Plant extract formula optimization method, medium and system in coffee deep processing

Abstract

The invention provides a method, medium and system for optimizing a plant extract formula in coffee deep processing, belongs to the technical field of coffee deep processing, and aims to optimize the use of plant extracts in a coffee deep processing formula through a machine learning model. The method comprises the following steps: collecting sources, extraction methods, main active ingredients and sensory evaluation and physical and chemical analysis data of various plant extracts; next, designing a series of initial formulas, adding different plant extracts into the coffee in proportion, and obtaining blind evaluation score data; based on the training data, a multi-branch embedded neural network model is established, and the score of coffee under the given plant extract ratio can be predicted. And carrying out score prediction on a basic formula to be optimized and a plurality of groups of adjustment formulas by utilizing the model. Finally, an optimization algorithm is adopted to select the formula with the highest score from all the formulas to serve as the optimal plant extract formula to be output, and the technical problem that an existing method is time-consuming and labor-consuming is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of coffee deep processing, and in particular, relates to a method, medium and system for optimizing a plant extract formula in coffee deep processing. Background Art

[0002] Different types of plant extracts can give coffee a unique flavor and taste, so adding plant extracts during the deep processing of coffee has become a widely used method. For example, vanilla extract can enhance the aroma of coffee, cinnamon extract can bring a mild sweetness, cocoa extract can give coffee a mellow cocoa flavor, mint extract can give people a refreshing feeling, and nut extract can enrich the taste of coffee. However, there are still some problems in the optimization of the ratio of plant extracts in the existing coffee deep processing technology. First of all, the sensory properties and physicochemical properties of different plant extracts vary greatly. How to effectively combine them to obtain ideal coffee quality has always been a difficult problem. At the same time, the ratio of plant extracts will not only affect the basic quality indicators of coffee such as aroma, acidity, sweetness, and bitterness, but also affect the complex sensory properties such as taste and aftertaste. There are complex interactions between these factors, which are difficult to grasp one by one. In addition, over time, the quality of coffee will change with the loss of volatile aroma components. The existing formula of plant extracts in coffee deep processing is often based on engineer experience and a large number of experiments to achieve formula optimization, which is time-consuming and labor-intensive.

[0003] Therefore, a systematic approach is urgently needed to optimize the formulation of plant extracts in coffee deep processing to meet the growing quality demands of consumers. Summary of the invention

[0004] In view of this, the present invention provides a method, medium and system for optimizing the formula of plant extracts in deep processing of coffee, which can solve the technical problems of existing methods that often rely on engineers' experience and a large number of experiments to achieve formula optimization, which is time-consuming and labor-intensive.

[0005] The present invention is achieved in that:

[0006] The first aspect of the present invention provides a method for optimizing the formula of plant extracts in coffee deep processing, which comprises the following steps:

[0007] S10. Collect a variety of plant extracts used for coffee deep processing, and record the source, extraction method, main active ingredients and their content of each extract;

[0008] S20. Obtain sensory evaluation data and physicochemical analysis data for each plant extract;

[0009] S30, obtaining a series of set initial formulas, wherein the initial formulas use the same batch of coffee beans and the same basic additives, and combine different plant extracts in different proportions to make coffee samples; and obtaining blind test score data of the coffee samples;

[0010] S40, establishing a training data set, including the plant extract ratio in each initial formula, sensory evaluation data and physicochemical analysis data of the corresponding plant extract, and blind test scoring data of coffee samples made from the initial formula; using the training data set to train a multi-branch embedded neural network model to obtain a plant extract coffee scoring model;

[0011] S50, obtaining a plant extract formula to be optimized, recorded as a basic formula, and using the plant extract coffee scoring model to obtain blind test scoring data of the basic formula using sensory evaluation data and physicochemical analysis data of the basic formula;

[0012] S60, obtaining multiple groups of adjusted formulas, and using the plant extract coffee scoring model to obtain blind test scoring data of the corresponding adjusted formulas using the sensory evaluation data and physical and chemical analysis data of each group of adjusted formulas;

[0013] S70, using an optimization algorithm to optimize the basic formula and each set of adjusted formulas according to the blind test score data, to obtain the optimal formula and output it.

