CN118036658A - Soft measurement method for lactobacillus fermentation process based on CO-MP-DSSM - Google Patents
Soft measurement method for lactobacillus fermentation process based on CO-MP-DSSM Download PDFInfo
- Publication number
- CN118036658A CN118036658A CN202410164563.8A CN202410164563A CN118036658A CN 118036658 A CN118036658 A CN 118036658A CN 202410164563 A CN202410164563 A CN 202410164563A CN 118036658 A CN118036658 A CN 118036658A
- Authority
- CN
- China
- Prior art keywords
- data
- fermentation process
- layer
- dssm
- soft measurement
- 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
- 238000000855 fermentation Methods 0.000 title claims abstract description 61
- 230000004151 fermentation Effects 0.000 title claims abstract description 61
- 241000186660 Lactobacillus Species 0.000 title claims abstract description 43
- 229940039696 lactobacillus Drugs 0.000 title claims abstract description 43
- 238000000691 measurement method Methods 0.000 title claims abstract description 24
- 238000000034 method Methods 0.000 claims abstract description 32
- 238000005259 measurement Methods 0.000 claims abstract description 24
- 230000000694 effects Effects 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims description 45
- 238000004590 computer program Methods 0.000 claims description 12
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000005070 sampling Methods 0.000 claims description 11
- JVTAAEKCZFNVCJ-UHFFFAOYSA-N lactic acid Chemical compound CC(O)C(O)=O JVTAAEKCZFNVCJ-UHFFFAOYSA-N 0.000 claims description 10
- 238000003860 storage Methods 0.000 claims description 10
- 210000004027 cell Anatomy 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 241000894006 Bacteria Species 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 6
- 235000014655 lactic acid Nutrition 0.000 claims description 5
- 239000004310 lactic acid Substances 0.000 claims description 5
- 230000006872 improvement Effects 0.000 claims description 4
- 241001052560 Thallis Species 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000002759 z-score normalization Methods 0.000 claims description 3
- 238000010923 batch production Methods 0.000 abstract description 13
- 230000002411 adverse Effects 0.000 abstract description 7
- 238000013136 deep learning model Methods 0.000 abstract description 7
- 239000002609 medium Substances 0.000 description 8
- 240000006024 Lactobacillus plantarum Species 0.000 description 3
- 235000013965 Lactobacillus plantarum Nutrition 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 235000013305 food Nutrition 0.000 description 2
- 229940072205 lactobacillus plantarum Drugs 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000000047 product Substances 0.000 description 2
- 239000002994 raw material Substances 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 229930182843 D-Lactic acid Natural products 0.000 description 1
- JVTAAEKCZFNVCJ-UWTATZPHSA-N D-lactic acid Chemical compound C[C@@H](O)C(O)=O JVTAAEKCZFNVCJ-UWTATZPHSA-N 0.000 description 1
- 230000001580 bacterial effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229960000074 biopharmaceutical Drugs 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000012467 final product Substances 0.000 description 1
- 239000001963 growth medium Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Landscapes
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
Abstract
The invention discloses a soft measurement method for a lactobacillus fermentation process based on CO-MP-DSSM, and belongs to the technical field of soft measurement based on data driving. The method collects normal fermentation process data of the lactobacillus, and marks pseudo tags on the non-tag data by using multi-player collaborative learning, so that adverse effects on a soft measurement model caused by difficulty in obtaining the tagged data are solved; then, the data are standardized, and the two-dimensional dynamic characteristics of the data are extracted through a two-dimensional sliding window; and finally, the strong nonlinearity and the two-dimensional dynamic characteristic of the data are further integrated through the deep learning model, so that the adverse effect of the strong nonlinearity and the two-dimensional dynamic characteristic of the batch process on the general prediction model is solved. The soft measurement method of the lactobacillus fermentation process based on the CO-MP-DSSM provided by the invention is improved aiming at the characteristics of the lactobacillus fermentation process, so that a better prediction effect is realized by the model, and a smaller prediction error is obtained by the predicted label.
