Big data analysis system for optimization and digital management of Maotai-flavor liquor blending formula

Through the big data analysis system, a regression mapping model of the ingredients and sensory indicators of sauce-flavored liquor formulas was constructed. Combined with association rules and collaborative filtering algorithms, the subjectivity and insufficient data of the traditional blending method were solved, and the precise control and personality of the formula was realized. It improves the quality consistency and production efficiency of sauce-flavored liquor.

CN118761329BActive Publication Date: 2024-11-19贵州轻工职业大学
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202411238590.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-05
Publication Date
2024-11-19
Estimated Expiration
2044-09-05

AI Technical Summary

Technical Problem

The traditional method of blending formula for sauce-flavored liquor has subjectivity and uncertainty in sensory evaluation, lacks quantitative indicators and data support, making it difficult to achieve precise control and optimization, and the existing data-driven methods have shortcomings in data quality and generalization capabilities. , it is difficult to meet market demand and product diversification trends.

Method used

Using a big data analysis system, including data units, modeling units, association units and matching units, to construct a regression mapping model of sauce-flavored liquor formula ingredients and sensory indicators, use Gaussian kernel functions to perform nonlinear modeling, and combine association rule mining and Collaborative filtering algorithm automatically generates a list of personalized formula optimization recommendations.

Benefits of technology

It realizes the intrinsic correlation between formula ingredients and sensory quality, ensures sensory consistency and personalized recommendation of formulas, improves production efficiency and product quality stability, and adapts to market demand and product diversification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118761329B_ABST
    Figure CN118761329B_ABST
Patent Text Reader

Abstract

The present application discloses a big data analysis system for optimizing and digitally managing the blending formula of Maotai-flavor liquor, the system comprising: a data unit for constructing a sensory index data set; a modeling unit for constructing a regression mapping model between the concentration of Maotai-flavor liquor formula ingredients and sensory indicators; an association unit for mining association rules between different raw material ingredients from a historical high-quality formula database using an association rule mining algorithm to obtain frequent item sets and their confidences; a matching unit for constructing a user formula rating matrix in combination with user preference information, and using the Pearson correlation coefficient to calculate formula similarity, so as to predict the target user's rating of the candidate formula using a project-based collaborative filtering algorithm; and automatically generating a personalized formula optimization recommendation list based on the rating prediction results and association rules. Through the scheme of the present application, it is possible to achieve sensory quality consistency control of Maotai-flavor liquor blending formulas and automatic recommendation of formula optimization.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of formula optimization management, and in particular to a big data analysis system for optimizing and digitally managing the blending formula of Maotai-flavor liquor. Background Art

[0002] As one of the four major flavors of Chinese liquor, Maotai-flavor liquor is highly favored for its unique Maotai flavor and long brewing history. However, with the continuous growth of market demand and the increasing taste of consumers, the optimization and management of the blending formula of Maotai-flavor liquor has become a key issue that needs to be urgently solved in the industry. Blending refers to the mixing of base liquors of different batches and years in a certain proportion to achieve specific sensory quality requirements. A scientific and reasonable blending formula can not only ensure the flavor consistency of the product, but also improve the fullness and coordination of the wine body, thereby improving consumer satisfaction and loyalty. Therefore, the optimization and management of the blending formula of Maotai-flavor liquor is of great significance to improving product competitiveness and expanding market share. At present, the formulation of the blending formula of Maotai-flavor liquor mainly depends on the experience and sensory evaluation of senior blending masters. Through years of practical accumulation, the blending master evaluates the base liquor with senses such as smell and taste, and determines the optimal blending ratio through repeated experiments based on the evaluation results and target flavor characteristics. Although this traditional blending method can guarantee the quality of the finished wine to a certain extent, it has certain limitations. First, the sensory evaluation of blending masters is subjective and uncertain, making it difficult to achieve standardization and traceability; second, the formulation process of blending formulas lacks quantitative indicators and data support, making it difficult to achieve precise control and optimization; third, the traditional blending method is inefficient and difficult to adapt to the growing market demand and product diversification trend. Therefore, a scientific, efficient, and intelligent method for optimizing and managing the blending formula of Maotai-flavor liquor is urgently needed to improve the consistency and stability of product quality, and improve production efficiency and management level.

[0003] In recent years, with the development of emerging technologies such as big data and artificial intelligence, the use of data-driven methods to optimize and manage the blending formula of Maotai-flavor liquor has attracted widespread attention in the industry. Some researchers have tried to establish a quantitative model between blending formula and sensory quality through correlation analysis between sensory evaluation data and physical and chemical indicators, and optimize the formula and predict the quality based on this model. Some scholars have also used machine learning algorithms, such as support vector machines and random forests, to build an intelligent recommendation system for blending formulas, which automatically generates the optimal blending formula based on the target flavor characteristics and product data on sale. These data-driven methods have overcome the limitations of traditional blending methods to a certain extent, and have provided new ideas and approaches for the optimization and management of Maotai-flavor liquor blending formulas. However, the existing data-driven methods still have some shortcomings in practical applications. On the one hand, the acquisition cost of sensory quality evaluation data of Maotai-flavor liquor is high, and there is a lack of unified standards for evaluation indicators, resulting in uneven data quality and difficulty in supporting accurate modeling and analysis; on the other hand, Maotai-flavor liquors from different production areas and different flavors have differences in flavor characteristics, consumer preferences, etc. The existing methods lack consideration of these differences, resulting in insufficient generalization of the obtained models and algorithms. In addition, since the blending formula of Maotai-flavor liquor involves the combination of multiple base liquor components, existing methods often use exhaustive search when exploring the formula space, which has low computational efficiency and is difficult to meet actual production needs.

