A Deep Learning-Based Online Soft Measurement Method and Device for Penicillin Fermentation Process

Abstract

This invention provides an online soft measurement method and apparatus for penicillin fermentation process based on deep learning, relating to the technical field of soft measurement modeling and application in industrial fermentation production processes. The method includes: establishing a penicillin fermentation process dataset by changing the initial values ​​of control parameters and state variables during the penicillin fermentation process; standardizing the penicillin fermentation process dataset, performing time window slicing, and proportionally dividing the training and testing sets required for the soft measurement deep learning network; and using the trained soft measurement deep learning network to predict the cell concentration value corresponding to each time window, thereby alleviating the technical problem that existing technologies cannot be widely adapted to soft measurement applications in biological fermentation.

Description

Technical Field

[0001] This invention relates to the technical field of soft measurement modeling and application in industrial fermentation production processes, and in particular to an online soft measurement method and device for penicillin fermentation processes based on deep learning. Background Technology

[0002] During fermentation, some key variables that serve as fundamental indicators of process behavior are often difficult to measure online. These variables are typically measured in the laboratory using offline sample analyzers. Such analytical procedures are costly and often require expensive and frequent maintenance. Furthermore, due to the significant delays in laboratory testing and analysis, the measured values ​​of these variables cannot serve as feedback signals for the control system. To address this issue, soft sensors are widely used for reliable online estimation of these important process variables. Soft sensors predict key variables by correlating them with readily available auxiliary variables.

[0003] In the field of bio-fermentation, such as penicillin fermentation production, some relatively mature soft measurement methods have been developed. However, due to the highly variable and complex biochemical reactions involved in the bio-reaction process, which exhibit nonlinearity and strong coupling characteristics, and the exponential growth of fermentation process data due to the development of high-throughput technology, the above methods are currently unable to meet the current situation of soft measurement in bio-fermentation. Summary of the Invention

[0004] The purpose of this invention is to provide an online soft measurement method and device for penicillin fermentation process based on deep learning, so as to alleviate the technical problem that the existing technology cannot be widely adapted to soft measurement applications in biological fermentation.

[0005] In a first aspect, embodiments of the present invention provide an online soft measurement method for penicillin fermentation processes based on deep learning, the method comprising:

[0006] A dataset of the penicillin fermentation process was established by changing the initial values ​​of the control parameters and state variables in the penicillin fermentation process.

[0007] The penicillin fermentation process dataset was standardized, time window slicing was performed, and the training and test sets required for the soft measurement deep learning network were divided proportionally.

[0008] The trained soft-measure deep learning network is used to predict the bacterial concentration value for each time window.

[0009] In conjunction with the first aspect, embodiments of the present invention provide a first possible implementation of the first aspect, wherein the method further includes:

[0010] The predicted bacterial concentration values ​​for each time window are evaluated based on the mean square error, mean absolute error, and mean absolute percentage error.

[0011] In conjunction with the first aspect, this embodiment of the invention provides a second possible implementation of the first aspect, wherein the steps of standardizing the penicillin fermentation process dataset, performing time window slicing, and proportionally dividing the training and test sets required for the soft-sensor deep learning network include:

[0012] The penicillin fermentation process dataset is divided into auxiliary variable data used as input to the soft measurement deep learning network and corresponding cell concentration data, and then 0-1 normalization is performed on each.

[0013] The auxiliary variable data is divided into time windows, and each time window corresponds to the output of a key variable data. At the same time, the auxiliary variable data in the initial stage is padded with 0.

[0014] The processed auxiliary variable data are divided into training and test sets according to a preset ratio.

[0015] In conjunction with the first aspect, embodiments of the present invention provide a third possible implementation of the first aspect, wherein, before the step of predicting the bacterial cell concentration value corresponding to each time window using a trained soft-sensor deep learning network, the method further includes:

[0016] Based on the training set and the test set, a trained soft measurement deep learning network is determined, wherein the trained soft measurement deep learning network is used to output the bacterial concentration corresponding to the time window.

