High-fiber coarse cereal extruded powder eGI prediction model based on machine learning, construction and optimization method and application

Abstract

The invention discloses a high-fiber coarse cereal extruded powder eGI prediction model based on machine learning, a construction and optimization method and application, and relates to the field of food processing intelligence. The method solves the problems that a prediction model established by an existing response surface method RSM is limited in precision, and R2 is usually lt; 0.75, and high-dimensional data cannot be processed. The method comprises the following steps: acquiring raw material characteristics, proportioning structure characteristics and extrusion process parameters, and integrating raw material chemical compositions, proportioning structure characteristics and extrusion process parameters; and constructing a high-dimensional feature space by adopting the integrated chemical composition of the raw materials, the matching structure features and the extrusion process parameters, and performing fitting analysis by adopting a neural network prediction model to obtain the eGI prediction model of the high-fiber coarse cereal extruded powder. And optimizing the eGI prediction model of the high-fiber coarse cereal extruded flour by adopting a grid search algorithm and a Bayesian optimization method to finish construction and optimization of the eGI prediction model of the high-fiber coarse cereal extruded flour. The method is also suitable for the food industry digital application field.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of food processing intelligentization, and particularly relates to a high-fiber coarse cereal extruded powder eGI prediction model based on machine learning and a construction and optimization method and application thereof. BACKGROUND

[0002] In recent years, the prevalence of diabetes and obesity has shown a continuous upward trend worldwide. Under this background, low glycemic index (GI) food, as an important strategy for the prevention and control of chronic diseases, is attracting widespread attention from the academic and industrial communities. Studies have shown that long-term intake of low GI food can significantly improve blood glucose control and reduce the risk of type 2 diabetes by 20-30%.

[0003] At the same time, coarse cereals are considered an ideal raw material for developing low GI food due to their high content of dietary fiber, resistant starch and other functional ingredients. Extrusion technology, as an efficient physical modification method, can change the structure of starch molecules through high temperature and high shear force, significantly reducing its digestion rate. Studies have shown that an optimized extrusion process can reduce the eGI value of coarse cereal products by 15-30%.

[0004] The key parameters of the extrusion process that affect the eGI value include: temperature (80-160℃); moisture content (12-20%) affecting starch gelatinization degree and Maillard reaction degree; screw speed (200-400rpm) determining material rheological properties and heat transfer efficiency; and shear force and residence time.

[0005] However, traditional extrusion process optimization faces major challenges. First, there is a complex nonlinear relationship between raw material ratios and process parameters, making it difficult to fully evaluate multi-factor interactions using single-factor or orthogonal test methods. Second, the prediction model established by traditional response surface method (RSM) has limited accuracy (R 2 Generally <0.75), and cannot handle high-dimensional data. More importantly, existing researches mainly focus on a single raw material or fixed process conditions, lacking a systematic formulation design method.

[0006] However, the low GI food on the current market still faces many challenges. Traditional low GI food mainly relies on the substitution of refined carbohydrates, such as using artificial sweeteners or fiber additives, which often have problems such as poor taste, high cost, and low consumer acceptance. Therefore, developing low GI food based on natural raw materials has become an important direction of current research.

[0007] Machine learning techniques are increasingly applied in food science. In the prediction of GI values, machine learning models have been shown to outperform traditional methods. For example, random forest algorithms have been shown to improve the accuracy of GI value prediction in cereal products by 25% compared to multiple linear regression. Deep learning models have successfully predicted in vitro digestion characteristics by analyzing the molecular structure of starch. Integrated learning methods have shown better generalization ability when considering the interaction effects of raw materials. However, current research still has obvious limitations. Most models only consider static formulations and ignore the effects of processing parameters, or only use laboratory-scale data and lack industrial validation. In addition, the lack of model interpretability also limits its application in practice. SUMMARY

[0008] The present application aims to solve the problem of the limited accuracy of the prediction model established by the existing response surface method (RSM). The RSM is a statistical method that uses a mathematical model to describe the relationship between the response variable and the independent variables. It is widely used in the design of experiments and the optimization of processes. However, the RSM has some limitations, such as the need for a large number of experiments, the difficulty in handling high-dimensional data, and the lack of interpretability. Therefore, there is a need for a new method to establish a prediction model for high-fiber mixed grain extruded powder eGI. 2 Generally <0.75, and cannot handle high-dimensional data. More importantly, current research focuses on single raw materials or fixed process conditions, and lacks a systematic formulation design method.