[0014] On the basis of the above technical solution, the method for optimizing the formula of plant extracts in coffee deep processing of the present invention can also be improved as follows:

[0015] The plant extracts include at least vanilla essence, cinnamon extract, cocoa extract, mint extract and nut extract; the vanilla essence is specifically extracted from vanilla pods by supercritical CO2 extraction, and the main active ingredient is vanillin, with a content of not less than 35%.

[0016] Furthermore, the sensory evaluation data include aroma intensity, aroma type, taste characteristics, and taste description; the physical and chemical analysis data include total polyphenol content, antioxidant capacity, solubility, pH value, and volatile aroma component composition.

[0017] Furthermore, the basic added ingredients are specifically caffeine, sugar, milk powder and stabilizer.

[0018] Furthermore, the blind test scoring data specifically refers to the blind test scores given by the tasting team, including scores for aroma, acidity, sweetness, bitterness, taste, and aftertaste.

[0019] Furthermore, the adjusted formula is a plurality of groups of adjusted formulas obtained by adjusting the ratios in the basic formula using a data augmentation method.

[0020] Furthermore, the specific steps of using the data augmentation method to adjust the ratio in the basic formula include: 1) randomly fine-tuning the original formula, and making small adjustments to the proportions of each ingredient while keeping the total amount unchanged; 2) adding Gaussian noise to the sensory evaluation and physical and chemical analysis data to simulate measurement errors; 3) generating a new virtual formula by interpolating or extrapolating the existing formula data.

[0021] Furthermore, the multi-branch embedded neural network model includes a mathematical fitting equation group module, a sensory feature branch network, a physical and chemical feature branch network, a formula ratio branch network, a timing feature branch network and a fusion network.

[0022] Among them, the mathematical fitting equation group module is specifically an equation group that contains multiple fitting formulas and takes into account the sensory evaluation data and physical and chemical analysis data of plant extracts, including a finished coffee aroma equation, a finished coffee acidity equation, a finished coffee sweetness equation, a finished coffee bitterness equation, a finished coffee taste equation, and a finished coffee aftertaste equation.

[0023] The finished coffee aroma equation is used to calculate the aroma intensity index of the finished coffee according to the aroma intensity, aroma type and volatile aroma component composition of the plant extract, which is specifically expressed as follows:

[0024]

[0025] In the formula, I a is the aroma intensity index of finished coffee; n is the number of plant extracts; w i is the weight coefficient of the i-th plant extract; A i is the aroma intensity of the i-th plant extract; t is the time variable; α, β are the time attenuation coefficients; m is the number of volatile aroma components; c j is the content of the jth volatile aroma component; V j is the aroma contribution coefficient of the jth volatile aroma component; γ is the adjustment coefficient; ε is the error term.

[0026] The finished coffee acidity equation is used to calculate the acidity index of the finished coffee according to the pH value of the plant extract, the total polyphenol content and the formula ratio, and is specifically expressed as follows:

[0027]

[0028] In the formula, I s is the acidity index of finished coffee; n is the number of plant extracts; r iis the formula ratio of the i-th plant extract; pH i is the pH value of the i-th plant extract; k is the adjustment coefficient; P i is the total polyphenol content of the i-th plant extract; λ is the concentration influence coefficient; C i is the concentration of the i-th plant extract in the final product; ε is the error term.

[0029] The finished coffee product sweetness equation is used to calculate the sweetness index of the finished coffee product according to the sugar content, sweetening substance content and formula ratio of the plant extract, which is specifically expressed as follows:

[0030]

[0031] In the formula, I d is the sweetness index of finished coffee; n is the number of plant extracts; r i is the formula ratio of the i-th plant extract; S i is the sugar content of the i-th plant extract; m is the number of sweet substances; D ij is the sweetness contribution coefficient of the jth sweet substance in the i-th plant extract; k is the concentration response coefficient; C ij is the concentration of the jth sweet substance in the ith plant extract; C0 is the perception threshold; ε is the error term.

