A pu'er tea intelligent blending method based on original chemical component features

Abstract

The application discloses a kind of based on original physicochemical component feature Pu'er tea intelligent matching method, comprising the following steps: determine to collect matching parameter information, control parameter;Establish the basic information of Pu'er tea raw materials and corresponding raw material aroma, taste physicochemical component feature library;According to physicochemical component feature, data preprocessing is carried out, and high-quality data set is provided for modeling;Collect the quantitative target value of user to the physicochemical component feature of matching Pu'er tea, and aroma, taste score is calculated based on original physicochemical component feature;Through original physicochemical component feature and matching target value, the Pu'er tea intelligent matching model based on random forest is constructed;According to matching target value and fitness function, establish matching fitness function selection knowledge base;Through genetic algorithm, matching model cross validation optimization is carried out by selecting fitness function, and the best matching scheme is formed;Best matching formula parameter output confirmation and matching control.The scientific, efficient tea matching is realized, and the quality and consistency of final product are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Pu'er tea, and particularly relates to a method for intelligent blending of Pu'er tea based on characteristics of raw physicochemical components. BACKGROUND

[0002] Pu'er tea, as a unique local tea, enjoys high reputation in the domestic and foreign markets. The production process of Pu'er tea involves multiple links such as spreading, fixation, fermentation, drying and storage of raw materials, among which, the blending of raw materials is an important factor determining the quality of the final product. The traditional blending method of Pu'er tea mainly relies on experienced tea experts, who make blending plans through their sensory evaluation and experience accumulation. This method has certain limitations, mainly in the following aspects:

[0003] 1. Strong subjectivity: Although the experience and sensory evaluation of tea experts can ensure the quality of blending to a certain extent, due to individual differences and subjective judgments, the blending results of different experts may have large differences, making it difficult to ensure the consistency of each blending.

[0004] 2. Low efficiency: The traditional blending method mainly relies on manual operation, which needs to go through multiple tests and adjustments to determine the final blending scheme, which not only consumes time and effort, but also is difficult to meet the needs of large-scale production.

[0005] 3. Poor stability: The quality of tea raw materials is affected by many factors such as climate, soil, picking time, etc. Even the same batch of tea may have certain fluctuations in quality. The traditional method is difficult to effectively respond to these changes, resulting in unstable quality of the final product.

[0006] 4. Difficult to quantify: The traditional blending method mainly relies on the experience and feeling of experts, which is difficult to quantify and standardize, and cannot form a systematic blending standard and specification, which is not conducive to the control and optimization of the production process.

[0007] With the development of big data technology and artificial intelligence technology, intelligent blending methods based on data driving have gradually become a research hotspot. Through comprehensive data collection on the physical and chemical properties of Pu'er tea raw materials, advanced machine learning algorithms can be used to scientifically analyze and optimize the blending process of Pu'er tea, thereby improving the efficiency and quality of blending. SUMMARY

[0008] The present application aims to provide a method for intelligent blending of Pu'er tea based on characteristics of raw physicochemical components, which realizes scientific and efficient tea blending and improves the quality and consistency of the final product by collecting, processing and analyzing the information of Pu'er tea raw materials.

[0009] The technical solution of the present application is as follows:

[0010] A Pu'er tea intelligent blending method based on original raw material physicochemical component characteristics, comprising the following steps:

[0011] Determine the collection of blending parameter information and control parameters;

[0012] Establish a Pu'er tea raw material basic information and corresponding raw material aroma, taste physicochemical component characteristic library;

[0013] According to the data preprocessing of the raw material aroma, taste physicochemical component characteristic library data, provide high-quality data set for modeling;

[0014] Collect the user's quantitative target value of the physicochemical component characteristics of the blended Pu'er tea, and comprehensively calculate the aroma and taste scores based on the original raw material physicochemical component characteristics;

[0015] Construct a Pu'er tea intelligent blending model based on random forest through the original raw material physicochemical component characteristics and the blending target value;

[0016] According to the blending target value and the fitness function, establish a blending fitness function selection knowledge base;

[0017] Through genetic algorithm, select the fitness function to optimize the cross-validation of the blending model, and form the best blending scheme;

[0018] Best blending formula parameter output confirmation and blending control.

