Penicillin fermentation process soft measurement method and device based on supervisory gating circulation unit deep network, processor and storage medium thereof

Abstract

The invention relates to a penicillin fermentation process soft measurement method based on a supervisory gating circulation unit deep network, and the method comprises the following steps: obtaining a penicillin fermentation process data set through a database, and dividing the data set into a training set and a test set required by a soft measurement deep learning network according to a proportion; constructing a penicillin fermentation process soft measurement model based on a supervisory gating circulation unit deep network, and performing model training on the integrated data by using the model; the trained model is used for predicting the outlet concentration of the product in the penicillin fermentation process, and a corresponding prediction result is given accordingly. By adopting the method disclosed by the invention, the learning of the hidden state and the cell state of the gating circulation unit is guided by additionally using the quality information as the input of the two gates and the intermediate unit, so that the soft measurement modeling of the penicillin fermentation process is realized, and the prediction precision of the product concentration in the penicillin fermentation process is improved; powerful guidance is provided for better generation of the fermentation production process.

Description

Technical Field

[0001] The present invention relates to the field of deep learning, and in particular to the field of soft measurement of fermentation processes, and specifically refers to a soft measurement method, device, processor and computer-readable storage medium of a penicillin fermentation process based on a supervised gated recurrent unit deep network. Background Art

[0002] The fermentation process is a process in which biological organisms participate and produce target products under appropriate fermentation conditions such as pH, temperature, and stirring. It is widely present in various industrial processes and has inherent strong nonlinearity and complexity of the mechanism process. There are great difficulties in its theoretical modeling and process control, especially the lack of suitable online sensors to analyze the quality of key products online. For example, the quality of key products in the penicillin fermentation process can only be analyzed occasionally in an offline manner, which greatly limits the control and optimization of the fermentation process.

[0003] As data-driven modeling methods have shown great advantages in industrial processes, in order to overcome the problem of difficulty in analyzing key product quality, more and more people are using data-driven modeling methods to estimate key product quality from other variables that can be measured online. Due to the strong nonlinearity between product quality and operating variables in the fermentation process, the data-driven modeling methods based on the fermentation process are mainly divided into statistical methods, shallow machine learning methods and deep learning methods. Among them, statistical methods mainly include principal component analysis (PCA) and partial least squares (PLS). Shallow machine learning methods refer to traditional machine learning models other than deep learning, including shallow artificial neural networks (ANN), support vector machines (SVM), and Gaussian process models (GPM). Deep learning mainly includes various deep neural networks with more than three layers, such as gated recurrent units (GRU). Deep neural networks have been proven to be able to approximate any nonlinear system and are widely used in process modeling.

[0004] The GRU network uses reset gates and update gates to process and model long sequences, overcoming the problem of gradient vanishing. However, the dynamics of quality variables are not learned for the hidden states and cell states in the basic GRU unit. In fact, in soft measurement applications, understanding the dynamic states related to quality is very important for quality prediction. Therefore, a supervised GRU (SGRU) network is proposed, which is more conducive to quality regression prediction. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a soft measurement method, device, processor and computer-readable storage medium for penicillin fermentation process based on a supervised gated recurrent unit deep network.

[0006] In order to achieve the above-mentioned object, the soft-sensing method, device, processor and computer-readable storage medium of penicillin fermentation process based on supervised gated recurrent unit deep network of the present invention are as follows:

[0007] The soft measurement method for penicillin fermentation process based on supervised gated recurrent unit deep network has the following main features:

[0008] (1) Obtaining a penicillin fermentation process data set from a database, and dividing the data set into a training set and a test set required by a soft sensor deep learning network in proportion;

[0009] (2) Construct a soft-sensing model of penicillin fermentation process based on a supervised gated recurrent unit deep network and use the model to train the integrated data;

[0010] (3) The trained model is used to predict the outlet concentration of penicillin fermentation products and the corresponding prediction results are given.

