Method and system for detecting fermentation process parameters of recombinant human interferon

Abstract

The application discloses a kind of detection methods and systems of recombinant human interferon fermentation process parameters, belong to biological process monitoring and artificial intelligence technical field, prediction is carried out by the form of Raman spectrum data, avoid the destruction to escherichia coli bacteria, to achieve the purpose of non-destructive prediction;And, using small-scale fermentation tank as experimental data group carries out training data acquisition, and it is applied to model training, to quickly train out process parameter prediction model, realizes the nonlinear feature extraction and calculation that simple linear model cannot realize, obtains accurate Raman spectrum characteristics, to improve prediction accuracy, and it is verified by the standard process parameter value of large-scale fermentation tank, obtain the final application model that can be directly applied to production, real-time, accurate process parameter detection is carried out to actual production.

Description

Technical Field

[0001] This invention belongs to the field of biological process monitoring and artificial intelligence technology, specifically relating to a method and system for detecting parameters in the fermentation process of recombinant human interferon. Background Technology

[0002] Recombinant human interferon has irreplaceable clinical value in antiviral and antitumor therapy. Its industrial production mainly relies on E. coli fermentation technology. However, this production process has long faced the bottleneck of lagging monitoring of key process parameters. Traditional quality control methods depend on manual timed sampling and offline analysis, such as using a spectrophotometer to determine cell density (OD600) and a biochemical analyzer to detect glucose concentration. These methods are not only time-consuming, lasting for several hours, but also destructive, resulting in a serious lack of process information and failing to reflect the dynamic changes in cell growth, substrate metabolism, and product synthesis within the fermenter in real time.

[0003] Fermentation, especially high-density fermentation of *E. coli*, is a highly dynamic and nonlinear complex process with an extremely short optimal process window. Offline detection introduces time blind spots, severely hindering process control and preventing timely intervention in metabolic deviations. This leads to large batch-to-batch quality fluctuations, unstable yields, and high risks associated with scale-up production, significantly restricting production cost control and market supply assurance. In recent years, process analysis techniques have offered new solutions, including Raman spectroscopy for process parameter detection. However, most existing Raman spectroscopy methods employ linear models such as partial least squares (PLS). These linear models struggle to resolve the high-dimensional, nonlinear characteristics of spectral signals in complex bio-fermentation systems, particularly for substances like recombinant human interferon (e.g., *E. coli*). For target products that cannot be directly quantified by a single characteristic peak and have low concentrations, these linear prediction models struggle to establish robust quantitative prediction relationships, and their prediction accuracy and robustness cannot meet the needs of actual production.

[0004] As mentioned above, how to provide a method and system for detecting recombinant human interferon fermentation process parameters that can achieve real-time, non-destructive prediction and improve the accuracy of process parameter prediction has become an urgent research topic in this field. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for detecting parameters in the fermentation process of recombinant human interferon, so as to solve the above-mentioned problems existing in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for detecting parameters in the fermentation process of recombinant human interferon, comprising: Using a small-scale fermenter as the experimental data set, the original fermentation broth Raman spectral data of Escherichia coli fermentation process containing recombinant human interferon were obtained from multiple experimental data sets. The original fermentation broth Raman spectral data were then preprocessed and physical constraint corrected to obtain the fermentation broth Raman spectral data. A preset process parameter prediction model structure is obtained, and the Raman spectral data of the fermentation broth is used as training data. The model is trained based on the process parameter prediction model structure to obtain the process parameter prediction model. Using a large-scale fermenter as the validation data set, the process parameter prediction model is used to process the validation data set. Through regression calculation of the process parameter prediction model, the predicted values ​​of process parameters corresponding to the validation data set are output, wherein the predicted values ​​of process parameters include predicted values ​​of cell density and predicted values ​​of glucose concentration. Obtain preset standard process parameter values ​​for the validation group, compare the predicted process parameter values ​​with the standard process parameter values ​​for the validation group to obtain model error values, and perform performance verification on the process parameter prediction model based on the model error values. The process parameter prediction model that passes the performance verification is used as the final application model to detect the fermentation process parameters of recombinant human interferon.

[0007] In one possible design, a small-scale fermenter is used as the experimental data set. Raw Raman spectral data of the fermentation broth during the E. coli fermentation process of recombinant human interferon are obtained from multiple experimental data sets. The raw fermentation broth Raman spectral data are then preprocessed and subjected to physical constraint corrections to obtain the fermentation broth Raman spectral data, including: Culture medium was injected into multiple small-scale fermenters to form a preliminary experimental data set; Obtain preset Raman spectral acquisition parameters, and according to the Raman spectral acquisition parameters, acquire spectral data for each of the pre-experimental data sets to obtain the original culture medium Raman spectral data; Each of the aforementioned pre-experimental data sets was inoculated with E. coli culture to obtain experimental data sets; For each of the experimental data groups, spectral data were acquired according to the Raman spectroscopy acquisition parameters to obtain the original fermentation broth Raman spectral data of the Escherichia coli fermentation process of recombinant human interferon in each experimental data group. The original culture medium Raman spectrum data is subtracted from the original fermentation broth Raman spectrum data to obtain the pre-fermentation broth Raman spectrum data; The Raman spectral data of the pre-fermentation broth are organized according to the standard time sequence and multiple Raman spectral vectors are formed. Each Raman spectral vector is a one-dimensional vector that includes timestamp information, and each element value in the Raman spectral vector represents the signal intensity value corresponding to a preset Raman shift. Obtain a preset Savitzky-Golay convolution smoothing coefficient, generate a sliding window based on the Savitzky-Golay convolution smoothing coefficient, and use the sliding window to perform convolution calculation on each of the Raman spectral vectors to complete the traversal of each of the Raman spectral vectors and obtain multiple corresponding smooth Raman spectral vectors. For each of the smoothed Raman spectral vectors, the corresponding arithmetic mean and standard deviation are calculated, and the standardized element value is calculated for each element value in each of the smoothed Raman spectral vectors using the following formula (1): (1) in, The first in the smoothed Raman spectral vector Each element value The arithmetic mean of the smoothed Raman spectral vector is given. The standard deviation corresponding to the smoothed Raman spectral vector is... The first in the smoothed Raman spectral vector The normalized element value corresponding to each element value; For each of the smoothed Raman spectral vectors, the standardized element values ​​corresponding to each element value in the smoothed Raman spectral vectors are integrated to form multiple corresponding standard smoothed Raman spectral vectors, wherein the arithmetic mean of each standard smoothed Raman spectral vector is 0, and the standard deviation of each standard smoothed Raman spectral vector is 1. A preset four-dimensional tensor structure is obtained, and the standard smoothed Raman spectral vectors are structured using the four-dimensional tensor structure to obtain Raman spectral data of the fermentation broth.