[0019] Further, the Pu'er tea intelligent blending model based on original raw material physicochemical component characteristics comprises the following steps:

[0020] Establish a target function: a blending scheme with the minimum product grade, maximum product taste characteristics, and maximum product aroma characteristics as the target;

[0021] Establish rules and constraints: the original raw material physicochemical component characteristics and the user's blending target value are input conditions;

[0022] Model establishment: use Python to establish an intelligent blending prediction model based on random forest algorithm;

[0023] Adaptive function: use the trained random forest model to calculate the aroma and taste scores of each blending scheme, and compare them with the user's target value. Through genetic algorithm, further optimize the Pu'er tea blending, and construct a fitness function to calculate and evaluate the closeness of each blending scheme calculated by the random forest model to the user's expected aroma, taste, and grade target;

[0024] Blending model optimization: the fitness function calculation, the blending fitness function coefficient characteristic library is used as the cross-validation input of the optimization model, and the model parameter optimization is carried out through genetic algorithm.

[0025] Further, the fitness function is:

[0026]

[0027] Wherein, S P , F P and G P are aroma, taste score and grade score predicted by the model of the blending scheme, T x , T y and T z are target scores provided by the user, w x , w y and w z are weight coefficients of aroma, taste and grade respectively, and satisfy w x +w y +w z =1, |S P -T x | is the absolute difference between the aroma score of the blending scheme and the target aroma score, |F P -T y | is the absolute difference between the taste score of the blending scheme and the target taste score, |G P -T z | is the absolute difference between the tea grade score of the blending scheme and the target grade score, and the value range of the fitness value F fitness is 0 to 1, and the value closer to 1 indicates that the blending scheme is closer to the target.

[0028] Further, the taste characteristics are scored by a scoring model, and the scoring model adopts nonlinear weighting, dynamic adjustment and synergistic effect; the scoring model is:

[0029] T comprehensive =w b f(T b )+w k (100-g(T k ))+w s f(T s )+w h f(T h )+λh(T s , T k ),

[0030] Wherein, f(x) is a nonlinear transformation of thickness, sweetness and aftertaste, g(T k ) is a nonlinear processing of bitterness, λh(T s , T k ) is a synergistic effect, h(T s , T k ) represents the interaction between sweetness and bitterness, and λ is a coefficient for adjusting the strength of synergistic effect, wb is a thickness weight, w k is a bitterness weight, w s is a bitterness weight, w h is a sweetness weight.

[0031] Further, the fitness function feature library is used to predict each blending scheme according to an intelligent blending model, combine user target values with expert knowledge, establish a blending fitness function feature library, and is used to optimize the blending model and seek the best formula solution.

[0032] Further, the blending fitness function feature includes a blending tea taste score, a blending tea aroma score, and a blending tea grade score.

[0033] Further, the main chemical components of Pu'er tea aroma are quantitatively scored according to a 100-point system, a weighted average method is adopted, each component is given a different weight w according to its contribution to the overall aroma, and the concentration is standardized to 0-100 points, and finally, the total aroma score is obtained by combining the scores of each component.

[0034] For each component, the standardized concentration score calculation formula is:

[0035]

[0036] where X is the actual concentration (unit: mg / L), Min represents the lower limit of the ideal concentration, Max represents the upper limit of the ideal concentration, and i represents the chemical component; if the actual concentration is lower than Min, the score is 0; if it is higher than Max, the score is 100.

[0037] The comprehensive aroma score calculation formula is:

[0038]

[0039] Further, the confirmation of the collected blending parameter information includes blending electronic scale information, blending material information, blending equipment information, and blending material weight information; and the control parameters involve blending formula parameters, including blending material weight, blending scale flow, blending equipment control, and blending material ratio.