[0011] Preferably, the step (1) specifically comprises the following steps:

[0012] (1.1) Set up the simulation process in the PenSim simulation platform to obtain a specific number of variable output data of the simulation platform;

[0013] (1.2) Divide the output data into training set and test set according to a specific ratio;

[0014] (1.3) and standardize the corresponding data in the following way:

[0015]

[0016] Among them, X′ is the data after standardization, X is the original data without standardization, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0017] Preferably, the step (2) specifically comprises the following steps:

[0018] (2.1) The reset gate, update gate and quality variable vector of the intermediate unit of the GRU network are used as input variables of the supervised gated recurrent unit SGRU;

[0019] (2.2) Randomly give the SGRU model a parameter, the input and output sequence is {x(1) ,x (2) ,...,x (T)} and {y (1) ,y (2) ,...,y (T)}, get the first prediction value of the SGRU model;

[0020] (2.3) According to the difference between the predicted value and the output value, the parameters of the SGRU model are adjusted so that the difference between the predicted value and the true value is reduced;

[0021] (2.4) Repeat steps (2.1) to (2.3) until the difference between the predicted value and the true value of the SGRU model is within the allowable error.

[0022] Preferably, the GRU network is represented as follows:

[0023] r t =σ(x t W r +h t-1 U r +b r )

[0024] z t =σ(x t W z +h t-1 U z +b z )

[0025]

[0026] Among them, r t represents the result of resetting the gate, z t represents the result of updating the gate, x t is the input at the current moment, h t-1 is the hidden state at the previous moment, represents the candidate hidden state, h t represents the final hidden state, σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, ⊙ represents element-wise multiplication, W r and U r are the weight matrices from input and hidden state to reset gate, W z and U z are the weight matrices from input and hidden state to update gate, W h and U h are the weight matrices from the hidden state adjusted by the input and reset gates to the candidate hidden state, and b r 、b z 、bh 、b y Both represent trainable bias terms, U y is the weight matrix of the output layer, It is the final output of the supervised gated recurrent unit SGRU.

[0027] Preferably, the supervisory gated recurrent unit SGRU is represented in the following manner:

[0028] r t =σ(x t W r +h t-1 U r +y t V r +b r )

[0029] z t =σ(x t W z +h t-1 U z +y t V z +b z )

[0030]

[0031] Among them, y t is the output at the current moment, V r is the weight matrix output to the reset gate, V z is the weight matrix output to the update gate, V y is the weight matrix of the output layer.

[0032] Preferably, the step (2) further comprises testing the SGRU model using the test data in the following manner:

[0033] (a) Assume that the test time series input is For each x from the first to the last sample in the test sequence (T+j) , its output value is initialized to the output value of the previous sample, that is,

[0034] (b) Replace the initialized data sample into the trained SGRU network to obtain x (T+j) The hidden state h(t) of (T+j) The final output is

[0035] (c) The root mean square error between the predicted and true values ​​of the test set is calculated, and the performance of the supervised gated recurrent unit SGRU, gated recurrent unit GRU, long short-term memory LSTM, and long short-term memory network SLSTM is evaluated.

[0036] Preferably, the step (3) is specifically as follows:

[0037] The test data is input into the trained SGRU model, and the mean square error RMSE is used as the evaluation indicator. The calculation formula is as follows:

[0038]

[0039] in, represents the real data, y i Represents the output of the SGRU model, and n represents the number of samples in the test set;

[0040] The smaller the evaluation indicator RMSE is, the closer the predicted value of the SGRU model is to the true value, that is, the better the prediction effect of the model is.

[0041] The device for soft measurement of penicillin fermentation process based on supervised gated recurrent unit deep network has the following main features: the device comprises:

[0042] a processor configured to execute computer executable instructions;

[0043] The memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the soft measurement method of penicillin fermentation process based on the supervised gated recurrent unit deep network are implemented.

[0044] The main feature of the processor for soft measurement of penicillin fermentation process based on supervised gated recurrent unit deep network is that the processor is configured to execute computer executable instructions to implement the steps of the above-mentioned soft measurement method for penicillin fermentation process based on supervised gated recurrent unit deep network.

[0045] The main feature of the computer-readable storage medium is that a computer program is stored thereon, and the computer program can be executed by a processor to implement the steps of the above-mentioned soft measurement method for penicillin fermentation process based on a supervised gated recurrent unit deep network.