[0009] To solve the above technical problems, the present application is achieved by the following technical solutions: Scheme one, the present application proposes a method for constructing a high-fiber mixed grain extruded powder eGI prediction model based on machine learning, which comprises the following steps: S1, obtain the characteristics of raw materials, the structure characteristics of the ratio, and the extrusion process parameters, and integrate the chemical composition of raw materials, the structure characteristics of the ratio, and the extrusion process parameters; S2, construct a high-dimensional feature space using the integrated chemical composition of raw materials, the structure characteristics of the ratio, and the extrusion process parameters in S1, and perform fitting analysis using a neural network prediction model to obtain a high-fiber mixed grain extruded powder eGI prediction model.

[0010] Further, a preferred embodiment is provided, in which the near-infrared rapid detection method is used to obtain the characteristics of starch, cellulose, and protein in S1.

[0011] Further, a preferred embodiment is provided, in which the D-optimal design method is used to set up a three-factor five-level ratio test scheme to obtain the structure characteristics of the ratio in S1, the three factors including material ratio, processing time, and processing temperature.

[0012] Further, a preferred embodiment is provided, in which the method for integrating the chemical composition of raw materials, the structure characteristics of the ratio, and the extrusion process parameters is an eGI experimental measurement method, which comprises the following steps: Step 1, perform in vitro-in vivo simulated digestion on the raw materials to obtain the calibration glucose release amount and the reference glucose release amount, respectively; Step 2, detecting the standard glucose release amount and the reference glucose release amount obtained in step 1, and drawing a standard glucose release amount-time curve and a reference glucose release amount-time curve, respectively; Step 3, calculating the iAUC of the standard glucose release amount-time curve and the reference glucose release amount-time curve, respectively, and calculating the eGI; eGI = (sample iAUC white bread iAUC) x 100 eGI = (white bread iAUC sample iAUC) x 100.

[0013] Further, a preferred embodiment is provided, and the method for fitting analysis in S2 adopts a neural network prediction model, and the method is as follows: An eGI prediction model is established, random forest tree algorithm is used to process high-dimensional nonlinear data, and gradient boosting tree algorithm is used to optimize the model.

[0014] Scheme two, an eGI prediction model of high-fiber mixed grain extruded powder obtained by the construction method in any one of scheme one, the eGI prediction model of high-fiber mixed grain extruded powder includes: dietary fiber content, extrusion temperature, amylose proportion and screw rotation speed.

[0015] Scheme three, an optimization method of the eGI prediction model of high-fiber mixed grain extruded powder based on machine learning, the eGI prediction model of high-fiber mixed grain extruded powder is constructed based on the construction method of the eGI prediction model of high-fiber mixed grain extruded powder based on machine learning in any one of scheme one, or the eGI prediction model is used to optimize the eGI model of high-fiber mixed grain extruded powder, the optimization method uses a grid search algorithm and a Bayesian optimization method to optimize the eGI prediction model of high-fiber mixed grain extruded powder, and the construction and optimization of the eGI prediction model of high-fiber mixed grain extruded powder are completed.

[0016] Scheme four, the construction method in any one of scheme one or the eGI prediction model of high-fiber mixed grain extruded powder in scheme two or the optimization method of the eGI prediction model of high-fiber mixed grain extruded powder in scheme three is applied to the eGI prediction of high-fiber mixed grain extruded powder.

[0017] Scheme four, a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the method in scheme one.

[0018] Scheme five, a computer device, including a memory and a processor, the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method in scheme one.

