CN101673096B - Soft-measuring method for density in concentration process of salvia miltiorrhiza injection production - Google Patents
Soft-measuring method for density in concentration process of salvia miltiorrhiza injection production Download PDFInfo
- Publication number
- CN101673096B CN101673096B CN2009101537003A CN200910153700A CN101673096B CN 101673096 B CN101673096 B CN 101673096B CN 2009101537003 A CN2009101537003 A CN 2009101537003A CN 200910153700 A CN200910153700 A CN 200910153700A CN 101673096 B CN101673096 B CN 101673096B
- Authority
- CN
- China
- Prior art keywords
- density
- model
- soft
- sample
- pred
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 88
- 238000002347 injection Methods 0.000 title claims abstract description 25
- 239000007924 injection Substances 0.000 title claims abstract description 25
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 23
- 235000011135 Salvia miltiorrhiza Nutrition 0.000 title claims abstract description 10
- 241000304195 Salvia miltiorrhiza Species 0.000 title abstract 5
- 239000007788 liquid Substances 0.000 claims abstract description 11
- 238000012216 screening Methods 0.000 claims abstract description 5
- 238000003908 quality control method Methods 0.000 claims abstract description 4
- 238000012549 training Methods 0.000 claims description 30
- 244000132619 red sage Species 0.000 claims description 21
- 238000013528 artificial neural network Methods 0.000 claims description 19
- 238000012628 principal component regression Methods 0.000 claims description 18
- WTPPRJKFRFIQKT-UHFFFAOYSA-N 1,6-dimethyl-8,9-dihydronaphtho[1,2-g][1]benzofuran-10,11-dione;1-methyl-6-methylidene-8,9-dihydro-7h-naphtho[1,2-g][1]benzofuran-10,11-dione Chemical compound O=C1C(=O)C2=C3CCCC(=C)C3=CC=C2C2=C1C(C)=CO2.O=C1C(=O)C2=C3CCC=C(C)C3=CC=C2C2=C1C(C)=CO2 WTPPRJKFRFIQKT-UHFFFAOYSA-N 0.000 claims description 16
- 239000012141 concentrate Substances 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 14
- 238000000691 measurement method Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 9
- 210000002569 neuron Anatomy 0.000 claims description 6
- 238000012417 linear regression Methods 0.000 claims description 5
- 230000001419 dependent effect Effects 0.000 claims description 4
- 238000011156 evaluation Methods 0.000 claims description 4
- 239000003182 parenteral nutrition solution Substances 0.000 claims description 4
- 238000010238 partial least squares regression Methods 0.000 claims description 3
- 238000013480 data collection Methods 0.000 claims 2
- 239000000498 cooling water Substances 0.000 claims 1
- 239000000243 solution Substances 0.000 abstract description 6
- 230000009286 beneficial effect Effects 0.000 abstract 1
- 239000007791 liquid phase Substances 0.000 abstract 1
- 238000000491 multivariate analysis Methods 0.000 abstract 1
- 239000012071 phase Substances 0.000 abstract 1
- 239000000523 sample Substances 0.000 description 43
- 238000005070 sampling Methods 0.000 description 6
- 239000003814 drug Substances 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 5
- 239000012467 final product Substances 0.000 description 4
- 235000014347 soups Nutrition 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 239000013598 vector Substances 0.000 description 4
- FFBHFFJDDLITSX-UHFFFAOYSA-N benzyl N-[2-hydroxy-4-(3-oxomorpholin-4-yl)phenyl]carbamate Chemical compound OC1=C(NC(=O)OCC2=CC=CC=C2)C=CC(=C1)N1CCOCC1=O FFBHFFJDDLITSX-UHFFFAOYSA-N 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000001704 evaporation Methods 0.000 description 2
- 230000008020 evaporation Effects 0.000 description 2
- 238000000513 principal component analysis Methods 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 description 1
- 240000007164 Salvia officinalis Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000001816 cooling Methods 0.000 description 1
- 239000006071 cream Substances 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010202 multivariate logistic regression analysis Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000004886 process control Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 235000005412 red sage Nutrition 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Images
Landscapes
- Feedback Control In General (AREA)
Abstract
The invention provides a soft-measuring method for density in concentrating process of salvia miltiorrhiza injection production, comprising the following steps: gathering the historical data of each sensor and densimeter in the concentration process of salvia miltiorrhiza, wherein the history data relate to concentrated solution density value and sensor data gathered on line in production process; selecting easily obtained procedure variables including gas phase temperature, liquid phase temperature, pressure and concentrated solution liquid level that have higher degree of correlation with density from each sensor data, and screening representative dataset; using a multivariate analysis method for building a soft-measuring model of density; and gathering the procedure variables on line, and using the soft-measuring model of density to carry out real-time estimation to control the concentration process. Aiming at the problem that the density in the concentration process of salvia miltiorrhiza injection production is difficultly monitored in real time, the invention provides a fast density soft-measuring method with high precision, which makes full use of the history data obtained by each sensor from the production process, and is beneficial to improvement of quality control on salvia miltiorrhiza injection production.