Description
Technical Field
The invention relates to the technical field of soft measurement based on data driving, in particular to a lactobacillus fermentation process soft measurement method based on CO-MP-DSSM (CO-TRAINING WITH multiple PLAYERS IN DEEP soft sensor modeling, multi-player collaborative learning deep soft measurement model).
Background
The batch process refers to a processing mode of adding raw materials in batches in a discrete mode, processing the raw materials through a series of processes designed in advance and finally outputting the products in batches, plays an important role in the industries of food processing, bio-pharmaceuticals and the like, and is increasingly widely used. Among them, lactobacillus fermentation is a typical batch process, and is a very popular research field. Because of the strong nonlinearity of the batch process and the two-dimensional dynamic characteristics in the time batch direction, and because of the expensive instruments and the complex production environment, it is very difficult to directly measure the quality of the product. Soft measurement, which is a computer technology, is an effective method for solving the problem of batch process by modeling and predicting the final product quality variable by inputting auxiliary variables. However, most of the conventional soft measurement models are based on statistical and machine learning methods, but most of the methods do not consider the characteristics of the batch process, and thus the effect is very limited.
Data acquisition is facilitated by advances in computer and sensor technology, and data-driven models are becoming increasingly more interesting. The deep learning model achieves excellent results in many fields and achieves effects far superior to those of the traditional method in the field of batch process soft measurement. However, conventional deep learning models are mostly used for other problems or continuous processes, and do not handle the strong nonlinearity and two-dimensional dynamics of batch processes well. Furthermore, since batch process tags require time-consuming and laborious offline measurements, acquisition of tagged data is difficult, which can degrade the effectiveness of the deep learning model. Therefore, there is a need to build a soft measurement model that takes into account strong nonlinearities, two-dimensional dynamics, and lack of tagged data to ensure the predictive effect of soft measurements of batch processes.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a soft measurement method for a lactobacillus fermentation process based on CO-MP-DSSM, which aims to solve the technical problem that the effectiveness of an established soft measurement model on quality variables is reduced because strong nonlinearity, two-dimensional dynamic characteristics and lack of labeled data are not considered when the traditional soft measurement method based on data driving is applied to the lactobacillus fermentation process.
In a first aspect, the invention provides a method for soft measurement of a lactobacillus fermentation process based on CO-MP-DSSM, comprising the following steps:
Step 1, collecting normal fermentation process data of lactobacillus;
Step 2, performing multi-player collaborative learning training on the plurality of regressors by using the data in the step 1, and predicting a pseudo tag at a moment when the concentration of the bacteria is not measured;
step 3, preprocessing the data obtained in the step 2 to obtain standardized data;
step 4, carrying out two-dimensional dynamic sliding window on the standardized data obtained in the step 3;
step 5, establishing GSTAE a network model for the data obtained in the step 4 after the two-dimensional dynamic sliding window is completed;
And 6, performing iterative training on the GSTAE network model in the step 5 until the set training times are reached.
In one embodiment of the present invention, the fermentation process data in the step 1 is three-dimensional data X (i×j×k) and y (i×1×l), wherein the three-dimensional data includes I batches, each batch has K sampling moments, each sampling moment has J auxiliary variables, and L is the number of times of acquiring the thallus concentration information.
In one embodiment of the present invention, the step 2 specifically includes:
Performing multi-player collaborative learning training on a plurality of kNN regressors by using the data in the step 1, and copying labeled data into a plurality of parts at the beginning, wherein each part corresponds to one regressor as a training set; then iterating, each regressor tries to predict label for each data in the label-free data set, and predicting k neighbors in the label data by using the retrained regressor; the prediction effect is improved, and the data with the largest improvement amplitude can be added into all the rest training sets; after training, the regressors respectively predict labels at the moment of not measuring the concentration of the thalli and average the labels, predict original non-label data and combine the original non-label data with labeled data to obtain y' (I multiplied by 1 multiplied by K);
the label represents a real quality variable at the moment in the fermentation process, and the pseudo label represents a staged prediction result of the model on the quality variable in the training process.