[0019] The present application has the advantages that: The high-fiber mixed grain extruded powder eGI prediction model based on machine learning, and the construction and optimization method and application, aims to construct the first eGI prediction model integrating raw material characteristics, ratio and extrusion process parameters, and has important theoretical value and application prospect. Mainly includes: multi-source data fusion innovation: integrating raw material chemical composition (near-infrared rapid detection), ratio structure characteristics (computer vision analysis), process parameters (industrial Internet of Things real-time collection), constructing high-dimensional feature space The high-fiber mixed grain extruded powder eGI prediction model based on machine learning, and the construction and optimization method and application, aims to construct the first eGI prediction model integrating raw material characteristics, ratio and extrusion process parameters, and has important theoretical value and application prospect. Mainly includes: multi-source data fusion innovation: integrating raw material chemical composition (near-infrared rapid detection), ratio structure characteristics (computer vision analysis), process parameters (industrial Internet of Things real-time collection), constructing high-dimensional feature space

[0020] The method disclosed by the application discloses a multi-element interaction mechanism of starch-fiber-protein in the extrusion process, which breaks through the limitation of the traditional trial and error method. The practical value is that the development cycle of low GI food is shortened from several months to several days, the research and development cost is reduced by more than 60%, and an intelligent solution is provided for the food industry.

[0021] The application is also applicable to the field of digital application of food industry. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The eGI value distribution box plot described in the eleventh embodiment; Figure 2 The process parameter and eGI correlation thermodynamic diagram described in the eleventh embodiment; Figure 3 The flowchart of the optimization method of the high-fiber mixed grain extruded powder eGI prediction model based on machine learning described in the eleventh embodiment; Figure 4 The starch-fiber reaction mechanism diagram described in the eleventh embodiment. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all the embodiments of the present application.

[0024] Embodiment one, the embodiment proposes a construction method of a high-fiber mixed grain extruded powder eGI prediction model based on machine learning, which comprises the following steps: S1, obtain the characteristics of raw materials, the structure characteristics of the ratio, and the extrusion process parameters, and integrate the chemical composition of raw materials, the structure characteristics of the ratio, and the extrusion process parameters; S2, construct a high-dimensional feature space using the integrated chemical composition of raw materials, the structure characteristics of the ratio, and the extrusion process parameters in S1, and perform fitting analysis using a neural network prediction model to obtain an eGI prediction model for high-fiber coarse grain extruded powder.

[0025] Embodiment II, the embodiment is a further limitation of the construction method of the eGI prediction model for high-fiber coarse grain extruded powder based on machine learning according to Embodiment I, and the near-infrared rapid detection method is used to obtain the characteristics of starch, cellulose, and protein in S1.

[0026] Embodiment III, the embodiment is a further limitation of the construction method of the eGI prediction model for high-fiber coarse grain extruded powder based on machine learning according to Embodiment I, and the structure characteristics of the ratio are obtained by setting a three-factor five-level ratio test scheme using a D-optimal design method in S1. The three factors include material ratio, processing time, and processing temperature.

[0027] Embodiment IV, the embodiment is a further limitation of the construction method of the eGI prediction model for high-fiber coarse grain extruded powder based on machine learning according to Embodiment I, and the method for integrating the chemical composition of raw materials, the structure characteristics of the ratio, and the extrusion process parameters is an eGI experimental measurement method, which includes the following steps: Step 1, in vitro-in vivo simulation digestion is performed on the raw materials to obtain the calibration glucose release amount and the reference glucose release amount, respectively; Step 2, the calibration glucose release amount and the reference glucose release amount obtained in Step 1 are detected, and the calibration glucose release amount-time curve and the reference glucose release amount-time curve are drawn, respectively; Step 3, the iAUC of the calibration glucose release amount-time curve and the reference glucose release amount-time curve is calculated, respectively, and the eGI is calculated; eGI=(sample iAUC white bread iAUC) x 100 eGI=(white bread iAUC sample iAUC) x 100.

[0028] Embodiment V, the embodiment is a further limitation of the construction method of the eGI prediction model for high-fiber coarse grain extruded powder based on machine learning according to Embodiment I, and the method for fitting analysis using a neural network prediction model in S2 is: An eGI prediction model is established, and a random forest tree algorithm is used to process high-dimensional nonlinear data, and a gradient boosting tree algorithm is used to optimize the model.

