A method for optimizing the preparation process of a composite plant powder with sugar control effect

Abstract

This invention belongs to the field of functional food processing technology and discloses an optimized method for the synergistic preparation process of compound plant powder with sugar control effect. The method includes: acquiring batch attribute data of each plant raw material in the current batch and target product standard data, calculating the attribute deviation value of each plant raw material relative to the target product standard data, and forming a multi-attribute deviation matrix; calling a pre-constructed synergistic preparation decision model to solve for a synergistic preparation plan that meets the target product standard data; preparing compound plant powder samples according to the synergistic preparation plan, obtaining a batch verification result set through testing, and adaptively correcting the synergistic preparation decision model in combination with the target product standard data to output an optimized synergistic preparation process scheme. This invention realizes the continuous evolution of synergistic preparation decision capability, and can comprehensively optimize the sugar control effect, sensory quality and processing performance of the product, achieving high-quality and stable production of compound plant powder.

Description

Technical Field

[0001] This invention relates to the field of functional food processing technology, and more specifically, to an optimized method for the synergistic preparation of compound plant powders with sugar-controlling effects. Background Technology

[0002] In recent years, compound plant powders prepared from natural plants such as mulberry leaves, bitter melon, and buckwheat have gained popularity due to their ability to effectively slow down carbohydrate absorption and inhibit [the absorption of certain substances]. The sugar-controlling effect of glucosidase activity is widely used in the functional food field. Traditional compound plant powder preparation usually adopts fixed feeding formulas and standardized crushing and mixing processes. However, as agricultural products, plant raw materials are significantly affected by natural factors such as place of origin, year, harvest season and climate. The intrinsic properties of different batches of raw materials (such as the content of functional components such as polyphenols / flavonoids, the intensity of bitterness caused by tannins, moisture content, etc.) naturally fluctuate drastically. Using fixed process parameters to process raw materials with fluctuations will inevitably lead to inconsistent sugar control efficacy of the final product, difficulty in uniform taste, and even problems such as clumping and deterioration of processing performance during preparation, making it difficult to meet the requirements of absolute product quality stability for industrial production.

[0003] To address batch-to-batch variations in raw materials, some companies rely on manual experience to make one-way fine-tuning adjustments to the formula during production (e.g., increasing the amount of active ingredient in a batch if it is low). However, this lack of systematic fine-tuning often leads to conflicts of multiple attributes. For example, blindly increasing the amount of a certain Chinese medicinal herb to ensure sugar control can cause a sudden increase in bitterness and loss of consumer acceptance. At the same time, existing preparation processes often separate the ingredient ratio from the particle size, failing to realize that the particle size of plant powders can also interact with the dissolution rate of active ingredients (sugar control) and the roughness on the tongue (sensory experience). Therefore, there is an urgent need for a multi-objective synergistic preparation process optimization method that can cope with batch-to-batch fluctuations in raw materials and comprehensively consider sugar control efficacy, sensory experience, and processing performance.

[0004] In view of this, the present invention proposes an optimized method for the synergistic preparation process of composite plant powders with sugar control function to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: an optimized method for the synergistic preparation process of composite plant powders with sugar-controlling effects, comprising: Step S1: Obtain batch attribute data and target product standard data for each plant raw material in the current batch, including target sugar control index, target sensory index and target processing index; Step S2: Based on the batch attribute data and the target product standard data, calculate the attribute deviation values ​​of each plant raw material in the current batch relative to the target sugar control index, target sensory index and target processing index, and form a multi-attribute deviation matrix for the current batch; Step S3: Call the pre-constructed collaborative preparation decision model, establish the synergistic influence relationship of each plant raw material on sugar control properties, sensory properties and processing properties, and solve the feeding ratio and crushing target mesh size to meet the standard data of the target product based on the multi-attribute deviation matrix, so as to form the collaborative preparation plan for the current batch. Step S4: According to the collaborative preparation plan, the plant raw materials of the current batch are dried, pulverized, sieved and homogenized to obtain composite plant powder samples. The actual sugar control index, actual sensory index and actual processing index of the composite plant powder samples are tested to form a batch verification result set. Step S5: Based on the deviation between the batch verification result set and the target product standard data, evaluate the compensation residual of the current batch, and adaptively correct the collaborative preparation decision model based on the compensation residual, and output the final collaborative preparation process optimization scheme for the current batch.

[0006] Furthermore, methods for forming the multi-attribute deviation matrix for the current batch include: Batch attribute data includes basic information about raw materials and physicochemical testing data of raw materials. The physicochemical testing data of raw materials includes sugar control attribute testing data, sensory attribute testing data and processing attribute testing data. The sugar control attribute detection data corresponding to each plant material are compared with the target sugar control index item by item, the sugar control attribute deviation value is calculated, and the sugar control deviation vector of each plant material is formed. The sensory attribute detection data corresponding to each plant material are compared with the target sensory indicators one by one, the sensory attribute deviation value is calculated, and a sensory deviation vector for each plant material is formed. The processing attribute detection data corresponding to each plant raw material are compared with the target processing indicators one by one, the processing attribute deviation value is calculated, and a processing deviation vector is formed for each plant raw material. The sugar control deviation vector, sensory deviation vector, and processing deviation vector of each plant material are concatenated in sequence to form the comprehensive deviation vector of each plant material; the comprehensive deviation vectors of all plant materials are arranged row by row to form the multi-attribute deviation matrix of the current batch.

[0007] Furthermore, methods for pre-constructing collaborative decision-making models include: Obtain historical production records for all historical production batches; historical production records include historical batch numbers, historical batch attribute data for each plant raw material, historical feed ratios, historical particle size parameters, and historical finished product test results; historical particle size parameters include the target mesh size for each plant raw material. Based on the pre-set knowledge base of synergistic influence relationships, a synergistic constraint encoding vector is constructed, and a model input feature vector is constructed based on the synergistic constraint encoding vector. The synergistic influence relationship knowledge base includes a table of sugar control synergistic relationships among plant raw materials, a table of sensory antagonistic relationships, and a table of processing interaction effects. For each historical production record, a corresponding historical finished product quality vector and a model output label vector are constructed sequentially. The model input feature vectors and model output label vectors from all historical production records are integrated to form a decision training sample set. A collaborative preparation decision network is constructed, which includes a bias-aware encoding layer, a collaborative constraint fusion layer, a ratio granularity joint decision layer, and a constraint verification output layer. Based on the decision training sample set, the collaborative preparation decision network is trained, and the trained collaborative preparation decision network is used as the collaborative preparation decision model.

[0008] Furthermore, methods for training collaborative decision-making networks include: The model input feature vector of each decision training sample is input into the collaborative preparation decision network, which outputs the corresponding feed ratio prediction value and crushing target mesh size prediction value. Combined with the corresponding model output label vector, the ratio prediction loss and particle size prediction loss corresponding to each decision training sample are calculated. Based on the predicted values ​​of the feed ratio and the target mesh size of the crushing, combined with the synergy coefficient in the sugar control synergy table, the antagonism intensity coefficient in the sensory antagonism table, and the interaction influence coefficient in the processing interaction influence table, the finished product quality prediction vector is calculated; the mean square error between the finished product quality prediction vector and the corresponding historical finished product quality vector is calculated to obtain the quality consistency auxiliary loss. Calculate the product of the quality consistency auxiliary loss and the preset quality consistency auxiliary loss weight, and combine it with the ratio prediction loss and granularity prediction loss to obtain the joint decision loss for each decision training sample; calculate the batch joint decision loss based on the joint decision loss of all decision training samples; iteratively update the model parameters in the collaborative preparation decision network based on the batch joint decision loss until the batch joint decision loss converges or the number of training iterations reaches the target, at which point the collaborative preparation decision network training is complete.

[0009] Furthermore, the methods for formulating the co-production plan for the current batch include: The multi-attribute deviation matrix of the current batch is concatenated and flattened into a one-dimensional vector to form the deviation input vector of the current batch. The combinations of plant raw materials involved in the feeding are determined, and the synergy coefficient, antagonism strength coefficient and interaction influence coefficient corresponding to each combination of plant raw materials are obtained from the synergy influence relationship knowledge base to construct the synergy constraint encoding vector of the current batch. The synergy constraint encoding vector of the current batch is concatenated with the deviation input vector of the current batch to form the model input feature vector of the current batch, and input into the synergistic preparation decision model to obtain the predicted feeding ratio and the predicted target mesh size of each plant raw material. The update and iteration process is executed to obtain the updated predicted values ​​of the feed ratio and the target mesh size of the crushing. Based on the absolute values ​​of the differences between each element of the predicted feed ratio and the target mesh size of the crushing before and after the update, it is determined whether the convergence condition is met. If the convergence condition is met, the update and iteration process is stopped, and the final predicted values ​​of the feed ratio and the target mesh size of the crushing are output to form the collaborative preparation plan for the current batch. If the convergence condition is not met, the update and iteration process is continued until the convergence condition is met or the number of update iterations reaches the preset maximum number of updates.

[0010] Furthermore, the method for preparing composite plant powder samples includes: Perform drying treatment on each plant material in the current batch: Preset a drying target moisture content mapping table, which includes the drying target moisture content and recommended drying temperature for each plant material; perform drying treatment on each plant material in the current batch according to the drying target moisture content and recommended drying temperature; The process involves implementing a graded pulverization strategy and sieving: a pre-defined pulverization strategy table is used, which includes the pulverization levels and parameters for each level corresponding to different target mesh sizes; based on the predicted target mesh size for each plant material, the corresponding pulverization level and parameters are obtained from the pulverization strategy table, and each plant material is pulverized step by step according to the pulverization level; based on the predicted target mesh size, each plant material is sieved to obtain qualified powder. Perform synergistic effect perception-based homogenization mixing: Based on the sensory antagonism table, determine whether there is a sensory antagonism relationship between the plant materials; if so, execute a step-by-step progressive mixing strategy to obtain a composite plant powder sample; if not, perform homogenization mixing operation on the qualified powders of all plant materials simultaneously to obtain a composite plant powder sample; perform online process control detection on the composite plant powder sample.