[0004] Therefore, there is an urgent need for a technical solution that can accurately characterize the intrinsic relationship between formula ingredients and sensory quality and achieve quantitative control of formula sensory consistency. Summary of the invention

[0005] In order to solve the deficiencies of the prior art, the embodiment of the present application provides a big data analysis system for optimizing and digitally managing the blending formula of Maotai-flavor liquor. The present application solves the technical problems of the prior art such as the sensory quality consistency control of the blending formula of Maotai-flavor liquor and the low quality of the automatic recommendation of formula optimization.

[0006] The embodiment of the present application provides a big data analysis system for optimizing and digitally managing the blending formula of Maotai-flavor liquor, comprising: a data unit, a modeling unit, an association unit and a matching unit; wherein the data unit is used to construct a raw material sensory evaluation index system including original liquor and spices, and collect sensory evaluation data of each batch of formulas to obtain a sensory index data set; the modeling unit is used to construct a regression mapping model between the concentration of Maotai-flavor liquor formula ingredients and sensory indicators, and use a Gaussian kernel function to map the formula ingredients to a high-dimensional space to model nonlinear relationships; the association unit is used to use an association rule mining algorithm to mine association rules between different raw material ingredients from a historical high-quality formula database to obtain frequent item sets and their confidence levels; the matching unit is used to construct a user formula rating matrix in combination with user preference information, and use the Pearson correlation coefficient to calculate formula similarity, so as to use a project-based collaborative filtering algorithm to predict the target user's rating of candidate formulas; according to the rating prediction results and the association rules, a personalized formula optimization recommendation list is automatically generated.

[0007] In a possible implementation method, a regression mapping model of the concentration of the formula ingredients of the sauce-flavored liquor and the sensory indicators is constructed, and the formula ingredients are mapped to a high-dimensional space using a Gaussian kernel function to model the nonlinear relationship, including: standardizing the formula ingredient concentration vector and the sensory evaluation index vector, including calculating the standardized value using the Z-score method; using the Gaussian kernel function to construct a support vector machine regression mapping model, and optimizing the penalty factor, ε parameter and γ parameter through a grid search combined with a cross-validation method to model the nonlinear relationship between the formula ingredients and the sensory indicators; using the mean square error, mean absolute error and determination coefficient to evaluate the model performance, and determining the degree of influence of each formula ingredient on the sensory indicators by calculating the prediction error increment after the feature disturbance.

[0008] In a possible implementation, an association rule mining algorithm is used to mine association rules between different raw material components from a historical high-quality formula database to obtain frequent item sets and their confidences, including: binary encoding the usage of raw material components in the historical high-quality formula database, and discretizing continuous concentration values ​​into three levels of low, medium, and high, with each component represented by three binary bits; setting minimum support and minimum confidence thresholds, using the Apriori algorithm to gradually generate candidate item sets, and filtering frequent item sets through a pruning strategy until new frequent item sets cannot be generated; generating association rules based on the obtained frequent item sets, calculating the support, confidence, and lift of each rule, and filtering out effective rules that meet the threshold requirements; and filtering the mined association rules in combination with domain expert knowledge.

[0009] In a possible implementation, a user recipe rating matrix is ​​constructed in combination with user preference information, and the recipe similarity is calculated using the Pearson correlation coefficient to predict the target user's rating of the candidate recipe using a project-based collaborative filtering algorithm, including: constructing a user recipe rating matrix, wherein the rating matrix includes rating data of multiple users on multiple recipes, and the sources of the rating data include professional wine taster ratings, consumer feedback, implicit ratings converted from sales data, and / or wine tasting event ratings; based on the similarity of recipe pairs, obtaining the target user's predicted rating of the candidate recipe; using a content-based recommendation method for new users, and searching for similar recipes for new recipes based on features including raw material composition and / or process parameters.

[0010] In a possible implementation, based on the similarity of the recipe pairs, the target user's predicted score for the candidate recipe is obtained, including: calculating the similarity of the recipe pairs, and introducing a penalty factor to adjust the similarity to reduce the similarity between recipe pairs that have only a few common scoring users; based on the scoring data of multiple recipes that are most similar to the candidate recipe, the target user's predicted score for the candidate recipe is calculated, and a user bias term and a recipe bias term are introduced to learn the bias term by minimizing the prediction error.

[0011] In a possible implementation, a personalized recipe optimization recommendation list is automatically generated based on the score prediction results, association rules and mapping models, including: using a support vector machine regression model to predict the sensory index of a new recipe to obtain a predicted sensory index vector; constructing a comprehensive scoring function for the sensory index vector, and constructing the recipe optimization problem as a constrained optimization problem; using a global optimization algorithm to solve the constrained optimization problem to obtain an optimized recipe, wherein a multi-starting point strategy is used to improve the solution efficiency; combining the association rule mining results, generating a new recipe combination based on the optimized recipe, and fine-tuning the support vector machine regression model and the comprehensive scoring function through the optimization algorithm to obtain a fine-tuned personalized recipe optimization recommendation list.

[0012] In a possible implementation, a comprehensive scoring function of the sensory index vector is constructed, and the recipe optimization problem is constructed as a constrained optimization problem, including: constructing a comprehensive scoring function of the sensory index vector, integrating multiple sensory indexes into a single score, and the scoring function includes linear terms, quadratic interaction terms, and threshold penalty terms; constructing the recipe optimization problem as a constrained optimization problem, setting the objective function to maximize the scoring function, and imposing total quantity constraints, cost constraints, inventory constraints, and process constraints.