[0032] The bitterness equation of the finished coffee product is used to calculate the bitterness index of the finished coffee product according to the bitter substance content, caffeine content and formula ratio of the plant extract, which is specifically expressed as follows:

[0033]

[0034] In the formula, I b is the bitterness index of finished coffee; n is the number of plant extracts; r i is the formula ratio of the i-th plant extract; m is the number of bitter substances; B ij is the bitterness contribution coefficient of the jth bitter substance in the i-th plant extract; C ij is the concentration of the jth bitter substance in the i-th plant extract; α, β are the caffeine influence coefficients; Caf i is the caffeine content in the i-th plant extract; ε is the error term.

[0035] The finished coffee taste equation is used to calculate the taste index of the finished coffee according to the solubility, viscosity and formula ratio of the plant extract, which is specifically expressed as follows:

[0036]

[0037] In the formula, I mis the taste index of finished coffee; n is the number of plant extracts; r i is the formula ratio of the i-th plant extract; Sol i is the solubility of the i-th plant extract; η i is the viscosity of the ith plant extract; a, b, c, d, e, f are fitting coefficients; ε is the error term.

[0038] The aftertaste equation of the finished coffee product is used to calculate the aftertaste index of the finished coffee product according to the volatile aroma component composition, persistence index and formula ratio of the plant extract, which is specifically expressed as follows:

[0039]

[0040] In the formula, I f is the aftertaste index of finished coffee; n is the number of plant extracts; r i is the formula ratio of the i-th plant extract; m is the number of volatile aroma components; V ij is the contribution coefficient of the jth volatile aroma component in the i-th plant extract; k j is the attenuation coefficient of the jth volatile aroma component; t is the time variable; P i is the persistence index of the ith plant extract; λ is the persistence attenuation coefficient; ε is the error term.

[0041] The sensory feature branch network is used to extract and learn the sensory features of plant extracts. The input is sensory evaluation data, the output is a sensory feature vector, and the structure is a multi-layer perceptron (MLP) including 3 hidden layers. Each layer uses a ReLU activation function, and the last layer uses a Sigmoid function to map the output to the [0, 1] interval.

[0042] The physicochemical characteristic branch network is used to extract and learn the physicochemical characteristics of plant extracts. The input is physicochemical analysis data, and the output is a physicochemical characteristic vector. The structure is a one-dimensional convolutional neural network (1D-CNN) including 2 convolutional layers and 1 fully connected layer, which is used to capture the local correlation between physicochemical parameters.

[0043] The formula ratio branch network is used to learn the impact of different plant extract ratios on the final product. The input is the formula ratio of the plant extract, and the output is the formula feature vector. The structure is a network with a self-attention mechanism, which is used to capture the interaction between different ingredients.

[0044] The temporal feature branch network is used to learn the characteristics of coffee quality changing over time. The input is sensory and physical and chemical data at different time points, and the output is a temporal feature vector. The structure is a long short-term memory network (LSTM) for modeling time series data.

[0045] The fusion network is used to integrate the outputs of each branch network and generate the final score prediction. The input is the output feature vector of each branch network, and the output is the predicted coffee quality score. The structure is a multi-layer perceptron with an attention mechanism, which is used to adaptively adjust the importance of different features and comprehensively obtain the final prediction result.

[0046] For the above equations and network parameter acquisition method:

[0047] 1) Sensory evaluation data (A i ,V j etc.) obtained through blind tasting and scoring by a professional tasting team;

[0048] 2) Physical and chemical analysis data (pH i ,P i ,C ij The pH value can be measured by laboratory instruments, such as pH meter, high performance liquid chromatograph (HPLC), gas chromatography-mass spectrometer (GC-MS), etc.

[0049] 3) Formula ratio (r i ) is obtained according to the actual production formula record;

[0050] 4) Time-related parameters (t, α, β, λ, etc.) are obtained by repeated testing of coffee samples at different time points;

[0051] 5) Various coefficients (w i ,k,D ij ,B ij The historical data is fitted by the least square method or other optimization algorithms (etc.);

[0052] 6) The weight parameters of the neural network are optimized on the training data through the back propagation algorithm.

[0053] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for optimizing the formula of plant extracts in deep processing of coffee.

[0054] A third aspect of the present invention provides a system for optimizing plant extract formula in coffee deep processing, comprising the above-mentioned computer-readable storage medium.