[0029] Embodiment six, the embodiment proposes an eGI prediction model of high-fiber mixed grain extruded powder obtained by the construction method of any one of embodiments one to five, the eGI prediction model of high-fiber mixed grain extruded powder includes: dietary fiber content, extrusion temperature, amylose proportion and screw speed.

[0030] Embodiment seven, the embodiment proposes an optimization method of an eGI prediction model of high-fiber mixed grain extruded powder based on machine learning, the eGI prediction model of high-fiber mixed grain extruded powder is constructed based on the construction method of the eGI prediction model of high-fiber mixed grain extruded powder based on machine learning of any one of embodiments one to five or the eGI prediction model in embodiment six, the optimization method optimizes the eGI model of high-fiber mixed grain extruded powder, the optimization method optimizes the eGI prediction model of high-fiber mixed grain extruded powder by using grid search algorithm and Bayesian optimization method, and the construction and optimization of the eGI prediction model of high-fiber mixed grain extruded powder are completed.

[0031] Embodiment eight, the embodiment proposes an application of the construction method of any one of embodiments one to five or the optimization method of the eGI prediction model of high-fiber mixed grain extruded powder in the eGI prediction of high-fiber mixed grain extruded powder.

[0032] Embodiment nine, the embodiment proposes a computer readable storage medium, the computer readable storage medium stores a computer program, and the computer program is executed by a processor to realize the steps of the method of any one of embodiments one to five.

[0033] Embodiment ten, the embodiment proposes a computer device, including a memory and a processor, the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method of any one of embodiments one to five.

[0034] Embodiment eleven, the embodiment proposes an example for explaining the above-mentioned embodiments one to ten, and the example is specifically: Reference Figures 1 to 4 To verify the above-mentioned embodiments, the embodiment mainly includes the following contents: I. Data collection and experimental design (quantitative research) 3.1.1 Raw material selection: representative and diversity control Raw material selection criteria: Five kinds of high-fiber mixed grains (oat, quinoa, barley, black bean and bitter buckwheat) are selected, covering three categories of grains, beans and pseudo-grains.

[0035] The dietary fiber content is required to be ≥8% (determined according to GB 5009.88-2014); Raw material pretreatment: uniformly ground through an 80-mesh sieve, and the moisture content is adjusted to 12±0.5% (controlled in a constant temperature and humidity box).

[0036] Ratio experiment design: Mixing design: D-optimal design (Design-Expert 12.0 software) is used, three factors (material ratio, processing time, and processing temperature) and five levels are set, and a total of 36 ratio schemes are generated.

[0037] Ratio range: any combination of A, B, C, D, E, and F makes the total one; Quality control: near-infrared spectroscopy (NIRS) is used to rapidly detect the composition of raw materials (total starch, amylose, and dietary fiber), and the RSD is ensured to be less than 3%.

[0038] Through mixing design, multi-dimensional variable space is covered, and the diversity of raw materials (starch type and fiber content difference) is ensured, avoiding experimental blind area.

[0039] 3.1.2 Extrusion process parameters: range determination and experimental scheme Parameter range determination: Temperature: 100-160℃ (6 gradients, reference Maillard reaction threshold 120℃ and starch gelatinization temperature interval).

[0040] Moisture content: 12-20% (every 2% as a gradient, calculated according to the glass transition temperature of the material).

[0041] Pressure: 10-20 MPa; Time: 50-150 s; Experimental scheme: Response surface design: Box-Behnken design is used to generate 210 process combinations (including 3 repetitions).

[0042] Process monitoring: melt pressure (0-10 MPa), specific mechanical energy (SME, kWh / kg), and torque fluctuation coefficient (CV<5%) are collected in real time.

[0043] Data collection: Process parameters: temperature, pressure, and speed data are recorded every 10 seconds through the PLC system.

[0044] Structural properties: micro-CT scanning (SkyScan 1272) is used to quantify porosity and pore size distribution (ImageJ analysis).

[0045] Model system construction: Two types of prediction models are constructed to adapt to different scenarios: 1. Generalized machine learning model: As described in embodiment five, random forest and other algorithms are used to fit for multiple high-fiber coarse grains (oats, quinoa, barley, etc.).

