A hanging ear coffee formula optimization method, medium and system

Abstract

The application provides a hanging ear coffee formula optimization method, medium and system, belongs to the hanging ear coffee formula technical field, including: first, collect the formula information and evaluation data of a plurality of hanging ear coffee, including aroma, taste, caffeine content and the like. Based on these data, a subjective and objective fusion model for predicting evaluation vectors is established. Next, the feature vector of each hanging ear coffee formula is constructed, including bean species, grinding degree, roasting degree, etc. Using data augmentation technology, the formulas are fine-tuned to obtain an augmented formula set. Then, a simpler and more efficient student model is trained using the knowledge distillation method to predict evaluation indicators based on the formula features. Finally, the student model is used to iteratively optimize the basic formula to be optimized, and the formula with the optimal evaluation index is finally selected as the final optimization result, which solves the technical problem that the existing hanging ear coffee formula design optimization needs to rely on the accumulation of a large amount of artificial experience.

Description

Technical Field

[0001] This invention belongs to the field of drip bag coffee formulation technology, specifically, it relates to a method, medium and system for optimizing drip bag coffee formulation. Background Technology

[0002] Coffee, as one of the world's most popular beverages, exhibits significant differences in taste due to variations in origin and roasting methods. For drip bag coffee, its unique brewing method brings out the true flavor of the coffee beans. Currently, numerous brands of drip bag coffee products are available on the market, differing in coffee bean type, grind size, roasting level, and other formulation parameters, resulting in a rich variety of flavor characteristics. Professional coffee tasters can score these drip bag coffee products based on their aroma, taste, and flavor through sensory evaluation. Simultaneously, some laboratories conduct objective tests on these samples, such as solubility and caffeine content. These subjective sensory evaluations and objective test indicators are crucial evidence reflecting the quality of drip bag coffee.

[0003] However, existing product development models still have some problems. First, subjective sensory evaluation and objective testing indicators are independent of each other, lacking an effective integration mechanism. Relying solely on evaluation information from one aspect cannot yield a comprehensive and objective quality judgment. Second, existing product optimization processes largely rely on empirical adjustments, making it difficult to systematically explore the impact of different combinations of formula parameters on the final quality. Furthermore, even after a high-quality formula has been determined, targeted adjustments are still needed in actual production based on batch variations of raw materials, requiring a significant accumulation of manual experience. In other words, the current optimization of drip bag coffee formula design relies heavily on the accumulation of manual experience. Summary of the Invention

[0004] In view of this, the present invention provides a method, medium and system for optimizing drip bag coffee recipes, which solves the technical problem that the optimization of existing drip bag coffee recipe design relies on the accumulation of a large amount of manual experience.

[0005] This invention is implemented as follows:

[0006] The first aspect of the present invention provides a method for optimizing drip bag coffee recipes, comprising the following steps:

[0007] S10. Collect samples of drip coffee bags from various brands, origins, and roasting levels to develop their recipes.

[0008] S20. Obtain evaluation data for each type of drip bag coffee, including subjective and objective indicators;

[0009] S30. Input the evaluation data of each type of drip bag coffee into the pre-trained subjective-objective fusion model to obtain the evaluation vector of each type of drip bag coffee.

[0010] S40. Construct the feature vector of drip bag coffee recipe;

[0011] S50. Establish data pairs of formulations and evaluation vectors as basic data pairs; use data seed fine-tuning augmentation to fine-tune the formulations to obtain an augmented formulation set, where the evaluation vector corresponding to each augmented formulation is the evaluation vector corresponding to the augmented seed; and establish augmented data pairs.

[0012] S60. Train a teacher model using basic data pairs and augmented data pairs to predict evaluation vectors based on recipe feature vectors.

[0013] S70. Design a simpler student model and use knowledge distillation to transfer knowledge from the teacher model to the student model. The training objectives of the student model include: minimizing the difference between the student model's predictions and the teacher model's predictions; and minimizing the difference between the student model's predictions and the true evaluation vectors.

[0014] S80. Obtain the basic formula to be optimized, use the data seed fine-tuning augmentation method to obtain the basic formula augmentation set, use the student model to calculate each formula data in the basic formula augmentation set to obtain the evaluation vector, and select multiple formulas with the best evaluation vector as candidate formulas.

[0015] S90. Obtain the evaluation data for each candidate formula and score it using the evaluation data experience scoring formula. Use the formula with the highest score as the optimized formula of the base formula to be optimized.

[0016] Based on the above technical solution, the method for optimizing drip bag coffee formula of the present invention can be further improved as follows:

[0017] The subjective indicators include aroma, mouthfeel, flavor, body, balance, and aftertaste.