[0013] In a possible implementation, the predicted score of the target user for the candidate formula is calculated based on the score data of multiple formulas most similar to the candidate formula, including:

[0014] ,

[0015] in, Indicates user About the recipe The prediction score of Representation and formulation Most similar A collection of recipes, Recipe and The similarity of Indicates user About the recipe 's rating.

[0016] In a possible implementation, constructing a comprehensive scoring function includes:

[0017] ,in, Recipe Total rating, Indicates The weight of each sensory index, Indicates Predicted sensory indicators, Indicates and The interaction weights between the sensory indicators are

[0018] Indicates The penalty weight of an indicator, Indicates the threshold of the i-th indicator.

[0019] In the big data analysis system for optimization and digital management of the blending formula of a sauce-flavor liquor provided above, the embodiment of the present application can accurately characterize the intrinsic relationship between the formula ingredients and the sensory quality by introducing the support vector machine regression model, and realize the quantitative control of the sensory consistency of the formula. Furthermore, in some embodiments, by integrating association rule mining and collaborative filtering, it is possible to make full use of historical high-quality formula knowledge and realize personalized automatic recommendation of formula optimization in combination with user preferences. Furthermore, in some embodiments, by constructing a digital management process for formula optimization, it is possible to realize digital management of the entire process from raw material procurement, blending production to sensory evaluation and quality control, thereby improving production efficiency and product quality stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0021] Figure 1 A schematic block diagram of a big data analysis system for optimizing and digitally managing a sauce-flavor liquor blending formula provided in an embodiment of the present application;

[0022] Figure 2 A flow chart of a big data analysis method for optimizing and digitally managing the blending formula of a sauce-flavor liquor provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] Various exemplary embodiments of the present application will now be described in detail with reference to the accompanying drawings. It should be noted that unless otherwise specifically stated, the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present application.

[0024] Those skilled in the art will appreciate that the terms "first", "second" and the like in the embodiments of the present application are only used to distinguish different steps, devices or modules, etc., and do not represent any specific technical meaning, nor do they represent the necessary logical order between them. It should also be understood that in the embodiments of the present application, "multiple" may refer to two or more, and "at least one" may refer to one, two or more. It should also be understood that for any component, data or structure mentioned in the embodiments of the present application, in the absence of explicit limitation or contrary revelation given in the context, it can generally be understood as one or more. In addition, the term "and / or" in the present application is only a kind of association relationship describing the associated objects, indicating that there can be three kinds of relationships, for example, A and / or B, which can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in the present application generally indicates that the associated objects before and after are an "or" relationship. It should also be understood that the description of each embodiment in the present application emphasizes the differences between the embodiments, and the same or similar parts can refer to each other. For the sake of brevity, they will not be repeated one by one.

[0025] At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship. The following description of at least one exemplary embodiment is actually only illustrative and is by no means intended to limit the present application and its application or use. The techniques, methods and devices known to ordinary technicians in the relevant fields may not be discussed in detail, but where appropriate, the techniques, methods and devices should be considered as part of the specification. It should be noted that similar numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0026] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0027] Figure 1 A schematic block diagram of a big data analysis system for optimizing and digitally managing the blending formula of a Maotai-flavor liquor provided in an embodiment of the present application. It should be understood that the system shown in the figure is exemplary and not restrictive. This means that the system architecture involved is not limited to a specific form or design, but is presented as an example. In other words, the architecture shown in the figure can be regarded as a way of expression to clearly describe related concepts and relationships, and does not exclude other forms of architecture. Therefore, when interpreting the architecture in the picture, it should be understood that the model is flexible and diverse, and its purpose is to provide an exemplary description rather than a restrictive provision on a specific form.

[0028] The present application discloses a big data analysis system for optimizing and digitally managing the blending formula of Maotai-flavor liquor, comprising: a data unit 101, a modeling unit 102, an association unit 103 and a matching unit 104; wherein the data unit 101 is used to construct a raw material sensory evaluation index system including original liquor and spices, and collect sensory evaluation data of each batch of formulas to obtain a sensory index data set; the modeling unit 102 is used to construct a regression mapping model between the concentration of Maotai-flavor liquor formula ingredients and sensory indicators, and use a Gaussian kernel function to map the formula ingredients to a high-dimensional space to model nonlinear relationships; the association unit 103 is used to use an association rule mining algorithm to mine association rules between different raw material ingredients from a historical high-quality formula database to obtain frequent item sets and their confidence levels; the matching unit 104 is used to construct a user formula rating matrix in combination with user preference information, and use the Pearson correlation coefficient to calculate formula similarity, so as to use a project-based collaborative filtering algorithm to predict the target user's rating of the candidate formula; according to the rating prediction results and the association rules, a personalized formula optimization recommendation list is automatically generated.

[0029] Specifically, the big data analysis system for optimizing and digitally managing the blending formula of Maotai-flavor liquor implements a big data analysis method for optimizing and digitally managing the blending formula of Maotai-flavor liquor as described below, and the above-mentioned units of the system specifically execute each step or function of the method.

[0030] Figure 2 A flow chart of a big data analysis method for optimizing and digitally managing the blending formula of a Maotai-flavor liquor provided in an embodiment of the present application. Figure 2 As shown, at step S201, a sensory evaluation index system for raw materials including base liquor and spices is constructed, and sensory evaluation data of each batch of recipes is collected to obtain a sensory index data set.