[0046] 2. High-precision quadratic response surface model: For specific, well-defined raw material systems (for example, a formula composed of six basic raw materials A, B, C, D, E, and F), experimental design (DOE) and least squares fitting are used to obtain a highly accurate mathematical model. The following is a specific embodiment of this model: Design space: Extrusion time (t): 50-150 seconds; Extrusion pressure (p): 10-20 MPa; Control temperature (T): 100-140°C; Food ratio: The ratio of the six foods (A, B, C, D, E, F) is generated by a Dirichlet distribution (all parameters are 1) to generate uniformly distributed points, and the sum of the ratios is 1; Experiment and fitting: Based on 210 experimental data points, the following quadratic response surface model formula is obtained by least squares fitting: GI = 80.64 - 25a - 15b - 5c - 20d - 10e - 0.05t - 0.1p - 0.05T +0.0001tp + 0.0001tT + 0.0001pT - 0.0001t 2 - 0.0001p 2 - 0.0001T 2 Where a, b, c, d, e represent the proportions of foods A, B, C, D, E, respectively, and the proportion of food F is f = 1 - a - b - c - d - e.

[0047] Model performance: The model goodness-of-fit (R 2 ) is 0.992, with extremely high prediction accuracy within the design space. The model coefficients are adjusted to ensure that the predicted value of the validation point is consistent with the actual value (for example: a=0.2, b=0.2, c=0.6, d=0, e=0, t=100s, p=15MPa, T=120°C, predicted GI=56.21).

[0048] 3.1.3 eGI determination: in vitro-in vivo correlation verification In vitro digestion process (modified ISO 20128:2022 method): Three-stage simulated digestion: Oral stage: a-amylase (150 U / mL, pH 6.9, 2 min); Stomach stage: pepsin (2000 U / mL, pH 2.0, 30 min); Small intestine stage: pancreatin (100 U / mL, pH 7.0, 180 min); Glucose assay: sample every 5 min, glucose oxidase method (GOPOD kit) to determine the release amount.

[0049] eGI calculation: Area under curve (iAUC): trapezoidal method to calculate the cumulative glucose release amount from 0 to 180 min.

[0050] Standardization formula: eGI = (sample iAUC / white bread iAUC) x 100.

[0051] In vitro-in vivo verification: Human test: 10 groups of samples were selected for clinical testing (n = 12 healthy subjects, ethical approval number: 2023-ETH-045); Correlation analysis: Pearson correlation coefficient r = 0.89 (p < 0.01), verifying the reliability of the in vitro method.

[0052] 3.2 Feature engineering and dataset construction 3.2.1 Input variables: feature selection strategy includes the following steps: Feature pool construction (4 categories, 31 original features): Raw material features: total starch, amylose, dietary fiber, protein (Kjeldahl nitrogen method); Process features: maximum temperature, average time, pressure; Interaction terms: fiber x temperature, starch x temperature pressure.

[0053] Feature selection: Random forest importance ranking: eliminate variables with importance <5% (such as raw material ash content); Collinearity processing: variance inflation factor (VIF) test, eliminate variables with VIF > 10 (such as linear combination of temperature and SME); Final feature set: 17 key variables are retained, covering raw materials, process, and structure.

[0054] 3.2.2 Data preprocessing: high-dimensional data optimization Missing value processing: Continuous variables: multiple imputation method (m = 5 imputations, MICE algorithm); Categorical variables: mode filling (such as equipment type label).

[0055] Data Standardization: Z-score Standardization: Suitable for continuous variables such as temperature and rotation speed. One-hot Encoding: Handle categorical variables (e.g., types of coarse grains).

[0056] Outlier Detection: Isolation Forest Algorithm: Set the anomaly score threshold to 0.65, and remove 3% outliers.

[0057] 3.2.3 Dataset Division: Overfitting Prevention Stratified Sampling: Divide the training set (70%), validation set (15%), and test set (15%) according to the eGI value distribution (45-85). Data Augmentation: SMOTE algorithm oversamples minority class samples (eGI<50) to balance the class distribution. Cross-validation: k=5-fold cross-validation to ensure model stability.