[0018] Furthermore, the objective indicators include solubility, soaking time, optimal water temperature range, caffeine content, total acidity, and total solids content.

[0019] Furthermore, the characteristic vector of drip bag coffee formulation includes coffee bean type, grind size, roasting degree, formulation ratio, and powder particle size distribution.

[0020] Furthermore, the subjective-objective fusion model employs a multilayer perceptron, with inputs including subjective indicator vectors and objective indicator vectors, and outputting a comprehensive evaluation vector.

[0021] Furthermore, the teacher model employs a multilayer perceptron structure, with the goal of minimizing the MSE loss between the predicted evaluation vector and the actual evaluation vector.

[0022] Furthermore, the training objective of the student model includes two parts: first, minimizing the difference between the student model's prediction results and the teacher model's prediction results; and second, minimizing the difference between the student model's prediction results and the true evaluation vector Ei.

[0023] Furthermore, the experience scoring formula is a weighted sum of subjective and objective indicator scores.

[0024] Specifically, step S10 includes collecting various samples from major drip bag coffee brands on the market, covering coffee samples from different origins and roasting levels. First, for mainstream drip bag coffee brands on the market, comprehensive coffee samples are collected. These samples should cover typical coffee bean varieties, different grind sizes, and roasting levels. By extensively collecting a large amount of sample data, relatively comprehensive drip bag coffee recipe information can be obtained, providing a basis for subsequent evaluation analysis and recipe optimization.

[0025] Specifically, step S20 includes the following steps: inviting experienced professional tasters to conduct sensory evaluations of the collected drip coffee samples, with a scoring range of 0 to 10 points. Evaluation indicators include six aspects: aroma, mouthfeel, flavor, body, balance, and aftertaste. Simultaneously, laboratory instruments are used to test objective indicators for each drip coffee sample, including solubility, steeping time, optimal water temperature range, caffeine content, total acidity, and total solids content. By combining sensory evaluation and experimental testing, comprehensive evaluation data for each drip coffee sample can be obtained, providing a foundation for subsequent training of the subjective-objective fusion model.

[0026] Specifically, step S30 is implemented by constructing a subjective-objective fusion model using a multilayer perceptron (MLP) structure. The model's input includes a subjective indicator vector and an objective indicator vector, and its output is a comprehensive evaluation vector. The subjective indicator vector consists of six sub-indicators: aroma, mouthfeel, flavor, body, balance, and aftertaste. The objective indicator vector consists of six objective test indicators. Through supervised learning, using actual expert evaluation data as training labels, the subjective-objective fusion model is trained to effectively integrate subjective sensory evaluations and objective test indicators, outputting a more comprehensive evaluation vector.

[0027] Specifically, step S40 includes constructing a feature vector for drip bag coffee recipes with five dimensions: coffee bean type, grind size, roast level, recipe ratio, and powder particle size distribution. The coffee bean type is represented by a numerical code; the grind size is quantified by visually inspected particle size in micrometers; the roast level is also represented by a numerical code; the recipe ratio is expressed as a percentage of each type of coffee bean; and the powder particle size distribution is described using two indicators: average particle size and standard deviation. This feature vector will serve as input data for the subsequent optimization model.

[0028] Specifically, step S50 includes generating an augmented formula feature vector by fine-tuning while retaining the original formula features. Specifically, the original formula feature vector F is randomly perturbed by a fine-tuning vector δ that follows a normal distribution to obtain the augmented formula feature vector F. aug =F + δ. This data augmentation method helps improve the generalization ability of subsequent models. The augmented recipe feature vectors and their corresponding true evaluation vectors constitute the augmented dataset, which will be used together with the original dataset for model training.

[0029] Specifically, step S60 is implemented by training a complex teacher model using a multilayer perceptron structure. The training objective of this teacher model is to minimize the MSE loss between its predicted evaluation vector and the true evaluation vector. By training the teacher model using basic data pairs and augmented data pairs, it can learn the complex mapping relationship between recipe features and evaluations, providing support for subsequent knowledge distillation.

[0030] Specifically, step S70 is implemented by training a simpler and more efficient student model using a knowledge distillation method. The training objectives of the student model include two parts: first, minimizing the difference between the student model's predictions and the teacher model's predictions; and second, minimizing the difference between the student model's predictions and the actual evaluations. Through this distillation training method, the student model can further improve its fitting ability and generalization performance while retaining the knowledge of the teacher model.

