CN112925202A - Fermentation process stage division method based on dynamic feature extraction - Google Patents
Fermentation process stage division method based on dynamic feature extraction Download PDFInfo
- Publication number
- CN112925202A CN112925202A CN202110078581.0A CN202110078581A CN112925202A CN 112925202 A CN112925202 A CN 112925202A CN 202110078581 A CN202110078581 A CN 202110078581A CN 112925202 A CN112925202 A CN 112925202A
- Authority
- CN
- China
- Prior art keywords
- matrix
- sampling
- moment
- stage
- time slice
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
Abstract
The invention discloses a stage division method based on dynamic feature extraction, which is used for solving the problem that the dynamic feature change in the capturing process of the existing method is not sensitive enough. Firstly, original data are unfolded along batches, partial least square analysis is carried out on each time slice matrix to obtain a scoring matrix of a process variable and a quality variable, and an AP algorithm is adopted to cluster the combined scoring matrix to realize the 1 st step of division; and then, extracting dynamic characteristics representing the dynamic property of the process by adopting an encoder-decoder model, carrying out 2 nd-step division on the dynamic characteristics by adopting an AP algorithm, and finally dividing the production process into different stable stages and transition stages by comprehensively analyzing two-step division results. After the stage division is completed, an attention LSTM quality prediction model can be further established for each divided stage respectively for expanding application.
Description
Technical Field
The invention relates to the technical field of process monitoring based on data driving, in particular to a fermentation process stage division method based on dynamic feature extraction, which is provided aiming at the multi-stage characteristics of the fermentation process. The invention realizes the 1 st step division by utilizing a joint scoring matrix; and then, extracting dynamic characteristics representing the dynamic property of the process by adopting an encoder-decoder model to realize the 2 nd step division, and finally dividing the production process into different stable stages and transition stages by comprehensively analyzing the two-step division results.
Background
The intermittent process is an extremely important production mode in the modern industrial process, and is widely applied to the production of various high-value-added products such as medicines, foods, biochemical engineering, semiconductors and the like. However, in certain production processes, such as fermentation processes, the quality variables are difficult to measure online. The manual off-line measurement is complex and time-consuming in operation, extra workload is brought to people, and in case of careless operation in the sampling process, the whole fermentation tank is possibly infected with mixed bacteria, so that great waste is brought to production. Therefore, the importance of online quality prediction is increasingly highlighted to improve product quality and process efficiency.
The multi-stage characteristic is a typical characteristic in the fermentation process, and obviously, the establishment of a single model for the whole production process is incomplete, so that a lot of domestic and foreign experts carry out a great deal of research on the stage division. Yu et al propose a stage division method based on a Gaussian Mixture Model (GMM), Camacho et al propose a linear local model approximation method to achieve the purpose of stage division, Lu et al propose an intermittent process sub-period division method based on K-means, but these methods belong to hard classification methods and ignore transition information between two stages. Compared with the operating state of each stable stage, the transition between the stages does not represent the main process operation mechanism, but is a general phenomenon and important process behavior, and shows a dynamic gradual change trend. On the basis, ZHao et al introduces fuzzy membership as a weight coefficient of two adjacent stable stages, then uses K-means to perform stage division, and Qi et al uses fuzzy C-means clustering (FCM) to perform stage division on time slices. However, the method does not consider the influence of quality variables on the stage division, and each sampling moment is independently regarded as a unit for analysis, so that the relation between the front and the back of different moments is ignored, and the method is not sensitive to capturing the dynamic characteristic change of the process. The multi-stage characteristics of the fermentation process are reflected by the dynamic property of the change of the characterization process to a great extent, and the fermentation process has different dynamic characteristics in different stages.