[0017] In conjunction with the first aspect, embodiments of the present invention provide a fourth possible implementation of the first aspect, wherein the step of determining the trained soft measurement deep learning network based on the training set and the test set includes:

[0018] Step a: First, the prepared auxiliary variable data is input into a one-dimensional convolutional neural network to extract high-dimensional spatial features of the fermentation process, expanding the input data dimension from 16 dimensions to 128 dimensions. This network contains three one-dimensional convolutional layers, each followed by a max-pooling layer. The calculation formula for the one-dimensional convolutional layer is as follows:

[0019]

[0020] Among them, c ij W is the i-th feature value obtained after performing a convolution operation between the j-th convolution kernel and the corresponding input data. j X represents the weights of the convolution kernel. i:i+h-1 For the input data from the i-th to the i+h-1-th rows that are convolved with the convolution kernel, bj Here, f(·) represents the bias of the convolution kernel, and f(·) is the ReLU activation function.

[0021] The output of the convolutional neural network is:

[0022] CNN 128×50 =conv(X) 16×50 )

[0023] Among them, X 16×50 The input consists of auxiliary variable data within a time window, and conv(·) represents the entire convolutional neural network (CNN). 128×50 This is the output of the entire convolutional neural network;

[0024] Step b: Input the high-dimensional data processed by the convolutional neural network into the long short-term memory recurrent neural network to mine the hidden long-term dependency characteristics and output the hidden layer states of the last LSTM network. The long short-term memory recurrent neural network contains two LSTM layers, each with a time step of 50 and 50 hidden layer states. The final output of the long short-term memory recurrent neural network is 50 hidden layer states that contain the spatiotemporal characteristics of the fermentation process.

[0025] The LSTM's memory cells and hidden state updates are as follows:

[0026] C (t) =f (t) ·C (t-1) +i (t) ·C' (t)

[0027] h (t) =O (t) ·tanh(C (t) )

[0028] Where t represents the time step, C (t) Let C' be the hidden state and f be the intermediate state used for computation within the basic unit. (t) For the Gate of Oblivion, i (t) For the input gate, h (t) For the state of the storage cell, O (t) For output gate, tanh(·) is the hyperbolic tangent activation function;

[0029] The LSTM network output is:

[0030] LSTM 50 =lstm(CNN) 128×50 )

[0031] Among them, CNN 128×50The input to the convolutional neural network is passed to the LSTM network, where lstm(·) is the entire long short-term memory recurrent neural network. 50 This is the output of the entire Long Short-Term Memory recurrent neural network;

[0032] Step c involves using the 50 hidden state outputs of the Long Short-Term Memory Recurrent Neural Network (LSTM) as input to the fully connected neural network. After calculation, the network outputs the value of the key variable to be predicted. The fully connected neural network comprises one input layer, two hidden layers, and one output layer. The hidden state calculation of the fully connected neural network is as follows:

[0033] O = f(W·X + b)

[0034] Where O is the output, W is the bias matrix, X is the input matrix, b is the bias matrix, and f(·) is the sigmoid activation function;

[0035] The output of the fully connected neural network is:

[0036] Cell concentration = fcn(LSTM) 50 )

[0037] LSTM 50 fcn(·) is the input passed from the LSTM network to the fully connected neural network, and fcn(·) is the entire fully connected neural network.

[0038] Secondly, embodiments of the present invention also provide an online soft measurement device for penicillin fermentation based on deep learning, the device comprising:

[0039] A module was established to create a dataset of the penicillin fermentation process by changing the initial values ​​of the control parameters and state variables in the penicillin fermentation process.

[0040] The training module performs standardization operations on the penicillin fermentation process dataset, completes time window slicing operations, and proportionally divides the training set and test set required for the soft measurement deep learning network.

[0041] The prediction module uses a trained soft-sensor deep learning network to predict the bacterial concentration value for each time window.

[0042] Thirdly, an embodiment provides an electronic device including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the steps of the method described in any of the foregoing embodiments.

[0043] Fourthly, an embodiment provides a machine-readable storage medium storing machine-executable instructions, which, when invoked and executed by a processor, cause the processor to perform the steps of the method described in any of the foregoing embodiments.

