Formulated nutritional food preparation control method, apparatus, device, and storage medium

Abstract

The application provides a formula nutritional food preparation control method, device, equipment and storage medium, and relates to the technical field of formula nutritional food preparation control. The method comprises the following steps: acquiring real-time particle size data in the formula nutritional food preparation process, inputting a trained osmotic pressure fitting model, and obtaining a fitted osmotic pressure; wherein the osmotic pressure fitting is trained based on sample characteristic particle size data and sample osmotic pressure corresponding to the formula nutritional food to be prepared; determining a risk intervention control scheme according to the fitted osmotic pressure and a preset early warning condition; wherein the process parameters in the risk intervention control scheme include one or more of prebiotic adjustment amount, lactose adjustment amount and homogenization pressure; and adjusting the process parameters according to the risk intervention control scheme. The application can predict the intestinal stress reaction of the product during the production process to realize real-time osmotic pressure reduction, avoid the intestinal stress risk of the formula nutritional food caused by improper osmotic pressure, and improve the accuracy of formula nutritional food preparation and product safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of formula nutrition food preparation control, and particularly relates to a formula nutrition food preparation control method, device, equipment and storage medium. BACKGROUND

[0002] During the period of infancy, the intestinal function has not yet developed perfectly, and is sensitive to the osmotic pressure of exogenous food. Intake of high-osmotic-pressure food is easy to cause intestinal stress reaction, accelerate intestinal peristalsis, and cause clinical symptoms such as diarrhea and vomiting. The degree of adaptation of food to the intestine can be reflected to a certain extent through indexes such as the rate of promoting intestinal peristalsis after eating, and the rate is negatively correlated with intestinal comfort.

[0003] In the research and development and production of infant formula nutrition food, osmotic pressure control is a key factor affecting the intestinal adaptability of the product. The existing technical system still has several obvious bottlenecks, which can be summarized as the following five aspects: first, the electrolyte composition is mostly used in a fixed ratio, which cannot be dynamically optimized according to the intestinal development state of infants of different months, and lacks individual adaptation ability; second, the application ratio of deep hydrolyzed protein has not been reasonably balanced, which not only reduces the sensitivity, but also significantly increases the osmotic pressure due to the increase of short peptide chains, and there is a technical contradiction between “sensitivity reduction” and “osmotic pressure control”; third, the addition of prebiotics is generally low, which cannot reach the concentration threshold of effectively promoting intestinal health, and the function is not fully realized; fourth, the current evaluation of intestinal discomfort caused by milk powder mostly relies on consumer feedback after the event, which is lagging behind in recognition, and not only affects the user experience, but also damages the brand trust.

[0004] At present, there is a lack of a scheme for predicting intestinal stress reaction of a product in a production process to realize real-time osmotic pressure reduction. SUMMARY

[0005] Embodiments of the present application provide a formula nutrition food preparation control method, device, equipment and storage medium to solve the problem of predicting intestinal stress reaction of a product in a production process to realize real-time osmotic pressure reduction.

[0006] In a first aspect, embodiments of the present application provide a formula nutrition food preparation control method, comprising:

[0007] obtaining real-time particle size data in a formula nutrition food preparation process;

[0008] inputting the real-time particle size data into a trained osmotic pressure fitting model to obtain a fitted osmotic pressure; wherein the osmotic pressure fitting model is trained based on sample characteristic particle size data and sample osmotic pressure corresponding to the formula nutrition food to be prepared;

[0009] determine a risk intervention control scheme according to the fitted osmotic pressure and preset early warning conditions, wherein the process parameters in the risk intervention control scheme include one or more of prebiotic adjustment amount, lactose adjustment amount and homogenization pressure;

[0010] adjust the process parameters according to the risk intervention control scheme.

[0011] In a possible implementation, the preset early warning conditions include preset low-risk early warning conditions, medium-risk early warning conditions and high-risk early warning conditions.

[0012] The risk intervention control scheme includes a medium-risk risk intervention control scheme and a high-risk risk intervention control scheme.

[0013] The process parameters corresponding to the medium-risk risk intervention control scheme include prebiotic adjustment amount and lactose adjustment amount.

[0014] The process parameters corresponding to the high-risk risk intervention control scheme include homogenization pressure.

[0015] In a possible implementation, the determining of the risk intervention control scheme according to the fitted osmotic pressure and preset early warning conditions includes:

[0016] When the fitted osmotic pressure meets the medium-risk early warning condition, a first difference between the fitted osmotic pressure and the upper limit of the low-risk early warning condition is calculated, and prebiotic addition amount and lactose reduction amount are determined according to the first difference; wherein the prebiotic addition amount and the lactose reduction amount are the same.

[0017] When the fitted osmotic pressure meets the high-risk early warning condition, a second difference between the fitted osmotic pressure and the upper limit of the low-risk early warning condition is calculated, and a homogenization pressure downshift value is determined according to the second difference.

[0018] In a possible implementation, before the acquiring of the real-time particle size data in the preparation process of the formula nutritional food, the method further includes:

[0019] acquiring a plurality of formula samples of the formula nutritional food to be prepared, and acquiring particle size distribution parameters and measured osmotic pressure values of the formula samples;

[0020] using a multivariate nonlinear regression analysis method, taking the particle size distribution parameters as independent variables and the osmotic pressure values as dependent variables, to construct an initial fitting model; wherein the particle size distribution parameters include D10, D50, D90, D10 2 , D50 2 and D10 x D50.

[0021] performing feature screening, five-step iterative optimization, cross-validation and parameter optimization on the initial fitting model to obtain the osmotic pressure fitting model.

[0022] In a possible implementation, the feature screening includes:

[0023] determining candidate particle size features associated with osmotic pressure; wherein the candidate particle size features include D10, D50, D90, D10 2 , D50 2 , and D10 x D50;

[0024] eliminating weakly correlated features from the candidate particle size features based on a significance test to obtain first candidate particle size features;

[0025] performing centering processing on features with moderate collinearity in the first candidate particle size features based on a variance inflation factor, so that a variance inflation factor of the processed features is less than or equal to 5;

[0026] calculating importance of each feature after the centering processing by using a random forest algorithm, and eliminating features with importance less than a set threshold;

[0027] comparing model performance with and without D10 x D50, and determining a final feature set according to model fitting goodness and information criterion values.

[0028] In a possible implementation, the five-step iterative optimization includes:

[0029] testing three types of linear models, including ordinary least squares, ridge regression, and Lasso regression, comparing model fitting goodness, residual standard deviation, and overfitting degree, and determining a basic model;

[0030] testing model performance after introducing quadratic terms for the final feature set, retaining quadratic terms that improve model fitting goodness of the basic model and reduce information criterion values, and forming a model combining linear terms and quadratic terms;

[0031] introducing candidate interaction terms by using forward stepwise regression, testing significance of the interaction terms, and retaining interaction terms that improve model fitting goodness and reduce information criterion values;

[0032] testing three types of regularization methods, including L1 regularization, L2 regularization, and elastic net, determining optimal regularization coefficients by grid search, and selecting a regularization method that minimizes error of a validation set and significantly improves coefficients of all core features.