Disclosure of Invention
The multi-stage characteristics of the fermentation process are reflected by the dynamics of the representation process change to a great extent, and if the process dynamic characteristic change is ignored, the whole production process cannot be accurately divided into a plurality of sub-stages, so the method is designed by adopting a design idea based on dynamic characteristic extraction, and the problem that the division of the fermentation process stages in the prior art is inaccurate is solved. The invention mainly has the following innovation points: 1) because the relation before and after the time slice is ignored in the time slice PLS modeling, and the traditional PLS method belongs to a static method, the method provided by the invention takes the static characteristics and the dynamic characteristics of the original data into consideration, and the dynamic characteristics extracted by an encoder-decoder model are adopted to carry out 2-time division to make up for the deficiency of 1-time division. 2) In most cases, it is difficult to know the exact number of stages for a complex, strange industrial process. The stage division method designed by the invention does not need to specify the exact number of stages and meets the requirement of practical application. 3) The whole operation stage can be divided into a stable stage and a transition stage by comprehensively analyzing the division results of the two steps, so that the division results conform to the actual production process.
The fermentation process stage division method based on dynamic feature extraction is characterized by comprising the following steps of:
i, carrying out data preprocessing;
the data includes historical process data and historical quality variable data, wherein the quality variable refers to product concentration, and the historical process data is J of a fermentation process for producing a productxThree-dimensional matrix X (I × J) composed of variables, I batches, and K sampling moments of each batchxXK). Preprocessing the three-dimensional data, specifically: three-dimensional process data X (I × J)xxK) is expanded along the batch direction to obtain K time slice data submatrices Xk(I×Jx) Wherein X iskA time slice data matrix representing the kth time instant, K being 1,2, …, K; after the development along the batch, each time slice matrix is normalized according to the following formula:
wherein J represents a process variable, J is 1,2, … J(ii) a K represents the sampling time, K is 1,2, … K;
xk,j，
and SkjAre all in the form of I x 1 dimensional data.
Represents the normalized k-th time slice matrix; x is the number ofk,jThe time slice matrix is a kth time slice matrix and consists of sampling values of the jth process variable at the kth sampling moment in all I batches;
the average value matrix of the kth time slice matrix is composed of the average sampling value of the jth process variable at the kth sampling moment in all I batches; sk,jThe standard deviation matrix is a kth time slice matrix and consists of the standard deviation of sampling values of the jth process variable at the kth sampling moment in all I batches; the same process is performed on the historical quality variable data Y (I × Jy × K), where Jy represents the quality variable number.
II, dividing in a first step;
1) after standardization, performing PLS regression analysis on each time slice data matrix to obtain a score vector T of a corresponding process variable data set X and a score vector U of a corresponding quality variable data set Y; and (4) arranging the two scoring matrixes into a matrix from left to right to obtain a combined scoring matrix of each time slice data matrix.
2) And calculating a similarity matrix S of the samples, and inputting the similarity matrix into an AP clustering algorithm for clustering. Considering that the 1 st partition is based on the static characteristics of data, and therefore there may be more jump points, for this reason, when a time constraint is added in calculating the similarity of the sample matrix at two time instants, the calculation formula is as follows:
S(i,k)＝-||(T,U)i-(T,U)k||*A-|k-j|*B
the A and B are determined by a cross-validation method, the values of the A and B are required to ensure that samples in the same stage are compact, samples in different stages are discrete, and the fluctuation phenomenon of the conversion part of two adjacent stages is allowed.
3) If the conversion part between two adjacent stages has no fluctuation, calculating a Silhouette value corresponding to each sampling moment, taking the average value of the Silhouette criterion of the sampling points as a judgment threshold value through multiple experiments in order to reasonably judge whether the clustering effect is good or not, and judging that the Silhouette value is good when the Silhouette value is greater than a set threshold value; otherwise it is inferior.
III, dividing;
1) and further processing the two-dimensional data obtained after the batch expansion according to a variable expansion mode, and then continuously sampling each batch of data by using a sliding window with the window width of d to obtain an input sequence required by an encoder-decoder model. And extracting dynamic characteristics by adopting a trained encoder-decoder model. And combining the dynamic characteristics of different batches at each moment into a dynamic characteristic time slice C.
2) And calculating a similarity matrix S among the dynamic characteristic time slice data matrixes, and inputting the similarity matrix into an AP clustering algorithm for clustering. Because the extracted dynamic characteristics have strong correlation with time, the similarity between the defined samples is as follows:
wherein the content of the first and second substances,
represents k1The dynamic feature matrix of the time of day,
represents k2A dynamic feature matrix of time instants.