[0008] In one possible design, the process parameter prediction model structure includes, in sequence, an input layer, a convolutional neural network layer, a flattening layer, a fully connected layer, and an output layer, wherein: The input layer is used to receive input Raman spectral data of the fermentation broth; The convolutional neural network layer is used to perform one-dimensional convolution processing on the Raman spectrum data of the fermentation broth transmitted by the input layer to obtain the Raman spectrum feature map of the fermentation broth, wherein the Raman spectrum feature map of the fermentation broth includes a first-level spectral feature map of the pooled fermentation broth, a second-level spectral feature map of the pooled fermentation broth, and a third-level spectral feature map of the pooled fermentation broth. The flattening layer is used to flatten the Raman spectral feature map of the fermentation broth output by the convolutional neural network layer into a one-dimensional Raman spectral feature vector of the fermentation broth. The fully connected layer is used to perform nonlinear combination and mapping on the Raman spectral feature vector of the fermentation broth output by the flattening layer, so as to establish the prediction and inference relationship between the Raman spectral feature vector of the fermentation broth and the process parameters, and calculate the predicted value of the process parameters based on the prediction and inference relationship between the Raman spectral feature vector of the fermentation broth and the process parameters. The output layer is used to perform regression output on the predicted values ​​of the process parameters output by the fully connected layer. The output layer includes two nodes, which are used to output the predicted values ​​of cell density and glucose concentration, respectively.

[0009] In one possible design, the convolutional neural network layer includes a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a third convolutional layer, and a third max-pooling layer connected in sequence, wherein: The first convolutional layer includes 16 convolutional kernels, each with a size of 3. The 16 convolutional kernels are used to perform one-dimensional convolution processing on the fermentation broth Raman spectral data transmitted from the input layer to extract the primary spectral features of the fermentation broth corresponding to the Raman spectral data. Based on the primary spectral features of the fermentation broth, 16 primary spectral feature maps of the fermentation broth are formed. Each primary spectral feature map of the fermentation broth retains the local correlation features of the Raman spectral data of the fermentation broth. The first maximum pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 16 primary spectral feature maps of the fermentation broth output by the first convolutional layer, so as to select the maximum value in each sliding window to form 16 primary spectral feature maps of the pooled fermentation broth, and output the 16 primary spectral feature maps of the pooled fermentation broth to the flattening layer. The second convolutional layer includes 32 convolutional kernels, each with a size of 3. The 32 convolutional kernels are used to perform one-dimensional convolution processing on the 16 primary spectral feature maps of the pooled fermentation broth output by the first maximum pooling layer to form 32 secondary spectral feature maps of the fermentation broth. The second maximum pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 32 secondary spectral feature maps of the fermentation broth output by the second convolutional layer, so as to select the maximum value in each sliding window to form 32 pooled fermentation broth secondary spectral feature maps, and output the 32 pooled fermentation broth secondary spectral feature maps to the flattening layer. The third convolutional layer includes 64 convolutional kernels, each with a size of 3. The 64 convolutional kernels are used to perform one-dimensional convolution processing on the 32 secondary spectral feature maps of the pooled fermentation broth output by the second maximum pooling layer to form 64 tertiary spectral feature maps of the fermentation broth. The third max-pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 64 tertiary spectral feature maps of the fermentation broth output by the third convolutional layer. The maximum value is selected in each sliding window to form 64 pooled fermentation broth tertiary spectral feature maps, and the 64 pooled fermentation broth tertiary spectral feature maps are output to the flattening layer.

[0010] In one possible design, the Raman spectral data of the fermentation broth is used as training data to train the model based on the process parameter prediction model structure, thereby obtaining the process parameter prediction model, including: A preset training sample ratio is obtained, the Raman spectral data of the fermentation broth is used as training data, and the training data is randomly divided into a training sample set and a test sample set according to the training sample ratio. Obtain a preset loss function, perform multiple rounds of forward propagation on the process parameter prediction model structure based on the training sample set to obtain the predicted values ​​of the process parameters of the training samples, and calculate the loss value of each round of forward propagation based on the loss function on the training sample set and the predicted values ​​of the process parameters of the training samples. Using the gradient descent iterative method, backpropagation is performed based on the loss value of each forward propagation round to update the model weight parameters of the process parameter prediction model structure. A preset maximum number of training rounds is obtained. Training is completed after the maximum number of training rounds is reached to obtain a preprocess parameter prediction model. The model prediction performance of the preprocess parameter prediction model is tested using the test sample set. After the preprocess parameter prediction model passes the model prediction performance test, the preprocess parameter prediction model is used as the process parameter prediction model.

[0011] In one possible design, a large-scale fermenter is used as the validation data set. The process parameter prediction model is used to process the validation data set. Through regression calculation of the process parameter prediction model, predicted values ​​of process parameters corresponding to the validation data set are output, including: At least one large-scale fermenter was injected with culture medium and inoculated with E. coli to obtain a validation data set. Spectral data were acquired from the validation data sets to obtain the pre-validation fermentation broth Raman spectra of the Escherichia coli fermentation process of recombinant human interferon in each validation data set. The Raman spectral data of the pre-validation fermentation broth were smoothed, standardized, and structured to obtain the Raman spectral data of the validation fermentation broth. The Raman spectral data of the validation fermentation broth are input into the process parameter prediction model. Regression calculations are performed through the process parameter prediction model to obtain the output of the process parameter prediction model. The output of the process parameter prediction model is then used as the predicted value of the process parameter corresponding to the validation data set.