[0033] In a possible implementation, the cross-validation includes:

[0034] using 10-fold cross-validation and leave-one-out method to verify stability of the model, and performing residual analysis to ensure that the residuals are normally distributed and have no systematic bias, to obtain target particle size features corresponding to the osmotic pressure fitting model.

[0035] In a possible implementation, the parameter optimization comprises:

[0036] A first-order to third-order polynomial model is tested to determine a second-order polynomial as the model complexity according to the goodness of fit and information criterion value of the test set.

[0037] A coarse grid and fine grid two-step search method is used to determine the optimal regularization coefficient. After the coarse grid search locks the coefficient range, the fine grid search obtains the coefficient value that minimizes the error of the test set, and the stability of the coefficient is verified under different data division ratios.

[0038] A super parameter combination matrix combining the polynomial order and the regularization coefficient is constructed, the goodness of fit and information criterion value of each combination are calculated, and the optimal super parameter combination is selected to determine the osmotic pressure fitting model based on the optimal super parameter combination.

[0039] In a second aspect, an embodiment of the present application provides a formula nutritional food preparation control device, comprising:

[0040] An acquisition module is configured to acquire real-time particle size data in a formula nutritional food preparation process.

[0041] An osmotic pressure fitting module is configured to input the real-time particle size data into a trained osmotic pressure fitting model to obtain a fitted osmotic pressure, wherein the osmotic pressure fitting model is trained based on sample characteristic particle size data and sample osmotic pressure corresponding to the formula nutritional food to be prepared.

[0042] A determination module is configured to determine a risk intervention control scheme according to the fitted osmotic pressure and a preset early warning condition, wherein the process parameters in the risk intervention control scheme include one or more of prebiotic adjustment amount, lactose adjustment amount, and homogenization pressure.

[0043] A control module is configured to adjust the process parameters according to the risk intervention control scheme.

[0044] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the method in the first aspect or any possible implementation manner of the first aspect when executing the computer program.

[0045] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the method in the first aspect or any possible implementation manner of the first aspect.

[0046] In the embodiment of the present application, by acquiring real-time particle size data in the preparation process of the formula nutritional food, inputting the real-time particle size data into the osmotic pressure fitting model trained based on the characteristic particle size and osmotic pressure of the corresponding formula nutritional food sample to obtain a fitted osmotic pressure, and combining the preset early warning condition to determine a risk intervention control scheme containing process parameters such as prebiotic adjustment amount, lactose adjustment amount or homogenization pressure and adjust the process parameters, real-time prediction of the osmotic pressure in the preparation process of the formula nutritional food and dynamic adjustment of the process can be realized, the intestinal stress risk of the formula nutritional food caused by improper osmotic pressure can be avoided, the precision of the formula nutritional food preparation and the product safety can be improved, and the defects of the traditional control relying on lagging feedback are solved. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is an application scenario diagram of the formula nutritional food preparation control method provided by an embodiment of the present application;

[0048] Figure 2 is an implementation flowchart of the formula nutritional food preparation control method provided by an embodiment of the present application;

[0049] Figure 3 is an implementation flowchart of the formula nutritional food preparation control method provided by another embodiment of the present application;

[0050] Figure 4 is a structural schematic diagram of the formula nutritional food preparation control device provided by an embodiment of the present application;

[0051] Figure 5 is a schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0052] The formula nutritional food preparation control method provided by the embodiment of the present application will be described in detail below with reference to the accompanying drawings.

[0053] Figure 1 is an application scenario diagram of the formula nutritional food preparation control method provided by an embodiment of the present application. As shown in Figure 1 , it includes a laser particle size analyzer, a control terminal and a formula nutritional food preparation device.

[0054] In the specific implementation process, the real-time particle size data in the preparation process of the formula nutritional food is detected by a laser particle size analyzer. The control terminal communicates with the laser particle size analyzer through wired or wireless mode to obtain the real-time particle size data, obtains the fitted osmotic pressure based on the real-time particle size data and the osmotic pressure fitting model, and determines the risk intervention control scheme based on the fitted osmotic pressure and the preset early warning condition. The formula nutritional food preparation equipment adjusts one or more process parameters such as prebiotic adjustment amount, lactose adjustment amount and homogenization pressure under the guidance of the risk intervention control scheme, so as to realize real-time osmotic pressure reduction in the preparation process of the formula nutritional food and avoid the intestinal stress risk of the formula nutritional food caused by improper osmotic pressure.

[0055] Figure 1 In the embodiment shown, the control terminal is independent of the formula nutritional food preparation equipment, which is convenient for remote control by the staff. In other possible implementations, the laser particle size analyzer is directly connected in communication with the controller arranged in the system of the formula nutritional food preparation equipment, without the need for an additional control terminal.

[0056] Figure 2 The implementation flowchart of the formula nutritional food preparation control method provided by an embodiment of the present application is shown in FIG. 1. Figure 2 As shown, the method comprises the following steps:

[0057] S201, obtaining real-time particle size data in the preparation process of the formula nutritional food.

[0058] The execution subject of each embodiment of the present application can be a server, a processor, a microprocessor or other devices with data processing function. In the actual implementation process, the specific implementation mode of the execution subject can be selected according to actual needs, and the present embodiment does not make special limitation as long as it is a device with data processing function. For the convenience of understanding, the control terminal shown in FIG. 1 is taken as the execution subject for illustration in the subsequent embodiments. Figure 1

[0059] In the actual implementation process, an online laser particle size analysis device is arranged to collect the particle size data of the formula nutritional food product particle samples flowing through the pipeline in real time. Optionally, the device is arranged at the key production node after the mixing and before the packaging of the infant formula nutritional food, and the device is in communication with the material conveying pipeline of the formula nutritional food production line.

[0060] ​To improve the detection efficiency, the online laser particle size analysis device periodically samples the real-time particle size data according to the set sampling frequency. For example, the characteristic particle size parameters (including D10, D50) of the formula nutritional food product are detected at a collection frequency of once every 5 minutes. In addition, each detection generates multiple sets of parallel data, and the average value calculated by the data processing module of the device is transmitted to the central processing unit of the formula nutritional food production control system as the real-time particle size data. Optionally, 3-5 sets of parallel data are generated for each detection.

[0061] S202, input the real-time particle size data into the trained osmotic pressure fitting model to obtain the fitted osmotic pressure; wherein the osmotic pressure fitting model is trained based on the sample characteristic particle size data and sample osmotic pressure corresponding to the formula nutritional food to be prepared.

[0062] The formula nutritional food production line can be used to prepare formula nutritional food for different age groups, so the osmotic pressure fitting model is different for formula nutritional food for different age groups. Taking 0-6 month old infant formula milk powder as an example, the training process is based on different batches of laboratory samples of the formula milk powder: first, prepare multiple groups of formula milk powder samples by simulating the production process in the laboratory, measure the characteristic particle size data and the corresponding measured osmotic pressure value of each group of samples after reconstitution, then use multivariate nonlinear regression analysis method, with characteristic particle size data as independent variable, measured osmotic pressure value as dependent variable, after characteristic screening, iterative optimization and cross-validation, the osmotic pressure fitting model is constructed.