[0031] Specifically, in the actual implementation process, the construction of a comprehensive and scientific sensory evaluation index system for Maotai-flavor liquor includes the following main dimensions: aroma, taste, style, coordination and aftertaste. Each dimension can be subdivided into multiple specific indicators. For example, aroma can be subdivided into Maotai aroma, mellow aroma, cellar aroma, etc.; taste can be subdivided into mellowness, softness, fullness, etc.; style can be subdivided into typicality, uniqueness, etc.; coordination can be subdivided into aroma coordination, taste balance, etc.; aftertaste can be subdivided into persistence, aftertaste, etc.

[0032] In one embodiment, a 100-point scoring standard may be used, and the weight of each indicator is assigned according to its importance. For example:

[0033] 1. Aroma (30 points): sauce aroma (10 points), mellow aroma (10 points), cellar aroma (10 points)

[0034] 2. Taste (25 points): body (10 points), softness (8 points), fullness (7 points)

[0035] 3. Style (20 points): Typicality (12 points), Uniqueness (8 points)

[0036] 4. Coordination (15 points): aroma coordination (8 points), taste balance (7 points)

[0037] 5. Aftertaste (10 points): persistence (6 points), aftertaste (4 points)

[0038] During the data collection process, a professional tasting team can be organized, usually consisting of 8-12 experienced wine tasters. Each batch of formula samples needs to undergo multiple rounds of blind tasting to ensure the objectivity and accuracy of the scoring. During the scoring process, the tasters need to score each indicator on a specially designed scoring sheet and record detailed tasting opinions.

[0039] In order to improve the reliability of the data, the following mathematical model can be used to process the original scoring data: represents the score of the jth sample on the kth indicator by the i-th taster, represents the weight of the kth indicator, n represents the number of tasters, and m represents the number of indicators. Then the comprehensive score of sample j is It can be calculated by the following formula: ,This formula first removes the highest and lowest scores of each indicator, and then calculates the weighted average of the remaining scores to reduce the impact of extreme values ​​and improve the stability and representativeness of the scores.

[0040] In addition, to ensure the quality and consistency of the data, tasters need to be trained and calibrated regularly. Standard samples can be set up, tasters can be organized to score regularly, and the performance of the tasting team can be evaluated by calculating the consistency coefficient of the scores (such as Kendall's harmony coefficient). If the consistency coefficient is lower than the preset threshold (for example, 0.7), additional training and discussion are required to improve the consistency of the scores.

[0041] At step S202, a regression mapping model of the concentration of the formula ingredients of the sauce-flavor liquor and the sensory index is constructed, and the formula ingredients are mapped to a high-dimensional space using a Gaussian kernel function to model the nonlinear relationship. This includes: standardizing the formula ingredient concentration vector and the sensory evaluation index vector, including calculating the standardized value using the Z-score method; using the Gaussian kernel function to construct a support vector machine regression mapping model, and optimizing the penalty factor, ε parameter and γ parameter through a grid search combined with a cross-validation method to model the nonlinear relationship between the formula ingredients and the sensory index; using the mean square error, mean absolute error and determination coefficient to evaluate the model performance, and determining the degree of influence of each formula ingredient on the sensory index by calculating the prediction error increment after feature perturbation.

[0042] Specifically, when constructing the support vector machine regression (SVR) mapping model, special attention should be paid to data preprocessing and the selection of model parameters. First, for the formula ingredient concentration vector , because the concentration ranges of different components may vary greatly, standardization is required. The Z-score standardization method can be used: ,in and are the mean and standard deviation of the concentration of the i-th component respectively. , also needs to be standardized similarly. When selecting the kernel function, the Gaussian kernel (RBF kernel) can be constructed as: ,in is the parameter of the kernel function, which controls the degree of mapping to high-dimensional space.

[0043] In practical applications, grid search combined with cross-validation can be used to select the optimal model parameters. The main parameters that need to be optimized include: 1. Penalty factor C: usually Search on the logarithmic scale

[0044] 2. :Usually in Search on the logarithmic scale

[0045] 3. :Usually in Search on the logarithmic scale

[0046] For example, you can set the parameter search grid as follows:

[0047] C_range = [1e-3, 1e-2, 1e-1, 1, 10, 100, 1000]

[0048] _range = [1e-4, 1e-3, 1e-2, 1e-1, 1]

[0049] _range = [1e-4, 1e-3, 1e-2, 1e-1, 1, 10]

[0050] A 5-fold cross validation was used to evaluate each parameter combination, and the parameter combination with the smallest mean square error (MSE) was selected as the final model parameter.

[0051] After the model training is completed, in order to evaluate the performance of the model, the following indicators can be used: 1. Mean square error (MSE):

[0052] 2. Mean Absolute Error (MAE):

[0053] 3. Coefficient of determination (R²):

[0054] in is the actual sensory index value, is the model prediction value, is the average value of the sensory index. In practical applications, a SVR model can be trained separately for each sensory index, because different sensory indexes may have different sensitivities and nonlinear relationships with formula ingredients.

[0055] In addition, in order to improve the interpretability of the model, feature importance analysis can be combined. Although the SVR model itself does not directly provide feature importance, it can be approximated by the following method: For each feature j, calculate the increase in prediction error after its perturbation: ,in Yes The j-th feature of is randomly shuffled. Normalized to get the importance score of feature j: ,This method can help identify the formulation ingredients that have the greatest impact on sensory indices, and provide guidance for subsequent formulation optimization.

