CN114334024A - Fermentation process soft measurement modeling method based on rapid component transfer learning - Google Patents
Fermentation process soft measurement modeling method based on rapid component transfer learning Download PDFInfo
- Publication number
- CN114334024A CN114334024A CN202111633686.4A CN202111633686A CN114334024A CN 114334024 A CN114334024 A CN 114334024A CN 202111633686 A CN202111633686 A CN 202111633686A CN 114334024 A CN114334024 A CN 114334024A
- Authority
- CN
- China
- Prior art keywords
- data
- model
- source domain
- domain
- working condition
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000005259 measurement Methods 0.000 title claims abstract description 33
- 238000000855 fermentation Methods 0.000 title claims abstract description 15
- 230000004151 fermentation Effects 0.000 title claims abstract description 15
- 238000013526 transfer learning Methods 0.000 title claims abstract description 15
- 229930182555 Penicillin Natural products 0.000 claims abstract description 47
- JGSARLDLIJGVTE-MBNYWOFBSA-N Penicillin G Chemical compound N([C@H]1[C@H]2SC([C@@H](N2C1=O)C(O)=O)(C)C)C(=O)CC1=CC=CC=C1 JGSARLDLIJGVTE-MBNYWOFBSA-N 0.000 claims abstract description 47
- 229940049954 penicillin Drugs 0.000 claims abstract description 47
- 238000009826 distribution Methods 0.000 claims abstract description 24
- 238000013508 migration Methods 0.000 claims abstract description 9
- 230000005012 migration Effects 0.000 claims abstract description 9
- 238000004458 analytical method Methods 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 30
- 230000008569 process Effects 0.000 claims description 23
- 230000006870 function Effects 0.000 claims description 17
- 238000013507 mapping Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 10
- 238000011156 evaluation Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 238000004088 simulation Methods 0.000 claims description 8
- 238000004422 calculation algorithm Methods 0.000 claims description 7
- CURLTUGMZLYLDI-UHFFFAOYSA-N Carbon dioxide Chemical compound O=C=O CURLTUGMZLYLDI-UHFFFAOYSA-N 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000012549 training Methods 0.000 claims description 4
- 238000005273 aeration Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 229910002092 carbon dioxide Inorganic materials 0.000 claims description 3
- 239000001569 carbon dioxide Substances 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000011112 process operation Methods 0.000 claims description 3
- 239000004576 sand Substances 0.000 claims description 3
- 239000000758 substrate Substances 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000010923 batch production Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012824 chemical production Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000002620 method output Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Landscapes
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
Abstract
The invention discloses a fermentation process soft measurement modeling method based on rapid component transfer learning, which comprises the following steps: 1) acquiring penicillin data; 2) preprocessing penicillin data and dividing a data set; 3) adapting the edge distribution of the source domain and the target domain; 4) establishing a soft measurement model based on a regularization extreme learning machine; 5) and (6) evaluating the performance of the model. The invention reduces the characteristic distribution distance of the source domain and the target domain based on the migration component analysis method, so that the characteristic distribution of the two groups of data is similar. And then establishing a soft measurement model on the mapped source domain data by adopting a regularization extreme learning machine method, and predicting the concentration of the penicillin in the target domain.
Description
Technical Field
The invention relates to the technical field of fermentation process soft measurement modeling, in particular to a fermentation process soft measurement modeling method based on rapid component transfer learning.
Background
In the chemical production process, stable operation of production equipment and product quality guarantee are greatly related to specific key variables. At present, a data-driven soft measurement modeling method is mostly adopted for a key index prediction task in an industrial process. The soft measurement modeling is dependent on methods such as statistical analysis or machine learning, and potential information in data is mined. The dependence on the internal mechanism or mathematical model of the industrial process is reduced, and the requirement on the prior knowledge of the process is greatly reduced. In a plurality of soft measurement modeling methods, an extreme learning machine is widely applied as a typical neural network algorithm by virtue of the advantages of quick training, simple structure, strong generalization capability and the like.