[0063] In actual implementation process, after the control terminal receives the real-time particle size data, the data is automatically input into the above osmotic pressure fitting model, the model processes the real-time particle size data through the built-in calculation logic, outputs the corresponding fitted osmotic pressure value, and executes the subsequent risk assessment based on the fitted osmotic pressure value.

[0064] S203, determine the risk intervention control scheme according to the fitted osmotic pressure and the preset warning condition; wherein the process parameters in the risk intervention control scheme include one or more of prebiotic adjustment amount, lactose adjustment amount and homogenization pressure.

[0065] In the control terminal, the osmotic pressure warning condition corresponding to the formula nutritional food to be prepared is preset. Taking 0-6 month old infant formula milk powder as an example, the warning condition is determined based on the osmotic pressure range of breast milk (260-300 mOsmol / kg) and the intestinal stress risk test data of infants and young children, and the fitted osmotic pressure value is used as the core judgment index.

[0066] Take the preset early warning conditions including low-risk, medium-risk and high-risk early warning conditions as an example for illustration. If the fitted osmotic pressure value is in the low-risk interval (≤290 mOsmol / kg), it is determined that the current milk powder osmotic pressure meets the intestinal tolerance requirement, and it is determined that the conventional production scheme does not need to adjust the process parameters. If the fitted osmotic pressure value is in the medium-risk interval (290-320 mOsmol / kg), it is determined that there is a slight risk of intestinal stress, and a medium-risk intervention control scheme containing a regulating amount of prebiotics and a regulating amount of lactose is determined, and the osmotic pressure is reduced by adjusting the ratio of prebiotics and lactose; if the fitted osmotic pressure value is in the high-risk interval (>320 mOsmol / kg), it is determined that there is a higher risk of intestinal stress, and a high-risk intervention control scheme containing a homogeneous pressure adjustment amount is determined, and the osmotic pressure is reduced by adjusting the homogeneous process parameters to change the particle size distribution of the milk powder.

[0067] In other possible implementations, the preset early warning conditions can be further refined, for example: into safe conditions, low-risk early warning conditions, medium-risk early warning conditions and high-risk early warning conditions.

[0068] S204, adjusting the process parameters according to the risk intervention control scheme.

[0069] In the specific implementation process, the control terminal converts the determined risk intervention control scheme into specific process control instructions, and sends them to the corresponding formula nutritional food preparation equipment control system to adjust the process parameters.

[0070] In this embodiment, by obtaining real-time particle size data in the formula nutritional food preparation process, inputting it into the osmotic pressure fitting model trained based on the characteristic particle size and osmotic pressure of the corresponding formula nutritional food sample to obtain the fitted osmotic pressure, and combining the preset early warning conditions to determine the risk intervention control scheme containing the process parameters such as the regulating amount of prebiotics, the regulating amount of lactose or the homogeneous pressure, and adjusting the process parameters, the real-time prediction and dynamic adjustment of the osmotic pressure in the formula nutritional food preparation process can be realized, the risk of intestinal stress of the formula nutritional food caused by improper osmotic pressure can be avoided, the precision and product safety of the formula nutritional food preparation can be improved, and the defects of traditional lagging feedback control can be solved.

[0071] Figure 3 is the implementation flowchart of the formula nutritional food preparation control method provided by another embodiment of the present application. As shown in Figure 3 In one possible implementation, the preset early warning conditions include preset low-risk early warning conditions, medium-risk early warning conditions and high-risk early warning conditions;

[0072] The risk intervention control scheme includes: a medium-risk risk intervention control scheme and a high-risk risk intervention control scheme;

[0073] Among them, the process parameters corresponding to the medium-risk risk intervention control scheme include: the regulating amount of prebiotics and the regulating amount of lactose;

[0074] The process parameters corresponding to the high-risk intervention control scheme include the homogenization pressure.

[0075] In the implementation process, if the fitting osmotic pressure meets the low-risk early warning condition, the routine production scheme is maintained, the operating parameters of the prebiotic adding device, the lactose adding device and the homogenizer are maintained unchanged by the formula nutritional food preparation equipment control system, and the formula nutritional food production continues according to the original process.

[0076] If the fitting osmotic pressure meets the medium-risk early warning condition, the medium-risk intervention control scheme is adopted, the prebiotic adding device increases the unit time addition amount of prebiotics according to the control instruction, the lactose adding device synchronously reduces the unit time addition amount of lactose, the total carbohydrate content is ensured to be stable, and the real-time particle size data and the fitting osmotic pressure value of the formula nutritional food product are continuously monitored after adjustment until the fitting osmotic pressure falls back to the low-risk interval.

[0077] If the fitting osmotic pressure meets the high-risk early warning condition, the high-risk intervention control scheme is adopted, the homogenizer control system reduces the homogenization pressure according to the control instruction, the particle size distribution of the formula nutritional food particles is adjusted by changing the particle size distribution, and then the osmotic pressure is affected. Optionally, during the adjustment process, the real-time particle size data are collected at a set frequency (such as once every 2 minutes) and the fitting osmotic pressure is calculated until the fitting osmotic pressure falls to the low-risk interval, and then the homogenization pressure is stabilized at the adjusted parameter value to ensure that the osmotic pressure of the formula nutritional food produced in the subsequent production meets the requirements.

[0078] In the embodiment, the formula nutritional food preparation control method divides the preset early warning conditions into three categories of low-risk, medium-risk and high-risk, and correspondingly sets the medium-risk intervention control scheme containing the prebiotic adjustment amount and the lactose adjustment amount, and the high-risk intervention control scheme containing the homogenization pressure, so that the risk intervention is more targeted, the problems of resource waste or insufficient intervention caused by unified intervention scheme are avoided, the risk control level in the formula nutritional food preparation process is further refined, and the adaptability and effectiveness of process adjustment under different risk levels are improved.

[0079] In a possible implementation manner, the risk intervention control scheme is determined according to the fitting osmotic pressure and the preset early warning condition, including:

[0080] When the fitting osmotic pressure meets the medium-risk early warning condition, a first difference value between the fitting osmotic pressure and the upper limit value of the low-risk early warning condition is calculated, and the prebiotic addition amount and the lactose reduction amount are determined according to the first difference value; wherein the prebiotic addition amount and the lactose reduction amount are the same.

[0081] When the fitting osmotic pressure meets the high-risk early warning condition, a second difference value between the fitting osmotic pressure and the upper limit of the low-risk early warning condition is calculated, and the homogenization pressure reduction value is determined according to the second difference value.

[0082] In the specific implementation process, the osmotic pressure range of the medium-risk early warning condition is 290-320 mOsmol / kg, and the upper limit of the low-risk early warning condition (290 mOsmol / kg) is used as the reference for calculating the first difference; the osmotic pressure range of the high-risk early warning condition is >320 mOsmol / kg, and the upper limit of the low-risk early warning condition (290 mOsmol / kg) is also used as the reference for calculating the second difference.