[0056] At step S203, an association rule mining algorithm is used to mine association rules between different raw material components from the historical high-quality formula database to obtain frequent item sets and their confidences, which include: binary encoding the usage of raw material components in the historical high-quality formula database, and discretizing the continuous concentration values ​​into three levels: low, medium, and high, with each component represented by three binary bits; setting minimum support and minimum confidence thresholds, using the Apriori algorithm to gradually generate candidate item sets, and filtering frequent item sets through a pruning strategy until new frequent item sets cannot be generated; generating association rules based on the obtained frequent item sets, calculating the support, confidence, and lift of each rule, and filtering out effective rules that meet the threshold requirements; and filtering the mined association rules in combination with domain expert knowledge.

[0057] Specifically, when implementing the Apriori algorithm to mine association rules, the historical high-quality recipe data must first be preprocessed and encoded. Suppose there is a database containing 100 high-quality recipes, each of which contains a variety of raw wine and spice ingredients. The usage of each ingredient can be encoded in binary form, with 1 indicating use and 0 indicating no use. For example:

[0058] Recipe 1: [1, 0, 1, 1, 0, 1, 0, 1, 0, 0]

[0059] Recipe 2: [1, 1, 0, 1, 1, 0, 1, 0, 1, 0]

[0060] Each position represents a specific raw liquor or flavor component. In practical applications, it may be necessary to discretize continuous concentration values. For example, the concentration values ​​can be divided into three levels: low, medium, and high: low concentration: 0-33%; medium concentration: 34-66%; high concentration: 67-100%.

[0061] In this way, each component can be represented by three binary bits, for example: [1, 0, 0] represents low concentration, [0, 1, 0] represents medium concentration, and [0, 0, 1] represents high concentration.

[0062] Next, you need to set the minimum support (min_support) and minimum confidence (min_confidence) thresholds. The choice of these two parameters needs to be adjusted according to the specific situation. For example, you can initially set min_support=0.1 (that is, it appears in at least 10% of the recipes) and min_confidence=0.7 (that is, the probability that the rule is valid is at least 70%).

[0063] The core idea of ​​the Apriori algorithm is to gradually generate candidate item sets and reduce the amount of calculation through pruning strategies. The specific implementation can be divided into the following steps. Generate 1-item frequent item sets, traverse all recipes, count the number of occurrences of each component, and retain the components with support greater than or equal to min_support. Generate k-item candidate item sets (k>1), and self-connect (k-1)-item frequent item sets to generate k-item candidate item sets. For example, if {A, B} and {A, C} are 2-item frequent item sets, then {A, B, C} is a 3-item candidate item set. Pruning, check whether all (k-1)-item subsets of each k-item candidate item set are frequent item sets. If not, delete the candidate item set. Support counting, traverse all recipes, and count the support of each candidate item set. Generate k-item frequent item sets, retain candidate item sets with support greater than or equal to min_support, and form k-item frequent item sets. Repeat the above steps until no new frequent item sets can be generated. Generate association rules. For each frequent item set I and its non-empty subset A, if conf(A→IA) ≥ min_confidence, then generate rule A→IA.

[0064] In order to improve the efficiency of the algorithm, some optimization strategies can be adopted, such as using a hash table to store candidate item sets to speed up the search process. Using the FP-growth algorithm instead of the Apriori algorithm to avoid generating a large number of candidate item sets by constructing an FP tree.

[0065] After obtaining the association rules, the lift of the rules can be calculated by the following formula to evaluate the effectiveness of the rules: , a lift greater than 1 indicates that the rule is effective, and the appearance of A does increase the probability of B appearing.

[0066] For example, the following association rules can be obtained:

[0067] 1. {High-concentration Daqu liquor, medium-concentration Zaoshu liquor} → {High-quality sensory evaluation}

[0068] Support = 0.15, Confidence = 0.8, Lift = 2.5

[0069] 2. {Medium-concentration Maotai-flavor liquor, low-concentration Qu-flavor liquor} → {Medium-quality sensory evaluation}

[0070] Support = 0.12, Confidence = 0.75, Lift = 1.8

[0071] These rules can provide valuable guidance for recipe optimization. For example, Rule 1 indicates that the combination of high-concentration Daqu liquor and medium-concentration Zaoshu liquor is likely to produce high-quality sensory evaluation, and the probability of this combination occurring is 2.5 times that of random chance.

[0072] In practical applications, the mined rules can be further screened and interpreted in combination with the knowledge of domain experts to ensure the practicality and interpretability of the rules.

[0073] Specifically, for the verification of raw material combination rules, experts can evaluate the raw material combination rules mined by the algorithm based on experience. For example, if the algorithm finds the rule of "high proportion of sorghum + medium proportion of wheat → strong sauce aroma", experts can verify it from the following aspects: The importance of sorghum: confirm the contribution of high proportion of sorghum to the formation of sauce aroma. The auxiliary role of wheat: explain how wheat affects the fermentation process and aroma formation. Raw material complementarity: analyze how the combination of sorghum and wheat affects the flavor characteristics of the final product.

[0074] For the explanation of process parameter rules, experts can provide more in-depth explanations of process parameter rules, such as "long fermentation + multiple steaming → mellow taste": Fermentation time effect: Detailed explanation of how long fermentation promotes the formation of flavor substances. The effect of steaming times: Explain how multiple steaming affects the layering and mellowness of the wine. Process synergy effect: Analyze how fermentation time and steaming times interact with each other to jointly affect the quality of the final product.