[0011] Furthermore, methods for forming batch verification result sets include: Samples were taken from the compound plant powder sample, and the corresponding actual sugar control index, actual sensory index and actual processing index were tested respectively; the actual sugar control index, actual sensory index and actual processing index were integrated to form a batch verification result set; Among them, the actual blood sugar control indicators include the actual total polyphenol content, the actual total flavonoid content, and the actual... - Glucosidase inhibition rate and actual dietary fiber content; actual sensory indicators include actual bitterness intensity value, actual grassy smell intensity value, actual color brightness value and actual particle roughness level; actual processing indicators include actual moisture content, actual angle of repose, actual moisture absorption rate and actual bulk density.

[0012] Furthermore, methods for evaluating the compensated residuals of the current batch include: The actual test values ​​of each indicator in the batch verification results set are compared with the corresponding target indicators in the target product standard data item by item, and the verification deviation value of each indicator is calculated. All verification deviation values ​​are arranged to form a verification deviation vector. Based on the verification deviation vector, the pass status of each indicator in the batch verification results set is determined and marked as a pass indicator or a fail indicator. If all indicators are determined to be pass indicators, the co-preparation plan is directly used as the co-preparation process optimization scheme. If any non-compliant indicators exist, the corresponding deviation source attribution dimension is determined based on the attribute dimension to which each non-compliant indicator belongs. For each non-compliant indicator, deviation source tracing analysis is performed to determine the dominant deviation factor. Based on the verification deviation value corresponding to the non-compliant indicator, the deviation direction is determined. The verification deviation values, deviation source attribution dimensions, deviation directions, and dominant deviation factors of all non-compliant indicators are integrated to form the compensation residual for the current batch. Among them, the deviation source attribution dimensions include sugar control dimension, sensory dimension, and processing dimension; the deviation direction includes positive exceedance and negative deficiency.

[0013] Furthermore, methods for performing deviation source analysis include: For the non-compliant indicators in the sugar control dimension, the synergistic correction amount of each combination of plant raw materials involved in the feeding is calculated. Based on the synergistic correction amount, the dominant raw materials of sugar control deviation are determined, and the corresponding deviation contribution amount is calculated. The deviation contribution amounts of all dominant raw materials are summarized to form the dominant factors of sugar control deviation. For the non-compliant indicators in the sensory dimension, obtain all combinations of plant raw materials that have an antagonistic relationship with the non-compliant indicators and participate in the feeding, and their corresponding antagonistic strength coefficients; determine the dominant raw materials for sensory deviation based on the antagonistic strength coefficients, and calculate the corresponding deviation contribution; summarize the deviation contribution of all dominant raw materials to form the dominant factors of sensory deviation. For non-conforming indicators in the processing dimension, all plant material combinations that interact with the non-conforming indicators and participate in the feeding are obtained, along with their corresponding interaction coefficients. The effective interaction coefficients for each plant material combination are then determined. Based on the effective interaction coefficients, the dominant raw materials for processing deviations are identified, and their corresponding deviation contribution and dominant particle size parameters are calculated. The deviation contribution and dominant particle size parameters of all dominant raw materials are then summarized to form the dominant factors of processing deviations.

[0014] Furthermore, methods for outputting optimized collaborative preparation process schemes include: A set of directional correction instructions is constructed based on the compensation residual, and conflict detection and coordination of correction instructions are performed on the set of directional correction instructions. The set of directional correction instructions includes proportion correction instructions and particle size correction instructions. Based on the coordinated set of directional correction instructions, the corrected feed ratio and corrected crushing target particle size of each plant material are calculated. The raw material name, unique code, target moisture content, recommended drying temperature, number of grinding stages, grinding parameters for each stage, mixing and addition order, and corrected feed mass of each plant raw material are obtained sequentially. These are then integrated with the corrected feed ratio and the corrected target particle size to form a detailed list of process parameters. The detailed list of process parameters for all plant raw materials is then integrated with the deviation traceability analysis results in the compensation residual to generate the final optimized synergistic preparation process scheme for the current batch. The method for detecting and coordinating conflict in the correction instructions is as follows: In the set of directional correction instructions, detect whether there are contradictory correction instructions for the same plant material; if so, perform conflict coordination: divide the contradictory correction instructions corresponding to the same plant material into increase instructions and decrease instructions according to the corresponding correction direction, and mark the increase instructions and decrease instructions as priority instructions and subordinate instructions respectively based on the verification deviation value; retain the priority instructions and perform attenuation processing on the subordinate instructions.

[0015] The technical effects and advantages of the optimized preparation process of a composite plant powder with sugar-controlling effect according to the present invention are as follows: By acquiring batch attribute data of each plant raw material in the current batch and target product standard data, the physicochemical testing information of plant raw materials in three aspects—sugar control attributes, sensory attributes, and processing attributes—was systematically collected. This provides a data foundation for addressing batch-to-batch attribute fluctuations of plant raw materials caused by natural factors such as origin, year, and harvest season. By calculating the multi-attribute deviation matrix of each plant raw material in the current batch relative to the target product standard, a structured and quantitative description of batch-to-batch differences in raw materials was achieved, providing precise input for subsequent collaborative preparation decisions. By constructing a synergistic influence relationship knowledge base, the synergistic effects, sensory antagonisms, and processing interactions among plant raw materials in three dimensions—sugar control efficacy, sensory quality, and processing performance—were systematically integrated, and prior knowledge of the domain was incorporated into the data. The collaborative constraint encoding vector is injected into the training and inference process of the collaborative preparation decision model, enabling the model to explicitly consider the multidimensional interaction effects between raw materials when generating feed ratio and crushing particle size decisions. This effectively overcomes the shortcomings of treating each raw material as an independent variable and ignoring the collaborative relationship between them. The feed ratio and particle size joint decision layer in the collaborative preparation decision network uses the feed ratio and crushing particle size as coupled decision variables for joint optimization. This effectively overcomes the shortcomings of separating ratio optimization and particle size setting into two independent steps and failing to capture the collaborative influence between ratio adjustment and particle size adjustment. The introduction of quality consistency auxiliary loss enables the model to learn the causal mapping relationship between process parameters and finished product quality while optimizing the prediction accuracy of process parameters, giving the model the ability to predict the consequences of process decisions. The graded gradient pulverization strategy avoids the degradation of functional components caused by excessive pulverization in a single step. A synergistic influence-based homogeneous mixing strategy determines the mixing order based on the sensory antagonism between raw materials, mitigating sensory quality degradation caused by localized high-concentration antagonistic effects. An online process control and detection mechanism ensures the consistency of powder mixing quality through real-time verification of mixing uniformity. Compensation residual assessment and deviation source analysis systematically identify the source structure and dominant factors of residual deviations between the finished product quality and target standards after the execution of the collaborative preparation plan. Based on a directional correction instruction set, precise directional correction of the feed ratio and pulverization particle size is achieved. The correction instruction conflict detection and coordination mechanism effectively solves the instruction contradictions that may arise when optimizing multiple attributes simultaneously. The online parameter fine-tuning mechanism of the collaborative preparation decision model allows the model to continuously absorb process correction feedback from each production batch, constantly accumulating experience in adapting to raw material batch fluctuations, achieving continuous evolution of collaborative preparation decision-making capabilities. Ultimately, it outputs a collaborative preparation process optimization scheme that comprehensively considers sugar control efficacy, sensory experience, and processing performance, providing a systematic intelligent process optimization method for the stable industrial production of composite plant powders. Attached Figure Description

[0016] Figure 1This is a flowchart of an optimized method for the synergistic preparation of composite plant powder with sugar-controlling effect, as described in Example 1 of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1:

[0018] Please see Figure 1 As shown in this embodiment, the optimized preparation process of a composite plant powder with sugar-controlling effect includes: Step S1: Obtain batch attribute data and target product standard data for each plant raw material in the current batch, including target sugar control index, target sensory index and target processing index.

[0019] Methods for obtaining batch attribute data for each plant material in the current batch include: Obtain the raw material testing reports for each plant raw material in the current batch from the raw material testing system (i.e., the laboratory information management platform used for physicochemical testing and quality analysis of incoming plant raw materials). Plant raw materials refer to natural plant-derived raw materials used to prepare composite plant powders, including but not limited to mulberry leaves, bitter melon, buckwheat, kudzu root, corn silk, and yam, which have sugar-controlling or auxiliary processing functions. The current batch refers to the batch of a complete set of plant raw materials used in this production. The raw material testing report for each plant raw material includes basic raw material information and physicochemical testing data. Specifically, the basic information of raw materials includes the unique code of the raw material, the name of the raw material, the place of origin, the harvest season, the date of entry into storage and the batch number of entry into storage; the unique code of the raw material is a predefined unique identification code of the plant raw material; the place of origin information includes the province of origin and the altitude of origin; the harvest season includes spring, summer, autumn and winter; the physicochemical test data of the raw materials includes sugar control property test data, sensory property test data and processing property test data; the basic information of the raw materials and the physicochemical test data of each plant raw material are integrated to form the batch attribute data of each plant raw material in the current batch.

[0020] The blood sugar control property testing data is used to record the content information of functional components related to blood sugar control efficacy in plant raw materials, specifically including total polyphenol content, total flavonoid content, etc. - Glucosidase inhibition rate and dietary fiber content; where total polyphenol content is the mass concentration of polyphenolic compounds in plant raw materials; total flavonoid content is the mass concentration of flavonoid compounds in plant raw materials; - The glucosidase inhibition rate of plant raw material extract at standard concentrations... - Percentage of inhibition of glucosidase activity; dietary fiber content is the total mass concentration of insoluble and soluble dietary fiber in plant materials; Sensory attribute testing data is used to record sensory characteristics of plant-based raw materials related to the product's taste experience. Specifically, this includes bitterness intensity, grassy aroma intensity, color brightness, and particle roughness level. The bitterness intensity value is the sensory score for bitterness determined by a pre-defined sensory evaluation panel using a standardized sensory scoring method. The grassy aroma intensity value is the sensory score for grassy odor determined by the same panel. The color brightness value is the brightness parameter of the raw material powder measured using a colorimeter. The particle roughness level is the degree of roughness of the raw material powder in the oral cavity, determined by the same panel using a standardized sensory scoring method, including fine, slightly coarse, medium coarse, and coarse. The sensory evaluation panel is a professional organization composed of trained evaluators who objectively and consistently evaluate the sensory characteristics of the product according to standardized procedures based on national standards or industry specifications. The composition of the sensory evaluation panel and the standardized sensory scoring methods are both implemented in accordance with national standards for food sensory evaluation. Processing property test data is used to record physical property information related to the powder processing performance of plant raw materials, specifically including raw material moisture content, angle of repose, moisture absorption rate, and bulk density. Among them, the raw material moisture content is the mass percentage of water in the plant raw material; the angle of repose is the angle between the generatrix of the cone formed by the powder in a free-stacking state and the base, which is used to reflect the flow properties of the plant raw material powder; the moisture absorption rate is the increase in the mass of moisture absorbed by the plant raw material powder per unit time under standard temperature and humidity conditions; and the bulk density is the apparent density of the plant raw material powder in a natural stacking state.