[0083] The control terminal compares the fitted osmotic pressure value with the preset early warning condition to determine the early warning level to which it belongs. If the fitted osmotic pressure value is in the range of 290-320 mOsmol / kg, it is determined to be a medium-risk early warning; if the fitted osmotic pressure value is >320 mOsmol / kg, it is determined to be a high-risk early warning. When it is determined to be a medium-risk early warning (for example, the fitted osmotic pressure value is 305 mOsmol / kg), the control terminal automatically calculates the first difference. The prebiotic addition amount is calculated according to Δosmotic pressure=K1×(prebiotic addition amount / kg). Wherein K1 is the adjustment coefficient, which can be selected as K1=-2.8 mOsmol / kg / kg.

[0084] The first difference (i.e. Δosmotic pressure) = fitted osmotic pressure value - upper limit of low-risk early warning condition, i.e. 305 mOsmol / kg - 290 mOsmol / kg = 15 mOsmol / kg, so the prebiotic addition amount is 5.4 kg. According to the correlation logic built in the system, the prebiotic addition amount and the lactose reduction amount are the same.

[0085] When it is determined to be a high-risk early warning (for example, the fitted osmotic pressure value is 335 mOsmol / kg), the central processing unit automatically calculates the second difference. The homogenization pressure reduction value is calculated according to Δosmotic pressure=K2×(homogenization pressure increment / bar), wherein K2 is the adjustment coefficient, which can be selected as K2=-2.5 mOsmol / kg / bar.

[0086] The second difference = fitted osmotic pressure value - upper limit of low-risk early warning condition, i.e. 335 mOsmol / kg - 290 mOsmol / kg = 45 mOsmol / kg, so the homogenization pressure increment is 18 bar.

[0087] In this embodiment, when the fitting osmotic pressure meets the medium-risk early warning condition, the first difference value between the fitting osmotic pressure and the upper limit of the low-risk early warning condition is calculated to determine the equivalent prebiotic addition amount and lactose reduction amount, and when the fitting osmotic pressure meets the high-risk early warning condition, the second difference value between the fitting osmotic pressure and the upper limit of the low-risk early warning condition is calculated to determine the homogenization pressure reduction value, so that the process parameter adjustment amount is directly related to the osmotic pressure deviation, avoiding the blindness of the adjustment amount, and ensuring that the adjustment of the prebiotic, lactose and homogenization pressure can accurately match the osmotic pressure control requirement, and the accuracy of risk intervention is improved. In addition, when the fitting osmotic pressure meets the high-risk early warning condition, the homogenization pressure reduction value is determined according to the upper limit of the low-risk early warning condition, so that the osmotic pressure can be adjusted at the fastest speed to ensure that the osmotic pressure is reduced to a reasonable range.

[0088] In other possible implementations, according to the fitting osmotic pressure and the preset early warning condition, a risk intervention control scheme is determined, including:

[0089] When the fitting osmotic pressure meets the medium-risk early warning condition, a first difference value between the fitting osmotic pressure and the upper limit of the low-risk early warning condition is calculated, and a prebiotic addition amount and a lactose reduction amount are determined according to the first difference value; wherein the prebiotic addition amount and the lactose reduction amount are the same.

[0090] When the fitting osmotic pressure meets the high-risk early warning condition, a second difference value between the fitting osmotic pressure and the upper limit of the medium-risk early warning condition is calculated, and a homogenization pressure reduction value is determined according to the second difference value.

[0091] In this embodiment, when the fitting osmotic pressure meets the high-risk early warning condition, the homogenization pressure reduction value is determined according to the upper limit of the medium-risk early warning condition, so that the system can be pulled back from the dangerous edge, and then the fitting osmotic pressure is updated, and whether further adjustment of the prebiotic and lactose is needed is further determined according to the fitting osmotic pressure and the early warning condition. The homogenization pressure reduction value determined based on the fitting osmotic pressure and the upper limit of the medium-risk early warning condition can make the control process smoother, and can effectively avoid the dramatic fluctuation of the parameters.

[0092] The above mainly introduces how to adjust the process parameters according to the real-time particle size data, and the training process of the osmotic pressure fitting model is introduced below.

[0093] In a possible implementation, before the real-time particle size data in the preparation process of the formula nutritional food is obtained, the method further includes:

[0094] A plurality of formula samples of the formula nutritional food to be prepared are obtained, and particle size distribution parameters and measured osmotic pressure values of the formula samples are obtained;

[0095] A multivariate nonlinear regression analysis method is used to construct an initial fitting model with the particle size distribution parameters as independent variables and the osmotic pressure values as dependent variables; wherein the particle size distribution parameters include D10, D50, D90, D10 2 , D50 2and D10 x D50;

[0096] The initial fitting model is subjected to feature screening, five-step iterative optimization, cross-validation and parameter optimization to obtain the osmotic pressure fitting model.

[0097] In the specific implementation process, the osmotic pressure value of the formula sample is measured by using an osmotic pressure instrument, each formula sample is tested in parallel for 3 times, and the average value is obtained. The characteristic particle size (at least including D10 and D50) of the formula sample is measured by using a laser particle size analyzer, each formula sample is tested in parallel for 3 times, and the average value is obtained. The measurement result accuracy is improved by obtaining the average value.

[0098] In this embodiment, before obtaining the real-time particle size data, by obtaining the multiple formula samples of the to-be-prepared formula nutritional food and the particle size distribution parameters and the measured osmotic pressure value thereof, an initial fitting model is constructed by using multivariate nonlinear regression analysis, taking the particle size distribution parameters as the independent variables and the osmotic pressure value as the dependent variable, and the osmotic pressure fitting model is obtained through feature screening, five-step iterative optimization, cross-validation and parameter optimization, which ensures that the construction of the osmotic pressure fitting model is based on the actual sample data of the to-be-prepared formula nutritional food, and the accuracy and reliability of the model are improved through multiple optimization links, which provides accurate model support for subsequent real-time osmotic pressure prediction and reduces the process adjustment deviation caused by model error.

[0099] In a possible implementation manner, the feature screening comprises:

[0100] Determine candidate particle size features associated with the osmotic pressure; wherein the candidate particle size features include: D10, D50, D90, D10 2 , D50 2 and D10 x D50;

[0101] Remove weakly correlated features from the candidate particle size features based on the significance test to obtain first candidate particle size features;

[0102] Centering process is performed on the moderately collinear features in the first candidate particle size features based on the variance inflation factor, so that the variance inflation factor of the processed features is ≤5;

[0103] The importance of each feature after the centering process is calculated by using the random forest algorithm, and features with importance lower than a set threshold are removed;

[0104] Compare the model performance of the model containing D10 x D50 and the model not containing D10 x D50, and determine the final feature set according to the model fitting degree and the information criterion value.

[0105] In the embodiment, the feature screening process first determines candidate particle size characteristics including D10, D50, etc., and then removes weakly correlated features through significance test to obtain first candidate particle size characteristics. The features with moderate collinearity are centered based on the variance inflation factor to ensure feature independence. Random forest algorithm is used to remove features with importance lower than a set threshold. Finally, the model performance with and without D10 x D50 is compared to determine the final feature set. The features closely related to the osmotic pressure, strong independence and effective for improving the model performance can be gradually screened out. The interference of redundant features or weakly correlated features on the fitting accuracy of the model is avoided, and the fitting model accuracy of the osmotic pressure is improved.