[0031] Specifically, step S80 is implemented by first using a data seed fine-tuning augmentation method to obtain a base formula augmentation set. Then, these augmented formulas are input into a trained student model to obtain a predicted evaluation vector for each formula. Finally, the top K formulas with the best predicted evaluation vectors are selected as the candidate formula set. This step allows for the selection of several high-quality candidate formulas from a large number of formula variations.

[0032] Specifically, step S90 includes: conducting actual sensory evaluations and experimental tests on the candidate formulation set selected in step S80 to obtain complete evaluation data for each formulation. Then, an empirical scoring formula is used to comprehensively score each formulation, and the formulation with the highest score is the final optimized formulation. This scoring formula includes a weighted sum of subjective and objective indicator scores, and the weighting coefficients can be set by expert experience or determined using the analytic hierarchy process (AHP).

[0033] Furthermore, in step S30, the calculation process of the subjective-objective fusion model can be represented as follows:

[0034] E = f(S,O);

[0035] In the formula, E is the evaluation vector; S is the subjective index vector; O is the objective index vector; and f is the pre-trained subjective-objective fusion model function.

[0036] The method for obtaining the subjective indicator vector S is as follows:

[0037] S = [s1,s2,s3,s4,s5,s6];

[0038] In the formula, s1 represents the aroma score; s2 represents the mouthfeel score; s3 represents the flavor score; s4 represents the body score; s5 represents the balance score; and s6 represents the aftertaste score. These scores are obtained through evaluation by professional tasters, with a range of 0-10 points.

[0039] The method for obtaining the objective indicator vector O is as follows:

[0040] O = [o1, o2, o3, o4, o5, o6];

[0041] In the formula, o1 is solubility (%); o2 is soaking time (seconds); o3 is optimal water temperature range (°C); o4 is caffeine content (mg / 100ml); o5 is total acidity (pH value); and o6 is total solids content (%). These indicators are obtained by laboratory instrument measurement.

[0042] Among them, the subjective-objective fusion model:

[0043] (1) Model structure:

[0044] The subjective-objective fusion model adopts a multilayer perceptron (MLP) structure, as follows:

[0045] h1 = σ(W1[S,O] + b1);

[0046] h2=σ(W2h1+b2);

[0047] E = W3h2 + b3;

[0048] In the formula, h1 and h2 are the hidden layer outputs; W1, W2, and W3 are the weight matrices; b1, b2, and b3 are the bias vectors; and σ is the activation function, which uses the ReLU function: σ(x) = max(0,x).

[0049] (2) Training dataset:

[0050] Training dataset D fusion ={(S i O i E i )|i=1,2,…,N}, where N is the sample size. S i O i E represents the subjective index vector and objective index vector of the i-th sample, respectively. i This is the comprehensive evaluation vector given by the experts.

[0051] (3) Training steps:

[0052] a. Initialize model parameters θ = {W1, W2, W3, b1, b2, b3};

[0053] b. Define the loss function:

[0054] c. Minimize the loss function using stochastic gradient descent (SGD) or the Adam optimizer:

[0055]

[0056] In the formula, η is the learning rate; t is the number of iterations.

[0057] d. Repeat step c until the model converges or the preset number of iterations is reached.

[0058] In step S40, the process of constructing the recipe feature vector is as follows:

[0059] F = [f1, f2, f3, f4, f5];

[0060] In the formula, F is the formula feature vector; f1 is the type of coffee bean (represented by numerical code); f2 is the grind size (visual particle size, unit: μm); f3 is the roasting degree (represented by numerical code, such as 1-10); f4 is the formula ratio (the proportion of various coffee beans, expressed as a percentage); and f5 is the powder particle size distribution (represented by average particle size and standard deviation).

[0061] In step S50, the process of fine-tuning and augmenting the data seed can be represented as follows:

[0062] F aug =F+δ;

[0063] In the formula, Faug F is the augmented formula feature vector; F is the original formula feature vector; δ is the fine-tuning vector, where each element follows a normal distribution N(0, σ). 2 ), where σ is the preset fine-tuning intensity.

[0064] In step S60, the training process of the teacher model can be represented as follows:

[0065]

[0066] In the formula, θ T For the teacher model parameters; N is the number of data pairs; E i F is the true evaluation vector for the i-th sample; i Let T(F) be the feature vector of the formula for the i-th sample; i ;θ T ) represents the prediction result of the teacher model for the i-th sample.

[0067] In step S70, the training process of the student model can be represented as follows:

[0068]

[0069] In the formula, θ S For student model parameters; S(F i ;θ S ) represents the prediction result of the student model for the i-th sample; α and β are weighting coefficients used to balance the importance of the two loss terms.