The penicillin fermentation process is a typical batch process, namely, the penicillin fermentation process has the characteristic of a multi-working-condition process. For the actual penicillin fermentation process, a soft measurement technology is mostly adopted to effectively monitor key variables such as the concentration of penicillin. The establishment of the soft measurement model is based on the auxiliary variables and the key variables, and in the actual multi-working-condition process, the collection of sufficient key variables, namely label values, for soft measurement modeling is time-consuming and labor-consuming, and even the situation that some working conditions are not label occurs. Meanwhile, in the traditional soft measurement modeling method, due to the fact that nonlinear relations among different working conditions are different and data distribution is different, a model established in a specific working condition cannot be directly used in other working conditions. Therefore, how to model the behavior of the unlabeled data is a question to be discussed.
Disclosure of Invention
In order to solve the problem of difficult modeling of unlabeled data under a certain working condition in the penicillin process, the invention provides a fermentation process soft measurement modeling method based on rapid component transfer learning. The distance of edge probability distribution between two working conditions is reduced by a Transfer Component Analysis (TCA) method, so that the characteristic distributions of the two working conditions are similar, and then a Regularized Extreme Learning Machine (RELM) method is adopted to quickly establish a model in a data characteristic alignment space to predict the penicillin concentration of the non-label data working condition.
The technical scheme of the invention is as follows:
a fermentation process soft measurement modeling method based on rapid component transfer learning comprises the following steps:
1) acquisition and pretreatment of penicillin data:
the penicillin data adopted by the invention is obtained by analog simulation of a Benchmark simulation platform (Pensim). In order to accelerate the convergence speed of the model and reduce the influence among different dimension data, the original data is normalized. In the modeling stage of the soft measurement model in the multi-working-condition process, one working condition is a source domain working condition with auxiliary variables, namely characteristic variables and key variables, and the other working condition is a target domain working condition with only the characteristic variables and no key variables.
2) Establishing a model of rapid component transfer learning:
based on the TCA method, the marginal probability distribution distance between the source domain and the target domain is reduced, thereby completing the feature migration. And in the space of the source domain and the target domain which are aligned through the TCA, establishing a regularization limit learning machine model based on the mapped source domain data, and quickly predicting the penicillin concentration of the working condition of the target domain.
3) And (3) evaluating model performance:
in order to evaluate the method proposed by the present invention more objectively, the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) of the evaluation index are introduced.
Further, the process of the step 1) is as follows:
step 1.1) obtaining penicillin data:
the penicillin data adopted by the invention is obtained by analog simulation of a Benchmark simulation platform (Pensim). The penicillin concentration is a key variable to be predicted in the process, and six related variables are used as auxiliary variables, namely carbon dioxide concentration, aeration rate, substrate feeding temperature, stirrer power, culture dish volume and pH value.
Step 1.2) data normalization treatment:
in order to accelerate the convergence speed of the model, reduce the training time of the model and simultaneously reduce the influence among different dimensional data, the data is normalized, and the formula is as follows:
in the formula, x is data after normalization processing; a is the collected raw data; a isminIs the minimum value in the original data; a ismaxIs the maximum value in the raw data.
Step 1.3) determining working condition data of a source domain and a target domain:
and randomly selecting one working condition from the different working condition data sets after the normalization processing as a source domain working condition data set, and randomly selecting one working condition from the rest working conditions as a target domain working condition data set. The source domain data has both characteristic variables and key variables, denoted as { X }S,YSThe target domain data only has characteristic variables and is marked as { X }T}。
Further, the process of step 2) is as follows:
step 2.1) feature migration of the source domain and the target domain:
it is assumed that there is a feature map ψ such that the distribution P (ψ (X) of the source domain and the destination domain after mappingS))＝P(ψ(XT)). The distance between the source domain and the target domain after mapping is measured by Maximum Mean Difference (MMD), which is expressed as:
where m is the number of source domain samples, n is the number of target domain samples, the above equation represents the distance of two distributions measured in the regenerated kernel hilbert space, and H represents the regenerated kernel hilbert space.