[0106] In a possible implementation, the five-step iterative optimization includes:

[0107] Test ordinary least squares, ridge regression, and Lasso regression three types of linear models, compare model fitting goodness, residual standard deviation, and overfitting degree, and determine the base model;

[0108] For the final feature set, test the model performance after introducing the quadratic term, and retain the quadratic term that improves the model fitting goodness of the base model and reduces the information criterion value, to form a linear and quadratic term combined model;

[0109] Introduce candidate interaction terms using forward stepwise regression, test the significance of the interaction terms, and retain the interaction terms that improve the model fitting goodness and reduce the information criterion value;

[0110] Test L1 regularization, L2 regularization, and elastic network three types of regularization methods, determine the optimal regularization coefficient through grid search, and select the regularization method that minimizes the validation set error and makes all core feature coefficients significant.

[0111] In the embodiment, the five-step iterative optimization process first determines the base model by testing three types of linear models, then tests the introduction of quadratic terms for the final feature set and retains the quadratic terms that improve the model performance to form a linear and quadratic term combined model, then introduces and retains significant interaction terms through forward stepwise regression, and finally tests three types of regularization methods and determines the optimal regularization coefficient and method. The model structure and parameters can be optimized step by step, the fitting ability of the model for osmotic pressure can be gradually improved, the risk of overfitting can be controlled, the model can ensure good generalization ability while ensuring fitting goodness, and the reliability of the osmotic pressure fitting model is improved.

[0112] In a possible implementation, the cross-validation includes:

[0113] 10-fold cross-validation and leave-one-out method are used to verify the stability of the model, and residual analysis is performed to ensure that the residuals are normally distributed and have no systematic bias, so as to obtain the target particle size characteristics corresponding to the osmotic pressure fitting model.

[0114] In the present embodiment, the cross-validation process verifies the stability of the model by using 10-fold cross-validation and leave-one-out validation, and simultaneously performs residual analysis to ensure that the residuals are normally distributed and have no systematic bias. This can verify the consistency of the model performance on different data subsets from different angles, avoid the problem of insufficient stability of the model due to data partition bias, and exclude the influence of systematic bias on the accuracy of the model through residual analysis. Finally, the target particle size characteristics corresponding to the osmotic pressure fitting model are determined to ensure the stability and accuracy of the model for real-time osmotic pressure prediction.

[0115] In one possible implementation, the parameter optimization includes:

[0116] Test the 1st to 3rd order polynomial model, and determine the 2nd order polynomial as the model complexity according to the evaluation criteria of the fitting goodness of the validation set and the information criterion value;

[0117] Determine the optimal regularization coefficient by using the two-step search method of coarse grid and fine grid. After locking the coefficient range by coarse grid search, the coefficient value that minimizes the error of the validation set is obtained by fine grid search, and the stability of the coefficient is verified under different data partition ratios;

[0118] Construct a hyperparameter combination matrix combining the polynomial order and the regularization coefficient, calculate the fitting goodness of the validation set and the information criterion value of each combination, and select the optimal hyperparameter combination to determine the osmotic pressure fitting model based on the optimal hyperparameter combination.

[0119] In the present embodiment, the parameter optimization process first tests the 1st to 3rd order polynomial model and determines the 2nd order polynomial as the model complexity according to the evaluation criteria of the fitting goodness of the validation set and the information criterion value. Then, the optimal regularization coefficient is determined by using the two-step search method of coarse grid and fine grid, and its stability is verified. Finally, a hyperparameter combination matrix is constructed to select the optimal hyperparameter combination to determine the osmotic pressure fitting model. This can accurately determine the polynomial complexity and regularization coefficient of the model, avoid overfitting or underfitting problems while ensuring the fitting ability of the model, and ensure that the selected hyperparameter combination can optimize the performance of the model through hyperparameter combination screening, thereby further improving the precision and generalization ability of the osmotic pressure fitting model and providing more reliable osmotic pressure prediction results for subsequent formulation of nutritional food preparation process adjustment.

[0120] The following describes the osmotic pressure fitting model training process of the above embodiments in combination with a specific embodiment:

[0121] 1. Multi-round feature screening and interactive verification

[0122] Based on the theoretical hypothesis (the influence of particle size distribution on osmotic pressure), the candidate particle size characteristics corresponding to the initial fitting model are determined, and 6 candidate particle size characteristics are initially included: D10, D50, D90 (particle size parameters), D10 2 , D502 (D10, D50), D10xD50 (interaction term), to form the initial feature pool.

[0123] First round: single-feature significance screening

[0124] Linear correlation of each feature with osmotic pressure was analyzed by t-test, and weakly correlated features with p>0.05 were removed:

[0125] D90 had a p-value of 0.12 (not significant) and a Pearson correlation coefficient r=0.91 with D50 (multicollinearity risk), so it was removed; the remaining features (D10, D50, D10 2 , D50 2 , D10xD50) all met p<0.01 and were retained for the next round.

[0126] Second round: multicollinearity test

[0127] The variance inflation factor (VIF) between features was calculated to determine the degree of collinearity:

[0128] D10 and D10 2 had a VIF of 8.7 (>5, moderate collinearity), which was reduced to VIF=3.2 by "centering" (subtracting the mean value from the feature value); D50 and D50 2 had a VIF of 9.1, which was also reduced to VIF=2.9 by centering, ensuring feature independence.

[0129] Third round: feature importance ranking and redundancy removal

[0130] The feature importance (Mean Decrease Accuracy, MDA) was calculated using a random forest model, and ranked by importance:

[0131] D50 2 (MDA=0.32) > D10xD50 (MDA=0.28) > D50 (MDA=0.21) > D10 (MDA=0.15) > D10 2 (MDA=0.04);

[0132] Since the MDA value of D10 2 was very low (<0.05), and the model R 2 only decreased by 0.8% (from 72.9% to 72.1%) after removal, D10 2 was removed, and the core features were finally retained: D10, D50, D50 2 , D10xD50.

[0133] Fourth round: interaction term effectiveness verification

[0134] Model performance comparison between "with interaction term" and "without interaction term":

[0135] Without D10 x D50 interaction term, model R 2 = 72.1%, AIC = 128.5;

[0136] After adding D10 x D50 interaction term, R 2 = 79.8%, AIC = 116.2 (AIC reduction ≥10, indicating that the interaction term significantly improves the model's explanatory power), and the final feature set {D10, D50, D50 2 , D10 x D50} is determined, laying the foundation for subsequent model construction.