[0040] Further, the data preprocessing is used for original recipe chemical component feature library feature parameter standardization processing and the preprocessing and feature engineering of the collected data, so as to extract the numerical representation of the tea original recipe chemical component features.

[0041] Further, the Pu'er tea raw material chemical composition feature library comprises raw material origin, year, grade raw material basic information and raw material chemical composition feature detection data, the raw material chemical composition detection data comprises taste features and aroma features, the taste features comprise caffeine, tea polyphenols, catechins, water-soluble sugars, amino acids and water extract substance components, and the aroma features comprise volatile aromatic substances, polyphenol compounds, amino acids, aldehyde sugars, sulfur-containing compounds, fatty acid oxidation products, carotenoid derivatives and microbial metabolite substance components.

[0042] The Pu'er tea raw material taste features comprise four dimensions of "bitterness", "sweetness", "aftertaste" and "thickness", each dimension is quantified according to a 100-point system to obtain a user target value; the Pu'er tea aroma features are comprehensive evaluation results according to the concentration of the raw material substance water extract chemical components, and are quantified according to a 100-point system to obtain a user target value.

[0043] Compared with the prior art, the present application has the following advantages:

[0044] The present application provides a Pu'er tea intelligent blending method based on raw material chemical composition features, an intelligent blending model is established through raw material feature index quantification and blended tea index quantification, the fitness of the blending model is optimized according to the user target value and the blended Pu'er tea features, and the minimum finished product grade, the maximum finished product taste features and the maximum finished product aroma features of the blended Pu'er tea are taken as targets to intelligently calculate the best raw material composition from the Pu'er tea raw material basic information, so that the best blended Pu'er tea formula is finally obtained.

[0045] The present application can not only accurately blend the blended Pu'er tea meeting the customer individual product customization target value, but also uses big data technology to drive the Pu'er tea blending based on objective data, reduces the influence of human subjective factors, and improves the blending efficiency through automatic data processing and optimization algorithm, and ensures the consistency of the quality and characteristics of each batch of tea through the scientific blending scheme. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 It is a flowchart of the method of the present application.

[0047] Figure 2 It is an intelligent blending implementation flowchart of the Pu'er tea of the present application.

[0048] Figure 3 It is an intelligent blending algorithm implementation flowchart of the Pu'er tea of the present application. DETAILED DESCRIPTION

[0049] It is to be understood that the terms "first" and "second" and similar

[0050] The features and advantages of the present application will be further described in the following detailed description of the embodiments.

[0051] Please refer to Figures 1-2 A Pu'er tea intelligent blending method based on raw material physicochemical component characteristics, as shown in Figure 1 、 Figure 2 and Figure 3 , comprises the following steps:

[0052] Determine the collection of blending parameter information and control parameters;

[0053] Establish a Pu'er tea raw material basic information and corresponding raw material aroma, taste physicochemical component characteristic library;

[0054] According to the data preprocessing of the raw material aroma, taste physicochemical component characteristic library data, provide high-quality data set for modeling;

[0055] Collect the user's quantitative target value of the physicochemical component characteristics of the blended Pu'er tea, and calculate the aroma and taste scores based on the raw material physicochemical component characteristics;

[0056] Construct a Pu'er tea intelligent blending model based on random forest through the physicochemical component characteristics of raw materials and blending target values;

[0057] According to the blending target value and the fitness function, establish a blending fitness function selection knowledge base;

[0058] Through genetic algorithm, select the fitness function to carry out cross-validation optimization of the blending model, and form the best blending scheme;

[0059] Best blending formula parameter output confirmation and blending control.