[0046] The soft measurement method, device, processor and computer-readable storage medium of the penicillin fermentation process based on the supervised gated recurrent unit deep network of the present invention are adopted. The quality information is additionally used as the input of two gates and the intermediate unit to guide the learning of the hidden state and the cell state of the gated recurrent unit, thereby realizing soft measurement modeling of the penicillin fermentation process, providing a method for online real-time measurement of the concentration of the product of the penicillin fermentation process, improving the prediction accuracy of the concentration of the product of the penicillin fermentation process, and providing powerful guidance for the better generation of the fermentation production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a diagram of the supervisory gated recurrent unit SGRU layer structure of the present invention.

[0048] Figure 2 This is a model framework diagram of the soft measurement method for penicillin fermentation process based on the supervised gated recurrent unit deep network of the present invention.

[0049] Figure 3 A box plot of a specific embodiment of the soft sensing method for penicillin fermentation process based on a supervised gated recurrent unit deep network of the present invention. DETAILED DESCRIPTION

[0050] In order to more clearly describe the technical content of the present invention, further description is given below in conjunction with specific embodiments.

[0051] Before describing in detail embodiments according to the present invention, it should be noted that, hereinafter, the terms "comprises", "includes" or any other variations are intended to cover non-exclusive inclusion, whereby a process, method, article or apparatus comprising a series of elements includes not only these elements, but also other elements not explicitly listed or inherent to such process, method, article or apparatus.

[0052] See also Figure 1 and 2 As shown, the soft measurement method of penicillin fermentation process based on the supervised gated recurrent unit deep network, wherein the method comprises the following steps:

[0053] (1) Collecting online measurable variable values ​​in the penicillin fermentation process through a database as input of soft sensor modeling samples, and obtaining penicillin concentration values ​​obtained through offline measurement as output of soft sensor modeling samples; performing standardization operations on the penicillin fermentation process data set, and dividing the training set and test set required by the soft sensor deep learning network in proportion;

[0054] (2) A soft-sensing model for penicillin fermentation process based on a deep network of supervised gated recurrent units was constructed, in which the quality variable vector was additionally used as the input of two gates and the intermediate unit. The integrated data was then trained using the supervised gated recurrent unit (SGRU) model.

[0055] (3) The trained supervised gated recurrent unit (SGRU) model is used to predict the outlet concentration of penicillin fermentation products, and the prediction results are given.

[0056] As a preferred embodiment of the present invention, the step (1) specifically comprises the following steps:

[0057] (1.1) Set up the simulation process in the PenSim simulation platform to obtain the output data of 15 variables of the simulation platform;

[0058] (1.2) Divide the output data into training set and test set according to the ratio of 7:3;

[0059] (1.3) and standardize the corresponding data in the following way:

[0060]

[0061] Among them, X′ is the data after standardization, X is the original data without standardization, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0062] As a preferred embodiment of the present invention, the processing process of step (2) is:

[0063] (2.1) The supervised gated recurrent unit SGRU is based on the relevant algorithm to build a fermentation process outlet concentration prediction model; the supervised gated recurrent unit SGRU is mainly based on the quality variable vector being used as the input of the two gates and the intermediate unit of the GRU; the GRU network is one of the most classic variants of RNN, and there is a reset gate r t and update gate z t Two gating mechanisms. Candidate hidden state at the current moment Mainly includes the current input x t , r t Decide Contains x t The ratio is calculated based on the current input and the previous hidden state adjusted by the reset gate. The current hidden state h t Using z t At the same time, forgetting and selective memory are performed, and the hidden state at the previous moment and the candidate hidden state are weighted combined through the update gate to obtain the hidden state at the current moment. t and the previous hidden state h t-1, the main equation of the GRU basic unit is as follows:

[0064] r t =σ(x t W r +h t-1 U r +b r )

[0065] z t =σ(x t W z +h t-1 U z +b z )

[0066]

[0067] Among them, r t represents the result of resetting the gate, z t represents the result of updating the gate, x t is the input at the current moment, h t-1 is the hidden state at the previous moment, represents the candidate hidden state, h t represents the final hidden state, σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, ⊙ represents element-wise multiplication, W r and U r are the weight matrices from input and hidden state to reset gate, W z and U z are the weight matrices from input and hidden state to update gate, W h and U h are the weight matrices from the hidden state adjusted by the input and reset gates to the candidate hidden state, and b r 、b z 、b h 、b y Both represent trainable bias terms, U y is the weight matrix of the output layer, It is the final output of the supervised gated recurrent unit SGRU.