[0012] In one possible design, preset validation group standard process parameter values ​​are obtained. The predicted process parameter values ​​are compared with the validation group standard process parameter values ​​to obtain a model error value. The performance of the process parameter prediction model is validated based on the model error value. The process parameter prediction model that passes the performance validation is used as the final application model to detect the parameters of the recombinant human interferon fermentation process, including: Using the offline OD600 measurement method, multiple sets of offline process parameter values ​​are generated for the verification data set. The average value of each set of offline process parameter values ​​is calculated to obtain the calculation result, and the calculation result is used as the standard process parameter value of the verification set. A preset prediction error threshold is obtained, the predicted value of the process parameter is compared with the standard process parameter value of the verification group to obtain the model error value, and the performance of the process parameter prediction model is verified using the prediction error threshold and the model error value to obtain the verification result; If the model error value exceeds the prediction error threshold, the process parameter prediction model is considered to have failed the performance verification. The experimental data set is then reset, training data is obtained again, and the process parameter prediction model is trained again. The retrained process parameter prediction model is then subjected to performance verification. The verification result is considered to be that the process parameter prediction model has passed the performance verification. If the model error value does not exceed the prediction error threshold, the process parameter prediction model is considered to have passed performance verification. The process parameter prediction model that has passed performance verification is used as the final application model, and the final application model is deployed to the production control computing node to detect the parameters of the recombinant human interferon fermentation process.

[0013] Secondly, the present invention provides a system for detecting parameters in the fermentation process of recombinant human interferon, comprising: The training data acquisition unit is used to acquire the original fermentation broth Raman spectrum data of Escherichia coli fermentation process of recombinant human interferon from multiple experimental data sets, using a small-scale fermenter as the experimental data set, and to perform data preprocessing and physical constraint correction on the original fermentation broth Raman spectrum data to obtain fermentation broth Raman spectrum data. The prediction model training unit is used to acquire a preset process parameter prediction model structure, use the Raman spectrum data of the fermentation broth as training data, and train the model based on the process parameter prediction model structure to obtain the process parameter prediction model. The validation data calculation unit is used to process the validation data set using the process parameter prediction model with a large-scale fermenter as the validation data set. Through regression calculation of the process parameter prediction model, the unit outputs the process parameter prediction values ​​corresponding to the validation data set, wherein the process parameter prediction values ​​include cell density prediction values ​​and glucose concentration prediction values. The application model validation unit is used to obtain preset validation group standard process parameter values, compare the predicted process parameter values ​​with the validation group standard process parameter values ​​to obtain model error values, and perform performance validation on the process parameter prediction model based on the model error values, so as to use the process parameter prediction model that passes the performance validation as the final application model to detect the fermentation process parameters of recombinant human interferon.

[0014] Thirdly, the present invention provides an electronic device comprising a memory, a processor, and a transceiver connected in sequence and communication, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the method for detecting parameters of the recombinant human interferon fermentation process as described in the first aspect or any possible design of the first aspect.

[0015] Fourthly, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform the method for detecting parameters of the recombinant human interferon fermentation process as described in the first aspect or any possible design of the first aspect.

[0016] Fifthly, the present invention provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform a method for detecting parameters of the recombinant human interferon fermentation process as described in the first aspect or any possible design of the first aspect.

[0017] Beneficial Effects: This invention provides a method and system for detecting parameters in the fermentation process of recombinant human interferon, comprising: First, using a small-scale fermenter as the experimental data set, obtaining the original fermentation broth Raman spectral data of the E. coli fermentation process of recombinant human interferon from multiple experimental data sets, and performing data preprocessing and physical constraint correction on the original fermentation broth Raman spectral data to obtain fermentation broth Raman spectral data; Second, obtaining a preset process parameter prediction model structure, using the fermentation broth Raman spectral data as training data, and training the model based on the process parameter prediction model structure to obtain the process parameter prediction model; Then, using a large-scale fermenter as validation data. The validation data set is processed using the process parameter prediction model. Through regression calculation by the process parameter prediction model, predicted process parameter values ​​corresponding to the validation data set are output, including predicted cell density and predicted glucose concentration. Finally, preset standard process parameter values ​​for the validation set are obtained, and the predicted process parameter values ​​are compared with the standard process parameter values ​​to obtain the model error value. The performance of the process parameter prediction model is validated based on the model error value, and the process parameter prediction model that passes the performance validation is used as the final application model to detect the fermentation process parameters of recombinant human interferon. Predictions are made using Raman spectroscopy data, avoiding damage to E. coli cells and achieving non-destructive prediction. Furthermore, small-scale fermenters are used as the experimental dataset for training data acquisition, which is then applied to model training to quickly develop a process parameter prediction model. This enables nonlinear feature extraction and calculation that simple linear models cannot achieve, obtaining accurate Raman spectral features to improve prediction accuracy. The model is then validated using standard process parameter values ​​from large-scale fermenters, resulting in a final application model that can be directly applied to production, enabling real-time and accurate process parameter detection in actual production. Attached Figure Description