[0060] The Pu'er tea intelligent blending model based on raw material physicochemical component characteristics comprises the following steps:

[0061] Objective function is established: the blending scheme with the minimum product grade, the maximum product taste characteristics, and the maximum product aroma characteristics is the target;

[0062] Establish rules and constraints: the original recipe chemical component characteristics and the user blending target value are input conditions;

[0063] Model establishment: an intelligent blending prediction model based on random forest algorithm is established by using Python;

[0064] Adaptive function: the aroma and taste scores of each blending scheme are calculated by using the trained random forest model, and compared with the user target value. The genetic algorithm is further used to optimize the Pu'er tea blending, and the fitness function is used to calculate the closeness of each blending scheme to the user's expected aroma, taste, and grade targets;

[0065] Blending model optimization: the fitness function calculation, the blending fitness function coefficient feature library is used as the cross-validation input of the optimization model, and the genetic algorithm is used to optimize the model parameters.

[0066] The fitness function is:

[0067]

[0068] Among them, S P , F P , and G P are the aroma, taste, and grade scores of the blending scheme predicted by the model, T x , T y , and T z are the target scores provided by the user, w x , w y , and w z are the weight coefficients of aroma, taste, and grade respectively, and satisfy w x +w y +w z =1, |S P -T x | is the absolute difference between the aroma score of the blending scheme and the target aroma score, |F P -T y | is the absolute difference between the taste score of the blending scheme and the target taste score, |G P -T z | is the absolute difference between the tea grade score of the blending scheme and the target grade score, and the value range of the fitness value F fitness is 0 to 1. The value closer to 1 indicates that the blending scheme is closer to the target.

[0069] The taste characteristics are scored by the scoring model, which adopts nonlinear weighting, dynamic adjustment, and synergistic effect. The scoring model is:

[0070] T comprehensive =w b f(T b )+w k (100-g(T k ))+w s f(T s )+w h f(T h )+λh(T s ,T k ),

[0071] where f(x) is a nonlinear transformation of thickness, sweetness, and astringency, g(T k ) is a nonlinear processing of bitterness, λh(T s , T k ) is a synergistic effect, h(T s , T k ) represents the interaction between sweetness and bitterness, λ is a coefficient for adjusting the strength of the synergistic effect, w b is the thickness weight, w k is the bitterness weight, w s is the bitterness weight, and w h is the astringency weight.

[0072] The fitness function feature library is used to predict each blending scheme according to the intelligent blending model, combine the user target value, and fuse expert knowledge to establish a blending fitness function feature library for optimizing the blending model and seeking the best formula solution.

[0073] The blending fitness function features include blending tea taste score, blending tea aroma score, and blending tea grade score.

[0074] The main chemical components of Pu'er tea aroma are quantitatively scored according to a 100-point system, and a weighted average method is used, each component is given a different weight w according to its contribution to the overall aroma, and its concentration is standardized to 0-100 points, finally, the total aroma score is obtained by combining the scores of each component;

[0075] For each component, the normalized concentration score calculation formula is:

[0076]

[0077] where X is the actual concentration (unit: mg / L), Min represents the lower limit of the ideal concentration, Max represents the upper limit of the ideal concentration, and i represents the chemical component; if the actual concentration is lower than Min, the score is 0; if it is higher than Max, the score is 100;

[0078] The comprehensive aroma score calculation formula is:

[0079]

[0080] The confirmation of the collection of the blending parameter information includes: blending electronic scale information, blending material information, blending equipment information, and blending material weight information; the control parameter involves blending formula parameters, including: blending material weight, blending scale flow, blending equipment control, and blending material ratio.

[0081] The data preprocessing is used for the standardization processing of the characteristic parameter of the raw material physicochemical component feature library and the preprocessing and feature engineering of the collected data, so as to extract the numerical representation of the tea raw material physicochemical component features.

[0082] The raw material physicochemical component feature library of Pu'er tea contains raw material origin, year, grade raw material basic information, and detection data of raw material physicochemical component features. The detection data of the raw material physicochemical component includes: taste characteristics and aroma characteristics. The taste characteristics include: caffeine, tea polyphenols, catechins, water-soluble sugars, amino acids, and water extract substances. The aroma characteristics include: volatile aromatic substances, polyphenol compounds, amino acids, aldehyde sugars, sulfur-containing compounds, fatty acid oxidation products, carotenoid derivatives, and microbial metabolite substances.