[0068] (2.2) The main disadvantage of GRU is that the hidden state and unit state in the GRU unit do not learn the dynamics of the quality variable. To overcome this problem, a notable feature of the supervised gated recurrent unit SGRU is that the quality variable vector is additionally used as the input of the two gates and the intermediate unit, and both the input variable and the quality variable are used to learn the shared dynamic hidden feature state. In this way, the hidden state contains the dynamic information related to the quality variable. The formula of the supervised gated recurrent unit SGRU is as follows:

[0069] r t =σ(x t W r +h t-1 U r +y t V r +b r )

[0070] z t =σ(x t W z +h t-1 U z +y t V z +b z )

[0071]

[0072] (2.3) Use the training data to train the supervised gated recurrent unit SGRU, select hyperparameters for the supervised gated recurrent unit SGRU, and obtain a trained supervised gated recurrent unit SGRU model.

[0073] As a preferred embodiment of the present invention, the step (3) is specifically as follows:

[0074] The test data is input into the trained SGRU model, and the mean square error RMSE is used as the evaluation indicator. The calculation formula is as follows:

[0075]

[0076] in, represents the real data, y i Represents the output of the SGRU model, and n represents the number of samples in the test set;

[0077] The smaller the evaluation indicator RMSE is, the closer the predicted value of the SGRU model is to the true value, that is, the better the prediction effect of the model is.

[0078] In practical application, the processing process of this technical solution is as follows:

[0079] (1) Obtaining the penicillin fermentation process data set

[0080] Step 1.1: Set up the Pensim simulation platform. The simulation data used in this study includes 15 input variables closely related to the penicillin fermentation process. To predict the quality variables, a data set of 120 batches was used, each batch consisting of 100 time points, totaling 12,000 samples.

[0081] Step 1.2: Divide the data into training set and test set according to the ratio of 7:3.

[0082] Step 1.3: Process each data according to the standardized formula.

[0083] (2) Perform supervised gated recurrent unit (SGRU) model training. The process is as follows:

[0084] Step 2.1: Randomly give the SGRU model a parameter w,u,v, and the input and output sequence is {x (1) ,x (2) ,...,x (T)} and {y (1) ,y (2) ,...,y (T)}, and obtain the first prediction value of the model.

[0085] Step 2.2: According to the difference between the predicted value and the output value, adjust the parameters of the model so that the difference between the predicted value and the true value is reduced.

[0086] Step 2.3: Repeat steps 3.1 to 3.2 until the difference between the model's predicted value and the true value is within the allowable error.

[0087] (3) Use the test data to test the model. The process is as follows:

[0088] Step 3.1: Assume that the test time series input is To estimate x using the SGRU network (T+j) The quality value of the network is unknown. Therefore, the initialization strategy is adopted to first give a rough output estimate. For each x from the first to the last sample in the test sequence (T+j) , its output value is initialized to the output value of the previous sample, that is,

[0089] Step 3.2: Replace the initialized data sample into the trained SGRU network to obtain x (T+j) The hidden state h(t) of (T+j) The final output is

[0090] Step 3.3: Calculate the RMSE between the predicted and true values ​​of the test set to evaluate the performance of the supervised gated recurrent units SGRU and GRU, LSTM, and SLSTM.