3) Measuring the 'good and bad' degree of the current sample point divided into the corresponding stages by using a Silhouette criterion, taking the average value of the Silhouette criterion of the sampling points as a judgment threshold value through multiple experiments in order to reasonably judge whether the clustering effect is good or bad, and judging that the sample point is good when the Silhouette value is greater than the set threshold value; otherwise it is inferior.
And IV, comprehensively analyzing the two-step division results, judging which stage each sampling moment belongs to, and finishing the final stage division, wherein the steps are as follows:
for the sampling moment k, when the dynamic feature clustering effect is excellent, but the static feature clustering effect is inferior, the moment is in a reasonable stable stage, namely the moment is in the same stage as the last sampling moment;
for the sampling moment k, when the static characteristic clustering effect is excellent, but the dynamic characteristic clustering effect is poor, the moment has a transition trend, and the moment is still assigned to be a reasonable stable stage, namely the moment is in the same stage as the last sampling moment;
for the sampling moment k, when the dynamic feature clustering effect is poor and the static feature clustering effect is poor, the moment is in a transition stage, that is, the moment is in a different stage from the last sampling moment.
Advantageous effects
The method realizes the multi-stage division of the intermittent process, combines the score matrix U of the quality variable and the score matrix T of the process variable to obtain a combined score matrix in order to simultaneously consider the influence of the process variable and the quality variable on the production process stage division during the stage division, and uses an AP clustering algorithm to divide the combined score matrix of each time slice by the step 1; and then, extracting dynamic characteristics by adopting a trained encoder-decoder model, wherein the input degree of each moment can obtain the dynamic characteristics of the corresponding moment. And then combining the dynamic features obtained from all batches to obtain dynamic feature time slices, performing 2 nd-step division on all time slices by using an AP clustering algorithm, and finally comprehensively analyzing the two-step division results, so that the accuracy of stage division can be effectively improved, the accuracy of online quality prediction is improved, and the method has great significance for the quality prediction of the industrial process.
Drawings
FIG. 1 is a view developed along the batch direction;
FIG. 2 is a diagram showing an encoder-decoder model structure;
FIG. 3 is a diagram illustrating a data processing process;
FIG. 4 is a graphical representation of the staging results;
FIG. 5 is a graphical representation of the results of an attention-based LSTM prediction;
FIG. 6 is a graphical representation of the prediction results based on attention LSTM after one-step partitioning;
FIG. 7 is a graphical representation of the results of attention-based LSTM prediction after two-step partitioning;
FIG. 8 is a flow chart illustrating the present invention and its expanded applications.
Detailed Description
The Pensim simulation platform is a penicillin simulation platform with relatively great influence on the world, and related researches show that the Pensim simulation platform is practical and effective. The experimental simulation is carried out on the fermentation process of penicillin based on the platform, the fermentation time of each batch is set to be 400h, the sampling time interval is 1h, 11 process variables (including ventilation rate, stirring power, substrate feeding rate, substrate temperature, dissolved oxygen concentration, exhaust carbon dioxide concentration, pH value, temperature, reaction heat, cold water flow acceleration rate and substrate concentration) and 1 quality variable (product concentration) are selected for monitoring, 40 batches of penicillin fermentation processes are selected for training and modeling, and 20 batches of penicillin fermentation processes are predicted.
TABLE 1 penicillin fermentation Process variables
Table 1 Process variables of Penicillin fermentation
Based on the above description, according to the invention, the specific process is implemented as follows:
and I, carrying out data preprocessing. The process data for 40 batches selected herein is denoted X400×40×11Spreading the time slice data in the batch direction to obtain 400 time slice data submatrices X as shown in FIG. 1k(40×11)。
II, carrying out first-step division. 1) For time slice Xk(40 x 11) a normalization process was performed,PLS regression analysis is then performed on each time slice data pair to obtain corresponding score vectors T and U for X and Y, which are good characterizations of data X and Y.