Description
Technical field
The invention belongs to Chinese medicine production run measuring method, the flexible measurement method that relates to a kind of Chinese medicine production run, especially a kind of flexible measurement method of danshen injections concentration process concentrated solution density comprises the soft-sensing model based on multiple linear regression (MLR), principal component regression (PCR), offset minimum binary (PLS) and the multiple multivariable technique of artificial neural network (ANN).
Background technology
Basic steps such as tcm manufacturing process comprises extraction, concentrates, alcohol precipitation.Enrichment process is the important step that Chinese medicine is produced, and the soup that it obtains abstraction process carries out evaporation and concentration makes the density of soup reach the requirement of production, so this link is directly connected to the quality of tcm product.Traditional enrichment process mainly relies on the way of manual observation, by extracting soup and concentrating back soup ratio and estimate, is difficult to guarantee the stability of the medicine composition of different batches like this.A lot of pharmacy corporations have been used for on-line densimeter to monitor in real time the density of concentrate.Because densitometer is the sensor of contact, and concentration tank seals, and is unfavorable for the installation of density sensor.Adopt mostly and draw a circulation duct, densitometer is installed in the pipeline from concentration tank.Because the Chinese medicine concentrate is thick liquid, along with the increase of concentrated solution density, its flowability is relatively poor, causes concentrate to remain on the densitometer probe, and the sensitivity of probe is descended.General solution is regularly to clean, but sensor dismounting inconvenience, the cleaning frequency is higher, can influence the The real time measure of density in the production.
A kind of new technology that soft-measuring technique emerges at process control and detection range in recent years, it mainly is the thinking by indirect measurement, utilizes other metrical informations that are easy to obtain, and realizes the estimation of tested measurement by computing machine.Utilize this technology, we can set up the soft-sensing model of density according to the mathematical relation between process variable that easily records (temperature, pressure, feed liquor amount etc.) and the density.The present domestic patent achievement that does not have as yet at the soft measurement aspect of parenteral solution concentration process.
Summary of the invention
At the problems referred to above, the purpose of this invention is to provide a kind of flexible measurement method of density in concentration process of salvia miltiorrhiza injection production.Utilize the historical data in the red sage root concentration process, therefrom choose the process variable that be easy to obtain higher with the density degree of correlation, filter out representative data set, use multivariable technique, comprise that multiple linear regression (MLR), principal component regression (PCR), offset minimum binary (PLS) and artificial neural network (ANN) method set up the soft-sensing model of density, and utilize soft-sensing model that density is carried out real-time estimate, the control concentration process.The multiple flexible measurement method of density fast based on the multivariable analysis technology is provided, has made full use of the historical data that obtains in the production run, helped the quality control that danshen injections is produced.
The present invention realizes by following steps:
(1) data set { X that each sensor and densitometer obtain in the collection danshen injections production concentration process, D}: at first be to gather historical data, danshen injections is produced concentration process and is carried out in concentrating under reduced pressure evaporation receipts cream device, use sensor online acquisition parenteral solution to produce concentration process c (c 〉=3) individual batch, each batch gathered k sample (k 〉=30) constantly, obtain history data set { X, D}.Wherein, X is each sensing data, comprises upward discharge of well heater, feed liquor flow, evaporator vacuum tightness, gas phase temperature, liquidus temperature, liquid level, surge tank pressure and cooling, and D is the actual density value of concentrate, uses densitometer to collect.