In one embodiment of the present invention, the step 3 includes:
preprocessing the data obtained in the step 2, performing Z-score standardization processing on the data according to batch dimension after the data is transposed, and obtaining standardized data after the data is transposed again;
z-score normalization uses the following formula:
wherein sigma is the standard deviation of the elements in the matrix, N is the number of elements in the matrix of each variable, For the average number of elements in the matrix for each variable, x i and x are the elements in the matrix that need to be normalized, and s is the result of normalization of the corresponding elements.
In one embodiment of the present invention, the sliding window size of the two-dimensional dynamic sliding window in the step 4 is p×q, and the raw data is obtained by sliding the window, and the data shape is p×q× (I-p+1) × (K-q+1) ×j; the size and the number of the sliding windows are respectively compressed into one dimension, and the data is obtained in the shape of (p multiplied by q) multiplied by [ (I-p+1) multiplied by (K-q+1) ]multipliedby J, so that the data is convenient to train; the two-dimensional dynamic sliding window slides along the time direction, and moves one frame along the batch direction after the sliding is completed, and the sliding is restarted.
In one embodiment of the present invention, the GSTAE network model in the step 5 is a fully connected neural network, which includes a plurality of layers, the number of neurons in the first layer is equal to the number of variables, and then the number of neurons in each layer is gradually reduced; each layer and the previous layer are an input layer and a hidden layer of the self-encoder AE, and an attempt is made to reconstruct the input during training; each tier contains a gating unit for controlling the effect of the tier results on the final output.
In one embodiment of the present invention, the GSTAE network model includes a pre-training phase and a fine-tuning phase in a training phase;
in the pre-training phase, each layer of self-encoder of GSTAE network model tries to reconstruct the input and predict the quality change, the calculation method is as follows:
Wherein, And/>W 1 and W 2 are the weight matrices of the input layer to the hidden layer and the hidden layer to the output layer, respectively, and b 1 and b 2 are the bias vectors of the input layer to the hidden layer and the hidden layer to the output layer, respectively, f and/>, respectivelyThe activation functions are respectively from the input layer to the hidden layer and from the hidden layer to the output layer;
In the fine tuning stage, GSTAE the network model predicts the variable labels and tries to fit the true values, the loss is calculated as follows:
Wherein y i and The actual quality variable and the predicted quality variable, respectively.
In one embodiment of the present invention, the calculation formula of the GSTAE network model prediction variable label is as follows:
Where n is the number of layers of the GSTAE network model, W gi and W oi are the weight matrices of each layer gating cell, b gi and b oi are the bias vectors of each layer gating cell, σ is the sigmoid activation function, and h i is the process variable of the ith hidden layer.
In a second aspect, the present invention provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the CO-MP-DSSM based lactobacillus fermentation process soft measurement method when executing the computer program.
In a third aspect, the present invention provides a computer readable storage medium having stored thereon a computer program which when executed by a processor realizes the steps of the CO-MP-DSSM based lactobacillus fermentation process soft measurement method.
The invention has the beneficial effects that:
According to the method, the soft measurement method of the lactobacillus fermentation process based on the CO-MP-DSSM is used, firstly, the multi-player collaborative learning is used for marking the label-free data with the pseudo labels, and the adverse effect on the soft measurement model caused by the difficulty in acquiring the labeled data is solved; then, the data are standardized, and the two-dimensional dynamic characteristics of the data are extracted through a two-dimensional sliding window; and finally, the strong nonlinearity and the two-dimensional dynamic characteristic of the data are further integrated through the deep learning model, so that the adverse effect of the strong nonlinearity and the two-dimensional dynamic characteristic of the batch process on the general prediction model is solved. The soft measurement method of the lactobacillus fermentation process based on the CO-MP-DSSM provided by the invention is improved aiming at the characteristics of the lactobacillus fermentation process, so that a better prediction effect is realized by the model, and a smaller prediction error is obtained by the predicted label.
Drawings
FIG. 1 is a schematic overall flow diagram of a method for soft measurement of a lactobacillus fermentation process based on CO-MP-DSSM;
FIG. 2 is a block diagram of a CO-MP-DSSM network model of the present invention;
FIG. 3 is a flow chart of a two-dimensional dynamic sliding window processing method of the data of the present invention;
FIG. 4 is a graph comparing the proportion of tagged data to total data for a lactic acid bacteria fermentation soft measurement using the method of the present invention with a GSTAE method.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The label in the invention represents the real quality variable at the moment in the fermentation process, and the pseudo label represents the staged prediction result of the model on the quality variable in the training process.
