Training method of gel quality parameter prediction model and quality parameter prediction method

Abstract

The invention provides a training method of a gel quality parameter prediction model and a quality parameter prediction method, and the training method comprises the steps: obtaining a plurality of groups of first gel attribute information after preprocessing; each group of first gel attribute information comprises sample quality parameter information corresponding to sample gel and sample spectrum information under the corresponding sample quality parameter information; and taking the sample spectrum information of each group as the input of a preset model, and taking the corresponding sample quality parameter information as the output of the preset model, so as to train and obtain the gel quality parameter prediction model for predicting the target quality parameter information corresponding to any gel. The gel quality parameter prediction model trained through the method is high in fitting precision and excellent and stable in prediction capacity, the application requirement of near-infrared quantitative analysis is met, and rapid, lossless, stable and accurate detection of gel key quality attributes can be achieved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing, and in particular to a training method and a method for predicting gel quality parameters. Background Technology

[0002] As a common topical semi-solid dosage form, gel formulations have good user compliance and transdermal release properties. Their preparation process and quality control have a key impact on product stability and clinical efficacy.

[0003] Currently, the detection of gel viscosity in industrial production mainly relies on traditional rheological methods, the most typical of which is the rotational viscometer method. Although this method is widely adopted as an industry standard, it has a series of inherent drawbacks in actual production and quality control, severely hindering production efficiency and the advancement of intelligent manufacturing.

[0004] 1) Traditional methods are destructive testing methods. Each measurement requires a sufficient amount of sample to be taken from the production batch. The measurement process changes the physical state of the sample, making it impossible to return the sample to the production line, resulting in high costs and waste.

[0005] 2) The complete viscosity testing process includes multiple steps such as sampling, sample transportation, temperature control, instrument preparation, taking average values ​​from multiple measurements, and instrument cleaning. The analysis process is cumbersome and time-consuming, and the whole process usually takes a long time. It cannot provide real-time data. This "time lag" makes it impossible for production personnel to adjust process parameters in a timely manner based on the test results.

[0006] 3) High dependence on personnel skills: The accuracy of test results largely depends on the operator's proficiency. The sample loading method, the presence of air bubbles, the control of constant temperature time, the selection and cleaning status of the rotor, and the timing of reading can all introduce large human errors, resulting in poor reproducibility of results between different operators or different laboratories. Poor reproducibility is not conducive to accurate quality control and the establishment of standardization.

[0007] Therefore, there is an urgent need in this field to develop a rapid, non-destructive, online method for gel viscosity detection to overcome the limitations of traditional methods and improve production efficiency and product quality control. Summary of the Invention

[0008] The technical problem to be solved by this disclosure is to overcome the shortcomings of traditional gel viscosity detection methods, such as lag, cumbersome process, long time consumption, low efficiency and large error. It provides a training method and quality parameter prediction method for a gel quality parameter prediction model. The gel viscosity prediction model obtained by this method can quickly and accurately detect gel viscosity, thereby improving production efficiency and process control.

[0009] This disclosure solves the above-mentioned technical problems through the following technical solution:

[0010] According to a first aspect of this disclosure, a training method for a gel quality parameter prediction model is provided, the training method comprising:

[0011] Obtain several sets of first gel property information after preprocessing;

[0012] Wherein, each group of the first gel attribute information includes the sample quality parameter information corresponding to the sample gel and the sample spectral information under the corresponding sample quality parameter information;

[0013] The sample spectral information of each group is used as the input of a preset model, and the corresponding sample quality parameter information is used as the output of the preset model, so as to train a gel quality parameter prediction model for predicting the target quality parameter information corresponding to any gel.

[0014] Optionally, the step of obtaining several sets of preprocessed first gel property information includes:

[0015] Obtain sample quality information of several sample gels with different viscosities, as well as the initial spectral information of the sample gels under a preset light source;

[0016] Abnormal spectra are removed from the initial spectral information based on the first preset processing method to obtain the first spectral information;

[0017] The first spectral information is standardized based on the second preset processing method to obtain the sample spectral information.

[0018] Optionally, the first preset processing method includes at least one of the following: reverse interval partial least squares method, elimination of uninformed variables, Monte Carlo elimination of uninformed variables, and competitive adaptive reweighted sampling;

[0019] And / or,

[0020] The second preset processing method includes: mean centering and vector normalization;

[0021] And / or,

[0022] Before obtaining sample quality information for several sample gels of different viscosities, the training method further includes:

[0023] The sample gel was degassed.

[0024] Optionally, after the step of obtaining the target quality parameter information for predicting any gel, the training method further includes:

[0025] Based on the first gel attribute information, input the gel quality parameter prediction model to obtain the first predicted value of the gel quality parameter prediction model;

[0026] Based on the gel property information and the preset verification algorithm, the second predicted value of the gel quality parameter prediction model is obtained;

[0027] Based on the preset gel property information, the gel quality parameter prediction model is input to obtain the third predicted value of the gel quality parameter prediction model;

[0028] In response to the fact that the first predicted value, the second predicted value, and the third predicted value are all within a first preset range, it is determined that the gel quality parameter prediction model has passed the evaluation.

[0029] In response to the first predicted value, the second predicted value, or the third predicted value exceeding the first preset range, it is determined that the gel quality parameter prediction model has failed the evaluation, and several new sets of the first gel attribute information are reacquired to obtain a gel quality parameter prediction model that meets the prediction conditions based on the new first gel attribute information.

[0030] And / or,

[0031] The sample quality parameters include viscosity information.

[0032] According to a second aspect of this disclosure, a method for predicting gel quality parameters is provided, the prediction method comprising:

[0033] Obtain the actual spectral information of the gel to be tested;

[0034] The actual spectral information is input into the gel quality parameter prediction model trained by the training method of the gel quality parameter prediction model as described in the first aspect of this disclosure, to obtain the actual quality parameter information corresponding to the gel to be tested.