[0091] The evaluation results are shown in the table:

[0092] Table 1 RMSE comparison of SGRU, GRU, LSTM, and SLSTM

[0093]

[0094] From the above comparison results, it can be seen that the prediction effect of the present invention on the product concentration of penicillin fermentation process is better than that of the traditional prediction model. The model evaluation is as follows:

[0095] For performance comparison, GRU, LSTM, SLSTM, and SGRU models were developed to predict penicillin concentration. Figure 3 The prediction results of the four methods on the training and test data sets are given. According to the data distribution of the box plot, it can be considered that the SGRU model performs best in terms of median value and data range, while the LSTM model performs worst. As for the SLSTM model, its performance is between the two, but there may be some outliers. The basic units of GRU and LSTM are essentially unsupervised learning modules, in which the dynamic states are mainly learned from the input variables. Therefore, there may be information unrelated to quality in these dynamic states. The difference is that by introducing the quality variable into the basic SGRU unit, it can learn important dynamic information related to quality to a large extent for output prediction. From the perspective of prediction RMSE, SGRU can significantly improve the prediction accuracy with a smaller RMSE value on both the training and test data sets.

[0096] The device for soft measurement of penicillin fermentation process based on supervised gated recurrent unit deep network, wherein the device comprises:

[0097] a processor configured to execute computer executable instructions;

[0098] The memory stores one or more computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the soft measurement method of penicillin fermentation process based on the supervised gated recurrent unit deep network are implemented.

[0099] The processor for soft measurement of penicillin fermentation process based on supervised gated recurrent unit deep network, wherein the processor is configured to execute computer executable instructions to implement the steps of the above-mentioned soft measurement method for penicillin fermentation process based on supervised gated recurrent unit deep network.

[0100] The computer-readable storage medium stores a computer program thereon, and the computer program can be executed by a processor to implement the steps of the above-mentioned soft measurement method for penicillin fermentation process based on a supervised gated recurrent unit deep network.

[0101] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, segment or portion of code that includes one or more executable instructions for implementing the steps of a specific logical function or process, and the scope of the preferred embodiments of the present invention includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in the reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present invention belong.

[0102] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device.

[0103] A person of ordinary skill in the art may understand that all or part of the steps of the method for implementing the above-mentioned embodiment may be completed by instructing the relevant hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one of the steps of the method embodiment or a combination thereof.

[0104] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0105] In the description of this specification, the description with reference to the terms "an embodiment", "some embodiments", "example", "specific example", or "embodiment" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0106] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

[0107] The soft measurement method, device, processor and computer-readable storage medium of the penicillin fermentation process based on the supervised gated recurrent unit deep network of the present invention are adopted. The quality information is additionally used as the input of two gates and the intermediate unit to guide the learning of the hidden state and the cell state of the gated recurrent unit, thereby realizing soft measurement modeling of the penicillin fermentation process, providing a method for online real-time measurement of the concentration of the product of the penicillin fermentation process, improving the prediction accuracy of the concentration of the product of the penicillin fermentation process, and providing powerful guidance for the better generation of the fermentation production process.

[0108] In this specification, the present invention has been described with reference to specific embodiments thereof. However, it is apparent that various modifications and variations may be made without departing from the spirit and scope of the present invention. Therefore, the specification and drawings should be regarded as illustrative rather than restrictive.

Claims

1. A soft sensing method for penicillin fermentation process based on supervised gated recurrent unit deep network, characterized in that: The method comprises the following steps: (1) Obtaining a penicillin fermentation process data set from a database, and dividing the data set into a training set and a test set required by a soft sensor deep learning network in proportion; (2) Construct a soft-sensing model of penicillin fermentation process based on a supervised gated recurrent unit deep network and use the model to train the integrated data; (3) The trained model is used to predict the outlet concentration of penicillin fermentation products and the corresponding prediction results are given.

2. The soft sensing method for penicillin fermentation process based on supervised gated recurrent unit deep network according to claim 1 is characterized in that: The step (1) specifically comprises the following steps: (1.1) Set up the simulation process in the PenSim simulation platform to obtain a specific number of variable output data of the simulation platform; (1.2) Divide the output data into training set and test set according to a specific ratio; (1.3) and standardize the corresponding data in the following way: Among them, X′ is the data after standardization, X is the original data without standardization, μ is the mean of the original data, and σ is the standard deviation of the original data.