[0075] Refinement of sensory evaluation rules. For rules such as "high-concentration Daqu liquor + medium-concentration Zao Shao liquor → high-quality sensory evaluation", experts can further refine the sensory evaluation standards: Aroma characteristics: describe the specific aroma characteristics that may be produced by this combination, such as sauce aroma, mellow aroma, cellar aroma, etc. Taste level: explain how the mixture of different concentrations of liquor affects the richness and level of taste. Flavor balance: analyze how to achieve the best balance of aroma, taste, and feel through this combination.

[0076] Experts can evaluate the practicality of the rules based on actual production experience to assess the operability and economic feasibility of the rules: Production cost: Consider the impact of rule implementation on raw material costs and production costs. Process feasibility: Assess the difficulty of implementing the rules under existing equipment and process conditions. Product consistency: Analyze the impact of rule implementation on product consistency between batches.

[0077] Experts can judge whether the rules mined are valuable: Traditional verification: confirm whether the rules are consistent with traditional brewing experience. New discovery identification: identify new rules that may break through traditional cognition and design experimental verification. Potential application: explore the potential application of rules in new product development or process improvement.

[0078] Safety and compliance checks of rules: experts need to ensure that the rules comply with food safety standards and industry specifications: Raw material safety: ensure that the raw material combinations involved in the rules meet food safety requirements. Process compliance: verify whether the process parameters comply with industry standards and regulatory requirements. Quality control: evaluate the impact of rule implementation on product quality stability.

[0079] Through these in-depth expert analysis and interpretation, the results of data mining can be transformed into more valuable practical guidance, which not only ensures the scientific nature of the rules, but also ensures their applicability in actual production. This method of combining expert knowledge can significantly improve the practical value of association rules in the production of Maotai-flavor liquor and promote the organic combination of traditional craftsmanship and modern technology.

[0080] At step S204, a user recipe rating matrix is ​​constructed in combination with user preference information, and the recipe similarity is calculated using the Pearson correlation coefficient to predict the target user's rating of the candidate recipe using a project-based collaborative filtering algorithm. This includes: constructing a user recipe rating matrix, wherein the rating matrix includes rating data of multiple users on multiple recipes, and the sources of the rating data include professional wine taster ratings, consumer feedback, implicit ratings converted from sales data, and / or wine tasting ratings; calculating the similarity of recipe pairs, and introducing a penalty factor to adjust the similarity to reduce the similarity between recipe pairs with only a few common rating users; based on the rating data of multiple recipes that are most similar to the candidate recipe, calculating the target user's predicted rating of the candidate recipe, and introducing a user bias term and a recipe bias term to learn the bias term by minimizing the prediction error; using a content-based recommendation method for new users, and searching for similar recipes for new recipes based on features including raw material composition and / or process parameters.

[0081] Specifically, when implementing the item-based collaborative filtering algorithm, we first need to construct a user-recipe rating matrix . Assume there are 1000 users and 500 recipes, then is a matrix of 1000 × 500. In practice, this matrix is ​​usually sparse because most users have only tasted a few recipes.

[0082] Rating data can come from multiple channels, such as: 1. Ratings from professional wine tasters: This is the most authoritative source of ratings, but the number is limited. 2. Consumer feedback: A large amount of consumer rating data collected through online and offline channels. 3. Sales data: Sales can be converted into implicit ratings. 4. Wine tasting ratings: Rating data collected at various wine tasting events.

[0083] Next, calculate the similarity between recipes. Commonly used similarity calculation methods include Pearson correlation coefficient, cosine similarity, etc. Taking cosine similarity as an example, for recipes i and j, the similarity calculation formula is: ,in, is the set of users who rated both recipes i and j.

[0084] In practical applications, similarity calculation can be optimized. For example, considering the sparsity of rating data, a penalty factor can be introduced: ,in, and denote the number of users who rated recipes i and j, respectively. This can reduce the similarity between recipe pairs that have only a few common rating users.

[0085] After calculating the similarity between recipes, for the target recipe t, the score of user u on it can be predicted, wherein the predicted score of the target user on the candidate recipe is calculated based on the score data of multiple recipes that are most similar to the candidate recipe, including: ,in, Indicates user About the recipe The prediction score of Representation and formulation Most similar A collection of recipes, Recipe and The similarity of Indicates user About the recipe In practical applications, the selection of k needs to balance computational efficiency and prediction accuracy. The optimal k value can usually be determined by methods such as cross-validation.

[0086] To further improve prediction accuracy, user and recipe bias factors can be taken into account: in, is the global average rating, and are the bias terms for user u and recipe t, respectively. These bias terms can be learned by minimizing the prediction error: ,in, is a regularization parameter used to prevent overfitting.

[0087] In practical applications, we also need to consider the cold start problem, that is, how to deal with new users or new recipes. For new users, we can use content-based recommendation methods to recommend recipes that similar users like based on user attributes (such as age, gender, region, etc.). For new recipes, we can find similar existing recipes for recommendation based on their raw material composition, process parameters and other characteristics.

[0088] In addition, in order to improve the diversity and novelty of the recommendations, some randomness can be introduced. For example, in the final recommendation list, 80% of the high-scoring predicted recipes can be retained, and 20% of the positions can be left for randomly selected recipes. This can help users discover new recipes that they may like, and also help the system collect more rating data.