[0137] 2. Five-step iterative optimization and multi-dimensional cross-validation

[0138] The five-step optimization method of "basic model → step-by-step upgrade → risk control → stability verification" is adopted, and the optimal direction is determined through comparison experiments and error analysis at each step. The specific parameter adjustment process is as follows:

[0139] First step: Basic linear model optimization (determine the benchmark)

[0140] Test 3 types of linear models (ordinary least squares OLS, ridge regression, and Lasso regression), and compare the core indicators:

[0141] OLS model: R 2 = 63.5%, residual standard deviation = 18.2 mOsmol / kg, with slight overfitting (training set R 2 = 68.3%, test set R 2 = 63.5%, difference of 4.8%); Ridge regression (λ = 0.05): test set R 2 = 64.2%, residual standard deviation = 17.8 mOsmol / kg, overfitting degree reduced (difference of 3.2%); Lasso regression (λ = 0.05): eliminate D10 feature (coefficient compressed to 0), R 2 = 59.8%, performance decreased;

[0142] Based on the above, choose ridge regression as the basic model, R 2 = 64.2% as the initial benchmark.

[0143] Second step: Introduce quadratic term optimization (improve non-linear fitting ability)

[0144] For the retained features (D10, D50), test the introduction effect of quadratic terms (D10 2 , D50 2 ) one by one:

[0145] Only add D102 : Model R 2 = 65.7%, AIC = 132.1 (limited improvement); only add D50 2 : Model R 2 = 72.1%, AIC = 120.3 (R 2 improved by 7.9%, AIC significantly reduced); add both D10 2 + D50 2 : Model R 2 = 72.3%, AIC = 122.5 (R 2 improved slightly, AIC increased, redundancy exists);

[0146] In summary, only keep D50 2 quadratic term, the model is upgraded to "linear + D50 2 ", R 2 = 72.1%.

[0147] Step 3: Interaction term introduction optimization (capture variable synergy effect)

[0148] Use forward stepwise regression to introduce interaction terms, add 1 candidate interaction term (D10 x D50, D10 x D50 2 ) each time, and test significance:

[0149] Introduce D10 x D50: interaction term p-value = 0.003 (<0.01), model R 2 = 79.8%, AIC = 116.2 (R 2 improved by 7.7%); continue to introduce D10 x D50 2 : interaction term p-value = 0.15 (> 0.05), model R 2 = 80.1% (improved by 0.3%), AIC = 118.7 (increased by 2.5%);

[0150] In summary, only keep significant interaction term D10 x D50, model R 2 = 79.8%.

[0151] Step 4: Regularization parameter optimization (control overfitting)

[0152] Test 3 types of regularization methods (L1, L2, ElasticNet), determine the optimal regularization coefficient λ through grid search (search range: 10 -4 ~ 10 1 , 20 nodes on logarithmic scale):

[0153] L1 regularization: when λ = 0.08, the validation set error = 15.2 mOsmol / kg, but the D10 coefficient is compressed to 0, and the feature information is lost; Elastic net (a = 0.5): when λ = 0.06, the validation set error = 14.8 mOsmol / kg, but the model complexity increases (the number of parameters increases by 2); L2 regularization: when λ = 0.1, the validation set error = 14.5 mOsmol / kg (minimum), and all core feature coefficients are significant (p < 0.01), and the degree of overfitting is best controlled (training set R 2 = 82.5%, test set R 2 = 80.3%, gap 2.2%);

[0154] Based on the above, L2 regularization is selected, λ = 0.1, and the model R 2 = 80.3%.

[0155] Step 5: Cross-validation and stability verification (to ensure generalization ability)

[0156] Double verification strategy is adopted to test the stability of the model:

[0157] 10-fold cross-validation: the data set is divided into training set / validation set by 8:2, repeated 10 times, and the average R 2 = 85.1% ± 2.3% (standard deviation < 3%, good stability);

[0158] Leave-one-out method verification (LOOCV): a total of 120 samples, each time leaving 1 sample as the test set, and the final LOOCV R 2 = 84.7%, close to the result of 10-fold cross-validation (gap < 0.5%);

[0159] Residual analysis: the residual error of the validation set is normally distributed (Shapiro-Wilk test p = 0.23 > 0.05), and there is no systematic bias (residual mean = 0.32 mOsmol / kg, close to 0);

[0160] The final model structure is as follows: L2 regularization (λ = 0.1) + linear term (D10, D50) + quadratic term (D50 2 ) + interaction term (D10 x D50), test set R 2 = 85.1%, meeting all performance requirements.

[0161] 3. Parameter optimization: grid search and fine iteration

[0162] Fine-tune the optimization of 3 key hyperparameters, the specific process is as follows:

[0163] A. Polynomial order optimization (determine model complexity)

[0164] Test 1~3 order polynomial model, with "validation set R 2 +AIC value" as double evaluation criteria:

[0165] 1 order polynomial (only linear term): validation set R 2 =64.2%, AIC=135.6 (underfitting, unable to capture nonlinear relationship); 2 order polynomial (linear term + quadratic term + interaction term): validation set R 2 =80.3%, AIC=116.2 (lowest AIC, balanced fitting effect and complexity); 3 order polynomial (high order term D50 3 , D10 x D50 2 ): validation set R 2 =81.5%, but training set R 2 =92.3% (serious overfitting), AIC=120.8 (higher than 2 order).

[0166] Determine the polynomial order as 2 order in combination.

[0167] B. Regularization coefficient λ optimization (balance fitting and generalization)

[0168] Use "coarse grid + fine grid" two-step search method to determine the optimal λ:

[0169] First step: coarse grid search (range 10 -4 ~10 1 , 10 nodes on logarithmic scale): calculate the validation set RMSE corresponding to each λ, find that when λ is in the range of 10 -2 ~10 0 , RMSE shows a downward and then flat trend, and the preliminary locking range is 0.01~1;

[0170] Second step: fine grid search (range 0.01~1, interval 0.01, total 100 nodes): draw "λ-RMSE" curve, find that when λ=0.1, validation set RMSE=14.5 mOsmol / kg (minimum), and when λ>0.1, RMSE increases (model underfitting), and when λ<0.1, RMSE does not decrease significantly (overfitting risk increases);

[0171] Third step: stability verification: under different data division (training set / validation set=7:3, 9:1), the validation set RMSE fluctuation of λ=0.1 is less than 0.5 mOsmol / kg, and the parameter stability meets the standard;

[0172] Determine the regularization coefficient λ=0.1 in combination.

[0173] C. Hyperparameter combination verification (ensure global optimization)

[0174] The combination matrix of "polynomial order (1-2 order) x regularization coefficient (0.05-0.2)" is constructed, and there are 2x16=32 combinations of hyperparameters, and the validation set R of each combination is calculated 2 Compared with AIC:

[0175] Optimal combination: 2-order polynomial + λ=0.1, validation set R 2 =80.3%, AIC=116.2 (higher than other combinations, such as 2-order + λ=0.08, AIC=118.5, 1-order + λ=0.1, AIC=132.7);

[0176] In summary, the optimal hyperparameter combination is 2-order polynomial + L2 regularization (λ=0.1), which has consistent performance on the training set, validation set, and test set, without overfitting / underfitting risk.

[0177] D. Final model equation (R 2 =85.1%, p<0.01)

[0178] Osmotic pressure (mOsmol / kg)=450.2-1025.5xD50+1213.8xD50 2 +500.6xD10xD50

[0179] Wherein, D10 represents the particle size of 10% of the particle size distribution, unit: μm; D50 represents the particle size of 50% of the particle size distribution, unit: μm; particle size (D50) is the main influencing factor: when D50 increases, the osmotic pressure decreases quadratically; the interaction term D10xD50: indicates that the combination of small particle size D10 and D50 can more sensitively affect the osmotic pressure.