[0035] According to a third aspect of this disclosure, a training system for a gel quality parameter prediction model is provided, the training system comprising:

[0036] The sample data acquisition module is used to acquire several sets of first gel property information after preprocessing;

[0037] Wherein, each group of the first gel attribute information includes the sample quality parameter information corresponding to the sample gel and the sample spectral information under the corresponding sample quality parameter information;

[0038] The model training module is used to take the spectral information of the samples in each group as the input of the preset model and the corresponding sample quality parameter information as the output of the preset model, so as to train the gel quality parameter prediction model for predicting the target quality parameter information corresponding to any gel.

[0039] Optionally, the sample data acquisition module includes: a first acquisition unit, an abnormal spectrum removal unit, and a standardization unit;

[0040] The first acquisition unit is used to acquire sample quality information of several sample gels with different viscosities, as well as the initial spectral information of the sample gels under a preset light source.

[0041] The abnormal spectrum removal unit is used to remove abnormal spectra from the initial spectral information based on a first preset processing method to obtain the first spectral information;

[0042] The standardization unit standardizes the first spectral information based on a second preset processing method to obtain the sample spectral information.

[0043] Optionally, the first preset processing method includes at least one of the following: reverse interval partial least squares method, elimination of uninformed variables, Monte Carlo elimination of uninformed variables, and competitive adaptive reweighted sampling;

[0044] And / or,

[0045] The second preset processing method includes: mean centering and vector normalization;

[0046] And / or,

[0047] The sample data acquisition module further includes a defoaming unit, which is used to defoam the sample gel before acquiring sample quality information of several sample gels with different viscosities.

[0048] Optionally, the training system further includes a model evaluation module, which is used after obtaining the target quality parameter information for predicting any gel:

[0049] Based on the first gel attribute information, input the gel quality parameter prediction model to obtain the first predicted value of the gel quality parameter prediction model;

[0050] Based on the gel property information and the preset verification algorithm, the second predicted value of the gel quality parameter prediction model is obtained;

[0051] Based on the preset gel property information, the gel quality parameter prediction model is input to obtain the third predicted value of the gel quality parameter prediction model;

[0052] In response to the fact that the first predicted value, the second predicted value, and the third predicted value are all within a first preset range, it is determined that the gel quality parameter prediction model has passed the evaluation.

[0053] In response to the first predicted value, the second predicted value, or the third predicted value exceeding the first preset range, it is determined that the gel quality parameter prediction model has failed the evaluation, and several new sets of the first gel attribute information are reacquired to obtain a gel quality parameter prediction model that meets the prediction conditions based on the new first gel attribute information.

[0054] And / or,

[0055] The sample quality parameters include viscosity information.

[0056] According to a fourth aspect of this disclosure, a system for predicting gel quality parameters is provided, the system comprising:

[0057] The actual spectrum acquisition module is used to acquire the actual spectral information of the gel to be tested;

[0058] The actual information acquisition module is used to input the actual spectral information into the gel quality parameter prediction model trained by the training system of the gel quality parameter prediction model described in the third aspect of this disclosure, so as to obtain the actual quality parameter information corresponding to the gel to be tested.

[0059] According to a fifth aspect of this disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and for running on the processor, wherein the processor executes the computer program to implement a training method for a gel quality parameter prediction model according to a first aspect of this disclosure, and / or a method for predicting gel quality parameters according to a second aspect of this disclosure.

[0060] According to a sixth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a training method for a gel quality parameter prediction model according to a first aspect of this disclosure, and / or a method for predicting gel quality parameters according to a second aspect of this disclosure.

[0061] According to a seventh aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements a training method for a gel quality parameter prediction model according to a first aspect of this disclosure, and / or a method for predicting gel quality parameters according to a second aspect of this disclosure.

[0062] Based on common knowledge in the field, the above-mentioned preferred conditions can be combined arbitrarily to obtain various preferred embodiments of this disclosure.

[0063] The positive and progressive effects of this disclosure are as follows:

[0064] The training method for the gel quality parameter prediction model and the prediction method for gel quality parameters provided in this disclosure enable rapid, non-destructive, stable, and accurate detection of key quality attributes of gels through the trained gel quality parameter prediction model, providing a reliable technical means for real-time monitoring and quality control of the production process. Attached Figure Description

[0065] Figure 1 A flowchart illustrating the training method for the gel quality parameter prediction model provided in Embodiment 1 of this disclosure;

[0066] Figure 2 The near-infrared spectrum collected in Embodiment 1 of this disclosure;

[0067] Figure 3 This is a schematic diagram of the preprocessing process provided in Embodiment 1 of this disclosure;

[0068] Figure 4 A comprehensive comparison of the gel viscosity models processed by biPLS (reverse interval least squares) provided in Embodiment 1 of this disclosure;

[0069] Figure 5 A comprehensive comparison of the gel viscosity models treated with MCUVE (Monte Carlo - elimination of no-information variables) provided in Embodiment 1 of this disclosure;

[0070] Figure 6 A comprehensive comparison of the gel viscosity models processed by CARS (Competitive Adaptive Reweighted Sampling) provided in Embodiment 1 of this disclosure;

[0071] Figure 7 This document compares the model performance under different combinations of preprocessing methods provided in Embodiment 1 of this disclosure.

[0072] Figure 8 The optimal preprocessing method provided in Embodiment 1 of this disclosure is the model after preprocessing by mean centering combined with vector normalization;

[0073] Figure 9 This is a flowchart illustrating the evaluation model provided in Embodiment 1 of this disclosure;

[0074] Figure 10 This is a schematic diagram of the external validation results of the prediction set of the optimal gel viscosity prediction model provided in Embodiment 1 of this disclosure;

[0075] Figure 11 This is a schematic flowchart of the method for predicting gel quality parameters provided in Embodiment 2 of this disclosure;

[0076] Figure 12 This is a schematic diagram of the structure of the training system for the gel quality parameter prediction model provided in Embodiment 3 of this disclosure;

[0077] Figure 13 This is a schematic diagram of the structure of the gel quality parameter prediction system provided in Embodiment 4 of this disclosure;

[0078] Figure 14 This is a schematic diagram of the structure of the electronic device provided in Embodiment 5 of this disclosure. Detailed Implementation

[0079] The present disclosure is further illustrated below by way of embodiments, but the present disclosure is not limited to the scope of the embodiments described herein.