(2) Variables Selection and representative data screening: { X chooses the process variable higher with the density D degree of correlation among the D}, and removes unusual sample wherein, filters out representative data set { X from each sensor data set
m, D} is divided into training sample and forecast sample.Select main process variable, calculate the degree of correlation of each variable and density D, obtain main process variable X by correlation coefficient process
m=(T
g, T
l, P, L), T wherein
gBe gas phase temperature, T
lFor liquidus temperature, P are that evaporator vacuum tightness and L are the concentrate liquid level; Remove the unusual sample of data centralization again, keep representative data, it is divided into training sample and forecast sample.
(3) set up soft-sensing model: use multiple multivariable technique respectively, comprise multiple linear regression (MLR), principal component regression (PCR), offset minimum binary (PLS) and artificial neural network (ANN) method, with the process variable chosen in (2) as input, density value uses the training sample training to set up the soft-sensing model of density as output.With X
m=(T
g, T
l, P L) is independent variable, D is a dependent variable, uses training dataset { X
M_tr, D
TrSet up model, predictive data set { X
M_pred, D
PredCarry out model evaluation.
Use forecast sample that model is estimated;
1. the MLR soft-sensing model of setting up density is as follows:
D
tr=X
m_trb
MLR+e=b
1T
g+b
2T
l+b
3P+b
4L+e
Wherein, e is an error term.Calculate regression coefficient b by training sample
MLR=(X
M_tr TX
M_tr)
-1X
M_tr TD
Tr, the model of forecast set unknown sample is estimated as D
Pred_MLR=X
M_predb
MLR
2. the PCR soft-sensing model of setting up density is as follows:
D
tr=T
m_trb
PCR+e
Wherein, e is an error term, T
M_trFor major component is decomposed X
M_tr=T
M_trP
M_tr TWhat obtain must sub matrix.Calculating can get major component regression coefficient b
PCR=(T
M_tr TT
M_tr)
-1T
M_tr TD
TrTo the forecast set unknown sample, at first calculate its score matrix T
M_pred=X
M_predP
M_tr, be D by regression coefficient computation model estimated value then
Pred_PCR=T
M_predb
PCR=X
M_predP
M_trb
PCR
3. the PLS soft-sensing model of setting up density is as follows:
U
tr=T
m_trb+e
Wherein, e is an error term, T
M_trFor major component is decomposed X
M_tr=T
M_trP
M_tr TObtain sub matrix, U
TrBy the major component decomposing D
Tr=U
TrQ
Tr TObtain.Calculating can get regression coefficient b=(T
M_tr TT
M_tr)
-1T
M_tr TU
TrBy the major component decomposing D
Tr=U
TrQ
Tr TGet and to set up D
TrWith X
mBetween relation:
D
tr=T
m_trbQ
tr T+eQ
tr T=X
m_trP
m_trbQ
tr T+eQ
tr T=X
m_trb
PLS+e’
To the forecast set unknown sample, by PLS regression coefficient b
PLSCan the computation model estimated value be D
Pred_PLS=X
M_predb
PLS
4. with X
mAs input, density D is as output, and end user's artificial neural networks is set up the soft-sensing model of density to training sample.At first select network type, the neural network type that is used for forecast modeling commonly used has BP neural network, RBF neural network etc.
Secondly, determine network structure and transport function.Network number of plies n 〉=3 comprise input layer, hidden layer and output layer.Wherein, the input layer number is X
mThe number of middle variable, the output layer neuron number is 1, the neuron number that hidden layer number and hidden layer comprise is chosen optimal value by different values, the optional tansig function commonly used of the transport function between input layer, middle hidden layer, the output layer, purelin function, Gaussian function, threshold function etc.
At last, training network model.The initial weight of picked at random network is set learning rate and end condition, obtains network connection weights, threshold value by optimizing the learning algorithm training.The model that use trains is estimated forecast sample.
(4) model is applied to production line, the online real time collecting danshen injections is produced the process variable X of concentration process
m=(T
g, T
l, P L), uses the density value of density soft-sensing model real-time estimate concentrate, monitors concentration process in real time and carries out quality control, instructs and produces.