[0058] 3.3 Machine Learning Model Construction 3.3.1 Algorithm Selection: Data Adaptability Analysis Candidate Algorithms: Random Forest (RF): Handle high-dimensional nonlinear data.

[0059] Gradient Boosting Tree (XGBoost): Optimize prediction accuracy.

[0060] Support Vector Regression (SVR): Suitable for small sample high-dimensional space.

[0061] BP Neural Network: Capture complex nonlinear relationships.

[0062] Selection Basis: Data Volume: Medium-scale (n=210) → Prefer tree models (RF / XGBoost).

[0063] Feature Type: Mixed (continuous + categorical) → Use one-hot encoding compatible algorithms.

[0064] 3.3.2 Hyperparameter Optimization: Efficiency and Accuracy Balance Optimization Strategy: Grid Search: Coarse adjustment of key parameters (e.g., RF's n_estimators: 100-500).

[0065] Bayesian Optimization: Fine-tune interactive parameters (learning rate, maximum depth) for 50 iterations.

[0066] Computational acceleration: GPU parallel computing (NVIDIA A100), time-consuming from 18h to 2h.

[0067] 3.3.3 SHAP value analysis: Calculate feature contribution, visualize interaction effects (e.g. synergistic GI-lowering effect of temperature and fiber).

[0068] Local interpretable model (LIME): Model diagnosis, guidance of formulation and process parameter range setting and generation of explanation rules for specific sample models (e.g. eGI outliers) under offline conditions.

[0069] Sensitivity analysis: Perturb key parameters (±10% fiber content), observe eGI trend.

[0070] 5-fold cross-validation: Calculate average R 2 and RMSE.

[0071] Learning curve analysis: Ensure training / validation error convergence.

[0072] External validation: Independent dataset: Collect data from a production line of a certain enterprise (n=72), test model generalization ability.

[0073] Comparison with traditional methods: Multiple linear regression (MLR), partial least squares (PLS).

[0074] Robustness test: Noise injection: Add 5% Gaussian noise, evaluate model interference resistance.

[0075] Extreme condition test: Input process parameter boundary value (e.g. temperature 160℃ + moisture 12%).

[0076] 4. Data analysis and results 4.1 Data feature description eGI value distribution characteristics: The eGI values of 210 samples are skewed distribution (range 45.2-84.7, mean 62.3±8.5), among which: Low GI group (<55): 24.5%; Medium GI group (55-70): 58.3%; High GI group (>70): 17.2%.

[0077] Key parameter correlation:

[0078] Conclusion: eGI value is synergistically regulated by dietary fiber and extrusion temperature, which jointly explain 56.8% of the eGI variation (Adjusted R 2 = 0.568).

[0079] 4.2 Model performance comparison 4.2.1 Prediction accuracy comparison Model selection: compare RF, XGBoost, SVR, BPNN four algorithms (training set n=324); Evaluation index:

[0080] Error source: RF error is concentrated in the high fiber and low temperature combination (predicted value is 2.5-3.0 higher than the measured value), BPNN error increases sharply in the extreme process (temperature > 150℃) (RMSE reaches 6.8), that is, RF is optimal for modeling non-linear relationships due to its integrated learning characteristics, and XGBoost has the highest computing efficiency.

[0081] 4.2.2 Key influencing factor analysis Dietary fiber (SHAP mean = -0.32): > 15% when increased by 1%, eGI decreased by 1.8; Extrusion temperature (SHAP mean = -0.28): eGI decreased by 2.3 when increased by 10℃ in the interval of 80-120℃; Nonlinear effect: SHAP value slows down when temperature ≥ 140℃ (Maillard reaction saturation); Fiber and temperature interaction term (SHAP = 0.21): the GI-lowering effect of fiber is increased by 40% at high temperature; Mechanism explanation: high temperature promotes the formation of fiber-starch complexes (electron microscopy Figure 4 ), and inhibits the exposure of enzymatic sites.

[0082] 4.3 Sensitivity analysis Error rate distribution:

[0083] Low fiber (<8%) + high speed (>380rpm) combination: error rate increased to 15.7% (due to excessive shear damage to fiber structure); High amylose (>35%) + low temperature (<90℃): error is due to incomplete gelatinization (DSC verification); That is, the model is robust in the "high fiber and high temperature" interval (MAE <2.0), but needs to be used cautiously at the process boundary.