[0180] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0181] In the specific implementation process, in order to improve the accuracy and product safety of the formula nutritional food preparation control method, the osmotic pressure fitting model is guaranteed to achieve the fitting accuracy of the osmotic pressure of the formula nutritional food. In order to verify that the osmotic pressure fitting model has good fitting effect, the osmotic pressure fitting model provided in the above embodiment is verified.

[0182] Example 1: Model verification experiment

[0183] Formula 1: raw cow milk 64.6%, desalted whey powder 19.1%, edible vegetable blend oil 8.0%, lactose 4.8%, whey protein powder 0.26%, galacto-oligosaccharide 1.6%, arachidonic acid oil powder 0.5%, compound minerals 0.4%, docosahexaenoic acid oil powder 0.4%, compound vitamins 0.2%, calcium carbonate 0.1%, choline chloride 0.04%.

[0184] Preparation process 1: the formula is mixed, homogenized at 170 bar, sterilized, concentrated, and then prepared into a semi-finished product by spray drying. The semi-finished product is physically mixed with docosahexaenoic acid and arachidonic acid to produce the finished product.

[0185] Formula 2: raw cow milk 64.5%, desalted whey powder 19.4%, edible vegetable blend oil 8.6%, lactose 4.9%, whey protein powder 0.38%, galacto-oligosaccharide 0.6%, arachidonic acid oil powder 0.5%, compound minerals 0.33%, docosahexaenoic acid oil powder 0.4%, compound vitamins 0.2%, calcium carbonate 0.1%, choline chloride 0.06%, lactoferrin 0.03%.

[0186] Preparation process 2: the formula is mixed, homogenized at 170 bar, sterilized, concentrated, and then prepared into a semi-finished product by spray drying. The semi-finished product is physically mixed with docosahexaenoic acid, arachidonic acid, and lactoferrin to produce the finished product.

[0187] Formula 3: raw cow milk 62.9%, desalted whey powder 12.9%, edible vegetable blend oil 8.09%, lactose 11.2%, whey protein powder 0.55%, galacto-oligosaccharide 1.6%, anhydrous butter 0.9%, arachidonic acid oil powder 0.5%, compound minerals 0.62%, docosahexaenoic acid oil powder 0.4%, compound vitamins 0.16%, calcium carbonate 0.1%, choline chloride 0.06%, nucleotides 0.02%.

[0188] Preparation process 3: the formula is mixed, homogenized at 190 bar, sterilized, concentrated, and then prepared into a semi-finished product by spray drying. The semi-finished product is physically mixed with docosahexaenoic acid, arachidonic acid, and nucleotides to produce the finished product.

[0189] The Bland-Altman analysis method was used to verify the prediction consistency:

[0190]

[0191] For formula 1, formula 2, and formula 3, the predicted values are within the actual value fluctuation range, the mean deviation is 1.0 mOsmol / kg, and the 95% confidence interval is -1.08~3.08 mOsmol / kg. This indicates that the model can accurately predict the osmotic pressure according to the particle size, and confirms that the model has good prediction reliability.

[0192] Example 2: Risk intervention measures

[0193]

[0194] To reduce the osmotic pressure of Formula 2 and Formula 3, intervention measures are taken to reduce the osmotic pressure of milk powder, which helps to optimize the osmotic pressure index of milk powder and reduce the risk of osmotic stress by adjusting the formula ingredients (such as increasing prebiotics and reducing lactose in Formula 2) and production process (such as reducing the homogenization pressure in Formula 3) and other intervention measures.

[0195] The following is an embodiment of the device of the present application, for details not described in detail, can refer to the corresponding method embodiments described above.

[0196] Figure 4 The structure diagram of the formula nutritional food preparation control device provided by the embodiment of the present application is shown, only the part related to the embodiment of the present application is shown for the convenience of description, and the details are as follows:

[0197] As Figure 4 shown, the formula nutritional food preparation control device 4 comprises:

[0198] The acquisition module 401 is configured to acquire real-time particle size data in the preparation process of the formula nutritional food;

[0199] The osmotic pressure fitting module 402 is configured to input the real-time particle size data into the trained osmotic pressure fitting model to obtain a fitted osmotic pressure; wherein the osmotic pressure fitting model is trained based on the sample characteristic particle size data and the sample osmotic pressure corresponding to the formula nutritional food to be prepared;

[0200] The determination module 403 is configured to determine a risk intervention control scheme according to the fitted osmotic pressure and a preset warning condition; wherein the process parameters in the risk intervention control scheme include one or more of the prebiotic adjustment amount, the lactose adjustment amount and the homogenization pressure;

[0201] The control module 404 is configured to adjust the process parameters according to the risk intervention control scheme.

[0202] In one possible implementation, the determination module 403 is specifically configured to:

[0203] When the fitted osmotic pressure meets the medium-risk warning condition, a first difference value between the fitted osmotic pressure and the upper limit of the low-risk warning condition is calculated, and the prebiotic addition amount and the lactose reduction amount are determined according to the first difference value; wherein the prebiotic addition amount and the lactose reduction amount are the same;

[0204] When the fitted osmotic pressure meets the high-risk warning condition, a second difference value between the fitted osmotic pressure and the upper limit of the low-risk warning condition is calculated, and the homogenization pressure downshift value is determined according to the second difference value.

[0205] In a possible implementation, the formula nutritional food preparation control device 4 further includes a model training module, configured to acquire a plurality of formula samples of formula nutritional food to be prepared, and acquire a particle size distribution parameter and a measured osmotic pressure value of each formula sample; adopt a multivariate nonlinear regression analysis method, take the particle size distribution parameter as an independent variable, and take the osmotic pressure value as a dependent variable to construct an initial fitting model; wherein the particle size distribution parameter includes D10, D50, D90, D10 2 , D50 2 and D10 x D50; the initial fitting model is subjected to feature screening, five-step iterative optimization, cross-validation and parameter optimization to obtain an osmotic pressure fitting model.

[0206] In the embodiment, by acquiring real-time particle size data in the preparation process of the formula nutritional food, inputting the real-time particle size data into the osmotic pressure fitting model trained based on the characteristic particle size and the osmotic pressure of the corresponding formula nutritional food sample to obtain a fitted osmotic pressure, and combining a preset early warning condition to determine a risk intervention control scheme including a prebiotic adjustment amount, a lactose adjustment amount or a homogenization pressure and adjust the process parameters, real-time prediction of the osmotic pressure in the preparation process of the formula nutritional food and dynamic adjustment of the process can be realized, the risk of intestinal stress of the formula nutritional food caused by improper osmotic pressure can be avoided, the accuracy of the preparation of the formula nutritional food and the safety of the product can be improved, and the defects of traditional control relying on lagging feedback can be solved.