[0080] The prefixes such as "first" and "second" used in this disclosure are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes used to distinguish descriptive objects in this disclosure does not constitute a limitation on the described objects. The description of the described objects is given in the claims or the context of the embodiments, and should not be construed as an unnecessary limitation. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0081] Example 1

[0082] like Figure 1 As shown in the figure, this embodiment provides a training method for a gel quality parameter prediction model. The training method includes:

[0083] S11: Obtain several sets of first gel property information after preprocessing;

[0084] Among them, the first gel attribute information of each group includes the sample quality parameter information corresponding to the sample gel and the sample spectral information under the corresponding sample quality parameter information;

[0085] S12: Use the sample spectral information of each group as the input of the preset model and the corresponding sample quality parameter information as the output of the preset model to train a gel quality parameter prediction model for predicting the target quality parameter information of any gel.

[0086] In this embodiment, the sample quality parameter information is viscosity information.

[0087] In one embodiment, the viscosity value corresponding to the original gel formulation is used as a standard benchmark, and the percentage change of each other sample relative to the benchmark value is calculated. Based on the obtained primary data, distribution statistics are performed with a gradient of 10% each. A total of 100 gel samples with different viscosity values ​​were prepared and measured. The viscosity values ​​of these 100 samples show a large range relative to the viscosity benchmark of the original formulation, covering a distribution range of 40% to 360%. The overall sample covers gel systems with different consistency states from low to high, effectively supporting the diversity and stability of subsequent chemometric modeling work.

[0088] The gel viscosity data of the samples were obtained using existing technologies, such as a rotational rheometer. The measurement system was configured with a cone-plate geometry of 40.0 mm diameter, a cone angle of 1.00306°, and a gap of 0.029 mm. The test was conducted in isothermal mode at 25°C. The rheological properties of the gel at different shear rates were evaluated using the "flow ramp mode". The viscosity value at a shear rate of 80 s⁻¹ was used as the representative viscosity of the sample for subsequent near-infrared modeling analysis.

[0089] In this embodiment, the sample spectral information collected is near-infrared spectral information. The gel viscosity data of the sample gel is obtained; the gel viscosity data is matched one-to-one with the near-infrared spectrum; abnormal spectra are identified and eliminated; a calibration set is defined; the near-infrared spectrum of the calibration set is preprocessed; the preprocessed near-infrared spectrum is screened for characteristic wavelengths; and a gel viscosity prediction model is established using PLS (Partial Least Squares Regression) on the screened near-infrared spectral data.

[0090] The collected near-infrared spectra are as follows Figure 2 As shown in the figure, the horizontal axis of the spectrum represents wavelength in nanometers (nm), and the vertical axis represents the absorption intensity of the near-infrared spectrum in absorbance (Abs), indicating the absorption capacity of the gel sample for near-infrared light of different wavelengths. Several distinct absorption bands are visible in the spectrum, mainly including overtone and combination absorption characteristics of C–H and O–H groups. However, given that this spectral signal contains not only the compositional information of the gel matrix but also the absorption characteristics from background substances such as transparent quartz tubes, direct assignment and analysis of single characteristic peaks become complex and unrepresentative. Therefore, during the acquisition of the near-infrared spectrum of the sample, degassing treatment can be performed on the sample, for example, by centrifuging at 5000–6000 rpm for 5–10 s to quickly remove air bubbles from the gel, obtaining a homogeneous gel sample with sufficient bubble removal.

[0091] The phrase "corresponding one-to-one with the gel viscosity data and the near-infrared spectrum" means pairing the viscosity data and near-infrared spectral data of the same sample.

[0092] To ensure the accuracy and repeatability of near-infrared spectroscopy acquisition, standardized processing and loading of gel samples are necessary. After gel sample preparation, it is first transferred to a 50 mL centrifuge tube and sealed for storage. Since a large number of air bubbles are often present during gel preparation, these bubbles significantly interfere with the stability and penetration of the spectral signal. Therefore, the sample is degassed before spectral acquisition by rapidly removing air bubbles through centrifugation at 5000–6000 rpm for 5 s, resulting in a homogeneous gel sample with thorough bubble removal.

[0093] This disclosure requires the use of a dedicated transparent quartz test tube for near-infrared measurement as a container before spectral acquisition. Before each measurement, an empty test tube is placed in the sample chamber of the near-infrared instrument to acquire the background spectrum as a system reference. Then, a suitable amount of gel sample is slowly drawn using a 5mL disposable syringe. A burr-free needle is attached to the end of the syringe and inserted into the test tube. The gel is slowly injected until the test tube is full, ensuring that no new air bubbles are introduced during the process to avoid light scattering interference. After gel injection, the test tube is placed in the instrument's sample chamber, and near-infrared spectra are acquired according to the set parameters.

[0094] Since the detection area of ​​the probe in near-infrared transmission measurement is located in the lower part of the test tube, the occasional presence of air bubbles in the upper part of the test tube has no significant impact on the actual spectral acquisition. Furthermore, whether or not a cap is placed on the top of the test tube will not interfere with the contact process between the sample and the light beam, and therefore will not affect the spectral quality. During spectral acquisition, simply place the filled sample test tube vertically into the sample chamber of the near-infrared instrument. After the system stabilizes and completes the spectral scan (approximately twenty seconds), remove the test tube to obtain the near-infrared spectral data of the target sample.

[0095] This disclosure uses multivariate statistical modeling to uncover the potential correlation between spectral data and key quality properties (viscosity) of samples, in order to achieve rapid and non-destructive quantitative prediction.

[0096] The gel quality parameter prediction model trained using the training method provided in this disclosure enables rapid, non-destructive, stable, and accurate detection of key gel quality attributes, providing a reliable technical means for real-time monitoring and quality control of the production process.

[0097] like Figure 3 As shown, the steps for obtaining several sets of preprocessed first gel property information include:

[0098] S31: Obtain sample quality information of several sample gels with different viscosities, as well as the initial spectral information of the sample gels under the illumination of a preset light source;

[0099] S32: Based on the first preset processing method, abnormal spectra are removed from the initial spectral information to obtain the first spectral information;

[0100] S33: Standardize the first spectral information based on the second preset processing method to obtain sample spectral information.