Example 1
As shown in fig. 1-3, the embodiment provides a lactobacillus fermentation process soft measurement method based on CO-MP-DSSM, comprising the following steps:
Step 1, collecting normal fermentation process data of lactobacillus, wherein the data are three-dimensional data X (I multiplied by J multiplied by K) and y (I multiplied by 1 multiplied by L), wherein the data comprise I batches, each batch has K sampling moments, each sampling moment has J auxiliary variables, and L is the collection times of thallus concentration information;
Step 2, performing multi-player collaborative learning training on the plurality of regressors by using the data in the step 1, and predicting a pseudo tag at a moment when the cell concentration is not measured to obtain y' (i×1×k);
step 3, preprocessing the data obtained in the step 2, performing Z-score standardization processing on the data according to batch dimension after the data are transposed, and obtaining standardized data after the data are transposed again;
Step 4, carrying out two-dimensional dynamic sliding window on the standardized data obtained in the step 3, wherein the size of the sliding window is p multiplied by q, and the original data is subjected to window sliding to obtain the data shape of p multiplied by q multiplied by (I-p+1) multiplied by (K-q+1) multiplied by J; in order to facilitate training of the data, the size and the number of the sliding windows are respectively compressed into one dimension to obtain data with the shape of (p multiplied by q) multiplied by [ (I-p+1) multiplied by (K-q+1) ]multipliedby J;
Step 5, establishing GSTAE a network model for the data obtained in the step 4 after the two-dimensional dynamic sliding window is completed; setting model parameters including the number of layers of the CO-MP-DSSM network, the number of neurons, the number of pre-training times, the number of fine tuning times, weight parameters, the number of multi-player collaborative learning regressors and the number of collaborative learning CO-tracking neighbors;
And 6, performing iterative training on the GSTAE network model in the step 5 until the set training times are reached.
Optionally, the step 2 specifically includes: performing multi-player collaborative learning training on a plurality of kNN (k nearest neighbor) -based regressors by using the data in the step 1, and copying labeled data into a plurality of parts at the beginning, wherein each part corresponds to one regressor as a training set thereof; then iterating, each regressor tries to predict label for each data in the label-free data set, and predicting k neighbors in the label data by using the retrained regressor; the prediction effect is improved, and the data with the largest improvement amplitude can be added into all the rest training sets; after training, the regressors predict and average labels at the moment of not measuring the concentration of the bacterial cells, predict the original unlabeled data and combine the labeled data to obtain y' (I multiplied by 1 multiplied by K).
Optionally, the Z-score normalization in step 3 uses the following formula:
wherein sigma is the standard deviation of the elements in the matrix, N is the number of elements in the matrix of each variable, For the average number of elements in the matrix for each variable, x i and x are the elements in the matrix that need to be normalized, and s is the result of normalization of the corresponding elements.
Optionally, the sliding window size of the two-dimensional dynamic sliding window in the step 4 is p×q, and the original data is subjected to window sliding to obtain a data shape of p×q× (I-p+1) × (K-q+1) ×j; the size and the number of the sliding windows are respectively compressed into one dimension to obtain (p multiplied by q) multiplied by [ (I-p+1) multiplied by (K-q+1) ]multipliedby J, so that the data is convenient for training; the two-dimensional dynamic sliding window slides along the time direction, and moves one frame along the batch direction after the sliding is completed, and the sliding is restarted.