Transforming the MMD distance solving process into a kernel function learning process by using a kernel function, D (X)S,XT) Conversion to the following form:
tr(KL)-λtr(K)
wherein tr (·) is the inverse of the matrix, λ is the introduced parameter, K is the kernel matrix obtained by kernel function mapping, L is the matrix introduced by the MMD algorithm, and the calculation method of each element is as follows:
wherein D isSRepresenting the source domain, DTRepresenting the target domain, xiAnd xjRepresenting the characteristic variables of any two samples in the source domain or the target domain. Then, will findThe problem of solving tr (KL) -lambda tr (K) is converted into the following optimization problem:
s.t.WTKMKW＝Im
where M is the central matrix, μ is the introduced parameter, ImIs an m-dimensional identity matrix. W is a matrix of lower dimension than K, whose solution is (KLK + muI)-1KMK. Using TCA algorithm to convert source domain characteristic variable XSAnd target domain feature variable XTMapping to a new space to obtain a new source domain data characteristic XSnewAnd target domain data characteristics XTnew. The number of rows in matrix X is the number of samples and the columns are the total number of features.
Step 2.2) establishing a soft measurement model based on RELM:
after the above TCA process operation, XSnewAnd XTnewThe distribution in space is similar. With { XSnew,YSEstablishing a regularization extreme learning machine soft measurement model, wherein an optimization objective function of the regularization extreme learning machine soft measurement model is as follows:
s.t.j h(xi)β＝yi-ξi,i＝1,2,...,m
wherein x isSiIs a characteristic variable of the source domain sample, ySiFor the key variable of the source domain sample reality, h (-) is a function for solving the hidden layer matrix, beta is the output weight, xiiThe prediction error of the ith sample is shown, and gamma is the regularization coefficient of the model. Substituting the constraint term into a target loss function formula to convert into:
obtaining a model by Moore-Penrose (Moore-Penrose generalized inverse) methodOutput matrix of
H is the hidden layer matrix of the RELM model. Thus, for unlabeled regime XTnewThe label value obtained by the regularization limit learning machine is:
further, the process of the step 3) is as follows:
step 3.1) root mean square error RMSE evaluation:
the root mean square error is defined as follows:
in the formula: r is the total amount of the test set samples; y istRepresenting input samples xtThe true tag value of;
representing input samples xtThe predicted value of (2). The smaller the RMSE is, the better the prediction performance of the model is;
step 3.2) evaluation of average absolute error MAE value:
the mean absolute error can be expressed as:
the smaller the MAE, the smaller the prediction error of the model, and the more excellent the method can be demonstrated.
The invention has the following beneficial effects: the invention reduces the characteristic distribution distance of the source domain and the target domain based on the migration component analysis method, so that the two groups of data are distributed similarly. And then, a soft measurement model is quickly established on the mapped source domain penicillin data by adopting a regularization extreme learning machine method, and the concentration of the target domain penicillin is predicted. The method solves the problem that the soft measurement model is difficult to establish when the penicillin data has no label data under certain working condition.
Drawings
FIGS. 1(a), (b) results of predicting penicillin concentrations in condition 1 and condition 2, respectively, for the RELM model established for condition 1;
FIGS. 2(a) and (b) are the results of predicting penicillin concentrations under working conditions 1 and 2 by using the RELM model established under working condition 1 after being treated by the TCA method, respectively;
FIGS. 3(a), (b) results of predicting penicillin concentrations in condition 1 and condition 3, respectively, for the RELM model established for condition 1;
FIGS. 4(a) and (b) are the results of predicting penicillin concentrations under working conditions 1 and 3 by using the RELM model established under working condition 1 after being treated by the TCA method, respectively;
FIGS. 5(a), (b) results of predicting penicillin concentrations for working condition 2 and working condition 3, respectively, for the RELM model established for working condition 2;
FIGS. 6(a) and (b) are the results of predicting penicillin concentrations under working conditions 2 and 3 by the RELM model established under working condition 2 after being processed by the TCA method, respectively;
fig. 7(a), (b), and (c) are data distribution scatter diagrams before and after the TCA method is adopted when the working condition 1 is migrated to the working condition 2, the working condition 1 is migrated to the working condition 3, and the working condition 2 is migrated to the working condition 3, respectively;
FIG. 8 is a flow chart of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Referring to fig. 1 to 8, a fermentation process soft measurement modeling method based on rapid component transfer learning includes the following specific steps:
(1) acquisition and pretreatment of penicillin data:
step 1.1: obtaining penicillin data
The penicillin data adopted by the invention is obtained by analog simulation of a Benchmark simulation platform (Pensim). The penicillin concentration is a key variable to be predicted in the process, and six related variables are used as auxiliary variables, namely carbon dioxide concentration, aeration rate, substrate feeding temperature, stirrer power, culture dish volume and pH value.