[0021] Methods for obtaining standard data for target products include: The target product standard data is obtained from the product standard management system (i.e., a standardized management platform used to store and manage the quality standards and specifications of functional food products). The target product standard data specifies the quality requirements that the compound plant powder must meet in three aspects: sugar control efficacy, sensory quality, and processing performance. Specifically, it includes target sugar control indicators, target sensory indicators, and target processing indicators. The specific values ​​of each indicator in the target product standard data are all pre-set by those skilled in the art based on product efficacy requirements, consumer acceptance survey results, and processing technology feasibility. The target sugar control indicators include the target ranges for total polyphenol content, total flavonoid content, lower limit of inhibition rate, and dietary fiber content in the finished product. Each target range includes the minimum and maximum allowable values ​​for the corresponding indicator. The lower limit of inhibition rate is the target value for the compound plant powder under standard brewing conditions. -Minimum allowable value for glucosidase inhibition rate; The target sensory indicators include the upper limit of the bitterness intensity of the finished product, the upper limit of the grassy aroma intensity of the finished product, the target range of the color brightness of the finished product, and the target level of the particle roughness of the finished product. Among them, the upper limit of the bitterness intensity of the finished product is the highest permissible value of the bitterness sensory score after the compound plant powder is brewed; the upper limit of the grassy aroma intensity of the finished product is the highest permissible value of the grassy aroma sensory score after the compound plant powder is brewed; and the target level of the particle roughness of the finished product is the highest permissible level of the roughness of the compound plant powder in the oral cavity. The target processing indicators include the upper limit of the finished product moisture content, the upper limit of the finished product angle of repose, the upper limit of the finished product moisture absorption rate, and the target range of the finished product bulk density; among them, the upper limit of the finished product moisture content is the maximum moisture content allowed to ensure the storage stability of the composite plant powder; the upper limit of the finished product angle of repose is the maximum angle of repose allowed to ensure the filling flowability of the composite plant powder.

[0022] Step S2: Based on the batch attribute data and the target product standard data, calculate the attribute deviation values ​​of each plant raw material in the current batch relative to the target sugar control index, target sensory index and target processing index, and form a multi-attribute deviation matrix for the current batch.

[0023] Methods for calculating the property deviation values ​​of each plant material relative to the target sugar control index include: For each plant material, the sugar control attribute test data is compared with the target index in the target sugar control index, and the sugar control attribute deviation value is calculated. Specifically, for a blood sugar control detection indicator with a target range, the difference between the value of the corresponding blood sugar control detection indicator and the center value of the corresponding target range is calculated to obtain the absolute deviation of blood sugar control; where the center value of the target range is the average of the allowable minimum and maximum values ​​of the corresponding target range. The ratio between the absolute deviation of blood sugar control and the half-width of the target range is calculated to obtain the relative deviation of blood sugar control; each relative deviation of blood sugar control is used as the blood sugar control attribute deviation value of the corresponding blood sugar control detection indicator; where the half-width of the target range is half the difference between the allowable maximum and minimum values ​​of the corresponding target range. If the absolute value of the relative deviation of blood sugar control is less than or equal to one, it indicates that the corresponding blood sugar control detection indicator is within its target range; if the absolute value of the relative deviation of blood sugar control is greater than one, it indicates that the corresponding blood sugar control detection indicator deviates from its target range. For a blood sugar control test indicator with a target lower limit, calculate the difference between the corresponding value and the target lower limit, and then divide it by the target lower limit to obtain the lower limit deviation rate. Each lower limit deviation rate is used as the blood sugar control attribute deviation value of the corresponding blood sugar control test indicator. If the lower limit deviation rate is greater than or equal to zero, it indicates that the corresponding blood sugar control test indicator meets its target lower limit requirement; if the lower limit deviation rate is less than zero, it indicates that the corresponding blood sugar control test indicator has not met its target lower limit requirement.

[0024] Methods for calculating the sensory attribute deviation values ​​of each plant material relative to the target sensory index include: Each sensory indicator in the sensory attribute detection data corresponding to each plant raw material is compared with the target indicator in the target sensory indicator, and the sensory attribute deviation value is calculated. Specifically, for a sensory detection index with a target upper limit, the difference between the corresponding value and the target upper limit is calculated, and then divided by the target upper limit to obtain the upper limit deviation rate. Each upper limit deviation rate is used as the sensory attribute deviation value of the corresponding sensory detection index. If the upper limit deviation rate is less than or equal to zero, it indicates that the corresponding sensory detection index meets its target upper limit requirement; if the upper limit deviation rate is greater than zero, it indicates that the corresponding sensory detection index exceeds its target upper limit. For sensory detection indicators with target ranges, the same calculation method as for the deviation value of sugar control attributes is used to calculate the sensory attribute deviation value corresponding to each sensory detection indicator in turn. For the particle roughness level, a preset level value mapping table is used, which contains the values ​​corresponding to each particle roughness level. This table is preset by those skilled in the art according to the sensory evaluation system. For example, fineness corresponds to value 1, slightly coarse to value 2, medium coarse to value 3, and coarse to value 4. From the level value mapping table, the values ​​corresponding to the particle roughness level and the target particle roughness level of the finished product are obtained to obtain the detection level value and the target level value. The difference between the detection level value and the target level value is calculated and then divided by the target level value to obtain the level deviation rate. The level deviation rate is used as the sensory attribute deviation value of particle roughness.

[0025] Methods for calculating the processing attribute deviation values ​​of each plant raw material relative to the target processing index include: For each processing indicator in the processing attribute test data corresponding to each plant raw material, compare it item by item with the target indicator in the target processing indicator, and calculate the processing attribute deviation value; the calculation method of the processing attribute deviation value is the same as the calculation method of the sugar control attribute deviation value and the sensory attribute deviation value.

[0026] Methods for generating the multi-attribute deviation matrix for the current batch include: All sugar control attribute deviation values ​​corresponding to each plant raw material are arranged in the order of sugar control detection indicators in the sugar control attribute detection data to form a sugar control deviation vector for each plant raw material; all sensory attribute deviation values ​​corresponding to each plant raw material are arranged in the order of sensory detection indicators in the sensory attribute detection data to form a sensory deviation vector for each plant raw material; all processing attribute deviation values ​​corresponding to each plant raw material are arranged in the order of processing detection indicators in the processing attribute detection data to form a processing deviation vector for each plant raw material. The sugar control deviation vector, sensory deviation vector, and processing deviation vector of each plant material are concatenated sequentially to form a comprehensive deviation vector for each plant material. The comprehensive deviation vectors of all plant materials are arranged row by row according to the order of the material name to form the multi-attribute deviation matrix of the current batch. The number of rows in the multi-attribute deviation matrix is ​​equal to the number of plant material types, and the number of columns is equal to the total dimension of all attribute deviation values. The multi-attribute deviation matrix is ​​used to systematically describe the deviation of each plant material in the current batch from the target product standard in terms of sugar control attribute, sensory attribute, and processing attribute.

[0027] Step S3: Call the pre-constructed collaborative preparation decision model, establish the synergistic influence relationship of each plant raw material on sugar control attributes, sensory attributes and processing attributes, and solve the feeding ratio and crushing target mesh size to meet the standard data of the target product based on the multi-attribute deviation matrix, so as to form the collaborative preparation plan for the current batch.

[0028] Methods for pre-constructing collaborative preparation decision models include: The historical production records for all historical production batches are retrieved from the historical production database (i.e., the production data management platform used to store and manage raw material data, process parameters, and finished product test results for historical production batches of compound plant powder). Each historical production record includes a historical batch number, historical batch attribute data for each plant raw material, historical feed ratio, historical pulverization particle size parameters, and historical finished product test results. The historical feed ratio is the mass percentage of each plant raw material in the historical batch. The historical pulverization particle size parameters include the target mesh size for each plant raw material, which refers to the quantitative process parameter of the target powder fineness set in the production plan. The historical finished product test results include historical finished product sugar control test data, historical finished product sensory test data, and historical finished product processing test data. The data structure of each test data in the historical finished product test results corresponds to the structure of each target indicator in the target sugar control index, target sensory index, and target processing index, respectively.

[0029] A synergistic relationship table for sugar control, a sensory antagonism table, and a processing interaction table are pre-defined among plant raw materials. These tables are then integrated to form a synergistic influence relationship knowledge base. Based on this knowledge base, a synergistic constraint coding vector is constructed. Specifically, the blood sugar control synergy table is used to record the synergistic or antagonistic effects of different plant raw material combinations on blood sugar control efficacy. It contains multiple synergy records, each of which includes a raw material combination identifier, synergy type, and synergy coefficient. The synergy type includes synergistic enhancement and antagonistic reduction. The synergy coefficient is used to quantify the rate of change in blood sugar control efficacy after combining two plant raw materials. A synergy coefficient greater than one indicates synergistic enhancement, and a coefficient less than one indicates antagonistic reduction. Each synergy record is pre-set by a person skilled in the art based on pharmacological experimental data and compound efficacy verification results. The sensory antagonism table is used to record the interactive effects of different combinations of plant materials on sensory quality. It contains multiple antagonism records, each of which includes a material combination identifier, antagonistic attribute, and antagonistic strength coefficient. The antagonistic attribute is the name of the sensory indicator affected by the interaction. The antagonistic strength coefficient is used to quantify the change in the corresponding sensory indicator after the combination of two plant materials. Each antagonism record is pre-set by a person skilled in the art based on cross-experimental data of sensory evaluation. The processing interaction table records the interactive effects of different plant raw material particle sizes on the processing performance of the mixed powder. It contains multiple interaction records, each including a raw material combination identifier, a particle size combination range, affected processing indicators, and an interaction coefficient. The particle size combination range identifies the combination of the two plant raw materials at different target mesh sizes. The interaction coefficient quantifies the degree of deviation of the corresponding particle size combination from the affected processing indicators. Each interaction record is pre-set by those skilled in the art based on cross-experimental data of powder properties. For each historical production record, the combinations of plant raw materials involved in the feeding are determined according to the corresponding historical feeding ratio. For each combination of plant raw materials, the corresponding synergy coefficient is obtained from the sugar control synergy relationship table, the corresponding antagonism intensity coefficient is obtained from the sensory antagonism relationship table, and the corresponding interaction influence coefficient is matched from the processing interaction influence table according to the historical crushing particle size parameters. The synergy coefficient, antagonism intensity coefficient and interaction influence coefficient corresponding to all plant raw material combinations for each historical production record are arranged in a predetermined order to form the synergy constraint coding vector corresponding to each historical production record.