[0083] The raw material taste characteristics of Pu'er tea include: "bitterness", "sweetness", "aftertaste", and "thickness". Each dimension is quantified according to the user's target value of 100 points. The aroma characteristics of Pu'er tea are the comprehensive score results according to the concentration of the raw material substance water extract chemical components, which are quantified according to the user's target value of 100 points.

[0084] In another specific embodiment, a Pu'er tea intelligent blending method based on raw material physicochemical component features is shown in Figure 1 .

[0085] Obtaining Pu'er tea raw material basic information and physicochemical component feature data;

[0086] The raw material physicochemical component feature library contains basic elements such as raw material origin, year, grade, and raw material physicochemical component collection or detection data. In this example, the raw material is selected from the same origin and different tea mountains and different grades of large-leaf species sun-cured green tea in Yunnan Province to establish the raw material physicochemical component feature library. Specifically, in this example, the raw material physicochemical component feature data is shown in Table 1:

[0087] Table 1

[0088]

[0089]

[0090] Specifically, the raw material quality taste feature library data is shown in Table 2:

[0091] Table 2

[0092]

[0093]

[0094]

[0095] The aroma component characteristic score of Pu'er tea raw materials is calculated.

[0096] The aroma characteristics are quantified according to a 100-point system after data preprocessing. According to the comprehensive aroma score, the actual concentration data table 3 of the main chemical components of the raw material aroma component characteristics and the comprehensive aroma score table 4 are calculated:

[0097] The main chemical components affecting the aroma of Pu'er tea and their weights w (sum = 1):

[0098] Volatile aromatic substances (VOCs): 0.3

[0099] Polyphenols: 0.2

[0100] Amino acids: 0.2

[0101] Sugars: 0.1

[0102] Sulfur compounds: 0.1

[0103] Others: 0.1

[0104] The ideal concentration range of each chemical component is as follows:

[0105] Volatile aromatic substances (VOCs): [10, 50] mg / L

[0106] Polyphenols: [100, 300] mg / L

[0107] Amino acids: [20, 100] mg / L

[0108] Sugars: [5, 15] mg / L

[0109] Sulfur compounds: [0.1, 1] mg / L

[0110] Others: [0.5, 2] mg / L

[0111] Table 3

[0112]

[0113]

[0114] Table 4

[0115]

[0116] Calculate the characteristic score of the taste components of Pu'er tea raw materials;

[0117] The taste characteristics of Pu'er tea include four dimensions: "bitterness", "sweetness", "aftertaste", and "thickness". After data preprocessing, each dimension is quantified according to the user's target value on a 100-point scale.

[0118] Taste characteristic score example:

[0119] 1) Select a nonlinear transformation as a square function (i.e. ), the synergistic effect is the square of the difference between sweetness and bitterness, then adjust it with adaptive weight, the calculation formula is as follows:

[0120]

[0121] 2) Score calculation

[0122] Assume the weights are:

[0123] Thickness weight wb=0.3;

[0124] Bitterness weight wk=0.2;

[0125] Sweetness weight ws=0.25;

[0126] Aftertaste weight wh=0.25;

[0127] Synergistic effect intensity λ=0.1;

[0128] The score of a certain blending tea is as follows:

[0129] Thickness Tb=95;

[0130] Bitterness Tk=50;

[0131] Sweetness Ts=85;

[0132] Aftertaste Th=90;

[0133] Substitute these values into the formula:

[0134]

[0135]

[0136] The calculation result output: T comprehensive =85.6125.

[0137] Collect quantitative data on the quality component characteristics of blended Pu-erh tea from users; Table 5 shows the quantitative target values ​​for the blended Pu-erh tea formula characteristics selected by users (grades are based on a scale of 1 to 5, with 5 being the highest):

[0138] Table 5

[0139]

[0140] Based on the user's target value and the physicochemical composition characteristics of the raw materials, the raw materials are divided into 1000 samples as input for training the random forest model, and the objective function is the user's target value.