Step 1.2: data normalization processing
In order to accelerate the convergence speed of the model, reduce the training time of the model and simultaneously reduce the influence among different dimensional data, the data is normalized, and the formula is as follows:
in the formula, x is data after normalization processing; a is the collected raw data; a isminIs the minimum value in the original data; a ismaxIs the maximum value in the raw data.
Step 1.3: determining source domain and target domain condition data
And randomly selecting one working condition from the different working condition data sets after the normalization processing as a source domain working condition data set, and randomly selecting one working condition from the rest working conditions as a target domain working condition data set. The source domain data has both characteristic variables and key variables, denoted as { X }S,YSThe target domain data only has characteristic variables and is marked as { X }T}。
(2) Establishing a model of rapid component transfer learning:
step 2.1: feature migration of source domain and target domain
It is assumed that there is a feature map ψ such that the distribution P (ψ (X) of the source domain and the destination domain after mappingS))＝P(ψ(XT)). The distance between the source domain and the target domain after mapping is measured by Maximum Mean Difference (MMD), which is expressed as:
where m is the number of source domain samples, n is the number of target domain samples, the above equation represents the distance of two distributions measured in the regenerated kernel hilbert space, and H represents the regenerated kernel hilbert space.
Transforming the solving process of the MMD distance into a kernel function by adopting a kernel functionLearning Process of numbers, D (X)S,XT) Conversion to the following form:
tr(KL)-λtr(K)
wherein tr (·) is the inverse of the matrix, λ is the introduced parameter, K is the kernel matrix obtained by kernel function mapping, L is the matrix introduced by the MMD algorithm, and the calculation method of each element is as follows:
wherein D isSRepresenting the source domain, DTRepresenting the target domain, xiAnd xjRepresenting the characteristic variables of any two samples in the source domain or the target domain. Then, the problem of solving tr (KL) - λ tr (K) is converted into the following optimization problem:
s.t.WTKMKW＝Im
where M is the central matrix, μ is the introduced parameter, ImIs an m-dimensional identity matrix. W is a matrix of lower dimension than K, whose solution is (KLK + muI)-1KMK. Using TCA algorithm to convert source domain characteristic variable XSAnd target domain feature variable XTMapping to a new space to obtain a new source domain data characteristic XSnewAnd target domain data characteristics XTnew. The number of rows in matrix X is the number of samples and the columns are the total number of features.
Step 2.2: soft measurement model establishment based on RELM
After the above TCA process operation, XSnewAnd XTnewThe distribution in space is similar. With { XSnew,YSEstablishing a regularization extreme learning machine soft measurement model, wherein an optimization objective function of the regularization extreme learning machine soft measurement model is as follows:
s.t.j h(xi)β＝yi-ξi,i＝1,2,...,m
wherein x isSiIs a characteristic variable of the source domain sample, ySiFor the key variable of the source domain sample reality, h (-) is a function for solving the hidden layer matrix, beta is the output weight, xiiThe prediction error of the ith sample is shown, and gamma is the regularization coefficient of the model. Substituting the constraint term into a target loss function formula to convert into:
the Moore-Penrose generalized inverse method is adopted to obtain an output matrix of the model as
H is the hidden layer matrix of the RELM model. Thus, for unlabeled regime XTnewThe label value obtained by the regularization limit learning machine is:
(3) and (3) evaluating model performance:
step 3.1: root Mean Square Error (RMSE) evaluation
The root mean square error is defined as follows:
in the formula: r is the total amount of the test set samples; y istRepresenting input samples xtThe true tag value of;
representing input samples xtThe predicted value of (2). The smaller the RMSE is, the better the prediction performance of the model is;
step 3.2: mean absolute error MAE value evaluation
The mean absolute error can be expressed as:
the smaller the MAE, the smaller the prediction error of the model, and the more excellent the method can be demonstrated.