[0030] Based on the historical batch attribute data and target product standard data in each historical production record, calculate the multi-attribute deviation matrix corresponding to each historical production record and mark it as the historical multi-attribute deviation matrix; flatten the historical multi-attribute deviation matrix of each historical production record into a one-dimensional vector by row splicing to form the historical deviation input vector of the corresponding historical production record; splice the historical feeding ratio and historical crushing particle size parameters of each historical production record to form the historical process parameter vector of the corresponding historical production record; arrange the values ​​in the historical finished product test results of each historical production record in order to form the historical finished product quality vector of the corresponding historical production record. The historical deviation input vector and the collaborative constraint encoding vector of each historical production record are concatenated to form the model input feature vector of the corresponding historical production record, and the historical process parameter vector of each historical production record is used as the model output label vector. The model input feature vector and the model output label vector of each historical production record are integrated to form the decision training sample of the corresponding historical production record. The decision training samples of all historical production records are summarized to form the decision training sample set.

[0031] A collaborative preparation decision network is constructed, comprising a bias-aware encoding layer, a collaborative constraint fusion layer, a proportioning granularity joint decision layer, and a constraint verification output layer; specifically: The deviation-aware coding layer is used to perform non-linear encoding on the historical deviation input vector part of the model input feature vector, extract the implicit correlation features between the deviation patterns of each raw material, and output the deviation coding representation; the deviation-aware coding layer contains multiple fully connected layers and non-linear activation functions. The collaborative constraint fusion layer is used to perform cross-attention fusion of the deviation encoding representation and the collaborative constraint encoding vector, enabling the model to explicitly consider the synergistic effects, sensory antagonism, and processing interactions among raw materials when generating process decisions. Specifically, the collaborative constraint fusion layer adopts a cross-attention mechanism, using the deviation encoding representation as the query vector and the collaborative constraint encoding vector as the key and value vectors. It calculates the attention weight between the deviation encoding representation and the collaborative constraint encoding vector, and performs weighted aggregation on the collaborative constraint encoding vector based on the attention weight to obtain the collaborative perception feature representation. The deviation encoding representation and the collaborative perception feature representation are then concatenated to form the fused feature representation. The cross-attention mechanism is a well-known technology in this field, and the specific calculation process will not be elaborated here. It should be understood that the core role of the collaborative constraint fusion layer is to inject the domain prior collaborative knowledge among plant raw materials into the decision-making process in a structured manner, so that the model can not only make decisions based on raw material deviation data, but also perceive the interaction effects of different combinations of plant raw materials in the three dimensions of sugar control, sensory effects, and processing. This is different from the shortcomings of traditional optimization methods that treat each material raw material as an independent variable and ignore the collaborative relationships among plant raw materials. The joint decision layer for feed ratio and particle size is connected after the collaborative constraint fusion layer. It is used to output the predicted values ​​of feed ratio and crushing target mesh size based on the fusion feature representation. The joint decision layer for feed ratio and particle size includes a feed ratio prediction branch and a particle size prediction branch. The feed ratio prediction branch contains multiple fully connected layers and normalized activation functions to output the predicted feed ratio values ​​of each plant material, and the sum of all the predicted feed ratio values ​​is equal to one. The particle size prediction branch contains multiple fully connected layers to output the predicted value of the crushing target mesh size of each plant material. It should be understood that the joint decision layer for feed ratio and particle size uses the feed ratio and crushing target mesh size as coupled decision variables for joint output, so that the model can learn the synergistic influence relationship between feed ratio adjustment and crushing particle size adjustment, which is different from the traditional method that separates the feed ratio optimization and particle size setting into two independent steps. The constraint verification output layer is connected after the proportioning and particle size joint decision layer. It is used to perform feasibility constraint verification on the predicted values ​​of the feed proportions and the predicted values ​​of the target mesh size for crushing. Specifically, a preset feed proportion constraint set is provided, which includes the allowable range of feed proportions for each plant material. Each allowable range of feed proportions is preset by those skilled in the art based on the requirements of functional food formulation regulations and the feasibility of the production process. A preset particle size constraint set is provided, which includes the allowable range of the target mesh size for crushing each plant material. Each allowable range of the target mesh size for crushing is preset by those skilled in the art based on the capacity of the crushing equipment and the requirements of the powder properties. If any predicted feed proportion value exceeds the corresponding allowable feed proportion range, the corresponding predicted feed proportion value is truncated to the nearest boundary value of the corresponding allowable range, and all predicted feed proportion values ​​are re-normalized so that the sum of all predicted feed proportion values ​​equals one. If any predicted target mesh size for crushing exceeds the corresponding allowable target mesh size for crushing, the corresponding target mesh size is truncated to the nearest boundary value of the corresponding allowable range.

[0032] Based on the decision training sample set, the collaborative preparation decision network is trained, and the trained collaborative preparation decision network is used as the collaborative preparation decision model. Specifically, the model input feature vector of each decision training sample is input into the collaborative preparation decision network, and the corresponding predicted value of the feed ratio and the predicted value of the crushing target mesh size are output. The mean square error between each predicted value of the feed ratio and the historical feed ratio in the corresponding model output label vector is calculated to obtain the ratio prediction loss corresponding to each decision training sample. The mean square error between each predicted value of the crushing target mesh size and the historical crushing target mesh size in the corresponding model output label vector is calculated to obtain the particle size prediction loss corresponding to each decision training sample. A pre-set auxiliary loss weight for quality consistency is established, which is pre-set by those skilled in the art based on the correlation strength between process decisions and finished product quality. Based on the predicted values ​​of the feed ratio and the target mesh size of the crushed material, and combined with the synergy coefficient, antagonism strength coefficient, and interaction influence coefficient in the synergy influence relationship knowledge base, a finished product quality prediction vector is calculated. Specifically, for each sugar control detection index, the values ​​of the corresponding sugar control detection index for each raw material are weighted and summed according to the predicted feed ratio of each plant material to obtain the basic mixed sugar control value corresponding to each sugar control detection index. For each combination of plant materials involved in the feed, the corresponding synergy coefficient is obtained from the sugar control synergy relationship table, and the synergy coefficient for each plant material combination is calculated. The product of the predicted feed ratios is multiplied by the corresponding synergy coefficient to obtain the synergistic correction amount for each plant raw material combination. All synergistic correction amounts are summed and added to each basic mixed sugar control value to obtain the synergistically corrected sugar control estimate for each sugar control detection index. Using the same method, combined with the sensory antagonism table and the processing interaction table, the sensory estimate for each sensory detection index and the processing estimate for each processing detection index are calculated. All sugar control estimates, sensory estimates, and processing estimates are arranged to form a finished product quality prediction vector. The mean square error between the finished product quality prediction vector and the corresponding historical finished product quality vector is calculated to obtain the quality consistency auxiliary loss. The sum of the ratio prediction loss and the granularity prediction loss is calculated, and then the sum is added to the product of the quality consistency auxiliary loss and its weight to obtain the joint decision loss for each decision training sample. The quality consistency auxiliary loss introduces a supervisory signal from the finished product quality dimension, forcing the model to learn the causal mapping relationship between process parameters and finished product quality while optimizing the prediction accuracy of process parameters, thus enabling the model to predict the consequences of process decisions. The mean of the joint decision losses for all decision training samples is calculated to obtain the batch joint decision loss. Using the backpropagation algorithm and gradient descent optimization method, the model parameters in the collaborative preparation decision network are iteratively updated based on the batch joint decision loss until the batch joint decision loss converges to a preset loss convergence threshold or the number of training iterations reaches a preset maximum number of iterations, at which point the collaborative preparation decision network training is complete. The loss convergence threshold and the maximum number of iterations are preset by those skilled in the art based on the model training stability requirements. It should be noted that the backpropagation algorithm and gradient descent optimization method are well-known techniques in the field and will not be elaborated upon here.

[0033] The methods for formulating the co-production plan for the current batch include: The multi-attribute deviation matrix of the current batch is concatenated and flattened into a one-dimensional vector to form the deviation input vector of the current batch. Based on the raw material names of each plant material in the current batch, the combinations of plant materials participating in the feeding are determined. The synergy coefficient, antagonism strength coefficient, and interaction influence coefficient corresponding to each plant material combination are obtained from the synergy influence relationship knowledge base to construct the synergy constraint encoding vector of the current batch. For the particle size-related interaction influence coefficient, since the particle size parameters for the current batch have not yet been determined, the interaction influence coefficient corresponding to the median value of the allowable range of the target mesh size is used as the interaction influence coefficient in the corresponding synergy constraint encoding vector of the current batch. The deviation input vector of the current batch is concatenated with the synergy constraint encoding vector of the current batch to form the model input feature vector of the current batch. The model input feature vector of the current batch is input into the synergy preparation decision model, and the synergy preparation decision model outputs the predicted feeding ratio and predicted target mesh size of each plant material. The update iteration process is executed to obtain the updated predicted feed ratio and the predicted target mesh size. The absolute values ​​of the differences between each element of the predicted feed ratio and the predicted target mesh size are compared with a preset iteration convergence accuracy threshold. The convergence condition is then determined based on the comparison results. If the convergence condition is met, the update iteration process is stopped, and the final predicted feed ratio and the predicted target mesh size are output. If the convergence condition is not met, the update iteration process continues until the convergence condition is met or the number of iterations reaches the preset maximum number of updates. The update iteration process is as follows: based on collaborative... The predicted mesh size of each plant raw material output by the preparation decision model is re-obtained from the processing interaction influence table. The collaborative constraint coding vector is updated based on the re-obtained interaction influence coefficient. The updated collaborative constraint coding vector is then re-concatenated with the current batch deviation input vector and input back into the collaborative preparation decision model. The convergence condition is that the absolute value of the difference between each element of the predicted feed ratio and the predicted mesh size of the crushing target is less than the iterative convergence accuracy threshold. The iterative convergence accuracy threshold and the maximum number of updates are both preset by those skilled in the art according to the process decision accuracy requirements. The final predicted values ​​of the feed ratios of each plant material are integrated with the predicted values ​​of the target mesh size for crushing to form a collaborative preparation plan for the current batch. The collaborative preparation plan is used to guide the setting of the feed ratios and crushing particle sizes of each plant material in the current batch.