[0207] Figure 5 is a schematic diagram of an electronic device provided by an embodiment of the present application. As shown in Figure 5 , the electronic device 5 of this embodiment includes a processor 50 and a memory 51. The memory 51 stores a computer program 52. The processor 50 implements the steps in each of the method embodiments described above when executing the computer program 52. Alternatively, the processor 50 implements the functions of each module / unit in each of the device embodiments described above when executing the computer program 52.

[0208] For example, the computer program 52 can be divided into one or more modules / units, which are stored in the memory 51 and executed by the processor 50 to complete the present application. The one or more modules / units can be a series of computer program instruction segments that can complete a specific function, which are used to describe the execution process of the computer program 52 in the electronic device 5.

[0209] The electronic device 5 can include, but is not limited to, the processor 50 and the memory 51. Those skilled in the art can understand, Figure 5 that the electronic device 5 is only an example and does not constitute a limitation on the electronic device 5, and can include more or fewer components than those shown, or combine certain components, or different components, for example, the electronic device 5 can also include an input / output device, a network access device, a bus, etc.

[0210] The processor 50 can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0211] The memory 51 can be an internal storage unit of the electronic device 5, such as a hard disk or a memory of the electronic device 5. The memory 51 can also be an external storage device of the electronic device 5, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. Further, the memory 51 can include both the internal storage unit and the external storage device of the electronic device 5. The memory 51 is used to store the computer program 52 and other programs and data required by the electronic device 5. The memory 51 can also be used to temporarily store data that has been output or will be output.

[0212] For the convenience and brevity of description, only the above-mentioned division of functional modules / units is exemplified, and in actual application, the above-mentioned functions can be completed by different functional modules / units according to needs. The above-mentioned modules / units can be realized in the form of hardware, in the form of software, or in the form of combination of hardware and software.

[0213] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method in each method embodiment described above is realized.

[0214] The embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the method in each method embodiment described above is realized.

[0215] The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0216] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments. If there is no special description and logical conflict, the terms and / or descriptions of different embodiments are consistent and can be mutually referred to, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0217] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A method for controlling the preparation of a formulated nutritional food, characterized in that, The method comprises the following steps: obtaining real-time particle size data in the preparation process of the formula nutrient food; inputting the real-time particle size data into a trained osmotic pressure fitting model to obtain a fitted osmotic pressure; wherein the osmotic pressure fitting model is trained based on the sample characteristic particle size data and sample osmotic pressure corresponding to the formula nutrient food to be prepared; determining a risk intervention control scheme according to the fitted osmotic pressure and a preset warning condition; wherein the process parameters in the risk intervention control scheme include one or more of prebiotic adjustment amount, lactose adjustment amount and homogenization pressure; adjusting the process parameters according to the risk intervention control scheme; wherein the preset warning condition includes a preset low-risk warning condition, a medium-risk warning condition and a high-risk warning condition; the risk intervention control scheme includes a medium-risk risk intervention control scheme and a high-risk risk intervention control scheme; wherein the process parameters corresponding to the medium-risk risk intervention control scheme include prebiotic adjustment amount and lactose adjustment amount; the process parameters corresponding to the high-risk risk intervention control scheme include homogenization pressure; wherein the determination of the risk intervention control scheme according to the fitted osmotic pressure and the preset warning condition comprises: when the fitted osmotic pressure meets the medium-risk warning condition, calculating a first difference value between the fitted osmotic pressure and the upper limit of the low-risk warning condition, and determining the prebiotic addition amount and the lactose reduction amount according to the first difference value; wherein the prebiotic addition amount and the lactose reduction amount are the same; when the fitted osmotic pressure meets the high-risk warning condition, calculating a second difference value between the fitted osmotic pressure and the upper limit of the low-risk warning condition, and determining the homogenization pressure downshift value according to the second difference value; wherein, before obtaining the real-time particle size data in the preparation process of the formula nutrient food, the method further comprises: obtaining a plurality of formula samples of the formula nutrient food to be prepared, and obtaining the particle size distribution parameters and the measured osmotic pressure values of each formula sample; A multivariate nonlinear regression analysis method was used, with the particle size distribution parameters as independent variables and the measured osmotic pressure value as the dependent variable, to construct an initial fitting model; wherein the particle size distribution parameters include D10, D50, D90, and D10. 2 D50 2 and D10×D50; performing feature screening, five-step iterative optimization, cross-validation and parameter optimization on the initial fitting model to obtain the osmotic pressure fitting model.

2. The method of claim 1, wherein the nutritional formula is a medical food. The feature screening comprises: determining a candidate particle size feature associated with osmotic pressure; wherein the candidate particle size feature comprises: D10, D50, D90, D10 2 , D50 2 , and D10 x D50; eliminating weakly correlated features from the candidate particle size features based on significance test to obtain first candidate particle size features; centering the features with moderate collinearity in the first candidate particle size features based on the variance inflation factor, so that the variance inflation factor of the processed features is ≤5; calculating the importance of each feature after centering using a random forest algorithm, and eliminating features with importance lower than a set threshold; comparing the model performance with and without D10×D50, and determining the final feature set according to the model fitting goodness and information criterion value.

3. The method of claim 2, wherein the nutritional formula is a medical food. The five-step iterative optimization comprises: ​ testing three types of linear models including ordinary least squares, ridge regression and Lasso regression, comparing the model fitting goodness, residual standard deviation and overfitting degree, and determining a basic model; for the final feature set, test the model performance after introducing quadratic terms, and retain the quadratic terms that improve the model fitting goodness of the basic model and reduce the information criterion value, to form a model combining linear and quadratic terms; The forward stepwise regression method is used to introduce candidate interaction terms, and the interaction terms are tested for significance, and the interaction terms that improve the model fitting degree and reduce the information criterion value are reserved; Test L1 regularization, L2 regularization, elastic network three kinds of regularization method, through the grid search to determine the optimal regularization coefficient, select the regularization method that makes the validation set error minimum and all core feature coefficient significant.

4. The method of claim 3, wherein the nutritional formula is a medical food. The cross-validation includes: 10-fold cross-validation and leave-one-out method are used to verify the stability of the model, and residual analysis is performed to ensure that the residuals are normally distributed and have no systematic bias, so as to obtain the target particle size characteristics corresponding to the osmotic pressure fitting model.

5. A formula food preparation control device for performing the formula food preparation control method according to any one of claims 1 to 4, characterized by It includes: An acquisition module is configured to acquire real-time particle size data in a preparation process of a formula nutritional food; An osmotic pressure fitting module is configured to input the real-time particle size data into a trained osmotic pressure fitting model to obtain a fitted osmotic pressure, wherein the osmotic pressure fitting model is trained based on sample characteristic particle size data and sample osmotic pressure corresponding to the formula nutritional food to be prepared; A determination module is configured to determine a risk intervention control scheme according to the fitted osmotic pressure and a preset early warning condition, wherein the process parameters in the risk intervention control scheme include one or more of a prebiotic adjustment amount, a lactose adjustment amount, and a homogenization pressure; A control module is configured to adjust process parameters according to the risk intervention control scheme.

6. An electronic device, comprising: The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the method in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to realize the method in any one of claims 1 to 4.

Images