[0044] This invention provides a deep learning-based online soft measurement method and device for penicillin fermentation. The method designs a model architecture for online soft measurement of penicillin fermentation, inputting auxiliary variable data in the form of time windows. A convolutional neural network is used to extract high-dimensional salient features of the fermentation process data within the time window, and a long short-term memory recurrent neural network is used to mine the hidden long-term dependencies of the data. By effectively combining these two approaches, the spatiotemporal features of complex biochemical reaction processes are effectively extracted, achieving high-precision prediction of cell concentration during penicillin fermentation. This also provides an effective modeling approach for soft measurement of other complex reaction processes.

[0045] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0046] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0047] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0048] Figure 1 A flowchart of an online soft measurement method for penicillin fermentation based on deep learning is provided for an embodiment of the present invention.

[0049] Figure 2 A flowchart of another online soft measurement method for penicillin fermentation based on deep learning provided in an embodiment of the present invention;

[0050] Figure 3 This refers to the deep learning network architecture involved in the embodiments of the present invention;

[0051] Figure 4 This is a schematic diagram illustrating the effect of predicting cell concentration during penicillin fermentation based on embodiments of the present invention;

[0052] Figure 5 -a is a schematic diagram comparing the prediction performance of the embodiments of the present invention with that of a single neural network;

[0053] Figure 5 -b is a schematic diagram comparing the prediction error of the embodiment of the present invention with that of a single neural network;

[0054] Figure 6 A schematic diagram of the functional modules of an online soft measurement device for penicillin fermentation process based on deep learning, provided in an embodiment of the present invention;

[0055] Figure 7 A schematic diagram of the hardware architecture of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Currently, in the field of bio-fermentation, such as penicillin fermentation production, some relatively mature soft measurement methods include partial least squares, support vector machines, and backpropagation neural networks. However, due to the highly variable and complex biochemical reactions involved in the bio-reaction process, which exhibit nonlinearity and strong coupling characteristics, and the exponential growth of fermentation process data due to the development of high-throughput technology, the above methods are currently unable to adapt to the current situation of soft measurement in bio-fermentation.

[0058] Based on this, the present invention provides an online soft measurement method and device for penicillin fermentation process based on deep learning, which uses a deep learning network to extract the spatiotemporal characteristics contained in the penicillin fermentation process and thereby predicts the cell concentration value in the penicillin fermentation process.

[0059] To facilitate understanding of this embodiment, a detailed description of an online soft measurement method for penicillin fermentation based on deep learning, as disclosed in this embodiment of the invention, will be provided first. This method can be applied to intelligent control devices such as host computers, servers, and controllers. The relevant data used is generated using Pensim 2.0 software, which is a modular simulation software with non-structural dynamics as its core calculation method. It can customize the penicillin fermentation process according to the user's needs.

[0060] Figure 1 This is a flowchart of an online soft measurement method for penicillin fermentation based on deep learning, provided as an embodiment of the present invention.

[0061] Reference Figure 1 The method may include the following steps:

[0062] Step S102: By changing the initial values ​​of the control parameters and state variables in the penicillin fermentation process, a penicillin fermentation process dataset is established.

[0063] By changing the initial values ​​of state variables such as substrate concentration, dissolved oxygen concentration, cell concentration, culture medium volume, carbon dioxide concentration, pH value, and fermentation temperature, as well as the control parameters for temperature and pH control, a large number of correlation experiments were designed to establish a penicillin fermentation process dataset. This dataset includes various reaction modes of the penicillin fermentation process under normal operating conditions. The dataset includes the key mass variable to be predicted: cell concentration, as well as auxiliary variables related to cell concentration such as aeration rate, stirring rate, substrate acceleration, substrate loading temperature, culture medium volume, carbon dioxide concentration, pH value, reactor temperature, heat of reaction, acid flow acceleration rate, alkali flow acceleration rate, cold water flow rate, and hot water flow rate.

[0064] Step S104: Standardize the penicillin fermentation process dataset, complete the time window slicing operation, and divide the training set and test set required for the soft measurement deep learning network according to the proportion.