[0070] In step S80, the selection process for the candidate formulation can be represented as follows:

[0071] F selected =TopK({S(F j ;θ S )|F j ∈F aug_base},K);

[0072] In the formula, F selected For the selected set of candidate recipes; F aug_base TopK is the augmented set of the basic recipes; TopK is a function that selects the top K recipes with the best evaluation vectors, where K defaults to 3 to 10.

[0073] In step S90, the empirical scoring formula for the evaluation data can be expressed as follows:

[0074]

[0075] In the formula, Score is the final score; w k For subjective indicator weights; s k For subjective indicator scores; v mWeights for objective indicators; m Scoring is assigned to objective indicators. Weighting coefficients are determined either through expert experience or using the Analytic Hierarchy Process (AHP).

[0076] The final process for selecting the optimized formula is as follows:

[0077]

[0078] In the formula, F opt To ultimately optimize the formula; Score(F) j Formula F j The score.

[0079] A second aspect of the present invention provides a computer-readable storage medium storing program instructions that, when executed in a computer, perform the aforementioned method for optimizing a drip bag coffee recipe.

[0080] A third aspect of the present invention provides a drip bag coffee recipe optimization system, wherein the system includes the aforementioned computer-readable storage medium.

[0081] Compared with existing technologies, the beneficial effects of the drip bag coffee recipe optimization method, medium, and system provided by this invention are as follows: First, the solution of this invention effectively integrates subjective sensory evaluation and objective test indicators. By constructing a subjective-objective fusion model, these two types of evaluation information can be integrated to output a more comprehensive and objective quality assessment vector. This fusion mechanism can overcome the limitations of a single evaluation indicator and provide a more reliable basis for subsequent recipe optimization.

[0082] Secondly, this invention employs data augmentation and knowledge distillation techniques to fully utilize limited sample data and train a relatively simple and efficient student model. This model not only has strong fitting ability but also generalizes well to new formulation situations. Therefore, compared to traditional methods that rely on experience-based adjustments, this invention can more systematically explore the influence of different combinations of formulation parameters on the final quality.

[0083] Furthermore, the optimization process of this invention is based on model prediction, rather than direct manual evaluation. Once the optimal formulation is determined, in actual production, only rapid prediction using the model is needed to quickly adjust the formulation to adapt to differences in raw materials. This model-based optimization method greatly reduces the burden of accumulating manual experience.

[0084] In summary, the subjective and objective integration formula optimization method of this invention fully leverages the advantages of combining quantitative analysis and qualitative evaluation, and solves the technical problem that existing drip bag coffee formula design and optimization rely on the accumulation of a large amount of manual experience. Attached Figure Description

[0085] Figure 1 A flowchart of the method provided by the present invention. Detailed Implementation

[0086] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0087] like Figure 1 The diagram shown is a flowchart of a method for optimizing drip bag coffee recipes provided by this invention. This method includes the following steps:

[0088] S10. Collect samples of drip coffee bags from various brands, origins, and roasting levels to develop their recipes.

[0089] S20. Obtain evaluation data for each type of drip bag coffee, including subjective and objective indicators;

[0090] S30. Input the evaluation data of each type of drip bag coffee into the pre-trained subjective-objective fusion model to obtain the evaluation vector of each type of drip bag coffee.

[0091] S40. Construct the feature vector of drip bag coffee recipe;

[0092] S50. Establish data pairs of formulations and evaluation vectors as basic data pairs; use data seed fine-tuning augmentation to fine-tune the formulations to obtain an augmented formulation set, where the evaluation vector corresponding to each augmented formulation is the evaluation vector corresponding to the augmented seed; and establish augmented data pairs.

[0093] S60. Train a teacher model using basic data pairs and augmented data pairs to predict evaluation vectors based on recipe feature vectors.

[0094] S70. Design a simpler student model and use knowledge distillation to transfer knowledge from the teacher model to the student model. The training objectives of the student model include: minimizing the difference between the student model's predictions and the teacher model's predictions; and minimizing the difference between the student model's predictions and the true evaluation vectors.

[0095] S80. Obtain the basic formula to be optimized, use the data seed fine-tuning augmentation method to obtain the basic formula augmentation set, use the student model to calculate each formula data in the basic formula augmentation set to obtain the evaluation vector, and select the multiple formulas with the best evaluation vector as candidate formulas.

[0096] S90. Obtain the evaluation data for each candidate formula and score it using the evaluation data experience scoring formula. Use the formula with the highest score as the optimized formula of the base formula to be optimized.