(4) Prediction of penicillin data:
penicillin data of 3 working conditions are obtained, namely working condition 1, working condition 2 and working condition 3, and each working condition has 200 samples. The effectiveness of the method is verified through 3 pairs of working condition migration, namely the working condition 1 is migrated to the working condition 2, the working condition 1 is migrated to the working condition 3, and the working condition 2 is migrated to the working condition 3. FIGS. 1(a) and (b) show the results of predicting penicillin concentrations in conditions 1 and 2, respectively, using the RELM method alone to model condition 1. Aligning the characteristic distribution of the working condition 1 and the working condition 2 by a TCA method, then establishing an RELM model based on the mapped working condition 1 data, and respectively obtaining penicillin concentration prediction results of the working condition 1 and the working condition 2 in the graphs (a) and (b) of the graphs. As can be seen from the figure, the penicillin concentration under the working condition 2 predicted by the TCA and RELM methods is better fitted with the original data value, and meanwhile, the accuracy of the penicillin concentration prediction result under the working condition 1 is also ensured. Similarly, fig. 3(a) and (b) are the results of predicting penicillin concentrations in condition 1 and condition 3 by means of the RELM model established in condition 1, respectively, and fig. 4(a) and (b) are the results of predicting penicillin concentrations in condition 1 and condition 3 by means of the RELM model established in condition 1 after the TCA method, respectively. Fig. 5(a) and (b) are the results of prediction of penicillin concentrations in conditions 2 and 3 by means of the RELM model established in condition 2, respectively, and fig. 6(a) and (b) are the results of prediction of penicillin concentrations in conditions 2 and 3 by means of the RELM model established in condition 2 after the TCA method. At the same time, the RMSE and MAE values of the target domain prediction results for each set of operating conditions are recorded in table 1. From table 1, it can be seen that the TCA + rel method can significantly reduce the RMSE and MAE values of target domain prediction, further verifying the effectiveness of the TCA + rel method in the prediction of unlabeled penicillin data concentration.
Fig. 7(a), (b), and (c) are data distribution scatter diagrams before and after the working condition 1 is migrated to the working condition 2, the working condition 1 is migrated to the working condition 3, and the working condition 2 is migrated to the working condition 3, respectively, and it can be seen from the diagrams that the data distribution between the original two working conditions is far, and the TCA method draws the distance between the two working conditions, which explains well why TCA + RELM can improve the accuracy of the target domain penicillin concentration prediction.
TABLE 1
The method is based on the knowledge of transfer learning, and solves the problem of difficult modeling of unlabeled data under a certain working condition in the penicillin process. And reducing the distance between the characteristic distributions of the source domain and the target domain by a TCA method, then establishing a RELM model in a space with similar characteristic distributions, and predicting the concentration of the penicillin in the working condition of the target domain only with characteristic data. The method has high accuracy, universality and universality.
The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.
Claims (4)
1. A fermentation process soft measurement modeling method based on rapid component transfer learning is characterized by comprising the following steps:
1) acquisition and pretreatment of penicillin data:
penicillin data is obtained through Pensim simulation of a Benchmark simulation platform, and normalization processing is carried out on original data in order to accelerate the convergence speed of a model and reduce the influence among data of different dimensions; in the modeling stage of the soft measurement model in the multi-working-condition process, one working condition is a source domain working condition with auxiliary variables, namely characteristic variables and key variables, and the other working condition is a target domain working condition with only the characteristic variables and no key variables;
2) establishing a model of rapid component transfer learning:
based on a migration component analysis TCA method, reducing the marginal probability distribution distance between a source domain and a target domain, thereby completing feature migration; in the space of the source domain and the target domain which are aligned through the TCA, establishing a regularization limit learning machine model based on the mapped source domain data, and quickly predicting the penicillin concentration of the working condition of the target domain;
3) and (3) evaluating model performance:
and introducing an evaluation index root mean square error RMSE and an average absolute error MAE.