2) And after the data are processed, clustering the obtained joint score matrix by adopting an AP algorithm. And (3) comprehensively considering the similarity and time constraint between the matrixes, determining that the value of the parameter A is 2 and the value of the parameter B is 1 by adopting a cross-validation method, and defining the similarity of the AP clustering algorithm as follows:
the division results are shown in FIG. 4, and it is obvious from the observation of the division results that there is no fluctuation at times 1-43, 68-142, 197-266, and 301-400, and the fluctuation is very strong at times 44-67, 143-196, 267-300.
III, carrying out second-step division.
1) The two-dimensional data obtained after the batch expansion is further processed according to a variable expansion mode, then each batch of data is continuously sampled by a sliding window with the window width d to obtain an input sequence required by an encoder-decoder model, and the data expansion sampling process is shown in fig. 3. An encoder-decoder model is adopted to extract dynamic characteristics, the structure of the encoder-decoder model is shown in figure 2, LSTM is adopted as a basic unit in the model, the main parameters to be set comprise the number of nodes of an implicit layer and the size of a sampling time window, and a grid search method is adopted to alternately optimize and determine the two parameters. The number of hidden layer nodes is finally selected to be 24 and the sampling window width is 10.
2) And calculating a similarity matrix S' among the dynamic characteristic time slice data matrixes, and inputting the similarity matrix into an AP clustering algorithm for clustering. The similarity is defined here as:
the observation result shows that different from the similarity function of the static characteristic, the time constraint is not considered at the moment, the dynamic characteristic time slice can be ideally divided, and the strong fluctuation state when two adjacent stages are converted does not exist. Therefore, a Silhouette criterion is introduced to evaluate the clustering effect. Through several experiments, it was considered reasonable to set the threshold value to 0.7. A value of Silhouette lower than 0.7 is considered to be poor for clustering. Since the sampling width is 10, the dynamic features at the first 9 moments cannot be extracted, and the 1 st sub-stage moment is far greater than 9 by observing the static feature division result, so that the division result at the first 9 moments and the 10 th sub-stage are not divided into the same stage. According to the Silhouette value, the clustering effect at the time of 41-54, 77-175 and 238-275 is poor, and the clustering effect at other times is good.
And IV, finally, comprehensively analyzing and dividing 1-43, 55-142, 176-266, wherein the time 276-400 is a stable stage, and the time 44-54, 143-175, 267-275 is a transition stage.
According to the division results, an attention LSTM soft measurement model is established for each stage respectively for quality prediction experiment verification. And for the process variable acquired on line, judging which stage the moment belongs to according to the time point, and predicting the quality by using the model of the stage. In order to show the specific embodiment of the invention more clearly and intuitively, the simulation result of the invention is presented below. The penicillin fermentation process has a batch duration of 400h and a sampling interval of 1 h. The selection of 11 major process variables is shown in table 1, one mass variable (product concentration) is selected. 40 batches of penicillin fermentation processes were selected for training modeling and 20 batches for prediction. In order to ensure the consistency of the simulation environment and the actual production field environment, certain white noise interference is added to the training sample set.
TABLE 2 prediction accuracy index for penicillin concentration
Table 2 Prediction accuracy index of penicillin concentration
To verifyThe effectiveness of the method provided by the invention is compared with an attention LSTM integrated prediction model without stages and a method for directly establishing the attention LSTM integrated prediction model according to the first-step division result through experiments. Using the predicted Root Mean Square Error (RMSE) and R2Evaluation indexes as models:
as can be seen from fig. 5 to 7, in comparison with the method in which the classification is not considered and only one-step classification is considered, the classification method based on the dynamic feature extraction is improved in the accuracy of the quality prediction.
Claims (5)
1. The fermentation process stage division method based on dynamic feature extraction is characterized by comprising the following steps of:
i, carrying out data preprocessing;
the data includes historical process variable data indicating product concentration in the fermentation process and historical quality variable data indicating J of the fermentation process for producing a productxThree-dimensional matrix X (I × J) composed of variables, I batches, and K sampling moments of each batchxXK). Preprocessing the three-dimensional data, specifically: three-dimensional process data X (I × J)xxK) is expanded along the batch direction to obtain K time slice data submatrices Xk(I×Jx) Wherein X iskA time slice data matrix representing the kth time instant, K being 1,2, …, K; after the development along the batch, each time slice matrix is normalized according to the following formula:
wherein J represents a process variable, J ═ 1,2, … J; k represents the sampling time, K is 1,2, … K;
xk,j，
and Sk,jAre all in the form of I x 1 dimensional data.