[0018] Figure 1 A schematic flowchart illustrating the method for detecting parameters in the fermentation process of recombinant human interferon provided in this embodiment of the invention; Figure 2 This is a Raman spectrum overlay diagram of the Raman spectral data of the pre-fermentation broth provided in an embodiment of the present invention; Figure 3 A schematic diagram of a possible process parameter prediction model structure provided in an embodiment of the present invention; Figure 4 A schematic diagram of another possible process parameter prediction model structure provided in an embodiment of the present invention; Figure 5 A schematic diagram of the functional structure of the detection system for recombinant human interferon fermentation process parameters provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0020] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0021] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0022] Example: like Figure 1 As shown, the first aspect of this embodiment provides a method for detecting parameters in the fermentation process of recombinant human interferon, which may include, but is not limited to, the following steps: S1. Using a small-scale fermenter as the experimental data set, the original fermentation broth Raman spectral data of Escherichia coli fermentation process of recombinant human interferon was obtained from multiple experimental data sets, and the original fermentation broth Raman spectral data was preprocessed and physical constraint corrected to obtain the fermentation broth Raman spectral data. In one possible implementation, step S1 involves using a small-scale fermenter as the experimental data set, obtaining the original fermentation broth Raman spectral data of *E. coli* during the fermentation process of recombinant human interferon from multiple experimental data sets, and performing data preprocessing and physical constraint correction on the original fermentation broth Raman spectral data to obtain the fermentation broth Raman spectral data. This step can be broken down into, but is not limited to, the following steps S11-S10, specifically including: S11. Inject culture medium into multiple small-scale fermenters to form a preliminary experimental data set; S12. Obtain preset Raman spectral acquisition parameters, and according to the Raman spectral acquisition parameters, acquire spectral data for each of the pre-experimental data sets to obtain the original culture medium Raman spectral data; S13. Inoculate each of the pre-experimental data groups with Escherichia coli culture to obtain experimental data groups; S14. For each of the experimental data groups, spectral data are acquired according to the Raman spectral acquisition parameters to obtain the original fermentation broth Raman spectral data of the Escherichia coli fermentation process of recombinant human interferon in each experimental data group; S15. Subtract the original culture medium Raman spectrum data from the original fermentation broth Raman spectrum data to obtain the pre-fermentation broth Raman spectrum data; S16. The Raman spectral data of the pre-fermentation broth are organized in standard time sequence and multiple Raman spectral vectors are formed. The Raman spectral vector is a one-dimensional vector including timestamp information, and each element value in the Raman spectral vector represents the signal intensity value corresponding to a preset Raman shift. S17. Obtain a preset Savitzky-Golay convolution smoothing coefficient, generate a sliding window based on the Savitzky-Golay convolution smoothing coefficient, and use the sliding window to perform convolution calculation on each of the Raman spectral vectors to complete the traversal of each of the Raman spectral vectors and obtain multiple corresponding smooth Raman spectral vectors. S18. For each of the smoothed Raman spectral vectors, calculate the corresponding arithmetic mean and standard deviation, and use the following formula (1) to calculate the corresponding standardized element value for each element value in each of the smoothed Raman spectral vectors: (1) in, The first in the smoothed Raman spectral vector Each element value The arithmetic mean of the smoothed Raman spectral vector is given. The standard deviation corresponding to the smoothed Raman spectral vector is... The first in the smoothed Raman spectral vector The normalized element value corresponding to each element value; S19. For each of the smoothed Raman spectral vectors, integrate the standardized element values ​​corresponding to each element value in the smoothed Raman spectral vector to form multiple corresponding standard smoothed Raman spectral vectors, wherein the arithmetic mean of each standard smoothed Raman spectral vector is 0, and the standard deviation of each standard smoothed Raman spectral vector is 1. S10. Obtain a preset four-dimensional tensor structure, and use the four-dimensional tensor structure to perform structuring processing on each of the standard smoothed Raman spectral vectors to obtain fermentation broth Raman spectral data. The four-dimensional tensor structure includes the number of samples, the number of channels, and the length. The standard smoothed Raman spectral vectors are structurated into a four-dimensional tensor structure to adapt to subsequent convolution operations.

[0023] In specific applications, the cultivation process of *E. coli* in small-scale fermenters (5L) and large-scale fermenters (80L) is as follows: First, the interferon gene is cloned from human cells, and this gene is ligated with an *E. coli* expression vector to form a recombinant expression plasmid, which is then transformed into *E. coli*. Glycerol tube cultures stored at -70℃ or below are thawed at room temperature. Second, the *E. coli* cultures are inoculated into shake flasks, and the culture temperature is set to 37℃, pH 7.0, and the shaking speed to 250 rpm for overnight culture to obtain the culture medium. Then, the culture medium is inoculated into a small-scale fermenter (5L) or a large-scale fermenter (80L), both of which are equipped with a pH control probe (BIOSYSTEC), a dissolved oxygen sensor (BIOSYSTEC), and a Raman probe (AUSTAR). Using a 5L fermenter... Taking a small-scale fermenter as an example, 2.7 LLB of culture medium was injected into the fermenter, followed by 270 mL of overnight cultured bacterial solution. Simultaneously, samples of the fermentation broth were taken for microscopic examination to control contaminating microorganisms. The culture temperature was set at 37°C, the pH at 7.0, the initial aeration rate at 3 L / min compressed air, and the initial stirring speed at 300 rpm. During fermentation, the dissolved oxygen level in the fermenter was maintained at 30% by adjusting the stirring speed (maximum speed 800 rpm) and aeration rate. Subsequently, glucose was added in batches, and 50% ammonia water was used to adjust the pH. During fermentation, samples of the bacterial solution in the culture medium were taken, and the absorbance value at 600 nm was measured using a Raman probe as the original Raman spectral data of the fermentation broth.

[0024] It should be noted that in the method for detecting parameters of the recombinant human interferon fermentation process provided in this embodiment, the Raman spectrometer used for acquiring the corresponding Raman spectral data is an AUSTAR instrument. It is equipped with a laser probe operating at an excitation wavelength of 785 nm. The stainless steel probe is designed to collect spectral signals below the surface of the bioreactor liquid. This setup allows for non-invasive monitoring of key process parameters during cultivation. The Raman spectral data processed by the Raman spectrometer can be transmitted to a terminal computer via a local area network for analysis and visualization. In specific applications, the laser power of the AUSTAR instrument can be set to 450 milliwatts, and the integration time to 5000 milliseconds. During an 8-hour continuous fermentation process, Raman spectral data at a total of 225 time points were acquired at 2.1-minute intervals. Figure 2 As shown, the Raman spectra of the pre-fermentation broth collected at 8-hour intervals during fermentation are stacked, with the highest intensity peak appearing at approximately 418. The weakest characteristic peak appears at 752. At, in the spectral range of 107-3284 The spectrum shows multiple peaks of varying intensities, while 50–3500 The spectral range can be determined as the effective bandwidth, covering most of the information about E. coli culture, where the characteristic Raman spectrum of glucose is at 448. 515 1034 and 1125 This can be observed to be consistent with the results of offline studies. The cell density of E. coli itself does not have a direct Raman peak, but can be indirectly obtained by detecting changes in the intensity of characteristic peaks of biomacromolecules in the cell and metabolic changes through Raman spectroscopy.

[0025] Furthermore, after obtaining the Raman spectral data of the pre-fermentation broth in step S15, this embodiment also proposes, but is not limited to, a label association method, specifically: First, labels are added to the Raman spectral data of the pre-fermentation broth lacking offline label values ​​using interpolation (i.e., adding offline measured OD600 values ​​to the Raman spectral data of the pre-fermentation broth at preset key time points). Initial labels for the entire time series are constructed through cubic spline interpolation. Through subsequent data preprocessing and obtaining preset physical constraints to correct unreasonable values, smooth and metabolically consistent full-time series labels are finally generated and associated with the corresponding Raman spectral data for storage. This label management process increases the scale of relatively small Raman spectral data by tens of times, effectively enriching the training sample size for subsequent training. Simultaneously, a self-supervised learning dataset can be introduced to reduce the reliance on obtaining labeled data through numerous repetitive experiments, significantly reducing the risk of model overfitting due to insufficient data.

[0026] This data preprocessing process, through Savitzky-Golay smoothing and standardization, eliminates baseline drift caused by differences in spectral acquisition equipment and fermenter size. This is the foundation for the subsequent training of the process prediction model to be validated in large-scale fermenters and applied.