[0097] The specific implementation methods of the above steps are described in detail below:

[0098] The specific implementation of step S10 is as follows: Collect samples of drip coffee bags from various brands, origins, and roasting levels. First, a comprehensive collection of samples from major drip coffee bag brands on the market, including coffee samples from different origins and roasting levels, is required. These samples should cover a variety of typical coffee bean types, different grind sizes, and roasting levels. By collecting a large number of samples, relatively comprehensive drip coffee bag recipe data can be obtained, providing a foundation for subsequent evaluation, analysis, and optimization.

[0099] The specific implementation of step S20 is as follows: Obtain evaluation data for each type of drip coffee bag. First, the collected drip coffee bag samples need to undergo professional sensory evaluation and laboratory index testing. Sensory evaluation mainly includes scoring in six aspects: aroma, mouthfeel, flavor, body, balance, and aftertaste, with a scoring range of 0-10 points. This requires inviting experienced professional tasters to conduct the scoring. The sensory evaluation vector can be represented as:

[0100] S = [s1,s2,s3,s4,s5,s6];

[0101] Among them, s1 is the aroma score, s2 is the mouthfeel score, s3 is the flavor score, s4 is the body score, s5 is the balance score, and s6 is the aftertaste score.

[0102] Simultaneously, objective indicators need to be tested for each drip coffee sample, including solubility, steeping time, optimal water temperature range, caffeine content, total acidity, and total solids content. These objective indicators can be measured using laboratory instruments. The objective evaluation vector can be represented as:

[0103] O = [o1, o2, o3, o4, o5, o6];

[0104] Where o1 is solubility (%), o2 is soaking time (seconds), o3 is optimal water temperature range (°C), o4 is caffeine content (mg / 100ml), o5 is total acidity (pH value), and o6 is total solids content (%).

[0105] Sensory evaluation and experimental testing can provide comprehensive evaluation data for each drip bag coffee sample.

[0106] The specific implementation of step S30 is as follows: Constructing a subjective-objective fusion model. First, a multilayer perceptron (MLP) structure is needed to construct the subjective-objective fusion model. The input of this model includes a subjective index vector S and an objective index vector O, and the output is a comprehensive evaluation vector E. The calculation process of the subjective-objective fusion model can be represented as follows:

[0107] E = f(S,O);

[0108] Where f is the pre-trained subjective-objective fusion model function.

[0109] The model structure is as follows:

[0110] h1 = σ(W1[S,O] + b1);

[0111] h2=σ(W2h1+b2);

[0112] E = W3h2 + b3;

[0113] Where h1 and h2 are the hidden layer outputs, W1, W2, and W3 are the weight matrices, b1, b2, and b3 are the bias vectors, and σ is the activation function, using the ReLU function: σ(x) = max(0,x).

[0114] Training dataset D fusion Composed of several triples (S i O i E i Composed of ) where S i O i E represents the subjective index vector and objective index vector of the i-th sample, respectively. i This is the comprehensive evaluation vector given by the experts.

[0115] The training steps are as follows:

[0116] a. Initialize model parameters θ = {W1, W2, W3, b1, b2, b3};

[0117] b. Define the loss function:

[0118] c. Minimize the loss function using stochastic gradient descent (SGD) or the Adam optimizer:

[0119]

[0120] Where η is the learning rate and t is the number of iterations;

[0121] d. Repeat step c until the model converges or the preset number of iterations is reached.

[0122] Through such training, the subjective-objective fusion model can learn the complex mapping relationship between subjective sensory evaluation and objective test indicators, and output a more comprehensive evaluation vector.

[0123] The specific implementation of step S40 is as follows: Constructing the drip bag coffee recipe feature vector. The recipe feature vector F includes five aspects:

[0124] F = [f1, f2, f3, f4, f5];

[0125] Wherein, f1 is the type of coffee bean (represented by a numerical code), f2 is the grind size (visually estimated particle size, unit: μm), f3 is the roasting degree (represented by a numerical code, such as 1-10), f4 is the formula ratio (the proportion of various coffee beans, expressed as a percentage), and f5 is the powder particle size distribution (expressed as average particle size and standard deviation).

[0126] These five feature dimensions can comprehensively describe the formulation characteristics of a drip bag coffee. This feature vector will serve as the input for subsequent optimization models.