[0141] In this example, the implementation process of the intelligent blending model based on the physicochemical characteristics of raw materials is described.

[0142] The entire process is divided into two core parts: the matching prediction model training process and the matching optimization and improvement process. The matching prediction model is trained using a random forest, and optimization is performed using a genetic algorithm. The detailed algorithm implementation process is as follows:

[0143] Data Acquisition and Preprocessing:

[0144] Input: Collect relevant historical blending data and finished product aroma, taste, and grade ratings based on the physicochemical properties of Pu-erh tea raw materials (such as tea polyphenol content, caffeine, aroma components, etc.).

[0145] Data preprocessing: Missing values ​​are imputed, and numerical values ​​are normalized to ensure consistency of the input data. The dataset is divided into training and test sets.

[0146] Model training:

[0147] Random Forest Algorithm: The processed, combined dataset is input into the random forest model for training. Random forest makes predictions using a set of multiple decision trees, which can handle high-dimensional data well and avoid overfitting.

[0148] Input features: chemical composition data of blending raw materials, historical formulas.

[0149] Output labels: corresponding aroma score, taste score, and tea grade.

[0150] Model evaluation:

[0151] Cross-validation: Cross-validation is used to evaluate the performance of a model, avoid overfitting, and ensure generalization ability.

[0152] Performance metrics: The mean squared error (MSE) metric was used to evaluate the model’s ability to predict aroma, flavor and grade.

[0153] Hyperparameter tuning: Adjust the hyperparameters of the random forest model, such as the number of decision trees and the maximum depth, through random search.

[0154] Save the model:

[0155] Save the trained and validated random forest model as a baseline model for predicting Pu'er tea blending.

[0156] Blending optimization improvement process:

[0157] This section uses genetic algorithms to optimize the blending scheme, aiming to find the best blending scheme to make aroma, taste, and grade scores meet the user's target values.

[0158] Initialize the population:

[0159] Population initialization: Randomly generate several blending schemes, each composed of different raw material proportions, forming the initial population.

[0160] Chromosome representation: Each blending scheme can be represented by a vector, with each element in the vector representing the proportion of a certain raw material.

[0161] Fitness function calculation:

[0162] Predictive scoring: Input each blending scheme in the population into the random forest prediction model to obtain the corresponding aroma, taste, and grade scores.

[0163] Fitness function: Calculate the fitness value of each blending scheme based on the aforementioned fitness function model (the difference between the comprehensive aroma, taste, grade and user target value). The higher the fitness, the closer the scheme is to the user's target.

[0164] Selection operation:

[0165] Selection operator: Use the tournament selection algorithm to select blending schemes with higher fitness as parents based on fitness values, and pass them to the next generation.

[0166] Crossover operation:

[0167] Crossover operator: Perform crossover operation on the selected parent schemes to generate new child schemes. Through multi-element crossover, exchange part of the parent chromosome genes (raw material proportions) to produce new blending schemes.

[0168] Mutation operation:

[0169] Mutation operator: Perform mutation operation on the newly generated child schemes, randomly adjust the proportions of some raw materials to increase the diversity of the population and avoid local optimal solutions.

[0170] Fitness evaluation:

[0171] Evaluate new population: input the new population after crossover and mutation into the prediction model again, recalculate its fitness value.

[0172] Keep the best individual: keep the individual with the highest fitness value (blending scheme) to the next generation, ensure the evolution direction of the population towards the optimal solution.

[0173] Termination condition:

[0174] Convergence judgment: if the fitness value of the best individual in the population does not improve significantly for several generations, or reaches the set generation limit, the algorithm is considered to have converged, and the optimization is stopped.

[0175] Output the optimal scheme: get the blending scheme with the highest fitness value as the output of the optimal blending scheme.

[0176] As shown in Table 6, as the initial prediction scheme of the random forest.