Optionally, the GSTAE network model in the step 5 is a fully connected neural network, which includes a plurality of layers, the number of neurons in the first layer is equal to the number of variables, and then the number of neurons in each layer is gradually reduced; each layer and the previous layer are an input layer and a hidden layer of the self-encoder AE, and an attempt is made to reconstruct the input during training; in addition, each layer comprises a gating unit for controlling the influence of the layer result on the final output;
Optionally, the GSTAE network model includes a pre-training stage and a fine-tuning stage in a training stage;
in the pre-training phase, each layer of self-encoder of GSTAE network model tries to reconstruct the input and predict the quality change, the calculation method is as follows:
Wherein, And/>W 1 and W 2 are the weight matrices of the input layer to the hidden layer and the hidden layer to the output layer, respectively, and b 1 and b 2 are the bias vectors of the input layer to the hidden layer and the hidden layer to the output layer, respectively, f and/>, respectivelyThe activation functions are respectively from the input layer to the hidden layer and from the hidden layer to the output layer;
In the fine tuning stage, GSTAE the network model predicts the variable labels and tries to fit the true values, the loss is calculated as follows:
Wherein y i and The actual quality variable and the predicted quality variable, respectively.
Optionally, the calculation formula of the GSTAE network model prediction variable label is as follows:
Where n is the number of layers of the GSTAE network model, W gi and W oi are the weight matrices of each layer gating cell, b gi and b oi are the bias vectors of each layer gating cell, σ is the sigmoid activation function, and h i is the process variable of the ith hidden layer.
According to the lactobacillus fermentation process soft measurement method based on the CO-MP-DSSM, firstly, pseudo tags are marked on label-free data through multi-player collaborative learning, so that adverse effects on a soft measurement model caused by difficulty in acquiring the labeled data are solved; then, the data are standardized, and the two-dimensional dynamic characteristics of the data are extracted through a two-dimensional sliding window; and finally, the strong nonlinearity and the two-dimensional dynamic characteristic of the data are further integrated through the deep learning model, so that the adverse effect of the strong nonlinearity and the two-dimensional dynamic characteristic of the batch process on the general prediction model is solved. The soft measurement method of the lactobacillus fermentation process based on the CO-MP-DSSM provided by the invention is improved aiming at the characteristics of the lactobacillus fermentation process, so that a better prediction effect is realized by the model, and a smaller prediction error is obtained by the predicted label.
Example 2
This example provides a validation of the method provided in example 1 above on actual lactobacillus fermentation process data. The strain selected in the actual fermentation process is Lactobacillus plantarum, which is one of lactobacillus, and the strain number HuNHHMYL (Lactobacillus plantarum species of Lactobacillus of the family Lactobacillus) is derived from the biological technology center of food college of university in the south of the Yangtze river. The fermentation process of lactobacillus adopts MARS culture medium, and the fermentation equipment adopts bioreactor of Dibi company, and the model of the equipment is T & j-Atype. Each fermentation batch was inoculated at 4% and the fermentation period was 8 hours and fed 2 hours after the start of fermentation. The fermentation environment is described in the lactic acid bacteria fermentation literature (Zhang Anmin, wang, bao Huifang, et al. High-density fermentation of Lactobacillus plantarum Lp-2 [ J ]. J.China journal of bioengineering, 2009,29 (6): 68-73.). The fermentation temperature was set to 37 ℃, the pH value was set to 6.0, the sampling interval was 1min, and the mass variable sampling interval was 30min. 20 batches of data were co-fermented, 15 batches were used to train the model, and 5 batches of normal data were used to validate the model. The sampling variables were 8 in total, as shown in table 1 below:
TABLE 1 lactic acid bacteria fermentation process variables
As shown in fig. 1, the embodiment includes an offline modeling stage and an online quality prediction stage, and the implementation steps are as follows:
A. Offline modeling stage:
(1) Collecting normal fermentation process data, wherein the data are three-dimensional data X (20 multiplied by 7 multiplied by 480) and y (20 multiplied by 1 multiplied by 16), the total number of I batches is 20, the number of K sampling moments in each batch is 480, the number of J auxiliary variables in each sampling moment is 7, and the number of collection times L of the concentration information of the thalli is 16;15 batches were used to train the model, 5 batches of normal data were used to validate the model;
(2) Performing multi-player collaborative learning training using a kNN-based structure of a plurality of regressors, and predicting a pseudo tag at a time point when the cell concentration is not measured, thereby obtaining y' (15×1×480);
(3) Preprocessing the data obtained in the step (2), transposing the data, performing Z-score standardization processing according to batch dimension, and transposing again to obtain standardized data;
(4) And (3) carrying out two-dimensional dynamic sliding window on the standardized data in the step (3), wherein the sliding window is 3×3 in size, and the original data is subjected to window sliding to obtain the data with the shape of 3×3×13×478×7. In order to facilitate training of the data, the size and the number of the sliding windows are respectively compressed into one dimension to obtain data with the shape of 9 multiplied by 6214 multiplied by 7;
(5) Setting model parameters, wherein the number of layers of a CO-MP-DSSM network is 4, the number of neurons is 7, 6, 5 and 3, the pre-training frequency is 600, the fine tuning frequency is 300, the weight parameters are 0.5, the number of multi-player collaborative learning regressors is 3, and the number of collaborative learning CO-training neighbors is 10;
(6) Training the model to finish model iteration.