[0127] The specific implementation of step S50 is as follows: Data seed fine-tuning augmentation is performed. First, based on the original formulation feature vector F, an augmented formulation feature vector F is generated using a fine-tuning method. aug Specifically,

[0128] F aug =F+δ;

[0129] Where δ is a normal distribution N(0,σ) 2 The augmented formula feature vectors are given by σ, where σ is the preset adjustment strength. This allows for the generation of approximate formula variants through random perturbation while preserving the original formula features. These augmented formula feature vectors and their corresponding true evaluation vectors constitute the augmented dataset, which will be used together with the original dataset for subsequent model training. This data augmentation method helps improve the model's generalization ability.

[0130] The specific implementation of step S60 is as follows: Training the teacher model. First, a complex teacher model is trained using basic data pairs and augmented data pairs. The training objective of the teacher model is to minimize the predicted evaluation vector T(F). i ;θ T ) and the true evaluation vector E i The MSE loss between them can be expressed as:

[0131]

[0132] Where, θ T Here are the parameters for the teacher model, N is the number of data pairs, and F is... iLet be the recipe feature vector for the i-th sample. Through this training, the teacher model can learn the complex mapping relationship between recipe features and evaluation. This complex teacher model will then serve as the object of subsequent knowledge distillation.

[0133] The specific implementation of step S70 is as follows: Training the student model. To obtain a simpler and more efficient model, knowledge distillation is used to transfer knowledge from the teacher model to the student model. The training objectives of the student model include two parts:

[0134]

[0135] Where, θ S For student model parameters, S(F) i ;θ S ) represents the prediction result of the student model for the i-th sample, and α and β are weight coefficients.

[0136] Through this distillation training method, the student model can further improve its fitting and generalization abilities while retaining the knowledge of the teacher model.

[0137] The specific implementation of step S80 is as follows: Select the formulation to be optimized. First, use the data seed fine-tuning augmentation method to obtain the base formulation augmentation set F. aug_base Then, these augmenting formulas are fed into the trained student model S(F). i ;θ S In the process of obtaining the predicted evaluation vector for each formulation, the top K formulations with the best predicted evaluation vectors are selected as the candidate formulation set F. selected , can be represented as:

[0138] F selected =TopK({S(F j ;θ s )|F j ∈F aug_base},K);

[0139] Here, TopK is a function that selects the top K recipes with the best evaluation vectors.

[0140] This step allows you to select the best-quality alternative formulations from a large number of formulation variations.

[0141] The specific implementation of step S90 is: evaluating data through empirical scoring. For the candidate formula set F selected in step S80... selected Further collection of their actual evaluation data is needed. Then, an empirical scoring formula will be used to give each formula a comprehensive score:

[0142]

[0143] Among them, w k and v m The weighting coefficients for subjective and objective indicators, respectively, s k and o m The scores are assigned to the corresponding subjective and objective indicators. The weighting coefficients can be set by expert experience or determined using the Analytic Hierarchy Process (AHP).

[0144] The formula with the highest score is the final optimized formula F. opt , can be represented as:

[0145]

[0146] In summary, this method for optimizing drip bag coffee recipes fully utilizes both subjective sensory evaluation and objective testing indicators. Through a series of steps, including constructing a fusion model of subjective and objective factors, performing data augmentation, and training a teacher-student model, the optimal recipe is ultimately selected. The entire process embodies an optimization approach that combines quantitative analysis and qualitative evaluation, enabling a relatively comprehensive assessment of the quality characteristics of different recipes.

[0147] A second aspect of the present invention provides a computer-readable storage medium storing program instructions that, when executed in a computer, perform the aforementioned method for optimizing a drip bag coffee recipe.

[0148] A third aspect of the present invention provides a drip bag coffee recipe optimization system, wherein the system includes the aforementioned computer-readable storage medium.

[0149] Specifically, the principle of this invention is to construct a model framework that can effectively integrate subjective sensory evaluation and objective test indicators. Through comprehensive collection and evaluation of a large number of drip coffee samples, abundant subjective indicator data (such as aroma and taste) and objective indicator data (such as solubility and caffeine content) were obtained. This data provides the foundation for subsequent training of the subjective-objective fusion model.

[0150] The subjective-objective fusion model employs a multilayer perceptron (MLP) structure. Inputs include subjective and objective indicator vectors, and the output is a comprehensive evaluation vector. The model is trained using supervised learning, utilizing actual expert evaluation data as the training objective. By minimizing the difference between the model's predictions and actual evaluations, a complex mapping relationship between subjective perception and objective test indicators can be learned. This fusion model provides a more comprehensive and objective quality assessment, supporting subsequent formula optimization.