[0101] The first preset processing method includes at least one of the following: reverse interval partial least squares method, elimination of uninformed variables, Monte Carlo elimination of uninformed variables, and competitive adaptive reweighted sampling.

[0102] This disclosure performs internal and external cross-validation on a pre-defined model to obtain the root mean square error of cross-validation (RMSECV), coefficient of determination (R²cv), and coefficient of determination (R²pred) of the prediction set, and records the number of latent variables (LVs). The model is evaluated based on its parameters, and the optimal feature wavelength selection method is chosen. The unit of the root mean square error is the unit of the predicted quantity, and the unit for the viscosity prediction model is cp (viscosity units).

[0103] The inverse interval partial least squares (biPLS) method divides the entire band into several equally wide intervals and gradually eliminates bands with small contributions to the prediction during the modeling process, thereby retaining feature intervals highly correlated with the target variable and reducing redundant information and noise interference. In this study, the original number of variables in the entire band was 101. According to the modeling requirements, different combinations of 5 to 10 segments were set, and the model was constructed and its performance evaluated under each segment number condition. Finally, the model with the best performance under each segment condition was selected, providing a concise and information-rich set of feature wavelengths for subsequent modeling.

[0104] The model parameters after band selection using the reverse interval partial least squares method are shown in Table 1 below.

[0105] Table 1

[0106]

[0107] The combined metrics of the 6 models (RMSECV + RMSEP) are as follows: Figure 4As shown, the performance of the gel viscosity model is significantly improved after biPLS band selection. According to the comparison of model performance parameters, the gel viscosity model exhibits the best predictive performance when the number of segments is 10. Under this segmentation condition, the sum of model errors reaches its minimum, indicating that the retention of characteristic wavelengths and the removal of redundant information achieve the best balance, thus enabling the model to have high stability and prediction accuracy on both the training and validation sets.

[0108] Uninformation variable elimination (UVE) refers to the model parameters after band filtering using UVE, as shown in Table 2 below.

[0109] Table 2

[0110]

[0111] The Monte Carlo-Uninformation Variable Elimination (MCUVE) method was used to screen spectral bands. To systematically evaluate the impact of different parameter settings on model performance, the number of samplings was set from 100 to 500, with a step size of 100, and the sampling was gradually increased within this range for modeling observation. The sampling ratio was fixed at 70%, meaning that 70% of the samples were randomly selected in each iteration to build the model, and the remaining 30% of the samples were used for validation.

[0112] The number of sampling iterations was chosen to balance computational cost and result stability: fewer sampling iterations may lead to larger fluctuations in wavelength importance assessment, while more sampling iterations, although improving stability, will significantly increase computational cost. Therefore, an incremental range of 100–500 was selected for consideration. A sampling ratio of 70% was chosen to ensure that the training set has sufficient sample size to support robust modeling, while retaining enough validation samples to accurately assess the model's generalization performance, thereby reducing the risk of overfitting and improving the reliability of variable importance assessment.

[0113] The model parameters after band filtering using MCUVE are shown in Table 3 below.

[0114] Table 3

[0115]

[0116] The combined metrics of the 5 models (RMSECV + RMSEP) are as follows: Figure 5 As shown in the figure. The results show that the performance of all samples improved after processing compared to the unprocessed samples. The parameters of groups 3 and 4 of the viscosity model are completely identical, indicating that the model performance tends to be stable and without significant differences under specific combinations of sampling times and sampling ratios. Therefore, only one representative model is retained for analysis in the same parameter group. Comparing the results of each group, the viscosity model performs best when the number of sampling times is 400 and the sampling ratio is 70%.

[0117] Table 4 shows the characteristic wavelength selection results of the gel viscosity prediction model under the optimal MCUVE treatment. It can be seen that the model parameters obtained by MCUVE treatment are better than those of the UVE method. However, in terms of viscosity model performance, the MCUVE method still lags behind biPLS.

[0118] Table 4

[0119]

[0120] Competitive Adaptive Reweighted Sampling (CARS) uses CARS to screen wavelengths. To systematically evaluate the impact of different parameters on modeling performance, the number of iterations is set to 100 to 200 with a step size of 25, and the number is gradually increased within this range for modeling observation. The sampling ratio is fixed at 70%. The range of iterations aims to strike a balance between computational efficiency and the stability of variable selection: too few iterations may result in important wavelengths not being adequately screened, while too many iterations, although improving stability, will significantly increase the computational cost. Therefore, the range of 100–200 is selected with a step size of 25 to obtain stable and repeatable variable selection results at a reasonable computational cost. The sampling ratio is set to 70% to ensure that the training set has a sufficient sample size to support robust modeling, while retaining enough validation samples to reliably evaluate the model's generalization ability, thereby reducing the risk of overfitting and improving the effectiveness of wavelength selection.

[0121] The model parameters after using CARS band filtering are shown in Table 5 below.

[0122] Table 5

[0123]

[0124] The combined metrics of the 5 models (RMSECV + RMSEP) are as follows: Figure 6 As shown, the results indicate that the error of the gel viscosity model remained unchanged within the iteration range of 100–200, demonstrating high stability. Therefore, under the same parameter combination, only one representative model needs to be retained for subsequent analysis. Based on the comprehensive comparison results, the optimal parameters for the gel viscosity model are 200 iterations and a sampling ratio of 70%.

[0125] Table 6 shows the characteristic wavelength selection results of the optimal CARS-treated gel viscosity prediction model. After CARS characteristic wavelength selection, the prediction performance of the gel viscosity model is actually lower than that of other wavelength selection methods.

[0126] The possible reason for this phenomenon is that the CARS screening mechanism tends to converge quickly to a smaller number of wavelength variables through adaptive weighting and stepwise elimination strategies, which has a significant advantage in datasets with low noise and concentrated feature signals. However, the near-infrared spectral data of this invention may be affected by the complex matrix of the gel system, solvent evaporation, and slight batch-to-batch differences, causing some wavelengths that are weakly correlated with the target variable but contain important information to be mistakenly deleted in early iterations, thus resulting in insufficient overall information content of the model. In addition, CARS relies on random sampling during iteration, which may introduce additional fluctuations in multivariate weak signal scenarios, further weakening the stability and generalization ability of the model.