[0027] S2. Obtain a preset process parameter prediction model structure, use the Raman spectrum data of the fermentation broth as training data, and train the model based on the process parameter prediction model structure to obtain the process parameter prediction model. like Figure 3 As shown, in one possible implementation, in step S2, the process parameter prediction model structure includes an input layer, a convolutional neural network layer, a flattening layer, a fully connected layer, and an output layer connected in sequence, wherein: The input layer is used to receive input Raman spectral data of the fermentation broth; The convolutional neural network layer is used to perform one-dimensional convolution processing on the Raman spectrum data of the fermentation broth transmitted by the input layer to obtain the Raman spectrum feature map of the fermentation broth, wherein the Raman spectrum feature map of the fermentation broth includes a first-level spectral feature map of the pooled fermentation broth, a second-level spectral feature map of the pooled fermentation broth, and a third-level spectral feature map of the pooled fermentation broth. The flattening layer is used to flatten the Raman spectral feature map of the fermentation broth output by the convolutional neural network layer into a one-dimensional Raman spectral feature vector of the fermentation broth. The fully connected layer is used to perform nonlinear combination and mapping on the Raman spectral feature vector of the fermentation broth output by the flattening layer, so as to establish the prediction and inference relationship between the Raman spectral feature vector of the fermentation broth and the process parameters, and calculate the predicted value of the process parameters based on the prediction and inference relationship between the Raman spectral feature vector of the fermentation broth and the process parameters. The output layer is used to perform regression output on the predicted values ​​of the process parameters output by the fully connected layer. The output layer includes two nodes, which are used to output the predicted values ​​of cell density and glucose concentration, respectively.

[0028] like Figure 4 As shown, in one possible implementation, in step S2, the convolutional neural network layer includes a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a third convolutional layer, and a third max-pooling layer connected in sequence, wherein: The first convolutional layer includes 16 convolutional kernels, each with a size of 3. The 16 convolutional kernels are used to perform one-dimensional convolution processing on the fermentation broth Raman spectral data transmitted from the input layer to extract the primary spectral features of the fermentation broth corresponding to the Raman spectral data. Based on the primary spectral features of the fermentation broth, 16 primary spectral feature maps of the fermentation broth are formed. Each primary spectral feature map of the fermentation broth retains the local correlation features of the Raman spectral data of the fermentation broth. The first maximum pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 16 primary spectral feature maps of the fermentation broth output by the first convolutional layer, so as to select the maximum value in each sliding window to form 16 primary spectral feature maps of the pooled fermentation broth, and output the 16 primary spectral feature maps of the pooled fermentation broth to the flattening layer. The second convolutional layer includes 32 convolutional kernels, each with a size of 3. The 32 convolutional kernels are used to perform one-dimensional convolution processing on the 16 primary spectral feature maps of the pooled fermentation broth output by the first maximum pooling layer to form 32 secondary spectral feature maps of the fermentation broth. The second maximum pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 32 secondary spectral feature maps of the fermentation broth output by the second convolutional layer, so as to select the maximum value in each sliding window to form 32 pooled fermentation broth secondary spectral feature maps, and output the 32 pooled fermentation broth secondary spectral feature maps to the flattening layer. The third convolutional layer includes 64 convolutional kernels, each with a size of 3. The 64 convolutional kernels are used to perform one-dimensional convolution processing on the 32 secondary spectral feature maps of the pooled fermentation broth output by the second maximum pooling layer to form 64 tertiary spectral feature maps of the fermentation broth. The third max-pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 64 tertiary spectral feature maps of the fermentation broth output by the third convolutional layer. The maximum value is selected in each sliding window to form 64 pooled fermentation broth tertiary spectral feature maps, and the 64 pooled fermentation broth tertiary spectral feature maps are output to the flattening layer.

[0029] It should be noted that in the method for detecting recombinant human interferon fermentation process parameters provided in this embodiment, the three one-dimensional convolutional layers of the convolutional neural network are set with 16, 32 and 64 convolutional kernels respectively, with kernel sizes of 3, 3 and 3 respectively, to progressively extract key features of the fermentation broth Raman spectral data. Each convolutional layer is followed by a max pooling layer with pooling kernel sizes of 2, 2 and 2 respectively. This is to reduce the feature dimension and retain core information, so as to improve the generalization ability of the process parameter prediction model.

[0030] The first, second, and third max-pooling layers feed the generated primary, secondary, and tertiary spectral feature maps of the pooled fermentation broth to the flattening layer for feature flattening, forming a one-dimensional vector which is then input to the fully connected layer. In specific applications, the fully connected layer is configured with 128 and 64 neurons respectively to integrate spectral features and achieve a more complex and high-dimensional representation. Therefore, the process parameter prediction model established in this way possesses translation invariance and local sensing capability, effectively capturing stable modes in Raman spectral data, unaffected by absolute signal intensity or slight shifts.

[0031] In one possible implementation, step S2, using the Raman spectral data of the fermentation broth as training data, and training the model based on the process parameter prediction model structure to obtain the process parameter prediction model, can be decomposed into, but is not limited to, the following steps S21-S24, specifically including: S21. Obtain a preset training sample ratio, use the Raman spectral data of the fermentation broth as training data, and randomly divide the training data into a training sample set and a test sample set according to the training sample ratio; S22. Obtain a preset loss function, perform multiple rounds of forward propagation on the process parameter prediction model structure based on the training sample set to obtain the predicted values ​​of the process parameters of the training samples, and calculate the loss value of each round of forward propagation based on the loss function on the training sample set and the predicted values ​​of the process parameters of the training samples. S23. Using the gradient descent iterative method, backpropagation is performed based on the loss value of each forward propagation to update the model weight parameters of the process parameter prediction model structure; S24. Obtain the preset maximum number of training rounds, complete the training after the maximum number of training rounds is reached, obtain the pre-process parameter prediction model, and use the test sample set to perform model prediction performance test on the pre-process parameter prediction model, so that after the pre-process parameter prediction model passes the model prediction performance test, the pre-process parameter prediction model is used as the process parameter prediction model.

[0032] In practical applications, a unified configuration is adopted for training the process parameter prediction model. The preset training sample ratio is 4:1, the preset maximum training rounds are 100, and the batch size of training samples in each round is 32. The model weight parameters are optimized by gradient descent iterative method to establish a robust mapping from complex Raman spectral data to key biological parameters, so as to meet the low latency requirements of real-time online monitoring. This is very suitable for monitoring the dynamics of cell growth and nutrient consumption in complex fermentation fluids.