B. On-line quality prediction stage:
(7) 7 variables of lactic acid fermentation data in a certain batch and at a certain moment are collected, standardization is carried out according to the mean value and the variance in the step (3), and x after a sliding window is obtained according to the step (4).
(8) Inputting data into a deep learning model, predicting quality variables layer by layer through the model, and calculating through gating units of each layer of GSTAE network model to finally obtain the prediction of the model on the quality variables.
The invention comprises an offline modeling stage and an online quality prediction stage. In the off-line modeling stage, normal fermentation process data are collected, multi-player collaborative learning is carried out on the original data to predict pseudo tags, Z-score standardization is carried out on the data, and a 2D-GSTAE model is trained; the online quality prediction stage comprises the steps of carrying out standardized processing on online collected data, carrying out a two-dimensional sliding window on the data, and predicting a data label. The method solves the adverse effects on the soft measurement model caused by strong nonlinearity, two-dimensional dynamic characteristics and difficulty in acquiring the labeled data in the batch process, and achieves a good prediction effect.
Comparative example 1
GSTAE comparative experiments with the CO-MP-DSSM method
The method of example 2 was used for lactobacillus fermentation soft measurement, except that GSTAE was used for modeling using the CO-MP-DSSM provided by the present invention, and the quality variable prediction results are shown in table 2 and fig. 4. The ratios in the table refer to the ratio of total data to tagged data, rmse and R 2 are common indicators for measuring soft measurement model performance. rmse the calculation method is as follows:
where n represents the total number of tags, y i represents the i-th real tag, Representing the i-th predictive label. The calculation method of R 2 is as follows:
From the above formula, it can be seen that the lower rmse and higher R 2 represent better fitting of the model predicted data to the actual results. As can be seen from table 2 and fig. 4, the proposed method has lower rmse and higher R 2 compared to GSTAE method, and thus has better prediction effect.
Table 2: quality variable prediction results of lactic acid bacteria fermentation soft measurement by using the method and GSTAE method
The present invention also provides a computer device that may include a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, causes the processor to perform the steps of the CO-MP-DSSM based lactobacillus fermentation process soft-measurement method according to any of the embodiments described above.
The working process, working details and technical effects of the computer device provided in this embodiment can be referred to the above embodiments of the method for soft measurement of the lactobacillus fermentation process based on CO-MP-DSSM, and will not be described herein.
Furthermore, the invention also provides a computer readable storage medium, on which a computer program is stored, which when being executed by a processor, implements the steps of the CO-MP-DSSM based lactobacillus fermentation process soft measurement method according to any of the embodiments described above. Wherein the computer readable storage medium refers to a carrier for storing data, and may include, but is not limited to, a floppy disk, an optical disk, a hard disk, a flash memory, and/or a memory stick, etc., where the computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable devices.
The working process, working details and technical effects of the computer readable storage medium provided in this embodiment can be referred to the above embodiments of the method for soft measurement of the lactobacillus fermentation process based on CO-MP-DSSM, and will not be described herein.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (SYNCHLINK) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements and similar elements thereof may be made without departing from the spirit and principles of the present invention.