[0055] Compared with the prior art, the method, medium and system for optimizing the formula of plant extracts in coffee deep processing provided by the present invention have the following beneficial effects:

[0056] 1. A raw material database containing information on a variety of plant extracts has been established. By systematically collecting data on the sources, extraction methods, main active ingredients and their contents of various plant extracts, comprehensive knowledge support is provided for subsequent formula design and quality prediction.

[0057] 2. Extracted and learned the sensory and physical and chemical characteristics of plant extracts. Using a professional tasting team and laboratory instruments, we obtained the aroma intensity, aroma type, taste characteristics, taste description, total polyphenol content, antioxidant capacity, solubility, pH value, volatile aroma component composition and other indicators of plant extracts, laying the foundation for establishing an accurate quality prediction model.

[0058] 3. An intelligent quality prediction model based on mathematical modeling and machine learning was constructed. The model contains multiple submodules, which can effectively extract relevant information from multiple aspects such as sensory, physical and chemical, recipe ratio and time dynamics, and accurately predict the quality indicators of coffee such as aroma, acidity, sweetness, bitterness, taste and aftertaste.

[0059] 4. Intelligent optimization of the formula is achieved by using data augmentation and optimization algorithms. A large number of alternative formulas are generated by randomly fine-tuning the benchmark formula, adding noise, and interpolating and extrapolating. Then, the optimization algorithm is used to find the optimal formula that can meet the expected quality indicators among these alternatives, which greatly improves the efficiency and accuracy of formula optimization.

[0060] In general, the coffee deep-processing plant extract formula optimization method proposed in the present invention fully utilizes the advantages of mathematical modeling and machine learning, realizes a comprehensive characterization of the sensory and physicochemical properties of plant extracts, constructs an intelligent quality prediction model, and obtains the best formula solution through an optimization algorithm. It solves the technical problem that the existing methods often rely on engineers' experience and a large number of experiments to achieve formula optimization, which is time-consuming and labor-intensive. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 A flow chart of the method provided by the present invention;

[0062] Figure 2 Predicted distribution diagram of quality indexes of alternative formulas;

[0063] Figure 3 This is the sensory evaluation result of the optimal recipe coffee sample. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0065] like Figure 1As shown, it is a flow chart of a method for optimizing the formula of plant extracts in coffee deep processing provided by the present invention, and the method comprises the following steps:

[0066] S10. Collect a variety of plant extracts used for deep processing of coffee, and record the source, extraction method, main active ingredients and their content of each extract;

[0067] S20. Obtain sensory evaluation data and physicochemical analysis data for each plant extract;

[0068] S30, obtaining a set series of initial formulas, where the initial formulas use the same batch of coffee beans and the same basic additives, and combine different plant extracts in different proportions to make coffee samples; and obtaining blind test score data of the coffee samples;

[0069] S40, establishing a training data set, including the plant extract ratio in each initial formula, sensory evaluation data and physical and chemical analysis data of the corresponding plant extract, and blind test scoring data of coffee samples made from the initial formula; using the training data set to train a multi-branch embedded neural network model to obtain a plant extract coffee scoring model;

[0070] S50, obtaining a plant extract formula to be optimized, recorded as a basic formula, and using a plant extract coffee scoring model to obtain blind test scoring data of the basic formula using sensory evaluation data and physical and chemical analysis data of the basic formula;

[0071] S60, obtaining multiple groups of adjusted formulas, and using the plant extract coffee scoring model to obtain blind test scoring data of the corresponding adjusted formula using the sensory evaluation data and physical and chemical analysis data of each group of adjusted formulas;

[0072] S70, using an optimization algorithm to optimize the basic formula and each set of adjusted formulas according to the blind test score data, obtain the optimal formula and output it.

[0073] The specific implementation methods of the above steps are described in detail below:

[0074] The specific implementation of step S10 is: first, collect a variety of plant extracts for deep processing of coffee. These plant extracts include at least vanilla essence, cinnamon extract, cocoa extract, mint extract and nut extract. Among them, vanilla essence is extracted from vanilla pods by supercritical CO2 extraction, and the main active ingredient is vanillin, with a content of not less than 35%. For each plant extract, record their source, extraction method, main active ingredient and its content and other information. The purpose of this step is to establish a raw material database containing a variety of plant extracts to lay a foundation for subsequent formula optimization.