3. The soft sensing method for penicillin fermentation process based on supervised gated recurrent unit deep network according to claim 1 is characterized in that: The step (2) specifically comprises the following steps: (2.1) The reset gate, update gate and quality variable vector of the intermediate unit of the GRU network are used as input variables of the supervised gated recurrent unit SGRU; (2.2) Randomly give the SGRU model a parameter, the input and output sequence is {x (1) ,x (2) ,...,x (T) } and {y (1) ,y (2) ,...,y (T) }, get the first prediction value of the SGRU model; (2.3) According to the difference between the predicted value and the output value, the parameters of the SGRU model are adjusted so that the difference between the predicted value and the true value is reduced; (2.4) Repeat steps (2.1) to (2.3) until the difference between the predicted value and the true value of the SGRU model is within the allowable error.

4. The soft sensing method for penicillin fermentation process based on supervised gated recurrent unit deep network according to claim 3 is characterized in that: The GRU network is represented as follows: r t =σ(x t W r +h t-1 U r +b r ) z t =σ(x t W z +h t-1 U z +b z ) Among them, r t represents the result of resetting the gate, z t represents the result of updating the gate, x t is the input at the current moment, h t-1 is the hidden state at the previous moment, represents the candidate hidden state, h t represents the final hidden state, σ is the sigmoid activation function, tanh is the hyperbolic tangent activation function, ⊙ represents element-wise multiplication, W r and U r are the weight matrices from input and hidden state to reset gate, W z and U z are the weight matrices from input and hidden state to update gate, W h and U h are the weight matrices from the hidden state adjusted by the input and reset gates to the candidate hidden state, and b r , b z , b h , b y Both represent trainable bias terms, U y is the weight matrix of the output layer, It is the final output of the supervised gated recurrent unit SGRU.

5. The soft sensing method for penicillin fermentation process based on supervised gated recurrent unit deep network according to claim 4 is characterized in that: The supervised gated recurrent unit SGRU is represented as follows: r t =σ(x t W r +h t-1 U r +y t V r +b r ) z t =σ(x t W z +h t-1 U z +y t V z +b z ) Among them, y t is the output at the current moment, V r is the weight matrix output to the reset gate, V z is the weight matrix output to the update gate, V y is the weight matrix of the output layer.

6. The soft sensing method for penicillin fermentation process based on supervised gated recurrent unit deep network according to claim 5 is characterized in that: The step (2) further includes testing the SGRU model using the test data in the following manner: (a) Assume that the test time series input is For each x from the first to the last sample in the test sequence (T+j) , its output value is initialized to the output value of the previous sample, that is, (b) Replace the initialized data sample into the trained SGRU network to obtain x (T+j) The hidden state h(t) of (T+j) The final output is (c) The root mean square error between the predicted and true values ​​of the test set is calculated, and the performance of the supervised gated recurrent unit SGRU, gated recurrent unit GRU, long short-term memory LSTM, and long short-term memory network SLSTM is evaluated.

7. The soft sensing method for penicillin fermentation process based on supervised gated recurrent unit deep network according to claim 6 is characterized in that: The step (3) is specifically as follows: The test data is input into the trained SGRU model, and the mean square error RMSE is used as the evaluation indicator. The calculation formula is as follows: in, represents the real data, y i Represents the output of the SGRU model, and n represents the number of samples in the test set; The smaller the evaluation indicator RMSE is, the closer the predicted value of the SGRU model is to the true value, that is, the better the prediction effect of the model is.

8. A device for soft measurement of penicillin fermentation process based on a supervised gated recurrent unit deep network, characterized in that: The device comprises: a processor configured to execute computer executable instructions; A memory storing one or more computer executable instructions, wherein when the computer executable instructions are executed by the processor, the steps of the soft measurement method for penicillin fermentation process based on a supervised gated recurrent unit deep network described in any one of claims 1 to 7 are implemented.

9. A processor for soft measurement of penicillin fermentation process based on a deep network of supervised gated recurrent units, characterized in that: The processor is configured to execute computer executable instructions to implement the steps of the soft measurement method for penicillin fermentation process based on a supervised gated recurrent unit deep network according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program can be executed by a processor to implement the steps of the soft measurement method for penicillin fermentation process based on a supervised gated recurrent unit deep network as described in any one of claims 1 to 7.

Images