Characteristics of the present invention and useful effect are, its principle is the process variable of producing concentration process by the danshen injections of easy acquisition, use multivariable technique, the structure mathematical model realizes the soft measurement to the concentrated solution density that is difficult to the online in real time measurement.The soft-sensing model of being set up based on MLR calculates simple, is applicable to the simple system that linear relationship is good, and the model explanation ability is strong, has actual physics chemistry implication; The soft-sensing model of the PCR-based of being set up can maximum utilize useful information, by ignoring those submembers, can suppress the influence of noise to model, further improves the predictive ability of model; The soft-sensing model of being set up based on PLS is not only eliminated the noise information in the independent variable, and has removed the garbage in the dependent variable, is the combination of multiple linear regression and principal component analysis (PCA), can significantly improve the predictive ability of model; The soft-sensing model of being set up based on ANN is a nonlinear model, has self study, self-organization, adaptive ability, the function of very strong fault-tolerant ability and parallel processing information and height non-linear expression ability, and this is that other classic methods are not available.The soft-sensing model that trains can use the process variable that obtains of screening that density value is predicted, for the optimization regulation and control of the concentration process of parenteral solution provide the technology that more suitably realizes.This method is not only applicable to the danshen injections concentration process, and extends to other pharmacy procedure, has a good application prospect.
Description of drawings
Accompanying drawing 1 is the flexible measurement method theory diagram of danshen injections concentration process density.
Accompanying drawing 2 is the correlativity curve of density observed reading and PLS soft-sensing model predicted value.
Accompanying drawing 3 is the correlativity curve of density observed reading and neural network soft sensor model predicted value.
Accompanying drawing 4 is that the observed reading and the neural network soft sensor model predicted value of test set sample rate compares.
Embodiment
The present invention is further explained in conjunction with the accompanying drawings and embodiments, but the invention is not restricted to this example.Accompanying drawing is embodiments of the invention, and two embodiment use Linear PLS method and Nonlinear A NN method to set up the soft-sensing model of danshen injections concentration process density respectively.
Embodiment 1: the linear flexible measurement method (PLS) of danshen injections concentration process density
The principle of the inventive method is referring to Fig. 1.
History data set X, the concentration section that D} produces from certain drugmaker danshen injections is chosen 3 batches data, each batch sampling number is 60, sample choose the whole concentration process of even distribution.Wherein, X is that production line is obtained and more accurate process variable data by sensor easily, comprises 9 process variable: well heater (X
1), feed liquor flow (X
2), feed liquor semi-invariant (X
3), evaporator vacuum tightness (X
4), gas phase temperature (X
5), liquidus temperature (X
6), liquid level (X
7), surge tank pressure (X
8) and cool off and go up discharge (X
9); D is the actual density value of concentrate, uses on-line densimeter to collect.For reducing systematic error, data are carried out 4 point sampling average treatment, and the sampling number that obtains the data of each batch is 15, obtains X=(X
1, X
2..., X
9) be 45 * 9 matrixes, D is 45 * 1 vectors.
Select main process variable according to priori, and calculate 9 process variable X respectively by correlation coefficient process
1, X
2..., X
9With the degree of correlation of density D, keep main process variable X
m=(T
g, T
l, P, L), T wherein
gBe gas phase temperature, T
lFor liquidus temperature, P are that evaporator vacuum tightness and L are the concentrate liquid level.Secondly, remove the unusual sample (process incipient stage system's instability should be removed) of data centralization, keep representative data, the data that obtain at last: X
mBe 30 * 4 matrixes, D
mBe 30 * 1 vectors.Data set { X
m=(T
g, T
l, P, L), D
mTo be used for the foundation of follow-up soft-sensing model.Data set is divided into training sample and forecast sample, and method is as follows: according to the big minispread sample of density value, uniformly-spaced choose 2/3 as the training set sample, remaining 1/3 as forecast sample.Use training dataset { X
M_tr, D
TrSet up model, predictive data set { X
M_pred, D
PredCarry out model evaluation.