[0177] Table 6

[0178]

[0179] The fitness function example calculation, the target is:

[0180] Aroma target T s = 90 (using percentage, 100 is the highest);

[0181] Taste target T f = 85 (using percentage, 100 is the highest);

[0182] Tea grade target T g = 5 (grade using 1 to 5 points, 5 is the highest);

[0183] One of the results of the initial prediction of the blending scheme based on the random forest is:

[0184] Aroma score S p = 88;

[0185] Taste score F p = 82;

[0186] Grade score G p = 4;

[0187] Set the weight w s = 0.4, w f = 0.4, w g = 0.2, that is, the weights of aroma and taste are higher, and the weight of grade is relatively low.

[0188] Substitute these values into the fitness function:

[0189]

[0190] The fitness value of the initial matching scheme is 0.937, indicating that the scheme is very close to the user's target and is a better matching scheme.

[0191] In this example, through genetic algorithm optimization, after 20 generations of evolution, the optimal matching scheme is obtained, with an aroma score of 89.5, a taste score of 84.8, a grade score of 5, and a fitness value of 0.98, close to the user's expected target. The system gives the following matching scheme:

[0192] - Raw material 1 accounts for 40%

[0193] - Raw material 2 accounts for 40%

[0194] - Raw material 4 accounts for 20%

[0195] The system also suggests that the proportion of raw material 4 can be appropriately increased to further approach the user's target score.

[0196] The best matching scheme is obtained through the intelligent matching scheme of Pu'er tea, as shown in Table 7:

[0197] Table 7

[0198]

[0199] The matching production is controlled by the control system through the control of the matching formula control parameters.

[0200] Summary: According to the results in Table 8, through expert evaluation, the corresponding characteristic component content of intelligent matching Pu'er tea is compared and analyzed (wherein the aroma characteristics are considered to be volatile and oxidizable, with a final score of 85), the product matched by the Pu'er tea intelligent matching method based on the physicochemical components of raw materials meets the user's target value, and the method of the present application is effective and feasible.

[0201] Table 8

[0202]

[0203] The above-described embodiments only express the specific implementation of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the protection scope of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the technical concept of the present application, some modifications and improvements can be made, which are within the scope of protection of the present application.

Claims

1. A method for intelligent blending of Pu-erh tea based on the physicochemical characteristics of raw materials, characterized in that, Includes the following steps: Confirm the collected assembly parameter information and control parameters; Establish a database of basic information on Pu'er tea raw materials and corresponding characteristics of their aroma, taste, and physicochemical components. Data preprocessing was performed based on the feature database of raw material aroma and taste physicochemical components to provide a high-quality dataset for modeling. Collect users' quantitative target values ​​for the physicochemical characteristics of blended Pu'er tea, and calculate aroma and taste scores based on the physicochemical characteristics of the raw materials. A smart blending model for Pu'er tea based on random forests is constructed by combining the physicochemical characteristics of raw materials with blending target values; specifically, the following steps are included: Establish the objective function: a blending scheme with the objectives of minimizing the finished product grade, maximizing the finished product flavor characteristics, and maximizing the finished product aroma characteristics; Establish rules and constraints: The physicochemical composition characteristics of raw materials and the user's blending target value are the input conditions; Model building: A smart matching prediction model based on the random forest algorithm was built using Python; Fitness function: The aroma and taste scores of each blending scheme are calculated using a trained random forest model and compared with the user's target values. Further optimization of Pu'er tea blending is carried out through a genetic algorithm. A fitness function is constructed for evaluation: the random forest model calculates how close each blending scheme is to the user's expected aroma, taste and grade target. Model optimization: Fitness function calculation, fitness function coefficient feature library is used as cross-validation input for the optimized model, and model parameters are optimized through genetic algorithm; The fitness function is: ， in, , and These are the aroma, flavor, and grade scores predicted by the model for the blending scheme. , and It is a target rating provided by the user. , and These are the weighting coefficients for aroma, flavor, and grade, and they satisfy... , It is the absolute difference between the aroma score of the blending scheme and the target aroma score. It is the absolute difference between the flavor score of the blended recipe and the target flavor score. It is the absolute difference between the blended tea grade score and the target grade score, and the fitness value. The value ranges from 0 to 1. The closer the value is to 1, the closer the matching scheme is to the target. Based on the matching target value and fitness function, establish a feature library of matching fitness function coefficients; By using a genetic algorithm, a fitness function is selected for cross-validation optimization of the matching model to form the optimal matching scheme; Confirmation of optimal blending formula parameters and blending control.