Represents the normalized k-th time slice matrix; x is the number ofk,jThe time slice matrix is a kth time slice matrix and consists of sampling values of the jth process variable at the kth sampling moment in all I batches;
the average value matrix of the kth time slice matrix is composed of the average sampling value of the jth process variable at the kth sampling moment in all I batches; sk,jThe standard deviation matrix is a kth time slice matrix and consists of the standard deviation of sampling values of the jth process variable at the kth sampling moment in all I batches; the same process is performed on the historical quality variable data Y (I × Jy × K), where Jy represents the quality variable number.
II, dividing in a first step;
1) after standardization, performing PLS regression analysis on each time slice data matrix to obtain a score vector T of corresponding historical process data X and a score vector U of corresponding historical quality variable data Y; arranging the two scoring matrixes into a matrix from left to right to obtain a combined scoring matrix of each time slice data matrix;
2) calculating a similarity matrix S among all the joint score matrixes, inputting the similarity matrix S into an AP clustering algorithm for clustering, and completing primary stage division of all sampling moments by using static characteristics;
3) if the conversion part between two adjacent stages has no fluctuation, the Silhouette criterion is adopted to measure the degree of 'good and bad' of the score to the corresponding stage during each sampling;
III, dividing;
1) and further processing the two-dimensional data obtained after the batch expansion according to a variable expansion mode, and then continuously sampling each batch of data by using a sliding window with the window width of d to obtain an input sequence required by an encoder-decoder model. And extracting dynamic characteristics by adopting a trained encoder-decoder model. Combining the dynamic characteristics of different batches at each moment into a dynamic characteristic time slice C;
2) calculating a similarity matrix S 'among the dynamic characteristic time slice data matrixes, inputting the similarity matrix S' into an AP clustering algorithm for clustering, and completing secondary stage division of all sampling moments by using the dynamic characteristics;
3) measuring the 'good and bad' degree of the current sample point divided into the corresponding stages by adopting a Silhouette criterion;
and IV, comprehensively analyzing the two-step division results, judging which stage each sampling moment belongs to, and finishing the final stage division, wherein the steps are as follows:
for the sampling moment k, when the dynamic feature clustering effect is excellent, but the static feature clustering effect is inferior, the moment is in a reasonable stable stage, namely the moment is in the same stage as the last sampling moment;
for the sampling moment k, when the static characteristic clustering effect is excellent, but the dynamic characteristic clustering effect is poor, the moment has a transition trend, and the moment is still assigned to be a reasonable stable stage, namely the moment is in the same stage as the last sampling moment;
for the sampling moment k, when the dynamic feature clustering effect is poor and the static feature clustering effect is poor, the moment is in a transition stage, that is, the moment is in a different stage from the last sampling moment.
2. The fermentation process staging method based on dynamic feature extraction as claimed in claim 1, characterized in that:
in the second step of the partitioning process, a similarity calculation formula among different time slice matrixes related in the AP clustering algorithm is as follows:
wherein k is1、k2Respectively, represent different sampling instants at which the samples are taken,
represents k1A joint score matrix of the time slice data matrix,
represents k2A joint score matrix of the time slice data matrix; a and B are determined by cross-validation.
3. The fermentation process staging method based on dynamic feature extraction as claimed in claim 2, characterized in that:
the values of A and B are to ensure the compactness among samples in the same stage, the dispersion among samples in different stages and the fluctuation of the conversion part of two adjacent stages.
4. The fermentation process staging method based on dynamic feature extraction as claimed in claim 1, characterized in that:
the clustering method in the second step of the partitioning process is an AP clustering algorithm, wherein the similarity among samples involved in the AP clustering algorithm is as follows:
wherein the content of the first and second substances,
represents k1The dynamic feature matrix of the time of day,
represents k2A dynamic feature matrix of time instants.