The PLS soft-sensing model of setting up density is as follows:
D
tr=T
m_trbQ
tr T+eQ
tr T=X
m_trP
m_trbQ
tr T+eQ
tr T=X
m_trb
PLS+e’
=(T
g,T
l,P,L)(-0.064,0.121,-0.017,0.008)
T+1.151
Wherein, b
PLS=(0.064,0.121 ,-0.017,0.008)
TBe the PLS regression coefficient.The model of forecast set unknown sample only need X to be estimated
M_predBringing following formula into gets final product.Fig. 2 has provided the correlativity curve of density observed reading and PLS soft-sensing model predicted value, the coefficient R of training sample=0.9542 wherein, proofread and correct root-mean-square error RMSEC=2.36%, the coefficient R pred=0.9468 of forecast set sample, predicted root mean square error RMSEP=2.75%.As seen, results relevance is higher, and root-mean-square error meets the industry spot application requirements all less than 3%.
In addition, also set up other linear soft-sensing models of density:
1. MLR soft-sensing model:
D
tr=X
m_trb
MLR+e=-0.016T
g+0.146T
l-0.089P+0.007L+1.151,
Regression coefficient b
MLR=(0.016,0.146 ,-0.089,0.007)
T, the model of forecast set unknown sample only need X to be estimated
M_predBringing following formula into gets final product.
2. PCR soft-sensing model:
Wherein, T
M_trFor major component is decomposed X
M_tr=T
M_trP
M_tr TObtain sub matrix, P
M_trBe loading matrix, b
PCR=(0.024 ,-0.018,0.133 ,-0.104)
TBe the principal component regression coefficient.Model when this model is number of principal components m=4, if need the model of number of principal components less than 4 (they being m<4), can be with P in the following formula
M_trBack several row (4-m row) remove corresponding regression coefficient b
PCRBack several row (4-m is capable) remove and to get final product.The model of forecast set unknown sample only need X to be estimated
M_predBringing following formula into gets final product.
Embodiment 2: the non-linear flexible measurement method (ANN) of danshen injections concentration process density
History data set X, the concentration section that D} produces from certain drugmaker danshen injections is chosen 5 batches data, each batch sampling number is 80, sample choose the whole concentration process of even distribution.Wherein, X is that production line is obtained and more accurate process variable data by sensor easily, comprises 9 process variable: well heater (X
1), feed liquor flow (X
2), feed liquor semi-invariant (X
3), evaporator vacuum tightness (X
4), gas phase temperature (X
5), liquidus temperature (X
6), liquid level (X
7), surge tank pressure (X
8) and cool off and go up discharge (X
9); D is the actual density value of concentrate, uses on-line densimeter to collect.For reducing systematic error, data are carried out 4 point sampling average treatment, and the sampling number that obtains the data of each batch is 20, obtains X=(X
1, X
2..., X
9) be 100 * 9 matrixes, D is 100 * 1 vectors.
Select main process variable according to priori, and calculate 9 process variable X respectively by correlation coefficient process
1, X
2..., X
9With the degree of correlation of density D, keep main process variable X
m=(T
g, T
l, P, L), T wherein
gBe gas phase temperature, T
lFor liquidus temperature, P are that evaporator vacuum tightness and L are the concentrate liquid level.Secondly, remove the unusual sample (process incipient stage system's instability should be removed) of data centralization, keep representative data, the data that obtain at last: X
mBe 72 * 4 matrixes, D
mBe 72 * 1 vectors.Data set { X
m=(T
g, T
l, P, L), D
mTo be used for the foundation of follow-up soft-sensing model.Data set is divided into training sample, checking sample and test sample book, and method is as follows: according to the big minispread sample of density value, uniformly-spaced choose 1/2 as training sample, 1/4 as the checking sample, and remaining 1/4 as test sample book.Use training dataset { X
M_tr, D
TrSet up model, verify and test data set { X
M_pred, D
PredCarry out model evaluation.