[0034] Step S4: According to the collaborative preparation plan, the plant raw materials of the current batch are dried, pulverized, sieved and homogenized to obtain composite plant powder samples. The actual sugar control index, actual sensory index and actual processing index of the composite plant powder samples are tested to form a batch verification result set.

[0035] The methods for drying the plant materials in the current batch include: Based on the predicted feed ratio of each plant material in the collaborative preparation plan, the moisture content of each plant material is obtained from the batch attribute data; a target moisture content mapping table is preset, which includes the target moisture content and recommended drying temperature for each plant material; each target moisture content and recommended drying temperature are preset by those skilled in the art based on the heat sensitivity and thermal stability of the functional components of different plant materials; wherein, the target moisture content is the moisture content level that the plant material needs to reach before pulverization to ensure subsequent pulverization efficiency and powder quality; For each type of plant material, the corresponding moisture content is compared with the corresponding target moisture content. If the moisture content is greater than the target moisture content, the corresponding plant material needs to be dried. If the moisture content is less than or equal to the target moisture content, the corresponding plant material does not need to be dried and can proceed directly to the pulverizing process. For plant materials that need to be dried, the recommended drying temperature is obtained from the target moisture content mapping table, and the operating parameters of the drying equipment are set according to the recommended drying temperature and the target moisture content to perform the drying process until the moisture content of the corresponding plant material drops below the target moisture content. After drying, the moisture content of the dried plant material is retested. If the retested moisture content is still greater than the target moisture content, the drying process continues. If the retested moisture content is less than or equal to the target moisture content, the drying is deemed qualified.

[0036] Methods for implementing graded gradient pulverization strategies and sieve grading processes include: A pre-set pulverization and grading strategy table is provided, which includes the number of pulverization stages and pulverization parameters for different target mesh size ranges. Among them, the pulverization parameters for each stage include the type of pulverizing equipment, the target mesh size, and the pulverizing speed. The pulverization and grading strategy table is pre-set by those skilled in the art based on the performance of the pulverizing equipment and the characteristics of plant fibers. Gradient pulverization refers to dividing the pulverization process into multiple progressive stages, with each stage gradually reducing the particle size from coarse to the target particle size, in order to avoid local overheating and degradation of functional components caused by excessive pulverization at one time. For each type of plant material, based on the predicted target mesh size range, the corresponding pulverization level and parameters for each level are obtained from the pulverization grading strategy table. The plant material is then pulverized sequentially using the corresponding pulverization equipment type and pulverization speed, from the first level to the last. After each pulverization stage, the particle size of the pulverized product is measured to obtain the corresponding mesh size. If the mesh size is greater than or equal to the target mesh size for the corresponding level, the pulverization proceeds to the next level or is deemed complete. If the mesh size is less than the target mesh size for the corresponding level, pulverization continues at the current level until the requirements are met. Based on the predicted mesh size of the crushing target in the collaborative preparation plan, the crushed products of each plant material are subjected to sieving and grading. Specifically, for the crushed products of each plant material, a standard sieve with the same mesh size as the corresponding predicted mesh size is used for sieving. The crushed products that pass through the sieve are marked as qualified powder and enter the mixing process. The crushed products that do not pass through the sieve are returned to the crushing equipment for secondary crushing until all crushed products pass through the standard sieve with the corresponding mesh size.

[0037] Methods for performing synergistic effect-aware homogeneous mixing include: Based on the sensory antagonism table in the synergistic influence knowledge base, it is determined whether there is a sensory antagonism relationship between the plant materials. If there is a combination of plant materials with sensory antagonism, a step-by-step progressive mixing strategy is implemented. Specifically, a mixing priority rule is preset to determine the order in which the plant materials are added. The mixing priority rule is preset by a person skilled in the art based on the sensory antagonism mechanism and powder hybrid dynamics. The core principle of the mixing priority rule is: first, the combination of plant materials with no sensory antagonism or the smallest antagonism intensity coefficient is premixed to form a basic mixture; then, the plant materials with sensory antagonism with the basic mixture are added in order of increasing antagonism intensity coefficient, so that the highly antagonistic plant materials are uniformly dispersed in the fully mixed basic mixture environment, thereby alleviating the local high concentration antagonistic effect. For each addition of plant raw materials, the corresponding feed mass is calculated according to the predicted feed ratio of the corresponding plant raw materials in the collaborative preparation plan. Specifically, the total feed mass of the current batch is preset, which is pre-set by those skilled in the art based on the production batch plan. The feed mass of each plant raw material is obtained by multiplying the predicted feed ratio of each plant raw material by the total feed mass. According to the addition order determined by the mixing priority rule, the qualified powder of each plant raw material is added to the homogenizing mixing equipment in sequence. After each addition of new plant raw materials, a homogenizing mixing operation with a preset mixing time is performed to ensure that the newly added plant raw materials are fully and evenly mixed with the existing mixture. The mixing time is preset by those skilled in the art based on the performance of the homogenizing mixing equipment and the requirements for powder mixing uniformity. After all plant raw materials have been added, a final homogenizing mixing operation is performed to obtain a composite plant powder sample. If there is no combination of plant raw materials with sensory antagonism, then all qualified plant raw materials powders are added to the homogenizing mixing equipment at the corresponding feed weights, and the homogenizing mixing operation is performed to obtain a composite plant powder sample.

[0038] Online process control testing is performed on the composite plant powder samples. Specifically, after the final homogenization mixing operation is completed, random samples are taken from the composite plant powder samples for mixing uniformity testing. The method for testing mixing uniformity is as follows: multiple samples are taken from different locations of the composite plant powder samples, and the content of the characteristic functional components in each sample is tested. The coefficient of variation of the content of the characteristic functional components in multiple samples is calculated. The characteristic functional components are representative components selected by those skilled in the art from total polyphenols or total flavonoids according to the characteristics of the product formulation. The coefficient of variation is the ratio of the standard deviation to the mean of the content of the corresponding characteristic functional components in multiple samples. A preset mixing uniformity judgment threshold is set by those skilled in the art according to the powder mixing quality control standards. If the coefficient of variation is less than or equal to the mixing uniformity judgment threshold, the mixing uniformity is deemed qualified. If the coefficient of variation is greater than the mixing uniformity judgment threshold, the homogenization mixing operation is continued and samples are resampled and tested until the coefficient of variation is less than or equal to the mixing uniformity judgment threshold.

[0039] Methods for generating batch validation result sets include: Samples were taken from the compound plant powder samples, and the actual sugar control indicators, actual sensory indicators, and actual processing indicators were tested according to the testing methods specified in national or industry standards. The actual sugar control indicators included the actual total polyphenol content, the actual total flavonoid content, and the actual... - Glucosidase inhibition rate and actual dietary fiber content; actual sensory indicators include actual bitterness intensity value, actual grassy smell intensity value, actual color brightness value and actual particle roughness grade; actual processing indicators include actual moisture content, actual angle of repose, actual moisture absorption rate and actual bulk density. In response to the actual situation - Glucosidase inhibition rate, verified by enzyme activity inhibition under standard brewing conditions; specifically, preset standard brewing parameters, including brewing water temperature, brewing concentration, and brewing time; the standard brewing parameters are preset by those skilled in the art according to the product instructions; the compound plant powder sample is brewed according to the standard brewing parameters, and the brewed compound plant powder sample is then analyzed. - Detection of glucosidase inhibition rate to obtain actual... -Glucosidase inhibition rate; For the actual sensory indicators, a pre-designated sensory evaluation team scores them according to standardized sensory evaluation methods to obtain the actual bitterness intensity value, the actual grassy smell intensity value, and the actual particle roughness level. The actual sugar control indicators, actual sensory indicators, and actual processing indicators are integrated to form a batch validation result set. The batch validation result set is used to record the actual test results of the compound plant powder samples in terms of sugar control efficacy, sensory quality, and processing performance, providing experimental feedback for the adaptive correction of the subsequent collaborative preparation decision model.

[0040] Step S5: Based on the deviation between the batch verification result set and the target product standard data, evaluate the compensation residual of the current batch, and adaptively correct the collaborative preparation decision model based on the compensation residual, and output the final collaborative preparation process optimization scheme for the current batch.