[0127] Table 6

[0128]

[0129] In summary, the predictive performance of the gel viscosity model was improved after the characteristic wavelength screening. The model parameters obtained by the four characteristic wavelength screening methods were compared and shown in Table 7 below, and the optimal characteristic wavelength screening method was determined.

[0130] Table 7

[0131]

[0132] For near-infrared quantitative prediction models of gel viscosity, the modeling effects of four commonly used characteristic wavelength screening methods (biPLS, UVE, MCUVE and CARS) were compared above. The model performance was systematically evaluated by indicators such as root mean square error of cross-validation (RMSECV), root mean square error of prediction set (RMSEP), coefficient of determination (R²cv, R²p) and residual prediction bias (RPD).

[0133] In the gel viscosity model, the biPLS method achieved the best modeling results. Its cross-validation coefficient of determination (R²cv) was 0.9234, the prediction set coefficient of determination (R²p) reached 0.9821, the RMSEP was 99.13, and the RPD value was as high as 8.05, fully demonstrating its high prediction accuracy and good stability. In contrast, the models built by the other three methods performed significantly worse on the prediction set, especially the CARS method, which, although showing good cross-validation performance, had a large prediction set error and poor model stability.

[0134] In summary, the biPLS characteristic wavelength screening method can effectively screen out key wavelength variables related to gel viscosity, significantly improving the predictive performance and stability of the model, and is suitable as the preferred method for near-infrared spectroscopy quantitative analysis in this disclosure.

[0135] The second preset processing method includes: mean centering combined with vector normalization;

[0136] This disclosure utilizes mean centering (MC), vector normalization (VN), subtracting a straight line (SSL), eliminating constant offset (ECO), first derivative (FD), derivative, standard normal variable transformation (SNV), detrending (DT), multiplicative scattering correction (MSC), and min-max normalization (MMN) to optimize the model after processing the spectrum. A model is built using PLS, and internal and external cross-validation are performed to obtain the root mean square error of cross-validation (RMSECV) and the coefficient of determination (R²cv), and the number of latent variables (LVs) is recorded. The model is evaluated based on its parameters, and the optimal preprocessing method is selected.

[0137] Mean centering transforms absorbance data at each wavelength into a zero-centered distribution, i.e., subtracting the mean of all samples from the value of each wavelength variable. While this process does not alter the correlation structure between variables, it effectively eliminates systematic biases among spectral variables, thus highlighting relative differences between samples and enhancing the expressive power of multivariate statistical modeling methods such as PCA and Partial Least Squares (PLS). Therefore, mean centering is often used as a fundamental standardization technique. Introducing other preprocessing methods (such as SNV and MSC) on top of this allows for a more reasonable evaluation of the impact of different preprocessing strategies on model performance. This not only improves the fairness of comparisons but also enhances the stability and consistency of model training.

[0138] Based on the above considerations, this invention uniformly adopts mean centering as the basic preprocessing step in all modeling processes, and further systematically examines its combined effect with other preprocessing methods. By constructing a PLS regression model, leave-one-out cross-validation is used to determine the optimal number of latent variables (LVs), thereby screening out the optimal spectral preprocessing path suitable for gel systems.

[0139] The results of the screening of preprocessing methods are shown in Table 8 below.

[0140] Table 8

[0141]

[0142] In this embodiment, 15 models were selected from different combinations of preprocessing methods for gel viscosity modeling, and their key evaluation parameters were compared to screen for the best model. During the screening process, the root mean square error of cross-validation (RMSECV) and the root mean square error of prediction set (RMSEP) of each model were added together as a comprehensive index; the smaller the total error, the better the overall model performance. When the total errors of different models were close, further judgment was made based on practical application requirements. Since the models are mainly used to predict unknown samples, when the total errors are similar, the model with the lower RMSEP was prioritized as the optimal solution. Based on the above principles, the preprocessing model with balanced performance indicators and optimal predictive ability was finally determined. The comprehensive index of the 15 models is as follows: Figure 7 As shown.

[0143] Figure 7 This paper presents a comparison of model performance under different combinations of preprocessing methods. Based on the model parameter evaluation results, gel viscosity prediction model 9 performed best, and its optimal preprocessing method was MC+VN. In terms of preprocessing strategy, the gel viscosity prediction model first uses MC (mean centering) to eliminate background interference and improve data comparability. Then, VN is introduced on top of MC to further enhance the spectral characteristics' responsiveness to physical performance indicators, improving the model's sensitivity and discriminative power to viscosity changes. The gel viscosity prediction model after MC+VN preprocessing is shown below. Figure 8 As shown. Figure 8 A schematic diagram of the optimal viscosity prediction model established for the calibration set data. The horizontal axis represents the reference value of viscosity, and the vertical axis represents the viscosity prediction value obtained by the model.

[0144] The predicted accuracy of the gel viscosity model obtained after preprocessing has significant room for improvement, and further processing is carried out by combining the characteristic wavelength screening method.

[0145] In step S42, the methods for identifying and removing abnormal spectra are PCA (Principal Component Analysis) and diagnostic plots. More specifically, the methods for identifying and removing abnormal spectra introduce F-residuals and Hotelling's T² statistic (a core hypothesis testing statistic in multivariate statistical analysis) as anomaly identification indicators. The α-residual of the F-test is controlled within 0.05 to determine the critical value of the F-residual statistical test, and the confidence interval of Hotelling's T² is set to 95%. Spectral data exceeding the critical value of the F-residual statistical test and the set threshold of Hotelling's T² are judged as abnormal spectra and removed. If no abnormal spectra are identified, no removal is required.

[0146] To assess the overall distribution of the spectral data of the gel samples and identify any outliers, principal component analysis (PCA) was performed on the near-infrared spectral data of 100 gel samples. The analysis revealed that most samples clustered well in the principal component space, exhibiting a relatively consistent data structure, indicating a high degree of similarity in spectral characteristics among the samples.