[0075] The specific implementation of step S20 is: obtaining sensory evaluation data and physicochemical analysis data of each plant extract. The sensory evaluation data includes indicators such as aroma intensity, aroma type, taste characteristics, and taste description. The physicochemical analysis data includes parameters such as total polyphenol content, antioxidant capacity, solubility, pH value, and volatile aroma component composition. These data are obtained through professional tasting teams and laboratory instrument measurements. This step aims to comprehensively characterize the sensory and physicochemical properties of each plant extract and provide a basis for subsequent formula design and evaluation.

[0076] The specific implementation of step S30 is: obtain a series of set initial formulas. These initial formulas use the same batch of coffee beans and the same basic additives, such as caffeine, sugar, milk powder and stabilizers. Different plant extracts are combined in different proportions to make coffee samples. These coffee samples are blindly scored, and the scoring indicators include aroma, acidity, sweetness, bitterness, taste and aftertaste. The purpose of this step is to collect the initial formula and its corresponding quality evaluation data to provide support for subsequent model training and optimization.

[0077] The specific implementation of step S40 is: establish a training data set. The data set includes the ratio of plant extracts in each initial formula, the sensory evaluation data and physicochemical analysis data of the corresponding plant extracts, and the blind test score data of the coffee samples made from the initial formula. Then, a multi-branch embedded neural network model is trained using this training data set to obtain a plant extract coffee scoring model. The main components of this model include: a mathematical fitting equation group module, a sensory feature branch network, a physicochemical feature branch network, a formula ratio branch network, a time series feature branch network, and a fusion network. The purpose of this step is to establish a machine learning model that can predict the quality of the final product based on the sensory and physicochemical properties of the plant extract.

[0078] The specific implementation of step S50 is: obtain the basic formula to be optimized, and use the plant extract coffee scoring model trained in step S40 to calculate the corresponding blind test scoring data using the sensory evaluation data and physical and chemical analysis data of the basic formula. This step is to obtain the quality evaluation index of the benchmark formula and provide a reference for subsequent formula optimization.

[0079] The specific implementation method of step S60 is: use the data augmentation method to adjust the ratio in the basic formula to obtain multiple groups of adjusted formulas. Specifically, it includes: 1) randomly fine-tuning the original formula, and making small adjustments to the proportion of each component while keeping the total amount unchanged; 2) adding Gaussian noise to the sensory evaluation and physical and chemical analysis data to simulate measurement errors; 3) generating a new virtual formula, which is obtained by interpolating or extrapolating the existing formula data. Then, the plant extract coffee scoring model trained in step S40 is used to calculate the corresponding blind test scoring data of each group of adjusted formulas using the sensory evaluation data and physical and chemical analysis data. The purpose of this step is to generate a series of alternative formulas and their quality indicators for optimization, and provide input for the optimization algorithm in the next step.

[0080] The specific implementation of step S70 is: using an optimization algorithm to optimize the basic formula and the collection of each group of adjusted formulas according to the blind test score data, obtain the optimal formula and output it. The optimization algorithm used here can be a genetic algorithm, a particle swarm optimization algorithm, a simulated annealing algorithm, etc. This step aims to find the optimal formula that can meet the expected quality indicators from a large number of alternative formulas, and complete the optimization process of the entire coffee deep processing plant extract formula.

[0081] In the above steps, some key algorithms and principles involved are as follows:

[0082] 1. The mathematical fitting equation group module contains multiple fitting formulas for different quality indicators, such as the finished coffee aroma equation, finished coffee acidity equation, finished coffee sweetness equation, finished coffee bitterness equation, finished coffee taste equation and finished coffee aftertaste equation. These equations use the sensory and physical and chemical properties of plant extracts, as well as factors such as formula ratio, to predict the various quality indicators of the final product through mathematical modeling.

[0083] 2. The sensory feature branch network is a multi-layer perceptron (MLP) structure used to extract and learn the sensory features of plant extracts. The input is sensory evaluation data, and the output is a sensory feature vector. The network contains 3 hidden layers, each layer uses the ReLU activation function, and the last layer uses the Sigmoid function to map the output to the [0,1] interval. This network can effectively learn the potential features contained in the sensory evaluation data and provide sensory feature support for subsequent quality prediction.