[0065] For example, the penicillin fermentation process dataset can be divided into auxiliary variable data (used as input to the soft sensor deep learning network) and corresponding cell concentration data, and both can be standardized by 0-1 to reduce prediction errors caused by inconsistent data scales and accelerate the neural network training process. Then, the auxiliary variable data can be divided into time windows of 1 hour each, containing 50 data sets. Each time window corresponds to the output of a key variable, and the initial auxiliary variable data is padded with zeros. Finally, the processed auxiliary variable data can be divided into training and testing sets according to a preset ratio. Specifically, 80% of the total data is used for training and 20% for testing, for the training and testing sets required for soft sensor modeling.

[0066] In practical applications, before step S106, the method further includes: determining the trained soft measurement deep learning network based on the training set and the test set, wherein the trained soft measurement deep learning network is used to output the bacterial concentration corresponding to the time window.

[0067] Specifically, a soft measurement deep learning network was constructed, comprising a convolutional neural network (CNN), a long short-term memory recurrent neural network (LSTM), and a fully connected neural network (FCN), with the three networks connected in a cascaded manner. The input to the entire deep learning network is a time window, and the output is the bacterial concentration corresponding to that time window. The constructed deep learning soft measurement model was then trained using the prepared training set data. The training algorithm used was the Adam algorithm, with a learning rate set to 10^(-5) and a training batch size of 1000.

[0068] Based on the foregoing embodiments, the construction and training steps of this soft measurement deep learning network include:

[0069] Step a: First, the prepared auxiliary variable data is input into a one-dimensional convolutional neural network to extract high-dimensional spatial features of the fermentation process, expanding the input data dimension from 16 dimensions to 128 dimensions. This network contains three one-dimensional convolutional layers, each followed by a max-pooling layer. The calculation formula for the one-dimensional convolutional layer is as follows:

[0070]

[0071] Among them, c ij W is the i-th feature value obtained after performing a convolution operation between the j-th convolution kernel and the corresponding input data. j X represents the weights of the convolution kernel. i:i+h-1 For the input data from the i-th to the i+h-1-th rows that are convolved with the convolution kernel, b j Here, f(·) represents the bias of the convolution kernel, and f(·) is the ReLU activation function.

[0072] The output of the convolutional neural network is:

[0073] CNN 128×50 =conv(X) 16×50 )

[0074] Among them, X 16×50 The input consists of auxiliary variable data within a time window, and conv(·) represents the entire convolutional neural network (CNN). 128×50 This is the output of the entire convolutional neural network;

[0075] Step b: Input the high-dimensional data processed by the convolutional neural network into the long short-term memory recurrent neural network to mine the hidden long-term dependency characteristics and output the hidden layer state of the last LSTM network. The long short-term memory recurrent neural network contains two LSTM layers, each with a time step of 50 and 50 hidden layer states. The final output of the long short-term memory recurrent neural network is 50 hidden layer states that contain the spatiotemporal characteristics of the fermentation process.

[0076] The LSTM's memory cells and hidden state updates are as follows:

[0077] C (t) =f (t) ·C (t-1) +i (t) ·C' (t)

[0078] h (t) =O (t) ·tanh(C (t) )

[0079] Where t represents the time step, C (t) Let C' be the hidden state and f be the intermediate state used for computation within the basic unit. (t) For the Gate of Oblivion, i (t) For the input gate, h (t) For the state of the storage cell, O (t) For output gate, tanh(·) is the hyperbolic tangent activation function;

[0080] The LSTM network output is:

[0081] LSTM 50 =lstm(CNN) 128×50 )

[0082] Among them, CNN 128×50 The input to the convolutional neural network is passed to the LSTM network, where lstm(·) is the entire long short-term memory recurrent neural network. 50 This is the output of the entire Long Short-Term Memory recurrent neural network;

[0083] Step c involves using the 50 hidden state outputs of the Long Short-Term Memory (LSTM) recurrent neural network as input to the fully connected neural network. After calculation, the network outputs the value of the key variable to be predicted. The fully connected neural network consists of one input layer, two hidden layers, and one output layer. The hidden state calculation of the fully connected neural network is as follows:

[0084] O = f(W·X + b)

[0085] Where O is the output, W is the bias matrix, X is the input matrix, b is the bias matrix, and f(·) is the sigmoid activation function;

[0086] The output of the fully connected neural network is:

[0087] Cell concentration = fcn(LSTM) 50 )

[0088] LSTM 50 fcn(·) is the input passed from the LSTM network to the fully connected neural network, and fcn(·) is the entire fully connected neural network.