Claims (10)
1. A lactobacillus fermentation process soft measurement method based on CO-MP-DSSM includes the following steps:
Step 1, collecting normal fermentation process data of lactobacillus;
Step 2, performing multi-player collaborative learning training on the plurality of regressors by using the data in the step 1, and predicting a pseudo tag at a moment when the concentration of the bacteria is not measured;
step 3, preprocessing the data obtained in the step 2 to obtain standardized data;
step 4, carrying out two-dimensional dynamic sliding window on the standardized data obtained in the step 3;
step 5, establishing GSTAE a network model for the data obtained in the step 4 after the two-dimensional dynamic sliding window is completed;
And 6, performing iterative training on the GSTAE network model in the step 5 until the set training times are reached.
2. The soft measurement method of lactobacillus fermentation process based on CO-MP-DSSM according to claim 1, wherein the fermentation process data in the step 1 is three-dimensional data X (ijxk) and y (i×1 xl), wherein the total comprises I batches, each batch has K sampling moments, each sampling moment has J auxiliary variables, and L is the number of times of acquiring thallus concentration information.
3. The method for soft measurement of lactobacillus fermentation process based on CO-MP-DSSM according to claim 2, wherein the step 2 specifically comprises:
Performing multi-player collaborative learning training on a plurality of kNN regressors by using the data in the step 1, and copying labeled data into a plurality of parts at the beginning, wherein each part corresponds to one regressor as a training set; then iterating, each regressor predicts label for each data in the label-free data set, and predicting k neighbors in the label-free data by using the retrained regressor; the data with improved prediction effect and maximum improvement amplitude is added into all the rest training sets; after training, the regressors respectively predict labels at the moment of not measuring the concentration of the thalli and average the labels, predict original non-label data and combine the original non-label data with labeled data to obtain y' (I multiplied by 1 multiplied by K);
the label represents a real quality variable at the moment in the fermentation process, and the pseudo label represents a staged prediction result of the model on the quality variable in the training process.
4. A CO-MP-DSSM based lactobacillus fermentation process soft measurement method according to claim 3, characterized in that step 3 comprises:
preprocessing the data obtained in the step 2, performing Z-score standardization processing on the data according to batch dimension after the data is transposed, and obtaining standardized data after the data is transposed again;
z-score normalization uses the following formula:
wherein sigma is the standard deviation of the elements in the matrix, N is the number of elements in the matrix of each variable, For the average number of elements in the matrix for each variable, x i and x are the elements in the matrix that need to be normalized, and s is the result of normalization of the corresponding elements.
5. The soft measurement method of the lactobacillus fermentation process based on the CO-MP-DSSM as claimed in claim 4, wherein the sliding window size of the two-dimensional dynamic sliding window in the step 4 is p×q, and the raw data is subjected to window sliding to obtain the data shape of p×q× (I-p+1) × (K-q+1) ×J; the size and the number of the sliding windows are respectively compressed into one dimension, and the data is obtained in the shape of (p multiplied by q) multiplied by [ (I-p+1) multiplied by (K-q+1) ]multipliedby J, so that the data is convenient to train; the two-dimensional dynamic sliding window slides along the time direction, and moves one frame along the batch direction after the sliding is completed, and the sliding is restarted.
6. The soft measurement method of lactobacillus fermentation process based on CO-MP-DSSM according to claim 5, wherein the GSTAE network model in step 5 is a fully connected neural network, which comprises a plurality of layers, the number of neurons in the first layer is equal to the number of variables, and then the number of neurons in each layer is gradually reduced; each layer and the previous layer are an input layer and a hidden layer of the self-encoder, and reconstruct the input during training; each tier contains a gating unit for controlling the effect of the tier results on the final output.
7. The method for soft measurement of a CO-MP-DSSM based lactic acid bacteria fermentation process according to claim 6, wherein the GSTAE network model includes a pre-training phase and a fine-tuning phase in a training phase;
in the pre-training stage, each layer of self-encoder of GSTAE network model reconstructs the input and predicts the quality change, and the calculation method is as follows:
Wherein, And/>W 1 and W 2 are the weight matrices of the input layer to the hidden layer and the hidden layer to the output layer, respectively, and b 1 and b 2 are the bias vectors of the input layer to the hidden layer and the hidden layer to the output layer, respectively, f and/>, respectivelyThe activation functions are respectively from the input layer to the hidden layer and from the hidden layer to the output layer;
in the fine tuning stage, GSTAE network models predict variable labels and fit true values, and the loss is calculated as follows:
Wherein y i and The actual quality variable and the predicted quality variable, respectively.