[0084] 3. The physical and chemical characteristics branch network is a one-dimensional convolutional neural network (1D-CNN) structure, which is used to extract and learn the physical and chemical characteristics of plant extracts. The input is the physical and chemical analysis data, and the output is the physical and chemical feature vector. The network contains 2 convolutional layers and 1 fully connected layer, which can capture the local correlation between physical and chemical parameters and provide physical and chemical feature support for quality prediction.

[0085] 4. The formula ratio branch network uses a self-attention mechanism to learn the impact of different plant extract ratios on the final product. The input is the formula ratio of the plant extract, and the output is the formula feature vector. This network can effectively capture the interaction between different ingredients and provide formula feature support for quality prediction.

[0086] 5. The temporal feature branch network adopts the long short-term memory network (LSTM) structure to learn the characteristics of coffee quality changing over time. The input is the sensory and physical and chemical data at different time points, and the output is the temporal feature vector. This network can model time series data and provide temporal dynamic feature support for quality prediction.

[0087] 6. The fusion network uses a multi-layer perceptron structure with an attention mechanism to integrate the outputs of each branch network and generate the final score prediction. The input is the output feature vector of each branch network, and the output is the predicted coffee quality score. This network can adaptively adjust the importance of different features and comprehensively obtain the final quality prediction result.

[0088] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the above-mentioned method for optimizing the formula of plant extracts in deep processing of coffee.

[0089] A third aspect of the present invention provides a system for optimizing plant extract formula in coffee deep processing, comprising the above-mentioned computer-readable storage medium.

[0090] Specifically, the principle of the present invention is:

[0091] 1. Comprehensively characterize the sensory and physicochemical properties of plant extracts. A detailed raw material database was established by systematically collecting basic information such as the source, extraction method, and content of the main active ingredients of various plant extracts. At the same time, sensory evaluation data such as aroma intensity, aroma type, taste characteristics, and taste description of plant extracts were obtained, as well as physicochemical analysis data such as total polyphenol content, antioxidant capacity, solubility, pH value, and volatile aroma component composition. These data provide the necessary basic support for subsequent quality prediction and formula optimization.

[0092] 2. Construct a quality prediction model based on mathematical modeling. A series of mathematical fitting equations were established for the quality indicators of finished coffee, such as aroma, acidity, sweetness, bitterness, taste and aftertaste. These equations can convert the sensory and physical and chemical properties of plant extracts, as well as factors such as formula ratio, into corresponding quality index prediction values. For example, the aroma equation of finished coffee combines the aroma intensity, volatile aroma component composition and time dynamic characteristics of plant extracts, and can accurately predict the aroma intensity of the final product. This mathematical modeling method provides a solid theoretical basis for quality prediction.

[0093] 3. Use a multi-branch neural network to achieve intelligent quality prediction. Based on the mathematical fitting equation group, the present invention constructs a multi-branch embedded neural network model. The model contains sub-modules such as sensory feature branch network, physical and chemical feature branch network, formula ratio branch network and time series feature branch network, which can fully explore the complex mapping relationship between various aspects of plant extracts and the quality of the final product. Through end-to-end joint optimization training, this neural network model can eventually achieve accurate prediction of coffee quality. Compared with a single mathematical model, this intelligent prediction method based on machine learning has stronger expression ability and generalization.

[0094] 4. Apply data augmentation and optimization algorithms to optimize the formula. In order to explore more feasible formula solutions, the present invention adopts data augmentation technology. A large number of alternative formulas are generated by means of random fine-tuning, adding noise, and interpolation and extrapolation of the benchmark formula. Then, optimization methods such as genetic algorithms, particle swarm optimization algorithms, or simulated annealing algorithms are used to find the optimal formula that can meet the expected quality indicators among these alternative solutions. This method that combines mathematical modeling, machine learning, and optimization algorithms can efficiently explore the best plant extract formula and provide valuable solutions for the coffee deep processing industry.

[0095] In summary, the coffee deep-processing plant extract formula optimization method proposed in the present invention fully utilizes the advantages of mathematical modeling and machine learning. First, by systematically collecting and analyzing the sensory and physicochemical property data of plant extracts, a detailed raw material knowledge base is established. Then, a mathematical fitting equation group and a multi-branch neural network model are used to construct an intelligent system that can comprehensively predict the quality of coffee. Finally, data augmentation and optimization algorithms are applied to intelligently explore the best formula solution that meets the expected quality indicators.