5. The fermentation process staging method based on dynamic feature extraction as claimed in claim 1, characterized in that:
the clustering effect in the first step of partitioning and the second step of partitioning is excellent, and the clustering indicates that the samples in the class are compact and the samples in different stages are discrete; otherwise, the clustering effect is inferior.
The specific judgment method comprises the following steps: calculating a Silhouette value corresponding to each sampling moment, taking an average value of a Silhouette criterion of sampling points as a judgment threshold value through multiple experiments in order to reasonably judge whether the clustering effect is good or not, and judging that the Silhouette value is good when the Silhouette value is larger than a set threshold value; otherwise it is inferior.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110078581.0A CN112925202B (en) | 2021-01-19 | 2021-01-19 | Fermentation process stage division method based on dynamic feature extraction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110078581.0A CN112925202B (en) | 2021-01-19 | 2021-01-19 | Fermentation process stage division method based on dynamic feature extraction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112925202A true CN112925202A (en) | 2021-06-08 |
| CN112925202B CN112925202B (en) | 2022-10-11 |
Family
ID=76165328
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110078581.0A Active CN112925202B (en) | 2021-01-19 | 2021-01-19 | Fermentation process stage division method based on dynamic feature extraction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112925202B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5862051A (en) * | 1996-02-03 | 1999-01-19 | Eastman Kodak Company | Method of controlling the effect of raw materials on manufacturability |
| WO2002036812A2 (en) * | 2000-11-03 | 2002-05-10 | Michael Korenberg | Nonlinear system identification for class prediction in bioinformatics and related applications |
| US20180259921A1 (en) * | 2015-11-26 | 2018-09-13 | Jiangnan University | Data-Difference-Driven Self-Learning Dynamic Optimization Method For Batch Process |
| CN108664009A (en) * | 2017-08-03 | 2018-10-16 | 湖州师范学院 | Divided stages based on correlation analysis and fault detection method |
| CN109086999A (en) * | 2018-08-02 | 2018-12-25 | 东南大学 | Filling production lines remote data acquisition analysis system and its exception analysis method |
| CN109116834A (en) * | 2018-09-04 | 2019-01-01 | 湖州师范学院 | A kind of batch process fault detection method based on deep learning |
| CN109308063A (en) * | 2018-12-03 | 2019-02-05 | 北京工业大学 | Fermentation process stage division method based on score matrix |
| CN110032799A (en) * | 2019-04-15 | 2019-07-19 | 北京工业大学 | A kind of the angle similarity divided stages and monitoring method of microbiological pharmacy process |
| US20200002710A1 (en) * | 2018-06-28 | 2020-01-02 | Trustees Of Boston University | Systems and methods for control of gene expression |
| CN111141860A (en) * | 2019-12-19 | 2020-05-12 | 中国农业科学院茶叶研究所 | Green tea aroma dynamic fingerprint analysis and aroma type discrimination method based on ion mobility spectrometry |
| CN111638707A (en) * | 2020-06-07 | 2020-09-08 | 南京理工大学 | Intermittent process fault monitoring method based on SOM clustering and MPCA |
| CN111832703A (en) * | 2020-06-29 | 2020-10-27 | 中南大学 | Sampling interval perception long-short term memory network-based process manufacturing industry irregular sampling dynamic sequence modeling method |
-
2021
- 2021-01-19 CN CN202110078581.0A patent/CN112925202B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5862051A (en) * | 1996-02-03 | 1999-01-19 | Eastman Kodak Company | Method of controlling the effect of raw materials on manufacturability |
| WO2002036812A2 (en) * | 2000-11-03 | 2002-05-10 | Michael Korenberg | Nonlinear system identification for class prediction in bioinformatics and related applications |
| US20180259921A1 (en) * | 2015-11-26 | 2018-09-13 | Jiangnan University | Data-Difference-Driven Self-Learning Dynamic Optimization Method For Batch Process |
| CN108664009A (en) * | 2017-08-03 | 2018-10-16 | 湖州师范学院 | Divided stages based on correlation analysis and fault detection method |
| US20200002710A1 (en) * | 2018-06-28 | 2020-01-02 | Trustees Of Boston University | Systems and methods for control of gene expression |