[0151] To further improve the model's generalization performance, this invention also employs data augmentation techniques. While retaining the original recipe features, several approximate recipe variants are generated through fine-tuning. These augmented recipes and their corresponding evaluation data together constitute the training set, which helps the model learn a wider range of recipe-quality relationships.

[0152] Furthermore, this invention proposes a model compression method based on knowledge distillation. First, a complex teacher model is used to learn the intricate mapping between recipe features and evaluations. Then, this knowledge is transferred to a simpler student model. By minimizing the differences between the student model's predictions, the teacher model's predictions, and the actual evaluations as the training objective, the student model's fitting ability and generalization performance can be further improved while retaining the teacher model's knowledge. The resulting student model not only has low computational complexity but also high prediction accuracy.

[0153] Based on a trained student model, this invention also designs a model-based recipe optimization process. First, a large number of recipe variations are generated using data augmentation. These variations are then input into the student model for evaluation and prediction, and those with the best predicted quality are selected as candidate recipes. Finally, actual evaluation data of these candidate recipes are collected and comprehensively scored, and the recipe with the highest score is selected as the final optimized recipe. This model-based optimization method can effectively explore the influence of recipe parameters on quality and guide actual recipe adjustments.

[0154] In summary, the present invention fully leverages the advantages of combining subjective sensory evaluation with objective testing indicators. Through a series of innovative technical means, such as constructing a subjective-objective fusion model, performing data augmentation, and training a teacher-student model, it achieves systematic analysis and optimization of drip bag coffee recipes.

[0155] To better understand and implement this invention, a specific application scenario is provided below: A coffee company aims to develop higher-quality products to meet consumer demand by optimizing the formula of drip coffee bags. Based on the subjective and objective fusion formula optimization method proposed in this invention, the company's R&D team carried out the following specific implementation work.

[0156] First, the team collected samples from 10 mainstream drip bag coffee brands on the market, covering different origins and roasting levels. These samples included typical coffee bean varieties such as Kenyan AA, Colombian waffle, and Brazilian yellow rhododendron, with grind sizes ranging from medium to coarse and roasting levels from light to dark. This extensive collection of sample data laid the foundation for subsequent evaluation, analysis, and recipe optimization.

[0157] Next, the R&D team invited three experienced professional coffee tasters to conduct sensory evaluations of the 10 drip coffee samples. Each taster scored independently, using six evaluation criteria: aroma, mouthfeel, flavor, body, balance, and aftertaste, with a score range of 0-10. Simultaneously, the team also conducted laboratory tests on these samples, obtaining six objective indicators: solubility, steeping time, optimal water temperature range, caffeine content, total acidity, and total solids content. Through a combination of sensory evaluation and experimental testing, a complete evaluation dataset for each drip coffee sample was obtained, as shown in Table 1.

[0158] Table 1. Coffee Sample Evaluation Dataset

[0159]

[0160]

[0161] With this comprehensive evaluation data, the R&D team then constructed a subjective-objective fusion model. This model employs a multilayer perceptron (MLP) structure, taking subjective indicator vector S and objective indicator vector O as inputs, and outputting a comprehensive evaluation vector E. The model's training process is as follows:

[0162] 1. Initialize the model parameters θ = {W1, W2, W3, b1, b2, b3}.

[0163] 2. Define the loss function: Where N is the sample size, E i Let be the true evaluation vector of the i-th sample.

[0164] 3. Minimize the loss function using the stochastic gradient descent (SGD) optimizer:

[0165]

[0166] Where η is the learning rate and t is the number of iterations.

[0167] 4. Repeat step 3 until the model converges.

[0168] After training, the subjective-objective fusion model can effectively integrate subjective sensory evaluations and objective test indicators, outputting a more comprehensive and objective evaluation vector.

[0169] To further improve the model's generalization performance, the research team also employed data augmentation techniques. While preserving the original formula features, they generated approximate formula variants through fine-tuning. Specifically, for the original formula feature vector F, we have:

[0170] F aug =F+δ;

[0171] Where δ is a normal distribution N(0,σ) 2 The fine-tuned vector of the augmented recipe is set to σ = 0.1. The resulting augmented recipe feature vector and the corresponding true evaluation vector constitute the expanded training dataset.

[0172] With the expanded training dataset, the research team then used knowledge distillation to train a simpler student model. The training objectives of this student model are as follows:

[0173]

[0174] Where, θ S For student model parameters, S(F) i ;θ S T(F) represents the prediction result of the student model for the i-th sample. i ;θ T The teacher model's predictions are represented by α and β, which are weighting coefficients with values ​​of 0.6 and 0.4, respectively. Through this distillation training method, the student model can retain the knowledge from the teacher model while further improving its fitting ability and generalization performance.