[0096] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: A coffee R&D team hopes to improve the quality of its flagship coffee product by optimizing the formula of plant extracts. The R&D personnel first established an internal raw material database covering information on a variety of plant extracts. The database contains 5 commonly used plant extracts, as shown in Table 1.

[0097] Table 1 Basic information of plant extracts

[0098] Extract Name source Extraction method Main active ingredients and content Vanilla extract Vanilla Bean Supercritical CO2 extraction Vanillin, content ≥35% Cinnamon Extract Cinnamon Bark Solvent Extraction Cinnamaldehyde, content ≥40% Cocoa Extract Cocoa beans Supercritical Fluid Extraction Cocoa phenol, content ≥ 25% Peppermint Extract Mint leaves Steam distillation Menthol, content ≥30% Nut Extracts Nuts Cold Pressed Fatty acids, content ≥ 50%

[0099] Next, the R&D staff invited a professional tasting team and laboratory to conduct a systematic sensory evaluation and physical and chemical analysis of these plant extracts. The specific results are shown in Tables 2 and 3.

[0100] Table 2 Sensory evaluation data of plant extracts

[0101] Extract Name Aroma intensity Aroma type Taste characteristics Taste Description Vanilla extract 4.7 Sweet, floral Rich, delicate Dense, smooth Cinnamon Extract 3.9 Mild, woody Rich, slightly bitter Thick, sticky Cocoa Extract 4.2 Rich, Chocolate Rich, slightly bitter Rich, smooth Peppermint Extract 3.5 Fresh, minty Cool, slightly sweet Refreshing, Cool Nut Extracts 3.8 Rich, Nutty Rich, slightly bitter Thick, sticky

[0102] Table 3 Physical and chemical analysis data of plant extracts

[0103]

[0104] Using the above data, the researchers designed a series of initial recipes and produced corresponding coffee samples for blind testing and scoring. The specific recipes and scoring results are shown in Table 4.

[0105] Table 4 Initial formula and its blind test score

[0106] vanilla Cinnamon cocoa Mint nut aroma acidity Sweetness Bitterness Taste Aftertaste 1 30 20 15 10 25 4.5 3.8 3.2 3.9 4.3 2 25 15 20 15 25 4.2 3.5 3.6 4.1 4.0 3 35 15 20 10 20 4.7 3.9 3.4 4.0 4.2 4 30 25 10 15 20 4.4 3.7 3.3 4.2 4.1 5 20 25 15 20 20 4.1 3.6 3.8 4.3 3.9

[0107] Based on the above data, researchers built a multi-branch embedded neural network model to predict coffee quality. The model contains four branch networks, which are used to extract and learn the sensory characteristics, physical and chemical characteristics, formula ratio characteristics, and time series characteristics of plant extracts. Then, these feature vectors are input into a fusion network, which adaptively adjusts the importance of different features through the attention mechanism, and finally outputs the corresponding coffee quality prediction value.

[0108] When training this neural network model, the researchers used the initial formula data in Table 4 as the training set. After 500 epochs of optimization, they finally achieved good prediction performance. Next, the researchers further used the data augmentation method to generate a series of candidate formulas and used the trained model to predict their quality. The specific results are as follows: Figure 2 and Figure 3 shown.

[0109] from Figure 2It can be seen that there are obvious differences in the quality indicators corresponding to different formulations. For example, some formulations can bring a stronger aroma, while others tend to produce higher acidity. This provides a good data basis for subsequent optimization.

[0110] The researchers further used genetic algorithms to optimize these candidate formulas and obtained an optimal formula that can meet the expected quality goals. The specific information of the optimal formula is shown in Table 5.