[0084] 4.4 Comparison advantage with traditional method Comparison of machine learning model and traditional statistical method: Control group: Response surface methodology (RSM) quadratic polynomial model; Test set: Independent process data set (n=72);

[0085] Nonlinear capture ability: RSM model R 2 Only 0.31, while RF accurately identifies the threshold through decision tree splitting (such as fiber > 15% splitting node); 5. Verification of the chemical mechanism of synergistic effect: Fiber wraps starch granules: High temperature extrusion (> 120℃) promotes the formation of complexes between dietary fiber (such as β-glucan) and amylose, which wrap starch granules through hydrogen bonding and hydrophobic interaction, reducing the contact area of α-amylase. Electron microscopy observation shows that when the fiber content is > 15%, the coverage of the complex increases by 40%, and the in vitro digestion rate decreases by 32%.

[0086] Maillard products inhibit enzyme activity: Maillard reaction in the high temperature section (140-160℃) generates products such as melanoidins, which bind to the active site of α-amylase through competitive binding (molecular docking simulation shows binding energy up to -8.2 kcal / mol), inhibiting starch hydrolysis. This effect is particularly pronounced in oat-black bean formulations (eGI reduction up to 26.3%)12.

[0087] Universal boundary conditions: Raw material adaptability: The model performs stably in high-fiber grains such as quinoa and barley (error rate < 5%), but has higher prediction error for starch types sensitive raw materials (such as glutinous rice) (MAE=4.8), as the proportion of amylopectin affects the wrapping efficiency4.

[0088] Process window: Synergistic effect is strongest at temperatures 120-140℃ and moisture 14-18%; temperatures > 160℃ result in caramelization leading to nutrient loss and weakening of the GI-lowering effect2.

[0089] That is, it first quantifies the "process-component-structure" ternary coupling mechanism for eGI regulation, providing molecular dynamics basis for low GI food design.

[0090] 5.1 Actual application scenario suggestions 5.1.1 Food industry: model integration and production optimization Implementation path: System integration architecture: API Interface Development: Encapsulate the RF model into a RESTful API, embed it into the enterprise MES (Manufacturing Execution System), receive raw material composition (NIRS detection data) and process parameters (PLC transmission) in real time, and output eGI prediction values ​​and process optimization suggestions.

[0091] Edge computing deployment: Deploy a lightweight model (TensorFlow Lite) on the extrusion device with a response latency of <0.5 seconds to meet real-time control requirements.

[0092] The specific implementation method is as follows: Meal replacement powder production line: A company shortened the formula development cycle from 6 months to 3 weeks by using a model, and the product eGI value was stably controlled below 55, increasing the customer repurchase rate by 25%.

[0093] 5.1.2 Personalized Nutrition: Dynamic Formula Optimization Metabolic feature input: Combine CGM (continuous glucose monitoring) data with user BMI and insulin sensitivity index to construct a personalized eGI correction coefficient (formula: eGI_personal = eGI_base × (0.8 + 0.2×HOMA-IR)4).

[0094] Multi-task learning framework: Simultaneously predicting eGI values ​​and postprandial blood glucose curves (R 2 =0.86), generating customized formulas for diabetic patients 7.

[0095] 3D printed nutrition bars: Based on user metabolic data, the ratio of inulin (slow-release fiber) and fast-digestible starch in the printing matrix is ​​dynamically adjusted to achieve gradient control of GI value (45-65 range) 5.

[0096] 5.1.3 Extending to complex whole-grain systems Knowledge graph construction: Integrate whole grain component databases (such as phenolic acid-starch interaction relationships) to establish a "component-structure-digestibility" knowledge graph, supporting multi-grain ratio reasoning 79.

[0097] Transfer learning application: Based on the existing model, the network layers were fine-tuned using small sample data (n=50) to adapt to new combinations such as quinoa-chickpea (prediction error <3.5%).