[0033] S3. Using a large-scale fermenter as the validation data set, the process parameter prediction model is used to process the validation data set. Through regression calculation of the process parameter prediction model, the predicted process parameters corresponding to the validation data set are output, wherein the predicted process parameters include predicted cell density and predicted glucose concentration. In one possible implementation, step S3 uses a large-scale fermenter as the validation data set. The process parameter prediction model is used to process the validation data set, and through regression calculations of the process parameter prediction model, predicted values ​​of the process parameters corresponding to the validation data set are output. This can be, but is not limited to, decomposed into the following steps S31-S34, specifically including: S31. Inject culture medium and inoculate with E. coli in at least one large-scale fermenter to obtain a validation data set; S32. Spectral data are acquired from the verification data set to obtain the pre-verification fermentation broth Raman spectral data of the Escherichia coli fermentation process of recombinant human interferon in each verification data set; S33. The Raman spectral data of the pre-validation fermentation broth are smoothed, standardized, and structured to obtain the Raman spectral data of the validation fermentation broth; S34. Input the Raman spectral data of the validation fermentation broth into the process parameter prediction model, perform regression calculation through the process parameter prediction model to obtain the output of the process parameter prediction model, and use the output of the process parameter prediction model as the process parameter prediction value corresponding to the validation data set.

[0034] S4. Obtain the preset standard process parameter values ​​of the validation group, compare the predicted process parameter values ​​with the standard process parameter values ​​of the validation group to obtain the model error value, and perform performance verification on the process parameter prediction model based on the model error value, so as to use the process parameter prediction model that has passed the performance verification as the final application model to detect the fermentation process parameters of recombinant human interferon.

[0035] In one possible implementation, step S4 involves obtaining preset validation group standard process parameter values, comparing the predicted process parameter values ​​with the validation group standard process parameter values ​​to obtain a model error value, and performing performance validation on the process parameter prediction model based on the model error value. The process parameter prediction model that passes the performance validation is then used as the final application model to detect the fermentation process parameters of recombinant human interferon. This can be broken down into, but is not limited to, the following steps S41-S44, specifically including: S41. Using the offline OD600 measurement method, generate multiple sets of offline process parameter values ​​corresponding to the verification data set, calculate the average value of each set of offline process parameter values, obtain the calculation result, and use the calculation result as the standard process parameter value of the verification set; S42. Obtain a preset prediction error threshold, compare the predicted value of the process parameter with the standard process parameter value of the verification group to obtain the model error value, and use the prediction error threshold and the model error value to perform performance verification on the process parameter prediction model to obtain the verification result; S43. If the model error value exceeds the prediction error threshold, the process parameter prediction model is considered to have failed the performance verification. The experimental data set is reset, training data is obtained again, and the process parameter prediction model is trained again. The retrained process parameter prediction model is then subjected to performance verification. The verification result is considered to be that the process parameter prediction model has passed the performance verification. S44. If the model error value does not exceed the prediction error threshold, the process parameter prediction model is considered to have passed performance verification. The process parameter prediction model that has passed performance verification is used as the final application model, and the final application model is deployed to the production control computing node to detect the parameters of the recombinant human interferon fermentation process.

[0036] It should be noted that the final application model constructed by the detection method of recombinant human interferon fermentation process parameters provided in this embodiment is not only a theoretically achievable model, but also a predictive model that has been validated by a large sample and on a large scale and can be directly connected to the actual production system and applied directly. In practical applications, it can not only provide real-time and accurate process detection capabilities and provide immediate insights into the current process, but also predict future development trends and support the implementation of adaptive control strategies, thereby significantly improving the intelligence level and production efficiency of the bioreactor process.

[0037] like Figure 5 As shown, the second aspect of this embodiment provides a hardware system for implementing the method for detecting parameters in the fermentation process of recombinant human interferon as described in the first aspect of the embodiment, including: The training data acquisition unit is used to acquire the original fermentation broth Raman spectrum data of Escherichia coli fermentation process of recombinant human interferon from multiple experimental data sets, using a small-scale fermenter as the experimental data set, and to perform data preprocessing and physical constraint correction on the original fermentation broth Raman spectrum data to obtain fermentation broth Raman spectrum data. The prediction model training unit is used to acquire a preset process parameter prediction model structure, use the Raman spectrum data of the fermentation broth as training data, and train the model based on the process parameter prediction model structure to obtain the process parameter prediction model. The validation data calculation unit is used to process the validation data set using the process parameter prediction model with a large-scale fermenter as the validation data set. Through regression calculation of the process parameter prediction model, the unit outputs the process parameter prediction values ​​corresponding to the validation data set, wherein the process parameter prediction values ​​include cell density prediction values ​​and glucose concentration prediction values. The application model validation unit is used to obtain preset validation group standard process parameter values, compare the predicted process parameter values ​​with the validation group standard process parameter values ​​to obtain model error values, and perform performance validation on the process parameter prediction model based on the model error values, so as to use the process parameter prediction model that passes the performance validation as the final application model to detect the fermentation process parameters of recombinant human interferon.

[0038] The working process, working details and technical effects of the system provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0039] like Figure 6 As shown, the third aspect of this embodiment provides an electronic device, including: a memory, a processor, and a transceiver that are sequentially and communicatively connected, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the method for detecting parameters of the recombinant human interferon fermentation process as described in the first aspect of the embodiment.

[0040] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0041] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee transceiver (a low-power LAN protocol based on the IEEE 802.15.4 standard), a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0042] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0043] The fourth aspect of this embodiment provides a storage medium that stores instructions for a method of detecting parameters of recombinant human interferon fermentation process as described in the first aspect of this embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the method of detecting parameters of recombinant human interferon fermentation process as described in the first aspect of this embodiment.