[0089] At step S205, a personalized recipe optimization recommendation list is automatically generated based on the score prediction results, association rules and mapping models. This includes: using the support vector machine regression model to predict the sensory index of the new recipe to obtain a predicted sensory index vector; constructing a comprehensive scoring function to integrate multiple sensory indicators into a single score, wherein the scoring function includes a linear term, a quadratic interaction term and a threshold penalty term; constructing the recipe optimization problem as a constrained optimization problem, setting the objective function as a maximization scoring function, and applying total quantity constraints, cost constraints, inventory constraints and process constraints; using a global optimization algorithm to solve the constrained optimization problem to obtain an optimized recipe, wherein a multi-starting point strategy is used to improve the solution efficiency; combining the association rule mining results to generate a new recipe combination, and fine-tuning it through the optimization algorithm to form a personalized recipe optimization recommendation list.

[0090] Specifically, after completing the previous steps (i.e., S201-S204), the next step is to integrate these analysis results to achieve automatic recommendation for formula optimization. This process involves multiple considerations, and it is necessary to balance multiple factors such as sensory quality, cost control, and production efficiency. First, based on the prediction results of the support vector machine regression model, the expected sensory quality of a new formula can be quickly evaluated. Assuming there are n sensory indicators, for a new formula , the predicted sensory index vector is: ,in, is the prediction result of the ith SVM regression model.

[0091] Next, we need to define a comprehensive scoring function that combines multiple sensory indicators into a single score. A simple approach is to use a weighted sum: ,in, is the weight of the i-th sensory index. These weights can be set based on expert experience or learned from market research data. However, simple linear weighting may not fully capture the relationship between sensory indicators. Therefore, a more accurate scoring function can be considered, for example, where a comprehensive scoring function is constructed, including: ,in, Recipe Total rating, Indicates The weight of each sensory index, Indicates Predicted sensory indicators, Indicates and The interaction weights between the sensory indicators are Indicates The penalty weight of an indicator, Represents the threshold of the i-th indicator. This function contains linear terms, quadratic interaction terms, and threshold penalty terms. Among them, It captures the interactions between indicators. and are the penalty weight and threshold of the ith indicator, respectively, which are used to ensure that certain key indicators do not fall below acceptable levels.

[0092] In practical applications, the specific form of the scoring function may be more complex and may require consideration of factors such as nonlinear relationships and multi-level thresholds. Determining the best scoring function form and parameters is a process that requires repeated trials and adjustments.

[0093] With the scoring function, the recipe optimization problem can be formalized as a constrained optimization problem:

[0094] ;

[0095] s :

[0096] ;

[0097] ;

[0098] ;

[0099] in, and are inequality and equality constraints, respectively. and are the upper and lower limit constraints for each component of the formula. These constraints can include:

[0100] 1. Total quantity constraint: The sum of the proportions of all components must equal 100%.

[0101] 2. Cost constraint: The total cost cannot exceed the preset upper limit.

[0102] 3. Inventory constraints: The usage of each component cannot exceed the current inventory.

[0103] 4. Process constraints: The proportions of certain components need to meet specific process requirements.

[0104] This is a nonlinear optimization problem that can be solved using various optimization algorithms, such as gradient descent, Newton's method, genetic algorithm, etc. Considering the nature of the problem and the possible existence of multiple local optimal solutions, global optimization algorithms are usually used, such as simulated annealing, particle swarm optimization, etc.

[0105] In practical applications, a multi-starting strategy can be adopted, that is, starting from multiple initial formulas and then selecting the best result. These initial formulas can come from the best historical formulas, expert recommended formulas, and high-scoring formulas recommended by collaborative filtering algorithms.

[0106] In addition, considering the particularity of liquor production, formula optimization also needs to consider the differences between batches. The idea of ​​robust optimization can be introduced to model the differences between batches as uncertainty, and then solve the robust optimization problem: ,in, Indicates uncertain parameters, is an uncertain set. Although the solution obtained in this way may not be the optimal solution in a specific case, it can perform well in various possible cases.

[0107] Finally, in order to improve the diversity and accuracy of recommendations, the results of association rule mining can be combined. For example, some new recipe combinations can be generated based on the mined rules, and then fine-tuned through the optimization algorithm. In this way, new recipe possibilities can be explored while ensuring sensory quality.

[0108] In actual operation, recipe optimization is an iterative process. After the system gives a recommended recipe, it needs to go through actual production and tasting, and then the results are fed back to the system to continuously update and optimize the model. This process can be achieved through online learning algorithms, allowing the system to continuously adapt to new data and changing market demands.

[0109] In general, the optimization of the blending formula of Maotai-flavor liquor combines technologies from multiple fields such as big data analysis, machine learning, and optimization theory, while fully considering industry characteristics and actual production needs. Through this digital and intelligent method, the efficiency and effect of formula optimization can be greatly improved, providing strong technical support for the quality improvement and personalized customization of Maotai-flavor liquor.

[0110] Furthermore, an embodiment of the present application also provides a big data analysis device for optimizing and digitally managing the blending formula of Maotai-flavor liquor, comprising: a processor, a memory, and a system bus; the processor and the memory are connected via the system bus; the memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, enable the processor to execute any of the above methods.

[0111] Furthermore, an embodiment of the present application also provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any of the above methods.