[0147] However, two sample points were observed in the PCA score plot to deviate significantly from the main population distribution, exhibiting large principal component scores, which are typical spectral outliers. Upon review, it was found that these two samples may have experienced abnormal operations or external interference during sample preparation or spectral scanning, resulting in significant differences in their spectral data compared to the other samples.

[0148] To ensure the representativeness of the data and the quality of the modeling, the two sample data were removed according to the outlier discrimination criteria, thereby improving the robustness and prediction accuracy of the model and avoiding interference from extreme values ​​in the modeling results.

[0149] After removing abnormal spectral information, in order to select a calibration set by using the Kennard-Stone algorithm, in a specific implementation, samples with large spectral differences can be selected into the calibration set to ensure the representativeness of the calibration samples; for example, the Kennard-Stone algorithm is used to divide the samples (including spectra and corresponding gel viscosity data) into a calibration set and a prediction set. The calibration set is used to train and build the model, and the prediction set is used for external verification when evaluating the model.

[0150] like Figure 9 As shown, after obtaining the target quality parameter information for predicting any gel, the training method further includes:

[0151] S91: Input the gel quality parameter prediction model based on the first gel property information, and obtain the first predicted value of the gel quality parameter prediction model;

[0152] S92: Based on gel property information and a preset verification algorithm, obtain the second predicted value of the gel quality parameter prediction model;

[0153] S93: Input the gel quality parameter prediction model based on the preset gel property information, and obtain the third predicted value of the gel quality parameter prediction model;

[0154] S94: In response to the fact that the first predicted value, the second predicted value and the third predicted value are all within the first preset range, the gel quality parameter prediction model is determined to have passed the evaluation.

[0155] S95: In response to the first predicted value, the second predicted value, or the third predicted value exceeding the first preset range, it is determined that the gel quality parameter prediction model has failed the evaluation, and several new sets of first gel attribute information are re-acquired to obtain a gel quality parameter prediction model that meets the prediction conditions based on the new first gel attribute information.

[0156] In one implementation, the model can be evaluated by calculating model parameters, and the calibration set coefficient of determination and root mean square error of the calibration set can be calculated using calibration set data. Internal cross-validation methods (such as leave-one-out cross-validation or K-fold cross-validation) are used for internal verification to calculate the cross-validation coefficient of determination and root mean square error of the cross-validation set. External verification is performed using the prediction set to calculate the prediction set coefficient of determination and root mean square error of the prediction set, as follows:

[0157] Root Mean Squared Error of Calibration (RMSEC):

[0158]

[0159] Root Mean Squared Error of Prediction (RMSEP):

[0160]

[0161] Prediction Residual Error Sum of Squares (PRESS) and Root Mean Squared Error of Cross Validation (RMSECV):

[0162]

[0163]

[0164] Coefficient of Determination (R²):

[0165]

[0166] Based on the input data, the coefficients of determination can be categorized into cross-validation coefficient of determination R²cv, calibration set coefficient of determination R²c, and prediction set coefficient of determination R²p. The smaller the root mean square error, the closer the coefficient of determination is to 1, indicating better predictive performance of the model.

[0167] Ratio of Performance to Deviation (RPD):

[0168]

[0169] SD (Standard Deviation) represents the standard deviation of the true values ​​of the sample, reflecting the inherent variability of the data. SEP (Standard Error of Prediction) represents the standard error of the predicted values, measuring the accuracy of the model's predictions. RPD is a parameter used to evaluate the predictive ability of the model; the higher the value, the stronger the predictive ability of the model. When RPD > 3, the model can be applied quantitatively; while when the ratio is between 2 and 3, the model has the potential to establish a predictive equation, but still needs improvement.

[0170] The model constructed in the example has a cross-validation R²cv of 0.9234, and the external validation results of the prediction set are as follows. Figure 10 , Figure 10 The figure shows the external validation results of the near-infrared spectroscopy prediction model for gel viscosity. The ten points in the figure represent ten data points from the test set. The horizontal axis represents the reference value of sample viscosity obtained by rheometer measurement, and the vertical axis represents the predicted value obtained by the best prediction model. As shown in the figure, the prediction set R²p reaches 0.9821, RMSEP is 99.13, and RPD value is as high as 8.05. The model meets the requirements (RPD>3), indicating that the model was successfully established and fully demonstrating its high prediction accuracy and good stability.

[0171] The implementation principle of the training method for the gel quality parameter prediction model in this disclosure is explained below with specific embodiments:

[0172] (1) The primary data were mapped one-to-one with the spectra, and abnormal spectra were identified by combining PCA and diagnostic chart methods. To this end, F residuals and Hotelling's T² statistic were introduced as anomaly identification indicators. The confidence interval of Hotelling's T² was set at 95%, and the α residual of the F test was controlled within 0.05. Spectral data exceeding the set thresholds of F residuals and Hotelling's T² were judged as abnormal spectra and were removed.

[0173] (2) After identifying and removing abnormal spectra, the dataset was divided using the Kennard-Stone algorithm. The Kennard-Stone algorithm is a uniform design method. Its core idea is to select samples with large spectral differences into the calibration set, thereby ensuring the representativeness of the calibration samples. The Kennard-Stone algorithm was used to divide the samples (including spectra and corresponding primary data) into 90 calibration sets and 10 prediction sets. The calibration sets were used to train and build the model, and the prediction sets were used for external validation during model evaluation. The number of prediction set samples was ensured to be 20 or more to ensure the reliability of external validation.