[0044] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0045] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0046] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the method for detecting parameters of the recombinant human interferon fermentation process as described in the first aspect of this embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0047] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting parameters in the fermentation process of recombinant human interferon, characterized in that, include: Using a small-scale fermenter as the experimental data set, the original fermentation broth Raman spectral data of Escherichia coli fermentation process containing recombinant human interferon were obtained from multiple experimental data sets. The original fermentation broth Raman spectral data were then preprocessed and physical constraint corrected to obtain the fermentation broth Raman spectral data. A preset process parameter prediction model structure is obtained, and the Raman spectral data of the fermentation broth is used as training data. The model is trained based on the process parameter prediction model structure to obtain the process parameter prediction model. Using a large-scale fermenter as the validation data set, the process parameter prediction model is used to process the validation data set. Through regression calculation of the process parameter prediction model, the predicted values ​​of process parameters corresponding to the validation data set are output, wherein the predicted values ​​of process parameters include predicted values ​​of cell density and predicted values ​​of glucose concentration. Obtain preset standard process parameter values ​​for the validation group, compare the predicted process parameter values ​​with the standard process parameter values ​​for the validation group to obtain model error values, and perform performance verification on the process parameter prediction model based on the model error values. The process parameter prediction model that passes the performance verification is used as the final application model to detect the fermentation process parameters of recombinant human interferon.

2. The method for detecting parameters in the fermentation process of recombinant human interferon according to claim 1, characterized in that, Using a small-scale fermenter as the experimental data set, raw Raman spectral data of the fermentation broth during the fermentation of recombinant human interferon in *E. coli* were obtained from multiple experimental data sets. The raw fermentation broth Raman spectral data were then preprocessed and subjected to physical constraint corrections to obtain the fermentation broth Raman spectral data, including: Culture medium was injected into multiple small-scale fermenters to form a preliminary experimental data set; Obtain preset Raman spectral acquisition parameters, and acquire spectral data for each of the pre-experimental data sets according to the Raman spectral acquisition parameters to obtain the original culture medium Raman spectral data; Each of the pre-experimental data sets was inoculated with E. coli culture to obtain the experimental data sets. For each of the experimental data groups, spectral data were acquired according to the Raman spectroscopy acquisition parameters to obtain the original fermentation broth Raman spectral data of the Escherichia coli fermentation process of recombinant human interferon in each experimental data group. The original culture medium Raman spectrum data is subtracted from the original fermentation broth Raman spectrum data to obtain the pre-fermentation broth Raman spectrum data; The Raman spectral data of the pre-fermentation broth are organized according to the standard time sequence and multiple Raman spectral vectors are formed. Each Raman spectral vector is a one-dimensional vector that includes timestamp information, and each element value in the Raman spectral vector represents the signal intensity value corresponding to a preset Raman shift. Obtain a preset Savitzky-Golay convolution smoothing coefficient, generate a sliding window based on the Savitzky-Golay convolution smoothing coefficient, and use the sliding window to perform convolution calculation on each of the Raman spectral vectors to complete the traversal of each of the Raman spectral vectors and obtain multiple corresponding smooth Raman spectral vectors. For each of the smoothed Raman spectral vectors, the corresponding arithmetic mean and standard deviation are calculated, and the standardized element value is calculated for each element value in each of the smoothed Raman spectral vectors using the following formula (1): （1） in, The first in the smoothed Raman spectral vector Each element value The arithmetic mean of the smoothed Raman spectral vector is given. The standard deviation corresponding to the smoothed Raman spectral vector is... The first in the smoothed Raman spectral vector The normalized element value corresponding to each element value; For each of the smoothed Raman spectral vectors, the standardized element values ​​corresponding to each element value in the smoothed Raman spectral vectors are integrated to form multiple corresponding standard smoothed Raman spectral vectors, wherein the arithmetic mean of each standard smoothed Raman spectral vector is 0, and the standard deviation of each standard smoothed Raman spectral vector is 1. A preset four-dimensional tensor structure is obtained, and the standard smoothed Raman spectral vectors are structured using the four-dimensional tensor structure to obtain Raman spectral data of the fermentation broth.

3. The method for detecting parameters in the fermentation process of recombinant human interferon according to claim 1, characterized in that, The process parameter prediction model structure includes, in sequence, an input layer, a convolutional neural network layer, a flattening layer, a fully connected layer, and an output layer, wherein: The input layer is used to receive input Raman spectral data of the fermentation broth; The convolutional neural network layer is used to perform one-dimensional convolution processing on the Raman spectrum data of the fermentation broth transmitted by the input layer to obtain the Raman spectrum feature map of the fermentation broth, wherein the Raman spectrum feature map of the fermentation broth includes a first-level spectral feature map of the pooled fermentation broth, a second-level spectral feature map of the pooled fermentation broth, and a third-level spectral feature map of the pooled fermentation broth. The flattening layer is used to flatten the Raman spectral feature map of the fermentation broth output by the convolutional neural network layer into a one-dimensional Raman spectral feature vector of the fermentation broth. The fully connected layer is used to perform nonlinear combination and mapping on the Raman spectral feature vector of the fermentation broth output by the flattening layer, so as to establish the prediction and inference relationship between the Raman spectral feature vector of the fermentation broth and the process parameters, and calculate the predicted value of the process parameters based on the prediction and inference relationship between the Raman spectral feature vector of the fermentation broth and the process parameters. The output layer is used to perform regression output on the predicted values ​​of the process parameters output by the fully connected layer. The output layer includes two nodes, which are used to output the predicted values ​​of cell density and glucose concentration, respectively.

4. The method for detecting parameters in the fermentation process of recombinant human interferon according to claim 3, characterized in that, The convolutional neural network layer includes a first convolutional layer, a first max-pooling layer, a second convolutional layer, a second max-pooling layer, a third convolutional layer, and a third max-pooling layer connected in sequence, wherein: The first convolutional layer includes 16 convolutional kernels, each with a size of 3. The 16 convolutional kernels are used to perform one-dimensional convolution processing on the fermentation broth Raman spectral data transmitted from the input layer to extract the primary spectral features of the fermentation broth corresponding to the Raman spectral data. Based on the primary spectral features of the fermentation broth, 16 primary spectral feature maps of the fermentation broth are formed. Each primary spectral feature map of the fermentation broth retains the local correlation features of the Raman spectral data of the fermentation broth. The first maximum pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 16 primary spectral feature maps of the fermentation broth output by the first convolutional layer, so as to select the maximum value in each sliding window to form 16 primary spectral feature maps of the pooled fermentation broth, and output the 16 primary spectral feature maps of the pooled fermentation broth to the flattening layer. The second convolutional layer includes 32 convolutional kernels, each with a size of 3. The 32 convolutional kernels are used to perform one-dimensional convolution processing on the 16 primary spectral feature maps of the pooled fermentation broth output by the first maximum pooling layer to form 32 secondary spectral feature maps of the fermentation broth. The second maximum pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 32 secondary spectral feature maps of the fermentation broth output by the second convolutional layer, so as to select the maximum value in each sliding window to form 32 pooled fermentation broth secondary spectral feature maps, and output the 32 pooled fermentation broth secondary spectral feature maps to the flattening layer. The third convolutional layer includes 64 convolutional kernels, each with a size of 3. The 64 convolutional kernels are used to perform one-dimensional convolution processing on the 32 secondary spectral feature maps of the pooled fermentation broth output by the second maximum pooling layer to form 64 tertiary spectral feature maps of the fermentation broth. The third max-pooling layer includes a pooling kernel of size 2. A sliding window is formed based on the pooling kernel to downsample the 64 tertiary spectral feature maps of the fermentation broth output by the third convolutional layer. The maximum value is selected in each sliding window to form 64 pooled fermentation broth tertiary spectral feature maps, and the 64 pooled fermentation broth tertiary spectral feature maps are output to the flattening layer.