Network type is selected the BP neural network, and network number of plies n=3 comprises hidden layer and output layer in the middle of the input layer, one.With X
mAs input, density D is as output, and the initial weight of picked at random network is set learning rate a=0.8, and end condition is maximum frequency of training N=1000, training sample is set up the soft-sensing model of density.Model can be represented with following functional relation:
D=f(X
m)=f(T
g,T
l,P,L)
Wherein, the input layer number is X
mThe number 4 of middle variable, the output layer neuron number is 1, and the neuron number of middle hidden layer is chosen optimal value by different values, and the transport function between input layer, middle hidden layer, the output layer is used tansig function and purelin function respectively:
g(x)=1-exp(-x)/1+exp(-x),f(x)=kx,
Thereby above-mentioned functional relation is:
D=f(X
m)=f(∑w
k1g(X
iw
ji+b
j)+a)
Wherein, X
iBe input variable, w
JiBe the weights between input layer i and the hidden node j, b
jBe the threshold value of hidden node j, w
K1Be the weights between hidden node k and the output layer node, a is the threshold value of output layer node.Parameter w
Ji, w
K1, b
jAll adopt error backpropagation algorithm (BP) and the training of Levevberg-Marquardt learning algorithm to obtain with a.
Fig. 3 has provided the correlativity curve of density observed reading and neural network soft sensor model predicted value, the coefficient R of training sample=0.9459 wherein, proofread and correct root-mean-square error RMSEC=2.51%, the coefficient R va=0.9135 of checking collection sample, checking root-mean-square error RMSEV=3.11%, the coefficient R pred=0.9348 of forecast set sample, predicted root mean square error RMSEP=2.75%.As seen, results relevance is higher, and root-mean-square error is all less than 4%.Fig. 4 has provided predicting the outcome of 18 samples, and accuracy is higher, satisfies the industry spot application demand.
Claims (3)
1. the flexible measurement method of a density in concentration process of salvia miltiorrhiza injection production is characterized in that this flexible measurement method may further comprise the steps:
(1) image data collection: each sensor and densitometer obtain in the collection danshen injections production concentration process data set { X, D};
(2) Variables Selection and representative data screening: { X chooses the process variable higher with the density D degree of correlation among the D}, and removes unusual sample wherein, filters out representative data set { X from each sensor data set
m, D} is divided into training sample and forecast sample;
(3) set up soft-sensing model: use multivariable technique, comprise multiple linear regression, principal component regression, offset minimum binary and Artificial Neural Network, with the process variable chosen in (2) as input, density value is as output, use the training sample training to obtain the soft-sensing model of density, and use forecast sample that model is estimated, wherein with X
m=(T
g, T
l, P L) is independent variable, D is dependent variable, wherein T
gBe gas phase temperature, T
lFor liquidus temperature, P are that evaporator vacuum tightness, L are the concentrate liquid level; Use training dataset { X
M_tr, D
TrSet up model; Concrete steps are:
With X
m=(T
g, T
l, P L) is independent variable, D is a dependent variable, uses training dataset { X
M_tr, D
TrSet up model, predictive data set { X
M_pred, D
PredCarry out model evaluation,
1. the MLR soft-sensing model of setting up density is as follows:
D
tr=X
m_trb
MLR+e=b
1T
g+b
2T
l+b
3P+b
4L+e
Wherein, e is an error term, calculates regression coefficient b by training sample
MLR=(X
M_tr TX
M_tr)
-1X
M_tr TD
Tr, the model of forecast set unknown sample is estimated as D
Pred_MLR=X
M_predb
MLR
2. the PCR soft-sensing model of setting up density is as follows:
D
tr=T
m_trb
PCR+e
Wherein, e is an error term, T
M_trFor major component is decomposed X
M_tr=T
M_trP
M_tr TObtain sub matrix, P
M_trBe loading matrix, calculating can get major component regression coefficient b
PCR=(T
M_tr TT
M_tr)
-1T
M_tr TD
Tr,, at first calculate its score matrix T to the forecast set unknown sample
M_pred=X
M_predP
M_tr, be D by regression coefficient computation model estimated value then
Pred_PCR=T
M_predb
PCR=X
M_predP
M_trb
PCR
3. the PLS soft-sensing model of setting up density is as follows:
U
tr=T
m_trb+e
Wherein, e is an error term, T
M_trFor major component is decomposed X
M_tr=T
M_trP
M_tr TObtain sub matrix, U
TrBy the major component decomposing D
Tr=U
TrQ
Tr TObtain, calculating can get regression coefficient b=(T
M_tr TT
M_tr)
-1T
M_tr TU