[0089] Step S106: Use the trained soft measurement deep learning network to predict the bacterial concentration value corresponding to each time window.

[0090] The deep learning network in this application can accurately represent highly variable functions and can handle large amounts of concurrent data. At the same time, the use of convolutional neural networks and long short-term memory recurrent neural networks can extract the spatiotemporal characteristics contained in the fermentation process data and use them to predict key variables in the fermentation process.

[0091] In some embodiments, after the prediction of cell concentration during penicillin fermentation is completed, the predictive effectiveness of the soft sensor model for cell concentration during penicillin fermentation is evaluated. The aforementioned method further includes:

[0092] Step 1.1) Evaluate the bacterial concentration value corresponding to each predicted time window based on the mean square error, mean absolute error, and mean absolute percentage error.

[0093] The evaluation metrics used in this paper are mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), as defined in the formula.

[0094]

[0095]

[0096]

[0097] Where n is the number of samples, y' od y represents the predicted bacterial cell concentration. od This represents the actual bacterial concentration.

[0098] This invention, compared to traditional methods such as partial least squares and support vector machines, introduces deep learning as the fundamental method for soft measurement modeling. The ability of deep learning networks to accurately represent height-varying functions makes them more adaptable to biological processes involving height variations and complex biochemical reactions. By organically combining fully connected neural networks, convolutional neural networks, and long short-term memory recurrent neural networks, this invention can better uncover the spatiotemporal characteristics inherent in fermentation process data compared to a single neural network, thereby improving prediction accuracy.

[0099] As an optional embodiment, this embodiment of the invention selects variables such as aeration rate, stirring rate, bottom flow acceleration, bottom flow temperature, culture medium volume, carbon dioxide concentration, pH value, reaction tank temperature, heat of reaction, acid flow acceleration rate, alkali flow acceleration rate, cold water flow rate, and hot water flow rate as auxiliary variables, with cell concentration being the key variable to be predicted. Figure 2As shown, this embodiment of the invention also provides an online soft measurement method for the penicillin fermentation process based on deep learning:

[0100] Step 1: Collect variable data required for modeling, including cell concentration, aeration rate, stirring rate, bottom stream acceleration, bottom stream temperature, and culture medium volume. The penicillin fermentation process lasted 400 hours, with a data sampling interval of 0.02 hours. Each batch of data contained 20,000 sets of state data. The entire dataset contains 100 penicillin fermentation processes.

[0101] Step 2: Standardize the penicillin fermentation process data by performing time window segmentation. The window size is 1 hour, containing 50 data sets. Each time window corresponds to one key variable. 80% of the total data is used for training, and 20% is used for testing.

[0102] Step 3, construct a soft measurement deep learning network, such as Figure 3 As shown, the deep learning network comprises a convolutional neural network, a long short-term memory recurrent neural network, and a fully connected neural network, all connected in a cascaded manner. The input to the entire deep learning network is a time window, and the output is the bacterial cell concentration corresponding to that time window.

[0103] Step 4: Train the built deep learning soft measurement model using the prepared training set data. The training algorithm used is the Adam algorithm, the learning rate is set to 10^(-5), and the training batch size is 1000.

[0104] Step 5: Use the constructed deep learning soft measurement model to predict the bacterial concentration value corresponding to each time window;

[0105] Step 6: After the prediction of cell concentration during penicillin fermentation is completed, the predictive effectiveness of the soft sensor model for cell concentration during penicillin fermentation is evaluated. The evaluation metrics used in this paper are mean squared error (MSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The results are shown in Table 1.

[0106] The results of the method of the present invention for predicting bacterial cell concentration are as follows: Figure 4 As shown, the comparison with convolutional neural networks, long short-term memory recurrent neural networks, etc. is as follows: Figure 5 As shown in Table 1, it can be seen that the deep learning-based soft measurement modeling method proposed in this invention has a better accuracy improvement compared with single neural network modeling and prediction.