[0112] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0113] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0114] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0115] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A big data analysis system for optimizing and digitally managing the blending formula of Maotai-flavor liquor, characterized in that: include: data unit, modeling unit, association unit and matching unit; wherein, The data unit is used to construct a raw material sensory evaluation index system including raw wine and spices, and collect sensory evaluation data of each batch of formula to obtain a sensory index data set; The modeling unit is used to construct a regression mapping model between the concentration of the formula components of the Maotai-flavor liquor and the sensory index, and use a Gaussian kernel function to map the formula components to a high-dimensional space to model the nonlinear relationship; The association unit is used to mine association rules between different raw material components from the historical high-quality formula database using an association rule mining algorithm to obtain frequent item sets and their confidences; The matching unit is used to construct a user recipe rating matrix in combination with user preference information, and calculate recipe similarity using the Pearson correlation coefficient, so as to predict the target user's rating of the candidate recipe using a project-based collaborative filtering algorithm; and automatically generate a personalized recipe optimization recommendation list based on the rating prediction results and association rules; Among them, according to the score prediction results, association rules and mapping models, a personalized recipe optimization recommendation list is automatically generated, including: The sensory index of the new formula is predicted using the support vector machine regression model to obtain the predicted sensory index vector; Constructing a comprehensive scoring function of the sensory index vector and constructing the recipe optimization problem as a constrained optimization problem; A global optimization algorithm is used to solve the constrained optimization problem to obtain an optimization formula, wherein a multi-starting point strategy is used to improve the solving efficiency; In combination with the association rule mining results, a new recipe combination is generated based on the optimized recipe, and the support vector machine regression model and the comprehensive scoring function are fine-tuned through the optimization algorithm to obtain a fine-tuned personalized recipe optimization recommendation list; Among them, the user recipe rating matrix is ​​constructed in combination with user preference information, and the recipe similarity is calculated using the Pearson correlation coefficient, so as to use the item-based collaborative filtering algorithm to predict the target user's rating of the candidate recipe, including: Constructing a user recipe rating matrix, wherein the rating matrix includes rating data of multiple users on multiple recipes, and the rating data sources include professional wine taster ratings, consumer feedback, implicit ratings converted from sales data, and / or wine tasting event ratings; Based on the similarity of the recipe pairs, obtain the target user's predicted rating for the candidate recipes; Adopt content-based recommendation methods for new users and find similar recipes for new recipes based on characteristics including raw material composition and / or process parameters; Among them, based on the similarity of the recipe pair, the target user's predicted score for the candidate recipe is obtained, including: Calculate the similarity of recipe pairs and introduce a penalty factor to adjust the similarity to reduce the similarity between recipe pairs with only a few common rating users; Based on the rating data of multiple recipes most similar to the candidate recipe, a predicted rating of the target user for the candidate recipe is calculated, and a user bias term and a recipe bias term are introduced to learn the bias term by minimizing the prediction error; The target user's predicted score for the candidate recipe is calculated based on the score data of multiple recipes that are most similar to the candidate recipe, including: , in, Indicates user About the recipe The prediction score of Representation and formulation Most similar A collection of recipes, Recipe and The similarity of Indicates user About the recipe 's rating.

2. The big data analysis system according to claim 1, characterized in that: in, A regression mapping model between the concentration of sauce-flavor liquor formula ingredients and sensory indicators was constructed, and the Gaussian kernel function was used to map the formula ingredients to a high-dimensional space to model nonlinear relationships, including: Standardizing the formula ingredient concentration vector and sensory evaluation index vector, including calculating the standardized value using the Z-score method; The Gaussian kernel function was used to construct a support vector machine regression mapping model, and the penalty factor, ε parameter and γ parameter were optimized by grid search combined with cross-validation method to model the nonlinear relationship between formula ingredients and sensory indicators. The model performance was evaluated using mean square error, mean absolute error and coefficient of determination, and the influence of each formula component on the sensory index was determined by calculating the prediction error increment after feature perturbation.

3. The big data analysis system according to claim 1, characterized in that: in, The association rule mining algorithm is used to mine the association rules between different raw material components from the historical high-quality formula database to obtain frequent item sets and their confidence, including: Binary code the usage of raw materials in the historical high-quality formula database, and discretize the continuous concentration value into three levels: low, medium, and high. Each ingredient is represented by three binary bits. Set the minimum support and minimum confidence thresholds, use the Apriori algorithm to gradually generate candidate item sets, and use the pruning strategy to filter frequent item sets until no new frequent item sets can be generated; Generate association rules based on the obtained frequent item sets, calculate the support, confidence and lift of each rule, and filter out effective rules that meet the threshold requirements; Combined with domain expert knowledge, the mined association rules are screened.

4. The big data analysis system according to claim 1, characterized in that: in, A comprehensive scoring function of the sensory index vector is constructed, and the recipe optimization problem is constructed as a constrained optimization problem, including: Constructing a comprehensive scoring function of the sensory index vector to integrate multiple sensory indexes into a single score, wherein the scoring function includes a linear term, a quadratic interaction term, and a threshold penalty term; The recipe optimization problem is constructed as a constrained optimization problem, the objective function is set as the maximization score function, and the total quantity constraints, cost constraints, inventory constraints and process constraints are imposed.

5. The big data analysis system according to claim 4, characterized in that: in, Construct a comprehensive scoring function, including: ,in, Recipe Total rating, Indicates The weight of each sensory index, Indicates Predicted sensory indicators, Indicates and The interaction weights between the sensory indicators are Indicates The penalty weight of an indicator, Indicates the threshold of the i-th indicator.

Citation Information

Patent Citations

  • Recommendation method for searching target user and matching target product for automobile industry

    CN106844787A

  • Association rule assessment method and apparatus based on machine learning

    WO2023024411A1