[0174] (3) A calibration set was used to train and build the model. After the dataset was partitioned, spectral preprocessing was performed. The effects of using Mean Centering (MC), Vector Normalization (VN), Subtract Straight Line (SSL), Eliminate Constant Offset (ECO), First Derivative (FD), Derivative, Standard Normal Variate Transformation (SNV), Detrending (DT), Multiplicative Scatter Correction (MSC), and Min-Max Normalization (MMN) to process the spectrum on the model were examined. The model was built using PLS, and internal cross-validation and external validation were performed to obtain the root mean square error of cross-validation (RMSECV) and the coefficient of determination (R²cv). The number of latent variables (LVs) was also recorded. The model was evaluated based on the model parameters, and the best preprocessing method was selected. During the screening process, the root mean square error of cross-validation (RMSECV) and the root mean square error of prediction set (RMSEP) of each model were summed as a comprehensive index; the smaller the total error, the better the overall model performance. When the total errors of different models were close, further judgment was made based on the actual application requirements. Since the models are mainly used to predict unknown samples, the model with the lower RMSEP was selected as the optimal solution when the total errors were similar. Spectral preprocessing using mean centering (MC) and vector normalization (VN) was preferred.

[0175] (4) After preprocessing, feature wavelengths are selected. Based on the best preprocessing method, four selection methods are examined: inverse interval partial least squares (biPLS), uninformed variable elimination (UVE), Monte Carlo-uninformed variable elimination (MCUVE), and competitive adaptive reweighted sampling (CARS). A PLS model is established, and internal cross-validation and external validation are performed on the model to obtain the root mean square error of cross-validation (RMSECV), the coefficient of determination of cross-validation (R2cv), and the coefficient of determination of the prediction set (R2pred). The number of latent variables (LVs) of the model is also recorded. The model is evaluated based on the model parameters, and the root mean square error of cross-validation, the root mean square error of the prediction set, the coefficient of determination, and the residual prediction bias are used as evaluation indicators to select the best feature wavelength selection method. The unit of the root mean square error is the unit of the predicted quantity, and the unit of the viscosity prediction model is cp. The inverse interval partial least squares (biPLS) method is selected as the best method for feature wavelength selection.

[0176] (5) After processing the data using the optimal preprocessing method and characteristic wavelength screening method, a model is established using PLS to obtain a gel quality parameter prediction model.

[0177] Example 2

[0178] like Figure 11 As shown, this embodiment provides a method for predicting gel quality parameters. The prediction method includes:

[0179] S111: Obtain the actual spectral information of the gel to be tested;

[0180] S112: Input the actual spectral information into the gel quality parameter prediction model trained by the gel quality parameter prediction model training method to obtain the actual quality parameter information corresponding to the gel to be tested.

[0181] The training method for the gel quality parameter prediction model and the prediction method for gel quality parameters provided in this disclosure enable rapid, non-destructive, stable, and accurate detection of key quality attributes of gels through the trained gel quality parameter prediction model, providing a reliable technical means for real-time monitoring and quality control of the production process.

[0182] Example 3

[0183] like Figure 12 As shown, this embodiment provides a training system for a gel quality parameter prediction model. The training system includes:

[0184] The sample data acquisition module 101 is used to acquire several sets of preprocessed first gel property information;

[0185] Among them, the first gel attribute information of each group includes the sample quality parameter information corresponding to the sample gel and the sample spectral information under the corresponding sample quality parameter information;

[0186] The model training module 102 is used to take the sample spectral information of each group as the input of the preset model and the corresponding sample quality parameter information as the output of the preset model, so as to train a gel quality parameter prediction model for predicting the target quality parameter information corresponding to any gel.

[0187] The sample data acquisition module 101 includes: a first acquisition unit 1011, an abnormal spectrum removal unit 1012, and a standardization unit 1013;

[0188] The first acquisition unit 1011 is used to acquire sample quality information of several sample gels with different viscosities, as well as the initial spectral information of the sample gels under the illumination of a preset light source.

[0189] The abnormal spectrum removal unit 1012 is used to remove abnormal spectra from the initial spectral information based on a first preset processing method to obtain the first spectral information;

[0190] The standardization unit 1013 standardizes the first spectral information based on the second preset processing method to obtain the sample spectral information.

[0191] The first preset processing method includes at least one of the following: reverse interval partial least squares method, elimination of uninformed variables, Monte Carlo elimination of uninformed variables, and competitive adaptive reweighted sampling;

[0192] The second preset processing method includes: mean centering and vector normalization;

[0193] The sample data acquisition module 101 also includes a degassing processing unit 1014, which is used to degas the sample gel before acquiring sample quality information of sample gels with different viscosities.

[0194] The training system also includes a model evaluation module 103, which is used to evaluate the target quality parameters for predicting any gel after obtaining the relevant parameters.

[0195] Based on the first gel property information, input the gel quality parameter prediction model to obtain the first predicted value of the gel quality parameter prediction model;

[0196] Based on gel property information and a preset verification algorithm, the second predicted value of the gel quality parameter prediction model is obtained.

[0197] Based on the preset gel property information, input the gel quality parameter prediction model to obtain the third predicted value of the gel quality parameter prediction model;

[0198] Since the first predicted value, the second predicted value, and the third predicted value are all within the first preset range, the gel quality parameter prediction model is determined to have passed the evaluation.

[0199] In response to the first predicted value, the second predicted value, or the third predicted value exceeding the first preset range, it is determined that the gel quality parameter prediction model has failed the evaluation, and several new sets of first gel attribute information are reacquired to obtain a gel quality parameter prediction model that meets the prediction conditions based on the new first gel attribute information.

[0200] Sample quality parameters include viscosity information.

[0201] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.

[0202] This disclosure enables rapid, non-destructive, stable, and accurate detection of key quality attributes of gels through a trained gel quality parameter prediction model, providing a reliable technical means for real-time monitoring and quality control of the production process.

[0203] Example 4

[0204] like Figure 13 As shown, a system for predicting gel quality parameters is provided. The prediction system includes:

[0205] The actual spectrum acquisition module 201 is used to acquire the actual spectral information of the gel to be tested.

[0206] The actual information acquisition module 202 is used to input the actual spectral information into the gel quality parameter prediction model trained by the training system of the gel quality parameter prediction model of the third aspect of this disclosure, so as to obtain the actual quality parameter information corresponding to the gel to be tested.

[0207] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs.

[0208] This disclosure enables rapid, non-destructive, stable, and accurate detection of key quality attributes of gels through a trained gel quality parameter prediction model, providing a reliable technical means for real-time monitoring and quality control of the production process.