2. The intelligent blending method for Pu'er tea based on the physicochemical characteristics of raw materials according to claim 1, characterized in that, Flavor characteristics are scored using a rating model that employs non-linear weighting, dynamic adjustment, and synergistic effects. The rating model is as follows: ， in, To perform non-linear transformations on thickness, sweetness, and aftertaste, For non-linear processing of bitterness, To introduce synergy, This indicates the interaction between sweetness and bitterness. It is a coefficient used to adjust the strength of the synergistic effect. For thickness weighting, Weighted by bitterness, Weighted by bitterness, The weight of the aftertaste.

3. The intelligent blending method for Pu'er tea based on the physicochemical characteristics of raw materials according to claim 1, characterized in that, The fitness function coefficient feature library is established by predicting each blending scheme based on the intelligent blending model, combining user target values ​​with expert knowledge, and is used to optimize the blending model and seek the best formulation solution.

4. The intelligent blending method for Pu'er tea based on the physicochemical characteristics of raw materials according to claim 3, characterized in that, The blending fitness function coefficient feature library includes blended tea taste score, blended tea aroma score, and blended tea grade score.

5. The intelligent blending method for Pu'er tea based on the physicochemical characteristics of raw materials according to claim 4, characterized in that, The main chemical components of Pu'er tea aroma were quantitatively scored out of 100 points, and a weighted average method was used, assigning different weights to each component based on its contribution to the overall aroma. The concentration of each component was standardized to a score of 0-100. Finally, the overall aroma score was obtained by combining the scores of each component. For each component, the standardized concentration score is calculated using the following formula: ， in This refers to the actual concentration (in mg / L). This indicates the lower limit of the ideal concentration. This indicates the upper limit of the ideal concentration. Represents chemical composition; if the actual concentration is lower than If the score is higher than 0, the score is 0; The score is 100; The formula for calculating the overall aroma score is as follows: 。 6. The intelligent blending method for Pu'er tea based on the physicochemical characteristics of raw materials according to claim 1, characterized in that, The confirmed and collected blending parameter information includes: blending electronic scale information, blending material information, blending equipment information, and blending material weight information; the control parameters involve blending formula parameters, including: blending material weight, blending scale flow rate, blending equipment control, and blending material ratio.

7. The intelligent blending method for Pu'er tea based on the physicochemical characteristics of raw materials according to claim 1, characterized in that, Data preprocessing is used for the standardization of feature parameters in the raw material physicochemical component feature library and the preprocessing and feature engineering of the collected data, so as to extract the numerical representation of the physicochemical component features of tea raw materials.

8. The intelligent blending method for Pu'er tea based on the physicochemical characteristics of raw materials according to claim 1, characterized in that, The Pu-erh tea raw material physicochemical component characteristic database contains basic information on the origin, year, and grade of raw materials, as well as detection data on the physicochemical component characteristics of the raw materials. The detection data on the physicochemical components of the raw materials include: taste characteristics and aroma characteristics. Taste characteristics include: caffeine, tea polyphenols, catechins, water-soluble sugars, amino acids, and water-soluble extracts. Aroma characteristics include: volatile aromatic substances, polyphenolic compounds, amino acids, aldoses, sulfur-containing compounds, fatty acid oxidation products, carotenoid derivatives, and microbial metabolites. The flavor characteristics of Pu'er tea raw materials include four dimensions: "bitterness", "sweetness", "aftertaste" and "body". Each dimension is quantified into a user target value on a 100-point scale. The aroma characteristics of Pu'er tea are based on a comprehensive score of the concentration of chemical components in the water extract of the raw materials, which is also quantified into a user target value on a 100-point scale.

Images