[0107] Table 1

[0108]

[0109] In summary, the online soft measurement method for penicillin fermentation based on deep learning proposed in this invention can complete soft measurement modeling of key variables in the penicillin fermentation process and improve prediction accuracy.

[0110] like Figure 6 As shown, this embodiment of the invention provides an online soft measurement device for penicillin fermentation based on deep learning, the device comprising:

[0111] A module was established to create a dataset of the penicillin fermentation process by changing the initial values ​​of the control parameters and state variables in the penicillin fermentation process.

[0112] The training module performs standardization operations on the penicillin fermentation process dataset, completes time window slicing operations, and proportionally divides the training set and test set required for the soft measurement deep learning network.

[0113] The prediction module uses a trained soft-sensor deep learning network to predict the bacterial concentration value for each time window.

[0114] The present invention provides an embodiment for implementing an electronic device. In this embodiment, the electronic device may be, but is not limited to, a personal computer (PC), a laptop computer, a monitoring device, a server, or other computer device with analysis and processing capabilities.

[0115] As an exemplary embodiment, see [reference]. Figure 7 The electronic device 110 includes a communication interface 111, a processor 112, a memory 113, and a bus 114. The processor 112, the communication interface 111, and the memory 113 are connected via the bus 114. The memory 113 is used to store a computer program that supports the processor 112 in executing the above-described method. The processor 112 is configured to execute the program stored in the memory 113.

[0116] The machine-readable storage medium mentioned in this article can be any electronic, magnetic, optical, or other physical storage device that can contain or store information such as executable instructions, data, etc. For example, machine-readable storage media can be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, storage drives (such as hard disk drives), any type of storage disk (such as optical discs, DVDs, etc.), or similar storage media, or combinations thereof.

[0117] Non-volatile media can be non-volatile memory, flash memory, storage drives (such as hard disk drives), any type of storage disk (such as optical discs, DVDs, etc.), or similar non-volatile storage media, or combinations thereof.

[0118] It is understood that the specific operation methods of each functional module in this embodiment can be referred to the detailed description of the corresponding steps in the above method embodiment, and will not be repeated here.

[0119] The computer-readable storage medium provided in the embodiments of the present invention stores a computer program. When the computer program code is executed, it can implement the method described in any of the above embodiments. For specific implementation, please refer to the method embodiments, which will not be repeated here.

[0120] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0121] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0122] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0123] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the scope of the technology disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention.

Claims

1. A deep learning-based online soft measurement method for penicillin fermentation process, characterized in that, The method includes: A dataset of the penicillin fermentation process was established by changing the initial values ​​of the control parameters and state variables in the penicillin fermentation process. The penicillin fermentation process dataset was standardized, time window slicing was performed, and the training and test sets required for the soft measurement deep learning network were divided proportionally. Using a trained soft-sensor deep learning network, the bacterial concentration value corresponding to each time window is predicted. Before the step of predicting the bacterial concentration value corresponding to each time window using the trained soft-sensor deep learning network, the following steps are also included: First, the prepared auxiliary variable data is input into a one-dimensional convolutional neural network to extract high-dimensional spatial features of the fermentation process, expanding the dimension of the input data from 16 to 128. The network contains three one-dimensional convolutional layers, each followed by a max pooling layer. The output of the convolutional neural network is: ； in, For the auxiliary variable data within a time window, For the entire convolutional neural network, This is the output of the entire convolutional neural network; The high-dimensional data processed by the convolutional neural network is input into the long short-term memory recurrent neural network to mine the hidden long-term dependency characteristics and output the hidden layer state of the last LSTM network. The long short-term memory recurrent neural network contains two LSTM layers, each with a time step of 50 and 50 hidden layer states. The final output of the long short-term memory recurrent neural network is 50 hidden layer states that contain the spatiotemporal characteristics of the fermentation process. The LSTM's memory cells and hidden state updates are as follows: ； ； in, Represents a time step. In hidden state, These are intermediate states used for calculation within the basic unit. For the Gate of Oblivion For input gate, The state of the storage unit. For output gate, It is the hyperbolic tangent activation function; The LSTM network output is: ； in, This is the input passed from the convolutional neural network to the LSTM network. For the entire long short-term memory recurrent neural network, This is the output of the entire Long Short-Term Memory recurrent neural network; The 50 hidden layer states output by the Long Short-Term Memory Recurrent Neural Network are used as input to the fully connected neural network. After calculation, the key variable values ​​to be predicted are output. The fully connected neural network consists of one input layer, two hidden layers, and one output layer. The output of the fully connected neural network is: ； in This is the input passed from the LSTM network to the fully connected neural network. This refers to the entire fully connected neural network.