[0209] Example 5

[0210] like Figure 14 As shown, Figure 14 This is a schematic diagram of the corresponding electronic device provided in Embodiment 5 of this disclosure. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the methods described in the above embodiments. Figure 14 The electronic device 30 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0211] like Figure 14 As shown, the electronic device 30 can be represented in the form of a general computing device, such as a server device. The components of the electronic device 30 may include, but are not limited to: at least one processor 31, at least one memory 32, and a bus 33 connecting different system components (including memory 32 and processor 31).

[0212] Bus 33 includes a data bus, an address bus, and a control bus.

[0213] The memory 32 may include volatile memory, such as random access memory (RAM) 321 and / or cache memory 322, and may further include read-only memory (ROM) 323.

[0214] The memory 32 may also include a program / utility 325 having a set (at least one) of program modules 324, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0215] The processor 31 performs various functional applications and data processing, such as the methods described in the above embodiments of this disclosure, by running computer programs stored in the memory 32.

[0216] Electronic device 30 can also communicate with one or more external devices 34 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 35. Furthermore, the model-generating device 30 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 36. Figure 14As shown, network adapter 36 communicates with other modules of the model-generated device 30 via bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated device 30, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0217] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0218] Example 6

[0219] This disclosure also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the training method for the gel quality parameter prediction model and the prediction method for the gel quality parameters provided in any of the above embodiments.

[0220] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0221] Example 7

[0222] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the training method for the gel quality parameter prediction model and the prediction method for the gel quality parameters described in any of the above embodiments.

[0223] The program code for executing the computer program product of this disclosure can be written in any combination of one or more programming languages, and the program code can be executed entirely on a user device, partially on a user device, as a stand-alone software package, partially on a user device and partially on a remote device, or entirely on a remote device.

[0224] While specific embodiments of this disclosure have been described above, those skilled in the art should understand that these are merely illustrative examples, and the scope of protection of this disclosure is defined by the appended claims. Those skilled in the art can make various changes or modifications to these embodiments without departing from the principles and essence of this disclosure, but all such changes and modifications fall within the scope of protection of this disclosure.

Claims

1. A training method for a gel quality parameter prediction model, characterized in that, The training method includes: Obtain several sets of first gel property information after preprocessing; Wherein, each group of the first gel attribute information includes the sample quality parameter information corresponding to the sample gel and the sample spectral information under the corresponding sample quality parameter information; The sample spectral information of each group is used as the input of a preset model, and the corresponding sample quality parameter information is used as the output of the preset model, so as to train a gel quality parameter prediction model for predicting the target quality parameter information corresponding to any gel.

2. The training method for the gel quality parameter prediction model according to claim 1, characterized in that, The step of obtaining several sets of preprocessed first gel property information includes: Obtain sample quality information of several sample gels with different viscosities, as well as the initial spectral information of the sample gels under a preset light source; Abnormal spectra are removed from the initial spectral information based on the first preset processing method to obtain the first spectral information; The first spectral information is standardized based on the second preset processing method to obtain the sample spectral information.

3. The training method for the gel quality parameter prediction model according to claim 2, characterized in that, The first preset processing method includes at least one of the following: reverse interval partial least squares method, elimination of uninformed variables, Monte Carlo elimination of uninformed variables, and competitive adaptive reweighted sampling; And / or, The second preset processing method includes: mean centering and vector normalization; And / or, Before obtaining sample quality information for several sample gels of different viscosities, the training method further includes: The sample gel was degassed.

4. The training method for the gel quality parameter prediction model according to any one of claims 1-3, characterized in that, After the step of obtaining the target quality parameter information for predicting any gel, the training method further includes: Based on the first gel attribute information, input the gel quality parameter prediction model to obtain the first predicted value of the gel quality parameter prediction model; Based on the gel property information and the preset verification algorithm, the second predicted value of the gel quality parameter prediction model is obtained; Based on the preset gel property information, the gel quality parameter prediction model is input to obtain the third predicted value of the gel quality parameter prediction model; In response to the fact that the first predicted value, the second predicted value, and the third predicted value are all within a first preset range, it is determined that the gel quality parameter prediction model has passed the evaluation. In response to the first predicted value, the second predicted value, or the third predicted value exceeding the first preset range, it is determined that the gel quality parameter prediction model has failed the evaluation, and several new sets of the first gel attribute information are reacquired to obtain a gel quality parameter prediction model that meets the prediction conditions based on the new first gel attribute information. And / or, The sample quality parameters include viscosity information.

5. A method for predicting gel quality parameters, characterized in that, The prediction method includes: Obtain the actual spectral information of the gel to be tested; The actual spectral information is input into the gel quality parameter prediction model trained by the training method of the gel quality parameter prediction model as described in any one of claims 1-4 to obtain the actual quality parameter information corresponding to the gel to be tested.

6. A training system for a gel quality parameter prediction model, characterized in that, The training system includes: The sample data acquisition module is used to acquire several sets of first gel property information after preprocessing; Wherein, each group of the first gel attribute information includes the sample quality parameter information corresponding to the sample gel and the sample spectral information under the corresponding sample quality parameter information; The model training module is used to take the spectral information of the samples in each group as the input of the preset model and the corresponding sample quality parameter information as the output of the preset model, so as to train the gel quality parameter prediction model for predicting the target quality parameter information corresponding to any gel.

7. A system for predicting gel quality parameters, characterized in that, The prediction system includes: The actual spectrum acquisition module is used to acquire the actual spectral information of the gel to be tested; The actual information acquisition module is used to input the actual spectral information into the gel quality parameter prediction model trained by the training system of the gel quality parameter prediction model according to claim 6, so as to obtain the actual quality parameter information corresponding to the gel to be tested.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and for running on the processor, characterized in that, When the processor executes the computer program, it implements the training method of the gel quality parameter prediction model according to any one of claims 1 to 4, and / or the gel quality parameter prediction method according to claim 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the training method of the gel quality parameter prediction model according to any one of claims 1 to 4, and / or the gel quality parameter prediction method according to claim 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the training method of the gel quality parameter prediction model as described in any one of claims 1 to 4, and / or the gel quality parameter prediction method as described in claim 5.

Images