[0041] Methods for evaluating the compensated residuals of the current batch include: The actual test values ​​of each indicator in the batch verification results are compared with the corresponding target indicators in the target product standard data item by item, and the verification deviation value of each indicator is calculated. The calculation method of the verification deviation value is the same as that of the attribute deviation value. All verification deviation values ​​are arranged in the order of actual sugar control indicators, actual sensory indicators and actual processing indicators to form a verification deviation vector. Based on the verification deviation vector, the pass / fail status of each indicator is determined. Specifically, for indicators with a target range, if the absolute value of the verification deviation is less than or equal to one, it is considered a pass / fail indicator; if the absolute value of the verification deviation is greater than one, it is considered a fail / fail indicator. For indicators with a target upper limit and the actual particle roughness level, if the verification deviation is less than or equal to zero, it is considered a pass / fail indicator; if the verification deviation is greater than zero, it is considered a fail / fail indicator. For indicators with a target lower limit, if the verification deviation is greater than or equal to zero, it is considered a pass / fail indicator; if the verification deviation is less than zero, it is considered a fail / fail indicator. If all indicators are deemed qualified, the compensation residual of the current batch is determined to be in a zero residual state. The collaborative preparation plan for the current batch is directly used as the final collaborative preparation process optimization scheme for the current batch, without the need for adaptive correction of the collaborative preparation decision model. If there are non-compliant indicators, the corresponding deviation source attribution dimension is determined according to the attribute dimension to which each non-compliant indicator belongs. Specifically, if the attribute dimension to which the non-compliant indicator belongs is an actual blood sugar control indicator, then the deviation source attribution dimension is the blood sugar control dimension; if the attribute dimension to which the non-compliant indicator belongs is an actual sensory indicator, then the deviation source attribution dimension is the sensory dimension; if the attribute dimension to which the non-compliant indicator belongs is an actual processing indicator, then the deviation source attribution dimension is the processing dimension. For each non-compliant indicator, a deviation source analysis was performed to determine the dominant deviation factor for that non-compliant indicator. Specifically, for non-compliant indicators in the sugar control dimension, the synergistic correction amount corresponding to each plant material combination was recalculated based on the predicted values ​​of the feed ratios of each plant material in the synergistic preparation plan and the synergistic coefficients in the synergistic influence relationship knowledge base. The recalculated synergistic correction amounts for each plant material combination were sorted from largest to smallest, and the plant material combination ranked first was selected, with each plant material in it identified as the dominant deviation material for sugar control deviation. The product of the predicted feed ratio of each dominant deviation material and the verification deviation value of the corresponding non-compliant indicator was calculated to obtain the deviation contribution of each dominant deviation material. The deviation contributions of all dominant deviation materials were summarized to form the dominant deviation factor for sugar control of the corresponding non-compliant indicator. For non-compliant indicators in the sensory dimension, all plant material combinations that have an antagonistic relationship with the corresponding non-compliant indicator and participate in the feed, along with their corresponding antagonistic strength coefficients, were obtained from the sensory antagonism relationship table. The plant material combinations were sorted from largest to smallest according to the antagonistic strength coefficients, and the plant material combination ranked first was selected, with each plant material in it identified as the dominant deviation material for sugar control deviation. The raw materials are used as the dominant raw materials for sensory deviation. Using the same method as for sugar control, the dominant factors for sensory deviation of the corresponding non-compliant indicators are calculated. For non-compliant indicators in the processing dimension, all combinations of plant raw materials that interact with the corresponding non-compliant indicators and participate in the feeding process are obtained from the processing interaction influence table, along with their corresponding interaction influence coefficients, and marked as selected interaction influence coefficients. Based on the predicted target mesh size of each plant raw material in the current batch, the corresponding particle size combination range is matched. Based on the matched particle size combination range, the effective interaction influence coefficient is obtained from the selected interaction influence coefficients corresponding to each plant raw material combination. The effective interaction influence coefficients corresponding to each plant raw material combination are sorted from largest to smallest, and the plant raw material combination ranked first is selected. Each plant raw material in this combination is determined as the dominant raw material for processing deviation, and the predicted target mesh size of the dominant raw material is used as the dominant particle size parameter. Using the same method as for sugar control, the deviation contribution of each dominant raw material is calculated. The deviation contribution of all dominant raw materials and the dominant particle size parameter are summarized to form the dominant factors for processing deviation of the corresponding non-compliant indicators. Based on the sign of the verification deviation value corresponding to the non-conforming indicator, the direction of the deviation is determined. Specifically, if the verification deviation value is positive, the deviation direction is positive exceedance; if the verification deviation value is negative, the deviation direction is negative undersufficiency. The verification deviation values, deviation source attribution dimensions, deviation direction, and dominant factors of all non-conforming indicators are integrated to form the compensation residual for the current batch. The compensation residual is used to systematically describe the residual deviation between the finished product quality and the target standard and its source structure after the execution of the collaborative preparation plan for the current batch, providing a directional correction basis for subsequent adaptive correction of the model.

[0042] Methods for adaptively modifying collaborative preparation decision models based on compensation residuals include: Based on the dimension of the source of deviation and the main raw material of deviation for each non-conforming indicator in the compensation residual, a set of directional correction instructions is constructed. Specifically, for each non-conforming indicator, a corresponding ratio correction instruction and particle size correction instruction are generated according to the direction of deviation and the main raw material of deviation. All ratio correction instructions and particle size correction instructions are summarized to form a set of directional correction instructions. The method for generating proportion correction instructions is as follows: For each non-compliant indicator, the corresponding proportion correction magnitude is calculated based on the deviation direction, the dominant ingredient of the deviation, and the verification deviation value. Specifically, the absolute value of the verification deviation value of the corresponding non-compliant indicator is obtained and marked as the absolute amount of the proportion deviation. The influence coefficient of the dominant ingredient of the deviation on the corresponding non-compliant indicator is obtained from the synergistic influence relationship knowledge base. Wherein, if the non-compliant indicator belongs to the sugar control dimension, the influence coefficient is the synergistic coefficient corresponding to the plant raw material combination where the dominant ingredient of the deviation is located in the sugar control synergistic relationship table; if the non-compliant indicator belongs to the sensory dimension, the influence coefficient is the antagonistic strength coefficient corresponding to the plant raw material combination where the dominant ingredient of the deviation is located in the sensory antagonism relationship table; a proportion correction sensitivity coefficient is preset, which is determined by the technology in this field. The technicians pre-set the sensitivity of the response to the finished product quality based on the adjustment of the feed ratio; calculate the product of the absolute amount of the ratio deviation and the ratio correction sensitivity coefficient, and then divide it by the influence coefficient to obtain the ratio correction baseline amount; obtain the predicted feed ratio of the main raw material in the co-preparation plan, calculate the product of the ratio correction baseline amount and the predicted feed ratio, and obtain the ratio correction range; if the unqualified indicator belongs to the sugar control dimension and the deviation direction is negative and insufficient, the ratio correction range is set to a positive value, indicating that the feed ratio of the main raw material in the deviation is increased; if the unqualified indicator belongs to the sugar control dimension and the deviation direction is positive and exceeds the standard, or if the unqualified indicator belongs to the sensory dimension and the deviation direction is positive and exceeds the standard, the ratio correction range is set to a negative value, indicating that the feed ratio of the main raw material in the deviation is reduced; The method for generating particle size correction instructions is as follows: For each non-conforming indicator, the corresponding particle size correction magnitude is calculated based on the deviation direction, the dominant raw material of the deviation, and the verification deviation value. Specifically, the absolute value of the verification deviation value of the corresponding non-conforming indicator is obtained and marked as the absolute amount of particle size deviation. A particle size correction sensitivity coefficient is preset, which is pre-set by those skilled in the art based on the response sensitivity of the particle size adjustment to the finished product quality indicators. If the non-conforming indicator belongs to the processing dimension, the effective interaction influence coefficient corresponding to the dominant raw material of the deviation is obtained, the product of the absolute amount of particle size deviation and the particle size correction sensitivity coefficient is calculated, and then divided by the effective interaction influence coefficient. The mutual influence coefficient is used to obtain the particle size correction reference amount. A preset particle size-processing response direction mapping table is used, which contains the direction of change of the value of each affected processing index as the target mesh size increases. The direction of change includes positive and negative changes. A positive change indicates that the value of the corresponding processing index increases as the target mesh size increases, while a negative change indicates that the value of the corresponding processing index decreases as the target mesh size increases. The particle size-processing response direction mapping table is preset by those skilled in the art based on the laws of powder properties. From the particle size-processing response direction mapping table, the corresponding values ​​of unqualified indicators are obtained. The direction of change; if the direction of change is positive and the deviation direction is negative and insufficient, the particle size correction benchmark is set to a positive value to obtain the particle size correction range, which means increasing the target mesh size of the raw material with the deviation to increase the corresponding processing index value; if the direction of change is positive and the deviation direction is positive and excessive, the particle size correction benchmark is set to a negative value to obtain the particle size correction range, which means decreasing the target mesh size of the raw material with the deviation to decrease the corresponding processing index value; if the direction of change is negative and the deviation direction is negative and insufficient, the particle size correction benchmark is set to a negative value to obtain the particle size correction range, which means decreasing the target mesh size of the raw material with the deviation to reduce the deviation. The target mesh size is set to increase the corresponding processing index value. If the change direction is negative and the deviation direction is positive and exceeds the standard, the particle size correction benchmark is set to a positive value to obtain the particle size correction range, which means increasing the target mesh size of the raw material that dominates the deviation to reduce the corresponding processing index value. If the unqualified index belongs to the sugar control dimension and the deviation direction is negative and insufficient, or the unqualified index is the actual particle roughness, the product of the absolute amount of particle size deviation and the particle size correction sensitivity coefficient is calculated to obtain the particle size correction range, and the particle size correction range is set to a positive value, which means increasing the target mesh size of the raw material that dominates the deviation to improve the dissolution rate of functional components or reduce the particle roughness. The system performs conflict detection and coordination on the targeted correction instruction set. Specifically, it checks for contradictory correction instructions targeting the same plant material. Conflicting correction instructions refer to adjustments that are opposite in direction regarding the feed ratio or target mesh size, specifically manifested as ratio conflicts and particle size conflicts. A ratio conflict occurs when, for the same plant material, there are both correction instructions requiring an increase in the feed ratio and instructions requiring a decrease. A particle size conflict occurs when, for the same plant material, there are both correction instructions requiring an increase in the target mesh size and instructions requiring a decrease. If contradictory correction instructions for the same plant material exist, conflict coordination is performed. The conflict coordination method is as follows: For contradictory correction instructions corresponding to the same plant material, classify them into increase instructions and decrease instructions based on their corresponding correction directions; compare the absolute values ​​of the verification deviation values ​​of the non-conforming indicators corresponding to all increase instructions, and take the absolute value of the verification deviation value with the largest value as the increase deviation value; compare the absolute values ​​of the verification deviation values ​​of the non-conforming indicators corresponding to all decrease instructions, and take the absolute value of the verification deviation value with the largest value as the decrease deviation value; if the increase deviation value is greater than the decrease deviation value, mark the increase instruction as the priority instruction and the decrease instruction as the subordinate instruction; if the increase deviation value is less than the decrease deviation value, mark the decrease instruction as the priority instruction and the increase instruction as the subordinate instruction; if the increase deviation value is equal to the decrease deviation value, then classify the increase instructions and decrease instructions into priority instructions and subordinate instructions according to the preset decision priority dimension; specifically, Based on the decision priority dimension, the decision priorities for increasing the deviation value of the corresponding non-compliant indicator and decreasing the deviation value of the corresponding non-compliant indicator are determined separately. Instructions corresponding to non-compliant indicators with higher decision priority are marked as priority instructions, and instructions corresponding to non-compliant indicators with lower decision priority are marked as subordinate instructions. The decision priority dimension is preset by those skilled in the art based on actual conditions; for example, the decision priority of the sugar control dimension is higher than that of the sensory dimension, and the decision priority of the sensory dimension is higher than that of the processing dimension. Priority instructions are retained, while subordinate instructions undergo attenuation processing. The attenuation processing method is as follows: a conflict attenuation coefficient is preset, which is preset by those skilled in the art based on a multi-attribute coordination strategy; the product of the correction magnitude of the subordinate instruction and the conflict attenuation coefficient is calculated, and the correction magnitude of the subordinate instruction is updated based on the product calculation result. Based on the coordinated set of directional correction instructions, the proportion correction amount and particle size correction amount for each plant material are calculated. Specifically, for each plant material, the correction magnitudes in all proportion correction instructions for the corresponding plant material are algebraically summed to obtain the proportion correction amount for the corresponding plant material; the correction magnitudes in all particle size correction instructions for the corresponding plant material are algebraically summed to obtain the particle size correction amount for the corresponding plant material; the predicted feed proportion values ​​for each plant material in the collaborative preparation plan are added to the corresponding proportion correction amounts to obtain the corrected feed proportion values; the corrected feed proportion values ​​are normalized so that the sum of the corrected feed proportion values ​​for all plant materials equals one; the predicted target mesh size for each plant material in the collaborative preparation plan is added to the corresponding particle size correction amounts to obtain the corrected target mesh size value; the corrected feed proportion values ​​and the corrected target mesh size value are respectively subjected to feasibility constraint checks in the constraint check output layer to ensure that both the corrected feed proportion values ​​and the target mesh size value are within the corresponding allowable ranges. Online parameter fine-tuning is performed on the collaborative preparation decision model. Specifically, a new training sample is formed by combining the model input feature vector of the current batch with the corrected feed ratio and the corrected crushing target value to create a new training sample. This new training sample is then added to the decision training sample set. A preset online fine-tuning learning rate is used, which is pre-set by those skilled in the art based on the requirements for controlling the update magnitude of the model parameters. The online fine-tuning learning rate is lower than the initial training learning rate to avoid excessive interference from single-batch data on the overall decision-making ability of the model. The model parameters of the collaborative preparation decision model are updated with a finite number of steps using the online fine-tuning learning rate, allowing the collaborative preparation decision model to absorb the process correction feedback of the current batch while maintaining its memory of historical production experience. The finite number of steps is pre-set by those skilled in the art based on the requirements for online learning stability.