2. The fermentation process soft measurement modeling method based on rapid component transfer learning according to claim 1, characterized in that the process of step 1) is as follows:
step 1.1) obtaining penicillin data:
the penicillin concentration is a key variable to be predicted in the process, and six correlation variables are used as auxiliary variables, namely carbon dioxide concentration, aeration rate, substrate feeding temperature, stirrer power, culture dish volume and pH value;
step 1.2) data normalization treatment:
in order to accelerate the convergence speed of the model, reduce the training time of the model and simultaneously reduce the influence among different dimensional data, the data is normalized, and the formula is as follows:
in the formula, x is data after normalization processing; a is the collected raw data; a isminIs the minimum value in the original data; a ismaxIs the maximum value in the original data;
step 1.3) determining working condition data of a source domain and a target domain:
randomly selecting one working condition from different working condition data sets after normalization processing as a source domain working condition data set, and randomly selecting one working condition from the rest working conditions as a target domain working condition data set; the source domain data has both characteristic variables and key variables, denoted as { X }S,YSThe target domain data is uniqueSign variable, denoted as { XT}。
3. The fermentation process soft measurement modeling method based on rapid component transfer learning according to claim 1, characterized in that the process of step 2) is:
step 2.1) feature migration of the source domain and the target domain:
it is assumed that there is a feature map ψ such that the distribution P (ψ (X) of the source domain and the destination domain after mappingS))＝P(ψ(XT) ); and measuring the distance between the source domain and the target domain after mapping by adopting the MMD with the maximum mean difference, wherein the distance is expressed as follows:
wherein m is the number of source domain samples, n is the number of target domain samples, the above formula represents the distance of two distributions measured in the regenerated kernel hilbert space, and H represents the regenerated kernel hilbert space;
transforming the MMD distance solving process into a kernel function learning process by using a kernel function, D (X)S,XT) Conversion to the following form:
tr(KL)-λtr(K)
wherein tr (·) is the inverse of the matrix, λ is the introduced parameter, K is the kernel matrix obtained by kernel function mapping, L is the matrix introduced by the MMD algorithm, and the calculation method of each element is as follows:
wherein D isSRepresenting the source domain, DTRepresenting the target domain, xiAnd xjA feature variable representing any two samples in the source domain or the target domain; then, the problem of solving tr (KL) - λ tr (K) is converted into the following optimization problem:
s.t.WTKMKW＝Im
where M is the central matrix, μ is the introduced parameter, ImIs an m-dimensional identity matrix, W is a matrix of lower dimension than K, the solution of which is (KLK + muI)-1KMK, the first p eigenvalues; using TCA algorithm to convert source domain characteristic variable XSAnd target domain feature variable XTMapping to a new space to obtain a new source domain data characteristic XSnewAnd target domain data characteristics XTnewThe row number of the matrix X is the sample number, and the columns are the total characteristic number;
step 2.2) establishing a soft measurement model based on a regularization extreme learning machine RELM:
after the above TCA process operation, XSnewAnd XTnewThe distribution in space is similar; with { XSnew,YSEstablishment of regularized extreme learning machine soft measurement model, YSThe key variable of the source domain is represented, and the optimization objective function of the key variable is as follows:
s.t.j h(xSi)β＝ySi-ξi,i＝1,2,...,m
wherein x isSiIs a characteristic variable of the source domain sample, ySiFor the key variable of the source domain sample reality, h (-) is a function for solving the hidden layer matrix, beta is the output weight, xiiRepresenting the prediction error of the ith sample, wherein gamma is a regularization coefficient of the model; substituting the constraint term into a target loss function formula to convert into:
the output matrix of the model obtained by adopting a Muel-Penrose generalized inverse Moore-Penrose method is
H is a hidden layer matrix of the RELM model; thus, for unlabeled regime XTnewThe label value obtained by the regularization limit learning machine is:
4. the fermentation process soft measurement modeling method based on rapid component transfer learning according to claim 1, characterized in that the process of step 3) is:
step 3.1) root mean square error RMSE evaluation:
the root mean square error is defined as follows:
in the formula: r is the total amount of the test set samples; y istRepresenting input samples xtThe true tag value of;
representing input samples xtThe predicted value of (2); the smaller the RMSE is, the better the prediction performance of the model is;
step 3.2) evaluation of average absolute error MAE value:
the mean absolute error can be expressed as:
the smaller the MAE, the smaller the prediction error of the model, and the more excellent the method can be demonstrated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111633686.4A CN114334024A (en) | 2021-12-28 | 2021-12-28 | Fermentation process soft measurement modeling method based on rapid component transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111633686.4A CN114334024A (en) | 2021-12-28 | 2021-12-28 | Fermentation process soft measurement modeling method based on rapid component transfer learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114334024A true CN114334024A (en) | 2022-04-12 |
Family
ID=81016358
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111633686.4A Pending CN114334024A (en) | 2021-12-28 | 2021-12-28 | Fermentation process soft measurement modeling method based on rapid component transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114334024A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116484723A (en) * | 2023-03-31 | 2023-07-25 | 昆明理工大学 | Dynamic multi-layer domain self-adaption based fermentation process soft measurement modeling method |
| CN117096070A (en) * | 2023-10-19 | 2023-11-21 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Semiconductor processing technology abnormality detection method based on field self-adaption |
-
2021
- 2021-12-28 CN CN202111633686.4A patent/CN114334024A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116484723A (en) * | 2023-03-31 | 2023-07-25 | 昆明理工大学 | Dynamic multi-layer domain self-adaption based fermentation process soft measurement modeling method |
| CN116484723B (en) * | 2023-03-31 | 2024-05-31 | 昆明理工大学 | Dynamic multi-layer domain self-adaption based fermentation process soft measurement modeling method |
| CN117096070A (en) * | 2023-10-19 | 2023-11-21 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Semiconductor processing technology abnormality detection method based on field self-adaption |
| CN117096070B (en) * | 2023-10-19 | 2024-01-05 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Semiconductor processing technology abnormality detection method based on field self-adaption |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114334024A (en) | Fermentation process soft measurement modeling method based on rapid component transfer learning | |
| CN108875933B (en) | Over-limit learning machine classification method and system for unsupervised sparse parameter learning | |
| CN114360652B (en) | Cell strain similarity evaluation method and similar cell strain culture medium formula recommendation method | |
| CN108334943A (en) | The semi-supervised soft-measuring modeling method of industrial process based on Active Learning neural network model | |
| CN107168063B (en) | Soft measurement method based on integrated variable selection type partial least square regression | |
| CN105868164B (en) | A kind of soft-measuring modeling method based on the linear dynamic system model for having supervision | |
| CN110046377B (en) | Selective integration instant learning soft measurement modeling method based on heterogeneous similarity | |
| CN111079856A (en) | CSJITL-RVM-based multi-period intermittent process soft measurement modeling method | |
| CN114330114B (en) | Beryllium bronze alloy corrosion rate prediction method based on quantum support vector machine | |
| CN114048682B (en) | Rolling bearing acoustic emission intelligent diagnosis method based on fusion of optimized wavelet basis and multidimensional depth characteristics | |
| CN115472233A (en) | Semi-supervised integrated industrial process soft measurement modeling method and system based on thermal diffusion label propagation | |
| CN117497038B (en) | Method for rapidly optimizing culture medium formula based on nuclear method | |
| CN101206727B (en) | Data processing apparatus, data processing method | |
| CN112651173B (en) | Agricultural product quality nondestructive testing method based on cross-domain spectral information and generalizable system | |
| Utami et al. | Hausman and taylor estimator analysis on the linear data panel model | |
| CN116776252A (en) | Industrial process soft measurement method and system for improving Mallow's Cp variable selection | |
| CN113988311A (en) | Quality variable prediction method, quality variable prediction device, terminal and storage medium | |
| Xiao et al. | Broiler growth performance analysis: from correlation analysis, multiple linear regression, to neural network | |
| TWI652481B (en) | Method for detecting drug resistance of microorganism | |
| Šizling et al. | Mathematically and biologically consistent framework for presence-absence pairwise indices | |
| CN112115646B (en) | Simulation method and system for oil refining chemical production process | |
| CN118609669B (en) | Mass spectrum detection method and system for microbial drug sensitivity, storage medium and electronic equipment | |
| CN112086143B (en) | Small molecule drug virtual screening method and device based on unsupervised domain adaptation | |
| CN107038147A (en) | A kind of industrial process flexible measurement method based on multi-sampling rate regression model | |
| CN114841000B (en) | Soft measurement modeling method based on modal common feature separation |
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 |