number, K (xi,vj) it is gaussian kernel function；
    Step 5, key stato variable prediction：Using the soft-sensing model trained, according to the newest of current tank to be predicted batch Input vector, after soft-sensing model is established, realized, and be embedded into intelligent controller using embedded type C Programming with Pascal Language, when The input vector of tank to be predicted batch, after measuring instrumentss read in intelligent controller, intelligent controller is calculated using hard measurement program The predicted value of key stato variable is obtained, and prediction result is sent on host computer through data channel and shown.
  2. 2. the flexible measurement method of stalk fermentation production ethanol process key parameters according to claim 1, it is characterised in that The described and closely related external variable of process is fermentation jar temperature t, fermentation tank pressure p, motor speed of agitator r, fermented liquid Product v, air mass flow q, CO2Release rate u, glucose feeding speed ρ, ammonia aqua stream rate of acceleration η, dissolved oxygen DO, zymotic fluid acid-base value PH, key stato variable are concentration of alcohol X, total sugar content P and cell concentration S.
  3. 3. the flexible measurement method of stalk fermentation production ethanol process key parameters according to claim 1, it is characterised in that The degree of association of the analysis external variable and key stato variable, it is specially：The fermentation jar temperature degree of association=0.927, fermentation tank The pressure degree of association=0.348, the motor speed of agitator degree of association=0.143, the fermentating liquid volume degree of association=0.8475, air mass flow Degree of association=0.563, CO2The release rate degree of association=0.946, glucose feeding Rate relating extent=0.725, the dissolved oxygen degree of association =0.296, the zymotic fluid acid-base value degree of association=0.787.
  4. 4. the flexible measurement method of stalk fermentation production ethanol process key parameters according to claim 1, it is characterised in that Input vector described in step 5 is fermentating liquid volume v, air mass flow q, CO2Release rate u, dissolved oxygen and zymotic fluid acid-base value pH；Output vector is concentration of alcohol X, total sugar content P and cell concentration S.
  5. 5. multi-model stalk fermentation process key state variable flexible measurement method according to claim 1, it is characterised in that Clustering is carried out using core Fuzzy c-means Clustering, degree of membership of each sample in nuclear space is calculated and carries out according to the following steps：
    If stalk fermentation collecting sample collection：X={ x1,x2,…,xn, k cluster is created, for similar to sample set X in data For one group；It is dissimilar that data are classified by seeking the minimum value J of object function as far as possible not at one group,It is constrained in：
    Wherein k is to cluster number, xiFor i-th of input data, vjFor cluster centre, uijFor data xiTo j-th cluster centre It is subordinate to angle value, m is weighted number, and U is subordinated-degree matrix more new formula, and V is cluster centre calculation formula.
  6. 6. the flexible measurement method of stalk fermentation production ethanol process key parameters according to claim 1, it is characterised in that The degree of membership of each submodel of basis described in step 4 obtains final multi-model soft-sensing model, real according to procedure below It is existing:
    By described object function, the classification of data sample is obtained, carries out DE-LSSVM data modeling：
    1) original sample is normalized：By the sample [x after normalizing1',y1'],…,[x'n,y'n] it is divided into N number of training Sample and M test sample,
    2) DE algorithm parameters are initialized：Current algebraically G=0 is made, is produced according to penalty coefficient C and the upper and lower limit of radial direction base width cs One group of random [C1,σ1]；
    3) by N number of training sample [x1',…,x'N] input as LS-SVM, current [C, σ] be used as parameter, trains LS-SVM to obtain [y is exported to corresponding1”,…,y”N]；
    4) output [y after training1”,…,y”N] and reality output [y1',…,y'N] square-error and result as DE algorithms Object function, whether error in judgement meet whether requirement or G are equal to Gmax；If meeting one of which, step 7) is gone to, Otherwise enter 5)；
    5) G=G+1；
    6) row variation, intersection, selection operation are entered to [C, σ], produces new [C, σ], return to step 3)；
    7) parameter for obtaining optimal penalty coefficient C and radial direction base width cs as LSSVM carries out soft sensor modeling；
    Obtaining every a kind of output is：
    <mrow> <mi>y</mi> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>k</mi> </munderover> <msub> <mi>u</mi> <mrow> <mi>i</mi> <mi>j</mi> </mrow> </msub> <msub> <mi>f</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>x</mi> <mo>)</mo> </mrow> </mrow>
    U in formulaijRepresent jth class fuzzy membership, f corresponding to i-th of samplei(x) i-th of submodel is represented.
CN201710685210.2A 2017-08-11 2017-08-11 Stalk fermentation produces the flexible measurement method of ethanol process key parameters Pending CN107341359A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710685210.2A CN107341359A (en) 2017-08-11 2017-08-11 Stalk fermentation produces the flexible measurement method of ethanol process key parameters

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710685210.2A CN107341359A (en) 2017-08-11 2017-08-11 Stalk fermentation produces the flexible measurement method of ethanol process key parameters

Publications (1)

Publication Number Publication Date
CN107341359A true CN107341359A (en) 2017-11-10

Family

ID=60216852

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710685210.2A Pending CN107341359A (en) 2017-08-11 2017-08-11 Stalk fermentation produces the flexible measurement method of ethanol process key parameters

Country Status (1)