2. The method according to claim 1, characterized in that, The method further includes: The predicted bacterial concentration values ​​for each time window are evaluated based on the mean square error, mean absolute error, and mean absolute percentage error.

3. The method according to claim 1, characterized in that, The steps of standardizing the penicillin fermentation process dataset, performing time window slicing, and proportionally dividing the training and test sets required for the soft-sensor deep learning network include: The penicillin fermentation process dataset was divided into auxiliary variable data used as input to the soft measurement deep learning network and corresponding cell concentration data, and then 0-1 normalized respectively. The auxiliary variable data is divided into time windows, and each time window corresponds to the output of a key variable data. At the same time, the auxiliary variable data in the initial stage is padded with 0. The processed auxiliary variable data are divided into training and test sets according to a preset ratio.

4. A deep learning-based online soft measurement device for penicillin fermentation process, characterized in that, The device includes: A module was established to create a dataset of the penicillin fermentation process by changing the initial values ​​of the control parameters and state variables in the penicillin fermentation process. The training module performs standardization operations on the penicillin fermentation process dataset, completes time window slicing operations, and proportionally divides the training set and test set required for the soft measurement deep learning network. The prediction module uses a trained soft-sensor deep learning network to predict the bacterial concentration value corresponding to each time window; wherein, the soft-sensor deep learning network includes a convolutional neural network (CNN), a long short-term memory recurrent neural network (LSTM), and a fully connected neural network (FCN), which are constructed by cascading the three networks; Before the step of predicting the bacterial concentration value corresponding to each time window using the trained soft-sensor deep learning network, the training module is also used for: First, the prepared auxiliary variable data is input into a one-dimensional convolutional neural network to extract high-dimensional spatial features of the fermentation process, expanding the dimension of the input data from 16 to 128. The network contains three one-dimensional convolutional layers, each followed by a max pooling layer. The output of the convolutional neural network is: ； in, For the auxiliary variable data within a time window, For the entire convolutional neural network, This is the output of the entire convolutional neural network; The high-dimensional data processed by the convolutional neural network is input into the long short-term memory recurrent neural network to mine the hidden long-term dependency characteristics and output the hidden layer state of the last LSTM network. The long short-term memory recurrent neural network contains two LSTM layers, each with a time step of 50 and 50 hidden layer states. The final output of the long short-term memory recurrent neural network is 50 hidden layer states that contain the spatiotemporal characteristics of the fermentation process. The LSTM's memory cells and hidden state updates are as follows: ； ； in, Represents a time step. In hidden state, These are intermediate states used for calculation within the basic unit. For the Gate of Oblivion For input gate, The state of the storage unit. For output gate, It is the hyperbolic tangent activation function; The LSTM network output is: ； in, This is the input passed from the convolutional neural network to the LSTM network. For the entire long short-term memory recurrent neural network, This is the output of the entire Long Short-Term Memory recurrent neural network; The 50 hidden layer states output by the Long Short-Term Memory Recurrent Neural Network are used as input to the fully connected neural network. After calculation, the key variable values ​​to be predicted are output. The fully connected neural network consists of one input layer, two hidden layers, and one output layer. The output of the fully connected neural network is: ； in This is the input passed from the LSTM network to the fully connected neural network. This refers to the entire fully connected neural network.

5. An electronic device, characterized in that, It includes a memory, a processor, and a program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the method as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed, implements the method described in any one of claims 1-3.

Images