[0175] With the trained student model in hand, the research team optimized a drip coffee recipe called "Kenya AA Dark Roast". First, based on the feature vector F of this recipe, they generated 100 approximate recipe variants F using the aforementioned data augmentation method. aug_base Then, these augmented formulations are input into the student model for evaluation prediction, and the top 5 formulations with the smallest predicted evaluation vector norm are selected as candidate formulations F. selected As shown in Table 2.

[0176] Table 2. Alternative Formulas

[0177]

[0178]

[0179] Next, the R&D team collected actual evaluation data for these five alternative formulations, as shown in Table 3.

[0180] Table 3 Evaluation Data Table

[0181]

[0182] Then, the R&D team used the following empirical scoring formula to comprehensively evaluate these 5 alternative formulas:

[0183] Score=0.15·s1+0.15·s2+0.15·s3+0.10·s4+0.10·s5+0.10·

[0184] s6+0.05·o1+0.05·o2+0.05·o3+0.05·o4+0.05·o5+0.05·o6;

[0185] Among them, s i and o j The scores are for subjective and objective indicators, respectively. Calculations show that Formula 3 has the highest overall score of 92.0, and therefore it was selected as the final optimized formula.

[0186] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing drip bag coffee recipes, characterized in that, Includes the following steps: S10. Collect samples of drip coffee bags from various brands, origins, and roasting levels to develop their recipes. S20. Obtain evaluation data for each type of drip bag coffee, including subjective and objective indicators; S30. Input the evaluation data of each type of drip bag coffee into the pre-trained subjective-objective fusion model to obtain the evaluation vector of each type of drip bag coffee. S40. Construct the feature vector of drip bag coffee recipe; S50. Establish data pairs for formulation and evaluation vectors as basic data pairs; The formula is fine-tuned using a data seed augmentation method to obtain an augmented formula set. The evaluation vector corresponding to each augmented formula is the evaluation vector corresponding to the augmented seed. And establish augmented data pairs; S60. Train a teacher model using basic data pairs and augmented data pairs to predict evaluation vectors based on recipe feature vectors. S70. Design a simpler student model and use knowledge distillation to transfer knowledge from the teacher model to the student model. The training objectives of the student model include: minimizing the difference between the student model's predictions and the teacher model's predictions; and minimizing the difference between the student model's predictions and the true evaluation vectors. S80. Obtain the basic formula to be optimized, use the data seed fine-tuning augmentation method to obtain the basic formula augmentation set, use the student model to calculate each formula data in the basic formula augmentation set to obtain the evaluation vector, and select multiple formulas with the best evaluation vector as candidate formulas. S90. Obtain the evaluation data for each candidate formula and score it using the evaluation data experience scoring formula. Use the formula with the highest score as the optimized formula of the base formula to be optimized.

2. The method for optimizing drip bag coffee formulation according to claim 1, characterized in that, The subjective indicators include aroma, mouthfeel, flavor, body, balance, and aftertaste.

3. The method for optimizing drip bag coffee formulation according to claim 2, characterized in that, The objective indicators include solubility, soaking time, optimal water temperature range, caffeine content, total acidity, and total solids content.

4. The method for optimizing drip bag coffee formulation according to claim 3, characterized in that, The characteristic vector of drip bag coffee recipes includes coffee bean type, grind size, roasting degree, recipe ratio, and powder particle size distribution.

5. The method for optimizing drip bag coffee formulation according to claim 4, characterized in that, The subjective-objective fusion model employs a multilayer perceptron, with inputs including subjective indicator vectors and objective indicator vectors, and outputting a comprehensive evaluation vector.

6. The method for optimizing drip bag coffee formulation according to claim 5, characterized in that, The teacher model employs a multilayer perceptron structure, and its goal is to minimize the MSE loss between the predicted evaluation vector and the actual evaluation vector.

7. The method for optimizing drip bag coffee formulation according to claim 6, characterized in that, The training objective of the student model includes two parts: first, minimizing the difference between the student model's prediction and the teacher model's prediction; and second, minimizing the difference between the student model's prediction and the true evaluation vector Ei.

8. The method for optimizing drip bag coffee formulation according to claim 7, characterized in that, The experience scoring formula is a weighted sum of subjective and objective indicator scores.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions, which, when executed in a computer, are used to perform a drip bag coffee recipe optimization method according to any one of claims 1-8.

10. A drip bag coffee recipe optimization system, characterized in that, It includes the computer-readable storage medium of claim 9.

Images