Tr, by the major component decomposing D
Tr=U
TrQ
Tr TGet and to set up D
TrWith X
mBetween relation:
D
tr=T
m_trbQ
tr T+eQ
tr T=X
m_trP
m_trbQ
tr T+eQ
tr T=X
m_trb
PLS+e’
To the forecast set unknown sample, by PLS regression coefficient b
PLSCan the computation model estimated value be D
Pred_PLS=X
M_predb
PLS
4. with X
mAs input, density D is as output, and end user's artificial neural networks is set up the soft-sensing model of density to training sample, and at first network type is selected the BP neural network,
Secondly, determine network structure and transport function, network number of plies n 〉=3 comprise input layer, hidden layer and output layer, and wherein, the input layer number is X
mThe number of middle variable, the output layer neuron number is 1, the neuron number that hidden layer number and hidden layer comprise is chosen optimal value by different values, and the transport function between input layer, middle hidden layer, the output layer is selected tansig function and purelin function respectively for use;
At last, the training network model, the initial weight of picked at random network is set learning rate and end condition, obtains network connection weights, threshold value by optimizing the learning algorithm training, uses the model that trains that forecast sample is estimated;
(4) the online acquisition process is produced and is concentrated variable X
m=(T
g, T
l, P L), uses soft-sensing model that density is carried out real-time estimate, the control concentration process; Concrete steps are:
Model is applied to production line, and the online real time collecting danshen injections is produced the process variable X of concentration process
m=(T
g, T
l, P L), uses the density value of step (3) soft-sensing model prediction concentrate, monitors concentration process in real time and carries out quality control.
2. the flexible measurement method of a kind of density in concentration process of salvia miltiorrhiza injection production according to claim 1 is characterized in that, the described data set of step (1) is gathered concrete steps and is:
Danshen injections is produced concentration process and is carried out in vacuum concentration equipment, use sensor online acquisition parenteral solution to produce concentration process c batch, c 〉=3, each batch gathered k sample constantly, k 〉=30, obtain data set X, D}, wherein, X is each sensing data, comprise heter temperature, feed liquor flow, evaporator vacuum tightness, gas phase temperature, liquidus temperature, concentrate liquid level, surge tank pressure and cooling water flow, D is the actual density value of concentrate, uses densitometer to collect.
3. the flexible measurement method of a kind of density in concentration process of salvia miltiorrhiza injection production according to claim 1 is characterized in that, the described data set screening of step (2) concrete steps are:
{ X, D} rule of thumb select main process variable to the garbled data collection, verify the degree of correlation of each variable and density D, keep main process variable X
m=(T
g, T
l, P, L), T wherein
gBe gas phase temperature, T
lFor liquidus temperature, P are that evaporator vacuum tightness and L are the concentrate liquid level; Remove the unusual sample of data centralization, keep representative data, it is divided into training sample and forecast sample.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009101537003A CN101673096B (en) | 2009-10-26 | 2009-10-26 | Soft-measuring method for density in concentration process of salvia miltiorrhiza injection production |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009101537003A CN101673096B (en) | 2009-10-26 | 2009-10-26 | Soft-measuring method for density in concentration process of salvia miltiorrhiza injection production |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101673096A CN101673096A (en) | 2010-03-17 |
CN101673096B true CN101673096B (en) | 2011-09-28 |
Family
ID=42020349
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009101537003A Expired - Fee Related CN101673096B (en) | 2009-10-26 | 2009-10-26 | Soft-measuring method for density in concentration process of salvia miltiorrhiza injection production |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101673096B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104636861B (en) * | 2014-12-30 | 2018-04-13 | 深圳市华星光电技术有限公司 | To the processing procedure of display into the method controlled in real time on line |
CN104678974B (en) * | 2015-03-03 | 2018-06-15 | 深圳市华星光电技术有限公司 | To product processing procedure into the method controlled in real time on line |
CN107064054B (en) * | 2017-02-28 | 2019-08-02 | 浙江大学 | A kind of near-infrared spectral analytical method based on CC-PLS-RBFNN Optimized model |