Country Link
CN (1) CN107341359A (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109063786A (en) * 2018-08-27 2018-12-21 浙江大学 The flexible measurement method of concentration of acetylene in coal pyrolysis in plasma producing acetylene reaction process
CN109554439A (en) * 2018-10-31 2019-04-02 江苏大学 A kind of Pichia pastoris fermentation process soft measuring method based on ONLINE-RLSSVM
CN111898301A (en) * 2020-07-31 2020-11-06 上海应用技术大学 Method for implementing soft measuring system
CN113444843A (en) * 2021-06-09 2021-09-28 华东理工大学 Visual online monitoring method for fuel ethanol production state
CN114417720A (en) * 2022-01-19 2022-04-29 江苏大学 Multi-model predictive control method for fermentation process of pichia pastoris

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101748186A (en) * 2009-12-29 2010-06-23 江苏大学 FSVM-based lysine fermentation process key state variable soft measuring method and system
CN104630410A (en) * 2015-02-10 2015-05-20 东北大学 Real-time dynamic converter steelmaking quality prediction method based on data analysis
EP3047475A2 (en) * 2013-09-19 2016-07-27 Tata Consultancy Services Ltd. System and method for evaluating a cognitive load on a user corresponding to a stimulus
CN106052753A (en) * 2016-05-18 2016-10-26 江苏大学 Straw fermentation fuel ethanol production process key state variable soft measuring method based on fuzzy support vector machine
CN106444377A (en) * 2016-10-09 2017-02-22 江苏大学 Soft measuring method and system for key variables of lysine fermentation process based on PSO-FSVM

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101748186A (en) * 2009-12-29 2010-06-23 江苏大学 FSVM-based lysine fermentation process key state variable soft measuring method and system
EP3047475A2 (en) * 2013-09-19 2016-07-27 Tata Consultancy Services Ltd. System and method for evaluating a cognitive load on a user corresponding to a stimulus
CN104630410A (en) * 2015-02-10 2015-05-20 东北大学 Real-time dynamic converter steelmaking quality prediction method based on data analysis
CN106052753A (en) * 2016-05-18 2016-10-26 江苏大学 Straw fermentation fuel ethanol production process key state variable soft measuring method based on fuzzy support vector machine
CN106444377A (en) * 2016-10-09 2017-02-22 江苏大学 Soft measuring method and system for key variables of lysine fermentation process based on PSO-FSVM

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
陈贵华 等: "基于模糊核聚类的乙烯裂解深度DE-LSSVM多模型建模", 《化工学报》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109063786A (en) * 2018-08-27 2018-12-21 浙江大学 The flexible measurement method of concentration of acetylene in coal pyrolysis in plasma producing acetylene reaction process
CN109554439A (en) * 2018-10-31 2019-04-02 江苏大学 A kind of Pichia pastoris fermentation process soft measuring method based on ONLINE-RLSSVM
CN111898301A (en) * 2020-07-31 2020-11-06 上海应用技术大学 Method for implementing soft measuring system
CN113444843A (en) * 2021-06-09 2021-09-28 华东理工大学 Visual online monitoring method for fuel ethanol production state
CN114417720A (en) * 2022-01-19 2022-04-29 江苏大学 Multi-model predictive control method for fermentation process of pichia pastoris
CN114417720B (en) * 2022-01-19 2023-04-28 江苏大学 Multi-model predictive control method for pichia pastoris fermentation process

Similar Documents

Publication Publication Date Title
CN107341359A (en) Stalk fermentation produces the flexible measurement method of ethanol process key parameters
CN101630376B (en) Soft-sensing modeling method and soft meter of multi-model neural network in biological fermentation process
CN102231057B (en) Method for carrying out soft-sensing on lysine fermenting process on basis of chaos particle swarm optimization (CPSO)
CN101748186A (en) FSVM-based lysine fermentation process key state variable soft measuring method and system
Karim et al. Global and local neural network models in biotechnology: application to different cultivation processes
CN104914723B (en) Industrial process soft-measuring modeling method based on coorinated training partial least square model
CN106052753A (en) Straw fermentation fuel ethanol production process key state variable soft measuring method based on fuzzy support vector machine
CN106444377A (en) Soft measuring method and system for key variables of lysine fermentation process based on PSO-FSVM
CN108090500B (en) Soft measurement method for key parameters in fermentation production process of edible fungi
CN103344530B (en) Soft measurement method for granularity of cement raw material grinded by ball mill
CN107782857A (en) Flexible measurement method based on the edible fungus fermented process key parameter for improving CS BPNN
CN106843172A (en) Complex industrial process On-line quality prediction method based on JY KPLS
CN110807554A (en) Generation method and system based on wind power/photovoltaic classical scene set
CN104462850A (en) Multi-stage batch process soft measurement method based on fuzzy gauss hybrid model
CN110245430A (en) Improve Bouc-Wen model hysteresis modeling method
CN110046377B (en) Selective integration instant learning soft measurement modeling method based on heterogeneous similarity
CN101929993A (en) Dynamic fuzzy neural network based penicillin fermentation process soft measuring modeling method
CN101673096B (en) Soft-measuring method for density in concentration process of salvia miltiorrhiza injection production
Zheng et al. Improved mahalanobis distance based JITL-LSTM soft sensor for multiphase batch processes
CN101587350A (en) Soft-sensing method of crucial biochemical quantity in penicillin fermentation process based on fuzzy neural inverse and system structure thereof
CN107766880B (en) Photosynthetic bacteria fermentation process key parameter soft measurement method based on BA-LSSVM
CN106599391A (en) Relevance vector machine soft measurement modeling method based on triangular angle value dynamic weighting
CN109033524A (en) A kind of chemical process concentration variable On-line Estimation method based on robust mixed model
Zhang Software sensors and their applications in bioprocess
CN104450499A (en) Early warning method of pichia pastoris fermentation process based on support vector machine

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
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20171110