[0111] Table 5 Optimal formulation

[0112] Plant Extracts Ratio(%) Vanilla extract 35 Cinnamon Extract 20 Cocoa Extract 18 Peppermint Extract 12 Nut Extracts 15

[0113] According to the optimal formula shown in Table 5, the R&D staff produced the corresponding coffee samples and organized a professional tasting team to conduct sensory evaluation. Figure 3 As shown. Figure 3 It can be seen that the coffee samples with the optimal formula have received high scores in various quality indicators, fully meeting the expected goals of the researchers. Among them, the aroma and taste indicators are particularly outstanding, thanks to the synergistic effect of vanilla essence and cinnamon extract. The addition of mint extract brings a refreshing feeling to the whole. In general, this formula optimization method based on mathematical modeling and machine learning has helped researchers successfully develop a high-quality coffee product that meets consumer preferences.

[0114] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing the formula of plant extracts in coffee deep processing, characterized in that: The following steps are involved: S10. Collect a variety of plant extracts used for coffee deep processing, and record the source, extraction method, main active ingredients and their content of each extract; S20. Obtain sensory evaluation data and physicochemical analysis data for each plant extract; S30, obtaining a series of set initial formulas, wherein the initial formulas use the same batch of coffee beans and the same basic additives, and combine different plant extracts in different proportions to make coffee samples; and obtaining blind test score data of the coffee samples; S40, establishing a training data set, including the plant extract ratio in each initial formula, sensory evaluation data and physicochemical analysis data of the corresponding plant extract, and blind test scoring data of coffee samples made from the initial formula; using the training data set to train a multi-branch embedded neural network model to obtain a plant extract coffee scoring model; S50, obtaining a plant extract formula to be optimized, recorded as a basic formula, and using the plant extract coffee scoring model to obtain blind test scoring data of the basic formula using sensory evaluation data and physicochemical analysis data of the basic formula; S60, obtaining multiple groups of adjusted formulas, and using the plant extract coffee scoring model to obtain blind test scoring data of the corresponding adjusted formulas using the sensory evaluation data and physical and chemical analysis data of each group of adjusted formulas; S70, using an optimization algorithm to optimize the basic formula and each set of adjusted formulas according to the blind test score data, to obtain the optimal formula and output it.

2. The method for optimizing the formula of plant extracts in coffee deep processing according to claim 1, characterized in that: The plant extracts include at least vanilla essence, cinnamon extract, cocoa extract, mint extract and nut extract; the vanilla essence is specifically extracted from vanilla pods by supercritical CO2 extraction, and the main active ingredient is vanillin, with a content of not less than 35%.

3. The method for optimizing the formula of plant extracts in coffee deep processing according to claim 2, characterized in that: The sensory evaluation data include aroma intensity, aroma type, taste characteristics, and taste description; the physical and chemical analysis data include total polyphenol content, antioxidant capacity, solubility, pH value, and volatile aroma component composition.

4. The method for optimizing the formula of plant extracts in coffee deep processing according to claim 3, characterized in that: The basic added ingredients are specifically caffeine, sugar, milk powder and stabilizer.

5. The method for optimizing the formula of plant extracts in coffee deep processing according to claim 4, characterized in that: The blind test scoring data specifically refers to the blind test scores given by the tasting team, including scores for aroma, acidity, sweetness, bitterness, taste, and aftertaste.

6. The method for optimizing the formula of plant extracts in coffee deep processing according to claim 5, characterized in that: The adjusted formula is a plurality of groups of adjusted formulas obtained by adjusting the proportions in the basic formula using a data augmentation method.

7. The method for optimizing the formula of plant extracts in coffee deep processing according to claim 6, characterized in that: The specific steps of adjusting the ratio in the basic formula by using the data augmentation method include: 1) randomly fine-tuning the original formula and making small adjustments to the proportion of each ingredient while keeping the total amount unchanged; 2) adding Gaussian noise to the sensory evaluation and physical and chemical analysis data to simulate measurement errors; 3) generating a new virtual formula by interpolating or extrapolating the existing formula data.

8. The method for optimizing the formula of plant extracts in coffee deep processing according to claim 7, characterized in that: The multi-branch embedded neural network model includes a mathematical fitting equation group module, a sensory feature branch network, a physical and chemical feature branch network, a formula ratio branch network, a time series feature branch network and a fusion network.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are run in a computer, they are used to execute the method for optimizing the formula of plant extracts in deep processing of coffee according to any one of claims 1 to 8.

10. A plant extract formula optimization system for coffee deep processing, characterized in that: A computer-readable storage medium comprising the computer-readable storage medium of claim 9.

Images