[0043] Methods for outputting the final optimized co-production process scheme for the current batch include: The corrected feed ratio and the corrected target mesh size are used as the final process parameters; the raw material name and unique code of each plant material are obtained from the batch attribute data; the target moisture content and recommended drying temperature of each plant material are obtained from the drying target moisture content mapping table; the number of grinding stages and grinding parameters of each stage are obtained from the grinding grading strategy table corresponding to the corrected target mesh size; the mixing order of each plant material is determined according to the mixing priority rule; the corrected feed mass is obtained by multiplying the corrected feed ratio of each plant material by the total feed mass. The unique raw material code, raw material name, final process parameters, corrected feed mass, target moisture content, recommended drying temperature, number of pulverization stages, pulverization parameters for each stage, and mixing order of each plant raw material are integrated to form a detailed list of process parameters for each plant raw material. The detailed list of process parameters for all plant raw materials is then integrated with the deviation traceability analysis results in the compensation residuals to generate the final optimized co-preparation process scheme for the current batch. The optimized co-preparation process scheme is used to guide the complete production process of the current batch of composite plant powder, including the drying conditions, particle size control, feed ratio, and mixing process arrangement of each plant raw material.

[0044] This embodiment systematically collects physicochemical testing information on plant raw materials in three aspects: sugar control attributes, sensory attributes, and processing attributes, by acquiring batch attribute data of each plant raw material in the current batch and target product standard data. This provides a data foundation for addressing batch-to-batch attribute fluctuations of plant raw materials caused by natural factors such as origin, year, and harvest season. By calculating the multi-attribute deviation matrix of each plant raw material in the current batch relative to the target product standard, a structured and quantitative description of batch-to-batch differences in raw materials is achieved, providing precise input for subsequent collaborative preparation decisions. By constructing a synergistic influence relationship knowledge base, the synergistic effects, sensory antagonisms, and processing interactions among plant raw materials in the three dimensions of sugar control efficacy, sensory quality, and processing performance are systematically integrated, and domain prior knowledge is incorporated. The knowledge is injected into the training and inference process of the collaborative preparation decision model in the form of collaborative constraint encoding vectors, so that the model can explicitly consider the multidimensional interaction effects between raw materials when generating feed ratio and crushing particle size decisions, effectively overcoming the shortcomings of treating each raw material as an independent variable and ignoring the collaborative relationship between raw materials; the ratio and particle size joint decision layer in the collaborative preparation decision network uses the feed ratio and crushing particle size as coupled decision variables for joint optimization, effectively overcoming the shortcomings of separating ratio optimization and particle size setting into two independent steps and failing to capture the collaborative influence relationship between ratio adjustment and particle size adjustment; the introduction of quality consistency auxiliary loss enables the model to learn the causal mapping relationship between process parameters and finished product quality while optimizing the prediction accuracy of process parameters, giving the model the ability to predict the consequences of process decisions; The graded gradient pulverization strategy avoids the degradation of functional components caused by excessive pulverization in a single step. A synergistic influence-based homogeneous mixing strategy determines the mixing order based on the sensory antagonism between raw materials, mitigating sensory quality degradation caused by localized high-concentration antagonistic effects. An online process control and detection mechanism ensures the consistency of powder mixing quality through real-time verification of mixing uniformity. Compensation residual assessment and deviation source analysis systematically identify the source structure and dominant factors of residual deviations between the finished product quality and target standards after the execution of the collaborative preparation plan. Based on a directional correction instruction set, precise directional correction of the feed ratio and pulverization particle size is achieved. The correction instruction conflict detection and coordination mechanism effectively solves the instruction contradictions that may arise when optimizing multiple attributes simultaneously. The online parameter fine-tuning mechanism of the collaborative preparation decision model allows the model to continuously absorb process correction feedback from each production batch, constantly accumulating experience in adapting to raw material batch fluctuations, achieving continuous evolution of collaborative preparation decision-making capabilities. Ultimately, it outputs a collaborative preparation process optimization scheme that comprehensively considers sugar control efficacy, sensory experience, and processing performance, providing a systematic intelligent process optimization method for the stable industrial production of composite plant powders. Example 2:

[0045] This application also provides an electronic device. The electronic device may include one or more processors and one or more memories. The memories store computer-readable code, which, when executed by the one or more processors, can perform the method described above for optimizing the synergistic preparation process of composite plant powders with sugar-controlling effects.

[0046] The method according to the embodiments of this application can also be implemented using the architecture of the electronic device shown in this application. The electronic device may include a bus, one or more CPUs, ROM, RAM, a communication port connected to a network, input / output, a hard disk, etc. The storage device in the electronic device, such as a ROM or hard disk, may store the optimized method for the synergistic preparation process of composite plant powders with sugar-controlling effects provided in this application. Furthermore, the electronic device may also include a user interface. Of course, the architecture shown in this application is merely exemplary; when implementing different devices, one or more components in the electronic device shown in this application may be omitted according to actual needs.

[0047] Example 3

[0048] Please refer to the accompanying drawings. One embodiment of this application discloses a computer-readable storage medium. The computer-readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, they can perform an optimization method for the synergistic preparation of compound plant powders with sugar-controlling effects, as described in the above-described embodiments of this application. The storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.

[0049] Furthermore, according to embodiments of this application, the processes described in the above-referenced flowcharts can be implemented as computer software programs. For example, this application provides a non-transitory machine-readable storage medium storing machine-readable instructions that can be executed by a processor to perform instructions corresponding to the method steps provided in this application, such as: an optimization method for the synergistic preparation process of composite plant powders with sugar-controlling effects. When this computer program is executed by a central processing unit (CPU), it performs the functions defined in the method of this application.

[0050] The above description is merely a preferred embodiment of the present invention and 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.

[0051] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0052] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. An optimized method for the synergistic preparation of composite plant powders with sugar-controlling effects, characterized in that, include: Step S1: Obtain batch attribute data and target product standard data for each plant raw material in the current batch, including target sugar control index, target sensory index and target processing index; Step S2: Based on the batch attribute data and the target product standard data, calculate the attribute deviation values ​​of each plant raw material in the current batch relative to the target sugar control index, target sensory index and target processing index, and form a multi-attribute deviation matrix for the current batch; Step S3: Call the pre-constructed collaborative preparation decision model, establish the synergistic influence relationship of each plant raw material on sugar control properties, sensory properties and processing properties, and solve the feeding ratio and crushing target mesh size to meet the standard data of the target product based on the multi-attribute deviation matrix, so as to form the collaborative preparation plan for the current batch. Step S4: According to the collaborative preparation plan, the plant raw materials of the current batch are dried, pulverized, sieved and homogenized to obtain composite plant powder samples. The actual sugar control index, actual sensory index and actual processing index of the composite plant powder samples are tested to form a batch verification result set. Step S5: Based on the deviation between the batch verification result set and the target product standard data, evaluate the compensation residual of the current batch, and adaptively correct the collaborative preparation decision model based on the compensation residual, and output the final collaborative preparation process optimization scheme for the current batch.

2. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 1, characterized in that, Methods for generating the multi-attribute deviation matrix for the current batch include: Batch attribute data includes basic information about raw materials and physicochemical testing data of raw materials. The physicochemical testing data of raw materials includes sugar control attribute testing data, sensory attribute testing data and processing attribute testing data. The sugar control attribute detection data corresponding to each plant material are compared with the target sugar control index item by item, the sugar control attribute deviation value is calculated, and the sugar control deviation vector of each plant material is formed. The sensory attribute detection data corresponding to each plant material are compared with the target sensory indicators one by one, the sensory attribute deviation value is calculated, and a sensory deviation vector for each plant material is formed. The processing attribute detection data corresponding to each plant raw material are compared with the target processing indicators one by one, the processing attribute deviation value is calculated, and a processing deviation vector is formed for each plant raw material. The sugar control deviation vector, sensory deviation vector, and processing deviation vector of each plant material are concatenated in sequence to form the comprehensive deviation vector of each plant material; the comprehensive deviation vectors of all plant materials are arranged row by row to form the multi-attribute deviation matrix of the current batch.

3. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 2, characterized in that, Methods for pre-constructing collaborative preparation decision models include: Obtain historical production records for all historical production batches; historical production records include historical batch numbers, historical batch attribute data for each plant raw material, historical feed ratios, historical particle size parameters, and historical finished product test results; historical particle size parameters include the target mesh size for each plant raw material. Based on the pre-set knowledge base of synergistic influence relationships, a synergistic constraint encoding vector is constructed, and a model input feature vector is constructed based on the synergistic constraint encoding vector. The synergistic influence relationship knowledge base includes a table of sugar control synergistic relationships among plant raw materials, a table of sensory antagonistic relationships, and a table of processing interaction effects. For each historical production record, a corresponding historical finished product quality vector and a model output label vector are constructed sequentially. The model input feature vectors and model output label vectors from all historical production records are integrated to form a decision training sample set. A collaborative preparation decision network is constructed, which includes a bias-aware encoding layer, a collaborative constraint fusion layer, a ratio granularity joint decision layer, and a constraint verification output layer. Based on the decision training sample set, the collaborative preparation decision network is trained, and the trained collaborative preparation decision network is used as the collaborative preparation decision model.

4. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 3, characterized in that, Methods for training collaborative decision-making networks include: The model input feature vector of each decision training sample is input into the collaborative preparation decision network, which outputs the corresponding feed ratio prediction value and crushing target mesh size prediction value. Combined with the corresponding model output label vector, the ratio prediction loss and particle size prediction loss corresponding to each decision training sample are calculated. Based on the predicted values ​​of the feed ratio and the target mesh size of the crushing, combined with the synergy coefficient in the sugar control synergy table, the antagonism intensity coefficient in the sensory antagonism table, and the interaction influence coefficient in the processing interaction influence table, the finished product quality prediction vector is calculated; the mean square error between the finished product quality prediction vector and the corresponding historical finished product quality vector is calculated to obtain the quality consistency auxiliary loss. Calculate the product of the quality consistency auxiliary loss and the preset quality consistency auxiliary loss weight, and combine it with the ratio prediction loss and granularity prediction loss to obtain the joint decision loss for each decision training sample; calculate the batch joint decision loss based on the joint decision loss of all decision training samples; iteratively update the model parameters in the collaborative preparation decision network based on the batch joint decision loss until the batch joint decision loss converges or the number of training iterations reaches the target, at which point the collaborative preparation decision network training is complete.

5. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 4, characterized in that, The methods for formulating the co-production plan for the current batch include: The multi-attribute deviation matrix of the current batch is concatenated and flattened into a one-dimensional vector to form the deviation input vector of the current batch. The combinations of plant raw materials involved in the feeding are determined, and the synergy coefficient, antagonism strength coefficient and interaction influence coefficient corresponding to each combination of plant raw materials are obtained from the synergy influence relationship knowledge base to construct the synergy constraint encoding vector of the current batch. The synergy constraint encoding vector of the current batch is concatenated with the deviation input vector of the current batch to form the model input feature vector of the current batch, and input into the synergistic preparation decision model to obtain the predicted feeding ratio and the predicted target mesh size of each plant raw material. The update and iteration process is executed to obtain the updated predicted values ​​of the feed ratio and the target mesh size of the crushing. Based on the absolute values ​​of the differences between each element of the predicted feed ratio and the target mesh size of the crushing before and after the update, it is determined whether the convergence condition is met. If the convergence condition is met, the update and iteration process is stopped, and the final predicted values ​​of the feed ratio and the target mesh size of the crushing are output to form the collaborative preparation plan for the current batch. If the convergence condition is not met, the update and iteration process is continued until the convergence condition is met or the number of update iterations reaches the preset maximum number of updates.

6. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 5, characterized in that, Methods for preparing composite plant powder samples include: Perform drying treatment on each plant material in the current batch: Preset a drying target moisture content mapping table, which includes the drying target moisture content and recommended drying temperature for each plant material; perform drying treatment on each plant material in the current batch according to the drying target moisture content and recommended drying temperature; The process involves implementing a graded pulverization strategy and sieving: a pre-defined pulverization strategy table is used, which includes the pulverization levels and parameters for each level corresponding to different target mesh sizes; based on the predicted target mesh size for each plant material, the corresponding pulverization level and parameters are obtained from the pulverization strategy table, and each plant material is pulverized step by step according to the pulverization level; based on the predicted target mesh size, each plant material is sieved to obtain qualified powder. Perform synergistic effect perception-based homogenization mixing: Based on the sensory antagonism table, determine whether there is a sensory antagonism relationship between the plant materials; if so, execute a step-by-step progressive mixing strategy to obtain a composite plant powder sample; if not, perform homogenization mixing operation on the qualified powders of all plant materials simultaneously to obtain a composite plant powder sample; perform online process control detection on the composite plant powder sample.

7. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 6, characterized in that, Methods for generating batch validation result sets include: Samples were taken from the compound plant powder samples, and the corresponding actual sugar control indicators, actual sensory indicators, and actual processing indicators were tested respectively. The actual sugar control indicators, actual sensory indicators, and actual processing indicators were integrated to form a batch verification result set. Among them, the actual blood sugar control indicators include the actual total polyphenol content, the actual total flavonoid content, and the actual... - Glucosidase inhibition rate and actual dietary fiber content; actual sensory indicators include actual bitterness intensity value, actual grassy smell intensity value, actual color brightness value and actual particle roughness level; actual processing indicators include actual moisture content, actual angle of repose, actual moisture absorption rate and actual bulk density.

8. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 7, characterized in that, Methods for evaluating the compensated residuals of the current batch include: The actual test values ​​of each indicator in the batch verification results set are compared with the corresponding target indicators in the target product standard data item by item, and the verification deviation value of each indicator is calculated. All verification deviation values ​​are arranged to form a verification deviation vector. Based on the verification deviation vector, the pass status of each indicator in the batch verification results set is determined and marked as a pass indicator or a fail indicator. If all indicators are determined to be pass indicators, the co-preparation plan is directly used as the co-preparation process optimization scheme. If any non-compliant indicators exist, the corresponding deviation source attribution dimension is determined based on the attribute dimension to which each non-compliant indicator belongs. For each non-compliant indicator, deviation source tracing analysis is performed to determine the dominant deviation factor. Based on the verification deviation value corresponding to the non-compliant indicator, the deviation direction is determined. The verification deviation values, deviation source attribution dimensions, deviation directions, and dominant deviation factors of all non-compliant indicators are integrated to form the compensation residual for the current batch. Among them, the deviation source attribution dimensions include sugar control dimension, sensory dimension, and processing dimension; the deviation direction includes positive exceedance and negative deficiency.

9. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 8, characterized in that, Methods for performing deviation source analysis include: For the non-compliant indicators in the sugar control dimension, the synergistic correction amount of each combination of plant raw materials involved in the feeding is calculated. Based on the synergistic correction amount, the dominant raw materials of sugar control deviation are determined, and the corresponding deviation contribution amount is calculated. The deviation contribution amounts of all dominant raw materials are summarized to form the dominant factors of sugar control deviation. For the non-compliant indicators in the sensory dimension, obtain all combinations of plant raw materials that have an antagonistic relationship with the non-compliant indicators and participate in the feeding, and their corresponding antagonistic strength coefficients; determine the dominant raw materials for sensory deviation based on the antagonistic strength coefficients, and calculate the corresponding deviation contribution; summarize the deviation contribution of all dominant raw materials to form the dominant factors of sensory deviation. For non-conforming indicators in the processing dimension, all plant material combinations that interact with the non-conforming indicators and participate in the feeding are obtained, along with their corresponding interaction coefficients. The effective interaction coefficients for each plant material combination are then determined. Based on the effective interaction coefficients, the dominant raw materials for processing deviations are identified, and their corresponding deviation contribution and dominant particle size parameters are calculated. The deviation contribution and dominant particle size parameters of all dominant raw materials are then summarized to form the dominant factors of processing deviations.

10. The optimized preparation process of a composite plant powder with sugar-controlling effect according to claim 9, characterized in that, Methods for outputting optimized collaborative preparation processes include: A set of directional correction instructions is constructed based on the compensation residual, and conflict detection and coordination of correction instructions are performed on the set of directional correction instructions. The set of directional correction instructions includes proportion correction instructions and particle size correction instructions. Based on the coordinated set of directional correction instructions, the corrected feed ratio and corrected crushing target particle size of each plant material are calculated. The raw material name, unique code, target moisture content, recommended drying temperature, number of grinding stages, grinding parameters for each stage, mixing and addition order, and corrected feed mass of each plant raw material are obtained sequentially. These are then integrated with the corrected feed ratio and the corrected target particle size to form a detailed list of process parameters. The detailed list of process parameters for all plant raw materials is then integrated with the deviation traceability analysis results in the compensation residual to generate the final optimized synergistic preparation process scheme for the current batch. The method for detecting and coordinating conflict in the correction instructions is as follows: In the set of directional correction instructions, detect whether there are contradictory correction instructions for the same plant material; if so, perform conflict coordination: divide the contradictory correction instructions corresponding to the same plant material into increase instructions and decrease instructions according to the corresponding correction direction, and mark the increase instructions and decrease instructions as priority instructions and subordinate instructions respectively based on the verification deviation value; retain the priority instructions and perform attenuation processing on the subordinate instructions.

Images