8. The soft measurement method of lactobacillus fermentation process based on CO-MP-DSSM according to claim 7, wherein the calculation formula of the GSTAE network model predictive variable label is as follows:
Where n is the number of layers of the GSTAE network model, W gi and W oi are the weight matrices of each layer gating cell, b gi and b oi are the bias vectors of each layer gating cell, σ is the sigmoid activation function, and h i is the process variable of the ith hidden layer.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that,
The processor, when executing the computer program, implements the steps of the CO-MP-DSSM based lactobacillus fermentation process soft measurement method according to any of the claims 1-8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that,
The computer program, when executed by a processor, implements the steps of the CO-MP-DSSM based lactobacillus fermentation process soft measurement method according to any of the claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410164563.8A CN118036658A (en) | 2024-02-05 | 2024-02-05 | Soft measurement method for lactobacillus fermentation process based on CO-MP-DSSM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410164563.8A CN118036658A (en) | 2024-02-05 | 2024-02-05 | Soft measurement method for lactobacillus fermentation process based on CO-MP-DSSM |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118036658A true CN118036658A (en) | 2024-05-14 |
Family
ID=90983596
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410164563.8A Pending CN118036658A (en) | 2024-02-05 | 2024-02-05 | Soft measurement method for lactobacillus fermentation process based on CO-MP-DSSM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118036658A (en) |
-
2024
- 2024-02-05 CN CN202410164563.8A patent/CN118036658A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109060001B (en) | Multi-working-condition process soft measurement modeling method based on feature transfer learning | |
| CN110378435B (en) | Apple leaf disease identification method based on convolutional neural network | |
| CN113449919B (en) | Power consumption prediction method and system based on feature and trend perception | |
| CN116975645A (en) | Industrial process soft measurement modeling method based on VAE-MRCNN | |
| CN114403486B (en) | Intelligent control method of airflow type cut-tobacco drier based on local peak value coding circulation network | |
| CN117139380A (en) | Camber control method based on self-learning of regulation experience | |
| CN114169459A (en) | Robust soft measurement method based on semi-supervised Bayesian regularization hybrid Student's t model | |
| CN110738363A (en) | photovoltaic power generation power prediction model and construction method and application thereof | |
| CN114676622A (en) | Short-term photovoltaic power prediction method based on self-encoder deep learning model | |
| CN116483036B (en) | Transformer-based self-encoder soft measurement modeling method | |
| CN106599391B (en) | Association vector machine soft measurement modeling method based on dynamic weighting of triangle angle values | |
| CN113469013A (en) | Motor fault prediction method and system based on transfer learning and time sequence | |
| CN115640743B (en) | Power distribution load-feeder intelligent matching method based on Bayesian optimization parameter neural network | |
| CN118036658A (en) | Soft measurement method for lactobacillus fermentation process based on CO-MP-DSSM | |
| CN117371573A (en) | Time sequence prediction method, device and medium based on TrAdaBoost-LSTM | |
| CN114707424B (en) | Chemical process soft measurement method based on quality-related slow characteristic analysis algorithm | |
| CN116703644A (en) | Attention-RNN-based short-term power load prediction method | |
| CN116484275A (en) | Aluminum electrolysis anode effect monitoring method based on self-adaptive space-time diagram convolutional neural network | |
| CN115759343A (en) | E-LSTM-based user electric quantity prediction method and device | |
| CN112529268B (en) | Medium-short term load prediction method and device based on manifold learning | |
| CN110826184B (en) | Variational Bayesian identification method for NARX model structure and parameters under time-varying lag | |
| CN116070818A (en) | Penicillin production abnormality detection method based on sliding window time sequence discriminant feature analysis | |
| CN113835964A (en) | Cloud data center server energy consumption prediction method based on small sample learning | |
| CN117350328B (en) | LSTM-CVAE-based lactobacillus fermentation process fault detection and diagnosis method | |
| Nurmalasari et al. | Classification for Papaya Fruit Maturity Level with 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 |