[0098] 5.1.4 Online Detection and Real-time Optimization NIRS-PLSR combined: Develop an online near-infrared spectroscopy detection module to analyze the moisture and fiber content of extruded powder in real time through partial least squares regression (PLSR) and adjust process parameters accordingly.

[0099] Digital twin system: Construct a digital twin of the extrusion process to simulate the rheological properties and eGI response under different ratios, reducing trial and error costs by 10.

[0100] 5.1.4 Lightweight Tool Development Mobile application: Develop a lightweight APP that integrates TensorFlow.js, supporting farmers or small and medium-sized manufacturers to input local grain ingredients and output optimized formulas (installation package <15MB) 6.

[0101] Blockchain traceability: Combining IoT sensors with blockchain, data from raw material planting to extrusion is recorded throughout the entire process, providing a trusted blockchain for low-GI food certification.

[0102] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. This is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0103] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for constructing an eGI prediction model for high-fiber extruded grain flour based on machine learning, characterized in that, The method includes the following steps: S1. Obtain the raw material characteristics, proportioning structure features and extrusion process parameters, and integrate the raw material chemical composition, proportioning structure features and extrusion process parameters; S2. A high-dimensional feature space is constructed using the raw material chemical composition, proportioning structure characteristics, and extrusion process parameters integrated in S1. A neural network prediction model is then used for fitting analysis to obtain the eGI prediction model for high-fiber extruded grain powder.

2. The method for constructing an eGI prediction model for high-fiber extruded grain flour based on machine learning as described in claim 1, characterized in that, In S1, the raw material characteristics of starch, cellulose, and protein are obtained using near-infrared rapid detection.

3. The method for constructing an eGI prediction model for high-fiber extruded grain flour based on machine learning as described in claim 1, characterized in that, The proportioning structure characteristics in S1 are obtained by setting up a three-factor, five-level proportioning test scheme using the D-optimal design method. The three factors include material proportion, processing time, and processing temperature.

4. The method for constructing an eGI prediction model for high-fiber extruded grain flour based on machine learning as described in claim 1, characterized in that, The method for integrating the chemical composition, proportioning characteristics, and extrusion process parameters of raw materials is the eGI experimental measurement method, which includes the following steps: Step 1: Perform in vitro-in vivo simulated digestion on the raw materials to obtain the calibrated glucose release and the reference glucose release. quantity; Step 2: Detect the calibration glucose release and reference glucose release obtained in Step 1, and plot the calibration glucose release and reference glucose release-time curves respectively; Step 3: Calculate the iAUC of the standard glucose release and the reference glucose release-time curves, and calculate the eGI; eGI = (sample iAUC, white bread iAUC) × 100 5. The method for constructing an eGI prediction model for high-fiber extruded grain flour based on machine learning as described in claim 1, characterized in that, The method used in S2 for fitting analysis using a neural network prediction model is as follows: An eGI prediction model was established, and the random forest tree algorithm was used to process high-dimensional nonlinear data, while the gradient boosting tree algorithm was used to optimize the model.

6. A high-fiber whole grain extruded flour eGI prediction model obtained by the construction method according to any one of claims 1-5, characterized in that, The eGI prediction model for high-fiber extruded grain powder includes: dietary fiber content, extrusion temperature, amylose ratio, and screw speed.

7. An optimization method for an eGI prediction model of high-fiber extruded grain flour based on machine learning, characterized in that, The high-fiber whole grain extruded powder eGI prediction model is constructed based on the machine learning-based eGI prediction model construction method of any one of claims 1-5, or the eGI prediction model of claim 6. The high-fiber whole grain extruded powder eGI model is then optimized. The optimization method uses a grid search algorithm and a Bayesian optimization method to optimize the high-fiber whole grain extruded powder eGI prediction model, thereby completing the construction and optimization of the high-fiber whole grain extruded powder eGI prediction model.

8. The application of the construction method according to any one of claims 1-5, or the eGI prediction model of high-fiber extruded grain powder according to claim 6, or the optimization method of the eGI prediction model of high-fiber extruded grain powder according to claim 7, in the eGI prediction of high-fiber extruded grain powder.

9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-5.

10. A computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method of any one of claims 1-5.

Images