CN107544286A (en) * | 2017-08-30 | 2018-01-05 | 浙江力太科技有限公司 | A kind of system identifying method in evaporization process |
CN109101758A (en) * | 2018-09-03 | 2018-12-28 | 江南大学 | Batch process process conditions design method based on T-PLS model |
CN111914214B (en) * | 2020-06-13 | 2023-10-17 | 宁波大学 | PTA production process soft measurement method based on reduced KPLS model |
CN111897298A (en) * | 2020-07-27 | 2020-11-06 | 浙江大学 | Method and system for monitoring acoustic emission in preparation process of traditional Chinese medicine particles in fluidized bed |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1210265A (en) * | 1998-06-26 | 1999-03-10 | 华南理工大学 | Online testing method for Kappa value of pulp in course of discontinuous sulfate cooking process |
CN1570629A (en) * | 2004-05-12 | 2005-01-26 | 浙江大学 | On-line soft measurement modeling method for 4-CBA content based on least square algorithm of restricted memory part |
CN101281182A (en) * | 2008-05-22 | 2008-10-08 | 沈阳东大自动化有限公司 | Method for soft measuring sodium aluminate solution component concentration |
-
2009
- 2009-10-26 CN CN2009101537003A patent/CN101673096B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1210265A (en) * | 1998-06-26 | 1999-03-10 | 华南理工大学 | Online testing method for Kappa value of pulp in course of discontinuous sulfate cooking process |
CN1570629A (en) * | 2004-05-12 | 2005-01-26 | 浙江大学 | On-line soft measurement modeling method for 4-CBA content based on least square algorithm of restricted memory part |
CN101281182A (en) * | 2008-05-22 | 2008-10-08 | 沈阳东大自动化有限公司 | Method for soft measuring sodium aluminate solution component concentration |
Non-Patent Citations (1)
Title |
---|
韦文祥 等.《神经元网络软测量模型在中药浓缩工段的应用》.《化工自动化及仪表》.2007,第34卷(第5期),第57-61页. * |
Also Published As
Publication number | Publication date |
---|---|
CN101673096A (en) | 2010-03-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101673096B (en) | Soft-measuring method for density in concentration process of salvia miltiorrhiza injection production | |
CN110084367B (en) | Soil moisture content prediction method based on LSTM deep learning model | |
CN112101480B (en) | Multivariate clustering and fused time sequence combined prediction method | |
CN104914723B (en) | Industrial process soft-measuring modeling method based on coorinated training partial least square model | |
US20220083839A1 (en) | Accuracy compensation method for discharge caustic alkali concentration measuring device in evaporation process | |
CN106845796A (en) | One kind is hydrocracked flow product quality on-line prediction method | |
CN108399434B (en) | Analysis and prediction method of high-dimensional time series data based on feature extraction | |
CN101863088A (en) | Method for forecasting Mooney viscosity in rubber mixing process | |
CN110957011B (en) | Online production parameter estimation method of continuous stirring reactor under unknown time-varying measurement noise | |
CN115495991A (en) | Rainfall interval prediction method based on time convolution network | |
CN106598918B (en) | Nonuniformity methods for calculating designed flood based on quantile estimate | |
CN113420500A (en) | Intelligent atmospheric and vacuum system | |
CN115618720A (en) | Soil salinization prediction method and system based on altitude | |
CN109886314B (en) | Kitchen waste oil detection method and device based on PNN neural network | |
CN109033524B (en) | Chemical process concentration variable online estimation method based on robust mixed model | |
CN110851897A (en) | Aqueduct stress-strain prediction method under multi-factor correlation | |
CN108204997A (en) | Normal line oil flash point on-line soft measurement method | |
CN116821695B (en) | Semi-supervised neural network soft measurement modeling method | |
CN117573668A (en) | Optimization method based on ultrasonic gas meter metering data | |
CN103389360A (en) | Probabilistic principal component regression model-based method for soft sensing of butane content of debutanizer | |
CN103279030B (en) | Dynamic soft measuring modeling method and device based on Bayesian frame | |
CN112801426A (en) | Industrial process fault fusion prediction method based on correlation parameter mining | |
CN116662925A (en) | Industrial process soft measurement method based on weighted sparse neural network | |
CN109059875A (en) | A method of drive perfect model to carry out moon scale Runoff Forecast | |
CN115270637A (en) | Underground drainage pipeline maximum stress prediction method based on GBRT |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110928 Termination date: 20151026 |
|
EXPY | Termination of patent right or utility model |