5. The method for detecting parameters in the fermentation process of recombinant human interferon according to claim 1, characterized in that, Using the Raman spectral data of the fermentation broth as training data, the model is trained based on the process parameter prediction model structure to obtain the process parameter prediction model, including: A preset training sample ratio is obtained, the Raman spectral data of the fermentation broth is used as training data, and the training data is randomly divided into a training sample set and a test sample set according to the training sample ratio. Obtain a preset loss function, perform multiple rounds of forward propagation on the process parameter prediction model structure based on the training sample set to obtain the predicted values ​​of the process parameters of the training samples, and calculate the loss value of each round of forward propagation based on the loss function on the training sample set and the predicted values ​​of the process parameters of the training samples. Using the gradient descent iterative method, backpropagation is performed based on the loss value of each forward propagation round to update the model weight parameters of the process parameter prediction model structure. A preset maximum number of training rounds is obtained. Training is completed after the maximum number of training rounds is reached to obtain a preprocess parameter prediction model. The model prediction performance of the preprocess parameter prediction model is tested using the test sample set. After the preprocess parameter prediction model passes the model prediction performance test, the preprocess parameter prediction model is used as the process parameter prediction model.

6. The method for detecting parameters in the fermentation process of recombinant human interferon according to claim 1, characterized in that, Using a large-scale fermenter as the validation data set, the process parameter prediction model is used to process the validation data set. Through regression calculation of the process parameter prediction model, predicted values ​​of process parameters corresponding to the validation data set are output, including: At least one large-scale fermenter was injected with culture medium and inoculated with E. coli to obtain a validation data set. Spectral data were acquired from the validation data sets to obtain the pre-validation fermentation broth Raman spectra of the Escherichia coli fermentation process of recombinant human interferon in each validation data set. The Raman spectral data of the pre-validation fermentation broth were smoothed, standardized, and structured to obtain the Raman spectral data of the validation fermentation broth. The Raman spectral data of the validation fermentation broth are input into the process parameter prediction model. Regression calculations are performed through the process parameter prediction model to obtain the output of the process parameter prediction model. The output of the process parameter prediction model is then used as the predicted value of the process parameter corresponding to the validation data set.

7. The method for detecting parameters in the fermentation process of recombinant human interferon according to claim 1, characterized in that, Obtain preset standard process parameter values ​​for the validation group; compare the predicted process parameter values ​​with the standard process parameter values ​​for the validation group to obtain model error values; perform performance validation on the process parameter prediction model based on the model error values; and use the process parameter prediction model that passes the performance validation as the final application model to detect the fermentation process parameters of recombinant human interferon, including: Using the offline OD600 measurement method, multiple sets of offline process parameter values ​​are generated for the verification data set. The average value of each set of offline process parameter values ​​is calculated to obtain the calculation result, and the calculation result is used as the standard process parameter value of the verification set. A preset prediction error threshold is obtained, the predicted value of the process parameter is compared with the standard process parameter value of the verification group to obtain the model error value, and the performance of the process parameter prediction model is verified using the prediction error threshold and the model error value to obtain the verification result; If the model error value exceeds the prediction error threshold, the process parameter prediction model is considered to have failed the performance verification. The experimental data set is then reset, training data is obtained again, and the process parameter prediction model is trained again. The retrained process parameter prediction model is then subjected to performance verification. The verification result is considered to be that the process parameter prediction model has passed the performance verification. If the model error value does not exceed the prediction error threshold, the process parameter prediction model is considered to have passed performance verification. The process parameter prediction model that has passed performance verification is used as the final application model, and the final application model is deployed to the production control computing node to detect the parameters of the recombinant human interferon fermentation process.

8. A detection system for parameters in the fermentation process of recombinant human interferon, characterized in that, The method for detecting parameters in the fermentation process of recombinant human interferon as described in any one of claims 1 to 7, comprising: The training data acquisition unit is used to acquire the original fermentation broth Raman spectrum data of Escherichia coli fermentation process of recombinant human interferon from multiple experimental data sets, using a small-scale fermenter as the experimental data set, and to perform data preprocessing and physical constraint correction on the original fermentation broth Raman spectrum data to obtain fermentation broth Raman spectrum data. The prediction model training unit is used to acquire a preset process parameter prediction model structure, use the Raman spectrum data of the fermentation broth as training data, and train the model based on the process parameter prediction model structure to obtain the process parameter prediction model. The validation data calculation unit is used to process the validation data set using the process parameter prediction model with a large-scale fermenter as the validation data set. Through regression calculation of the process parameter prediction model, the unit outputs the process parameter prediction values ​​corresponding to the validation data set, wherein the process parameter prediction values ​​include cell density prediction values ​​and glucose concentration prediction values. The application model validation unit is used to obtain preset validation group standard process parameter values, compare the predicted process parameter values ​​with the validation group standard process parameter values ​​to obtain model error values, and perform performance validation on the process parameter prediction model based on the model error values, so as to use the process parameter prediction model that passes the performance validation as the final application model to detect the fermentation process parameters of recombinant human interferon.

9. An electronic device, characterized in that, The device includes a memory, a processor, and a transceiver that are sequentially and communicatively connected. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the method for detecting parameters of the recombinant human interferon fermentation process as described in any one of claims 1 to 7.

10. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or the instructions are executed by the computer, they implement the method for detecting parameters of the recombinant human interferon fermentation process as described in any one of claims 1 to 7.

Images