| CN109086999A (en) * | 2018-08-02 | 2018-12-25 | 东南大学 | Filling production lines remote data acquisition analysis system and its exception analysis method |
| CN109116834A (en) * | 2018-09-04 | 2019-01-01 | 湖州师范学院 | A kind of batch process fault detection method based on deep learning |
| CN109308063A (en) * | 2018-12-03 | 2019-02-05 | 北京工业大学 | Fermentation process stage division method based on score matrix |
| CN110032799A (en) * | 2019-04-15 | 2019-07-19 | 北京工业大学 | A kind of the angle similarity divided stages and monitoring method of microbiological pharmacy process |
| CN111141860A (en) * | 2019-12-19 | 2020-05-12 | 中国农业科学院茶叶研究所 | Green tea aroma dynamic fingerprint analysis and aroma type discrimination method based on ion mobility spectrometry |
| CN111638707A (en) * | 2020-06-07 | 2020-09-08 | 南京理工大学 | Intermittent process fault monitoring method based on SOM clustering and MPCA |
| CN111832703A (en) * | 2020-06-29 | 2020-10-27 | 中南大学 | Sampling interval perception long-short term memory network-based process manufacturing industry irregular sampling dynamic sequence modeling method |
Non-Patent Citations (4)
| Title |
|---|
| XICHANG WANG ER AL.: "On-line quality prediction of batch processes using a new kernel multiway partial least squares method", 《CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS》 * |
| 于涛 等: "一种多阶段发酵过程的联合加权支持向量机建模方法", 《北京化工大学学报(自然科学版)》 * |
| 李征 等: "基于信息增量矩阵的多阶段间歇过程质量预测", 《化工学报》 * |
| 陈修哲 等: "基于多阶段MPLS法的发酵过程质量预测", 《计算机与应用化学》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112925202B (en) | 2022-10-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109060001B (en) | Multi-working-condition process soft measurement modeling method based on feature transfer learning | |
| CN111079836B (en) | Process data fault classification method based on pseudo label method and weak supervised learning | |
| CN107357269A (en) | Multistage process quality prediction method based on mixing MPLS | |
| CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
| CN110046377B (en) | Selective integration instant learning soft measurement modeling method based on heterogeneous similarity | |
| CN103678869A (en) | Prediction and estimation method of flight parameter missing data | |
| CN109308063A (en) | Fermentation process stage division method based on score matrix | |
| CN113031553B (en) | Intermittent process quality prediction method based on automatic clustering combined with partial least squares | |
| CN109886314B (en) | Kitchen waste oil detection method and device based on PNN neural network | |
| CN113311796B (en) | Fermentation process stage division method based on joint typical variable matrix | |
| CN117421618B (en) | Building energy consumption monitoring method and system | |
| CN116738866B (en) | Instant learning soft measurement modeling method based on time sequence feature extraction | |
| CN112925202B (en) | Fermentation process stage division method based on dynamic feature extraction | |
| CN113159225A (en) | Multivariable industrial process fault classification method | |
| CN113223634B (en) | Blast furnace molten iron silicon content prediction method based on two-dimensional self-attention enhanced GRU model | |
| CN113835964A (en) | Cloud data center server energy consumption prediction method based on small sample learning | |
| CN108363830B (en) | Functional clothes hanger-oriented principle scheme non-cooperative game decision method | |
| Haoguang et al. | Study on soft sensing technology of penicillin fermentation based on PLS and SVR | |
| Zhang et al. | Complex batch processes quality prediction using non-Gaussian dissimilarity measure based just-in-time learning model | |
| CN114841000B (en) | Soft measurement modeling method based on modal common feature separation | |
| CN113743652B (en) | Sugarcane squeezing process prediction method based on depth feature recognition | |
| CN117370870B (en) | Knowledge and data compound driven equipment multi-working condition identification and performance prediction method | |
| CN114035529B (en) | ATL-BMA-based nonlinear industrial process low-cost modeling method | |
| CN117648646B (en) | Drilling and production cost prediction method based on feature selection and stacked heterogeneous integrated learning | |
| US20230223099A1 (en) | Predicting method of cell deconvolution based on a convolutional neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |