Diagnostic method for wine astringency quality based on ga-bp neural network

Abstract

The application provides a diagnosis method of wine astringency quality based on a GA-BP neural network, which is based on separation of wine single / oligo / high polymer tannin fragments, proposes a wine matrix and tannin quantitative parameter, takes a descriptor of a tannin structure as a characteristic input value according to a quantitative description relationship, and realizes nonlinear fitting of the matrix data of the wine, the tannin quantitative parameter and the astringency quality by means of nonlinear mapping capacity, self-learning and self-adaptive capacity, generalization capacity and fault tolerance capacity of the BP neural network, so that a potential relationship between each characteristic parameter and the astringency quality can be accurately obtained; and the BP neural network is continuously optimized by a genetic algorithm, so that the prediction precision of the neural network on the astringency quality of the wine is realized.

Description

Technical Field

[0001] This invention relates to the field of wine sensory science, specifically to a diagnostic method for the astringency quality of wine based on a GA-BP neural network. Background Technology

[0002] Astringency is a key factor determining the sensory quality of wine. With the booming development of my country's wine industry, research on wine astringency requires more scientific evaluation. The assessment of wine astringency quality is an important basis for wine price and quality control, directly reflecting the state of tannins and matrix in the wine. It is significant for identifying methods to improve wine astringency and for classifying wine prices.

[0003] The assessment of astringency in wine has always been determined by a sensory panel, with the wine's matrix and tannins being key factors influencing the final astringency quality. Existing research focuses on the overall astringency performance of tannins in wine, neglecting the differences in astringency quality resulting from varying degrees of tannin polymerization. Furthermore, investigating tannin characteristics in isolation is a major reason why wine astringency is difficult to predict. Determining the characteristic parameters for predicting wine astringency is a pressing issue, with selecting effective tannin quantification features being particularly challenging. Summary of the Invention

[0004] To address the problems of the prior art, this invention provides a diagnostic method for the astringency quality of wine based on a GA-BP neural network. This method, based on the separation of mono / oligo / poly-tannin fragments in wine, proposes quantitative parameters for the matrix and tannins of wine, establishes a mathematical prediction model for the astringency quality of wine using a BP neural network algorithm, and optimizes the initial parameters of the BP neural network using a GA genetic algorithm to obtain the optimal solution, thereby improving the accuracy of the wine astringency quality prediction model.

[0005] This invention is achieved through the following technical solution:

[0006] Diagnostic methods for the astringency quality of wine based on GA-BP neural networks include:

[0007] S1, collect matrix parameters of the wine sample to be tested;

[0008] S2, after the wine sample to be tested is de-alcoholized, the tannins in the wine sample to be tested are separated into three parts: tannin monomers, oligotannins and high-polymer tannins using solid phase extraction separation technology, and then redissolved into model wine solutions to prepare corresponding model tannin solutions; each model tannin solution is detected and analyzed to determine the structure and content of substances in the three parts: tannin monomers, oligotannins and high-polymer tannins;

[0009] S3, Calculate the tannin quantification parameters based on the structure and content of substances in the three parts of tannin monomer, oligotannin and polymeric tannin;

[0010] S4. Input the matrix parameters and tannin quantification parameters into the BP neural network model optimized by the genetic algorithm. The BP neural network model optimized by the genetic algorithm predicts and outputs the astringency quality grade of the wine sample to be tested.

[0011] Preferably, in S1, the matrix parameters are one or more of the following: the pH of the matrix, the ethanol content in the matrix, and the titratable acid content in the matrix.

[0012] Preferably, in S2, each model tannin solution is detected and analyzed to determine the structure and content of substances in the three parts: tannin monomer, oligotannin, and polymeric tannin. Specifically:

[0013] The proportions of total phenols, total flavanols, and pigment tannins were determined by spectroscopic methods; the content of hydrolyzed tannins, condensed tannins, degree of polymerization, protozoaldin percentage, and galloyl acylated tannins were determined by high performance liquid chromatography.

[0014] Preferably, in S2, solid-phase extraction separation technology is used to separate the tannins in the wine sample into three parts: tannin monomers, oligotannins, and high-polymer tannins, specifically:

[0015] After injecting the wine sample to be tested into a C18 solid phase extraction column, it was dried with nitrogen gas, washed with distilled water, ethyl acetate and methanol respectively, and the fractions were collected and freeze-dried into powder to obtain three parts: tannin monomers, oligotannins and polytannins.

[0016] Preferably, in S3, the tannin quantification parameters are one or more of τ1, τ2, τ3, τ4, and τ5:

[0017] τ1=CT / HT

[0018] τ2=TF3 / TP3

[0019] τ3=TF2 / TP2

[0020] τ4=mDP3 / mDP2

[0021] τ5 = %P2*PT3 / %G3

[0022] Wherein, τ1 represents the content parameters of condensed tannins and hydrolyzed tannins, τ2 represents the quantitative parameter of polymeric tannin content, τ3 represents the quantitative parameter of oligomeric tannins, τ4 represents the quantitative parameter of tannin polymerization, τ5 represents the proportion of each tannin structure at the extended end, CT represents the content of condensed tannins, HT represents the content of hydrolyzed tannins, TF3 represents the total flavanol content of polymeric tannins, TP3 represents the total phenol content of polymeric tannins, TF2 represents the total flavanol content of oligomeric tannins, TP2 represents the total phenol content of oligomeric tannins, mDP3 represents the degree of polymerization of polymeric tannins, mDP2 represents the degree of polymerization of oligomeric tannins, %P2 represents the proportion of protozoanthracene in oligomeric tannins, PT3 represents the proportion of pigment tannins in polymeric tannins, and %G3 represents the proportion of gallic tannins in polymeric tannins.

[0023] Preferably, in S3, the BP neural network model optimized by the genetic algorithm is constructed using the following method:

[0024] (1) Collect sensory data of multiple wine samples and calculate the astringency quality grade of multiple wine samples based on the sensory data;

[0025] (2) Collect matrix parameters from multiple wine samples;

[0026] (3) After de-alcoholizing multiple wine samples, solid-phase extraction separation technology was used to separate the tannins in the wine samples into three parts: tannin monomers, oligotannins and polytannins. These were then redissolved into model wine solutions to prepare corresponding model tannin solutions. Each model tannin solution was tested and analyzed to determine the structure and content of tannin monomers, oligotannins and polytannins.

[0027] (4) Calculate the tannin quantification parameters based on the structure and content of tannin monomers, oligotannins and polymeric tannins;

[0028] (5) Divide the astringency quality grade, matrix parameters and tannin quantification parameters of multiple wine samples into training set and test set;

[0029] (6) Using the matrix parameters and tannin quantification parameters of the wine sample as inputs and the astringency quality grade of the wine sample as output, a genetic algorithm-optimized BP neural network model is constructed. The training set is used for training to obtain the genetic algorithm-optimized BP neural network model.

[0030] (7) The reliability of the prediction results of the BP neural network model optimized by the genetic algorithm was tested using a test set.

[0031] Furthermore, in step (1), the collection of sensory data specifically involves using the five-point oral sensory method and a linear scale to assess the intensity of astringency in the palate, cheeks, tongue, upper and lower lips, and periodontal areas.

[0032] Furthermore, the astringency quality grade of the wine sample is calculated as follows: Y = 0.80 + 0.14 * astringency intensity of upper and lower lips + 0.69 * astringency intensity of periodontal tissue + 0.1 * astringency intensity of palate - 0.62 * astringency intensity of tongue surface + 0.82 * astringency intensity of both cheeks. The calculated score is rounded down and ranges from 1 to 5. Y represents poor, second-best, average, good, and excellent, respectively.

[0033] Furthermore, step (5) also includes: normalizing the matrix data and tannin quantification parameters of the wine.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] This invention, based on quantitative descriptive relationships, uses tannin structure descriptors (tannin quantification parameters) and matrix data as feature input values. Leveraging the nonlinear mapping capability, self-learning and adaptive capabilities, generalization ability, and fault tolerance of a backpropagation (BP) neural network, it achieves nonlinear fitting of matrix data, tannin quantification parameters, and astringency quality in wine, accurately revealing the potential relationship between each feature parameter and astringency quality. By continuously optimizing the BP neural network using a genetic algorithm (GA), the prediction accuracy of the neural network for wine astringency quality is improved, significantly enhancing the prediction speed and accuracy of the BP neural network model optimized by the genetic algorithm (GA-BP neural network model). This method, through the establishment of a GA-BP neural network model, demonstrates high feasibility in predicting the astringency quality level of wine, enabling a more accurate evaluation of wine astringency quality and possessing high application value. It effectively solves the problem of locating and correcting astringency defects in wine, providing guidance for the diagnosis of wine astringency.

[0036] Furthermore, this invention proposes a new algorithm for calculating tannin quantification parameters (τ1, τ2, τ3, τ4, τ5), which reduces the amount of tannin feature extraction and also provides a new approach to tannin quantification parameters.

[0037] Furthermore, this invention simplifies the sensory process by using a five-point sensory method to evaluate the astringency quality level of wine. Sensory observers only need to evaluate the astringency intensity on the upper and lower lips, periodontal tissues, cheeks, tongue surface, and palate, and the astringency quality level can be obtained through regression calculation. Attached Figure Description

[0038] Figure 1 This is a flowchart of the wine astringency quality evaluation method of the present invention;

[0039] Figure 2 This is a structural diagram of the BP neural network of the present invention;

[0040] Figure 3The graph shows the prediction results of the genetic algorithm optimizing the BP neural network according to the present invention. Detailed Implementation

[0041] To further understand the present invention, the present invention will be described below with reference to embodiments. These descriptions are only for further explaining the features and advantages of the present invention and are not intended to limit the claims of the present invention.

[0042] The diagnostic method for astringency quality of wine based on GA-BP neural network described in this invention includes the following steps:

[0043] S1, collect matrix parameters of the wine sample to be tested, including one or more of the following: matrix pH, ethanol content in the matrix, and titratable acid content in the matrix.

[0044] S2. After the wine sample to be tested is de-alcoholized, the tannins in the wine sample are separated into three parts: tannin monomers, oligotannins, and polytannins using solid-phase extraction separation technology. These parts are then redissolved into model wine solutions to prepare corresponding model tannin solutions. Each model tannin solution is then tested and analyzed to determine the structure and content of substances in the three parts: tannin monomers, oligotannins, and polytannins.

[0045] This invention assumes that condensed tannins are straight-chain molecules linked by C4-C8 and do not have branched structures.

[0046] For example, the proportions of total phenols, total flavanols (epicatechin equivalent), and pigment tannins were determined by spectroscopic methods, while the contents of hydrolyzed tannins, condensed tannins, degree of polymerization of condensed tannins, protochlordine percentage, and galloyl acylated tannins were determined by high performance liquid chromatography.

[0047] S3, Calculate tannin quantification parameters based on the structure and content of tannin monomers, oligotannins, and polymeric tannins. The tannin quantification parameters include one or more of τ1, τ2, τ3, τ4, and τ5:

[0048] τ1=CT / HT

[0049] τ2=TF3 / TP3

[0050] τ3=TF2 / TP2

[0051] τ4=mDP3 / mDP2

[0052] τ5 = %P2*PT3 / %G3

[0053] Wherein, τ1 represents the content parameters of condensed tannins and hydrolyzed tannins, τ2 represents the quantitative parameter of polymeric tannin content, τ3 represents the quantitative parameter of oligomeric tannins, τ4 represents the quantitative parameter of tannin polymerization, τ5 represents the proportion of each tannin structure at the extended end, CT represents the content of condensed tannins, HT represents the content of hydrolyzed tannins, TF3 represents the total flavanol content of polymeric tannins, TP3 represents the total phenol content of polymeric tannins, TF2 represents the total flavanol content of oligomeric tannins, TP2 represents the total phenol content of oligomeric tannins, mDP3 represents the degree of polymerization of polymeric tannins, mDP2 represents the degree of polymerization of oligomeric tannins, %P2 represents the proportion of protozoanthracene in oligomeric tannins, PT3 represents the proportion of pigment tannins in polymeric tannins, and %G3 represents the proportion of gallic tannins in polymeric tannins.

[0054] S4. Input the matrix parameters and tannin quantification parameters into the BP neural network model optimized by the genetic algorithm. The BP neural network model optimized by the genetic algorithm predicts and outputs the astringency quality grade of the wine sample to be tested.

[0055] The genetic algorithm-optimized BP neural network model described in this invention is constructed using the following method:

[0056] (1) Collect sensory data of wine samples: Establish a sensory group and use the five-point oral sensory method to evaluate the astringency intensity of the palate, cheeks, tongue, upper and lower lips and periodontal area of ​​the oral cavity using a linear scale (0-5cm) to obtain sensory data, namely, the astringency intensity of the upper and lower lips, periodontal astringency intensity, palate astringency intensity, tongue astringency intensity and cheek astringency intensity; calculate the astringency quality grade of the wine samples based on the sensory data.

[0057] In this embodiment of the invention, the astringency quality grade of the wine sample is Y = 0.80 + 0.14 * astringency intensity of upper and lower lips + 0.69 * astringency intensity of periodontal tissue + 0.1 * astringency intensity of palate - 0.62 * astringency intensity of tongue surface + 0.82 * astringency intensity of both cheeks. The calculated score is rounded down and ranges from [1-5]. Y from 1 to 5 represents poor, second-best, average, good, and excellent, respectively.

[0058] (2) Collect matrix data of wine samples: Analyze at least one of pH, ethanol and titratable acid content in the matrix of wine according to GB / T 15038-2006.

[0059] (3) Collect tannin data from wine samples: After dealcoholization, the tannins in the wine samples were separated into three parts—tannin monomers, oligotannins, and polytannins—using a C18 solid-phase extraction column. These were then dissolved in a simulated wine solution to prepare a simulated tannin solution with the same volume as the wine sample. Each model tannin solution was tested and analyzed to determine the structure and content of the substances in the three parts: tannin monomers, oligotannins, and polytannins. The specific method for this step is the same as in S2 above.

[0060] (4) Tannin quantification description (i.e. tannin quantification parameters) calculation (τ1, τ2, τ3, τ4, τ5).

[0061] τ1=CT / HT

[0062] τ2=TF3 / TP3

[0063] τ3=TF2 / TP2

[0064] τ4=mDP3 / mDP2

[0065] τ5 = %P2*PT3 / %G3

[0066] Wherein, τ1 represents the content parameters of condensed tannins and hydrolyzed tannins, τ2 represents the quantitative parameter of polymeric tannin content, τ3 represents the quantitative parameter of oligomeric tannins, τ4 represents the quantitative parameter of tannin polymerization, and τ5 represents the proportion of each tannin structure at the extended end. CT represents the content of condensed tannins, HT represents the content of hydrolyzed tannins, TF3 represents the total flavanol content of polymeric tannins, TP3 represents the total phenol content of polymeric tannins, TF2 represents the total flavanol content of oligomeric tannins, TP2 represents the total phenol content of oligomeric tannins, mDP3 represents the degree of polymerization of polymeric tannins, mDP2 represents the degree of polymerization of oligomeric tannins, %P2 represents the proportion of protozoalcium nitrate in oligomeric tannins, PT3 represents the proportion of pigment tannins in polymeric tannins, and %G3 represents the proportion of gallic tannins in polymeric tannins.

[0067] (5) Data processing.

[0068] Step S501 involves dividing the data from multiple wine samples into training and testing sets. The training set is used for learning the GA-BP neural network model, while the testing set is used to verify the model's predictive accuracy. This data includes the wine's matrix data, tannin quantification parameters, and astringency quality grades.

[0069] Step S502, normalization processing: The matrix data and tannin quantitative parameters of the wine are normalized, while the astringency quality grade data are not normalized. The corresponding calculation formula is as follows:

[0070]

[0071] Where, x i Let x be any value in the sample data. max x is the maximum value in the sample data. min y is the minimum value in the sample data. i The value is the normalized value, and its range is [0, 1].

[0072] (6) A BP neural network model optimized by a genetic algorithm was constructed using the matrix parameters and tannin quantification parameters of the wine samples as inputs and the astringency quality grade of the wine samples as outputs. The BP neural network model and genetic algorithm optimization were designed.

[0073] Step S601: Construct a BP neural network.

[0074] Step S602: Genetic algorithm parameter design and optimization, initialization of the genetic algorithm population, and random generation of an initial population containing N individuals.

[0075] Step S603: Determine the fitness function SSE is the sum of squared errors between the predicted and actual values.

[0076] Step S604: Use the roulette wheel algorithm to select the chromosome with higher fitness.

[0077] Step S605, the crossover operation of the population genes is as follows.

[0078] Step S606, mutation operation: change the gene values ​​at certain loci of the individual strings in the population to generate new individuals, so that the genetic algorithm has local random search capability.

[0079] Step S607: When the fitness threshold is reached, the set number of iterations is reached, or the fitness of the best individual or group no longer increases, the algorithm can be terminated if any of these three conditions are met, and the optimal weight and threshold are output.

[0080] (7) Compare the real data of sensory quality levels in the training set and the test set with the predicted data of sensory quality levels obtained by the GA-BP neural network model to determine the accuracy of the prediction results.

[0081] Example

[0082] This embodiment uses 40 wine samples from both domestic and international sources as examples to demonstrate a wine astringency quality evaluation method based on a GA-BP neural network. Leveraging the nonlinear mapping capability, self-learning and adaptive capabilities, generalization ability, and fault tolerance of the BP neural network, it achieves nonlinear fitting of the wine's matrix, tannin parameters, and astringency quality. The BP neural network is continuously optimized using a genetic algorithm (GA) to improve the prediction accuracy of the astringency quality.

[0083] Reference Figure 1 The overall process of the wine astringency quality evaluation method of the present invention is as follows:

[0084] (1) Sensory data collection. A sensory group of 15 people was established. All members of the group had long-term experience in wine studies, including 8 males and 7 females. All members had undergone sensory training before the formal sensory evaluation. Tannins were used as the standard solution for astringency intensity in the sensory training. The sensory method required the use of the five-point oral sensory method to score the astringency intensity of the wine on the upper and lower lips, periodontal tissues, tongue surface, cheeks, and palate. The result was the average of the scores given by all sensory members. The astringency quality grade Y was calculated as follows: Y = 0.80 + 0.14 * astringency intensity of upper and lower lips + 0.69 * astringency intensity of periodontal tissues + 0.1 * astringency intensity of palate - 0.62 * astringency intensity of tongue surface + 0.82 * astringency intensity of cheeks. The calculated score was rounded down to the nearest integer, ranging from [1-5], representing poor, second-best, average, good, and excellent, respectively.

[0085] The quality grade of astringency is calculated based on the intensity of astringency at five points in the oral cavity, as shown in Table 1 below.

[0086] Table 1. Sensory data of 40 wine samples and calculation of wine astringency quality grade.

[0087]

[0088] (2) Collect matrix data of wine samples by fully automated wine component analyzer. The determination of pH, ethanol and titratable acid (g / L) in the matrix shall be performed in accordance with GB / T 15038-2006 standard.

[0089] (3) Collect tannin data from wine samples, including the proportion of total phenols, total flavanols and pigment tannins, hydrolyzed tannin content, condensed tannin content, degree of polymerization, percentage of protochlor, and proportion of galloyl acylated tannins.

[0090] In step S301, after the wine sample is de-alcoholized, the tannins in the wine sample are separated into three parts—tannin monomers, oligotannins, and polytannins—using a C18 solid-phase extraction column. These are then redissolved separately in model wine solutions to prepare corresponding model tannin solutions. Each model tannin solution is detected and analyzed. The pigment tannin percentage (P%) is determined using a UV spectrophotometer (A520 / A280), where A520 represents the spectrophotometric value of the wine sample at 520 nm, and A280 represents the spectrophotometric value of the wine sample at 280 nm. The total phenol percentage is determined using the Folin-Schönlein method, and the total flavanol percentage is determined using the DMACA reagent method.

[0091] Step S302 involves determining the content of hydrolyzed tannins, condensed tannins, the degree of polymerization of polymeric tannins, the degree of polymerization of oligomeric tannins, the percentage of proto-sterilizing tannins, and the proportion of galloyl acylated tannins using high-performance liquid chromatography (HPLC). The HPLC determination was performed using an HPLC1260 (Agilent) system equipped with a VWD UV detector and a Waters XBridgeShield RP18 3.5μm 4.6×250mm column. Mobile phase A consisted of 2% formic acid in water, and mobile phase B consisted of a mixed solution of acetonitrile containing 20% ​​of mobile phase A. The elution gradient was as follows: 0–5 min, 0–10% B; 5–10 min, 10–25% B; 10–40 min, 35–55% B; 55–65 min, 55–70% B; 65–80 min, 70–75% B; 80–90 min, 75–80% B; 90–100 min, 80–100% B; 100–115 min, 100–100% B; 115–120 min, 100–10% B; 120–125 min, 10–10% B.

[0092] (4) Calculation of tannin quantification parameters (τ1, τ2, τ3, τ4, τ5).

[0093] τ1=CT / HT

[0094] τ2=TF3 / TP3

[0095] τ3=TF2 / TP2

[0096] τ4=mDP3 / mDP2

[0097] τ5 = %P2*PT3 / %G3

[0098] Wherein, τ1 represents the content parameters of condensed tannins and hydrolyzed tannins, τ2 represents the quantitative parameter of polymeric tannin content, τ3 represents the quantitative parameter of oligomeric tannins, τ4 represents the quantitative parameter of tannin polymerization, and τ5 represents the proportion of each tannin structure at the extended end. CT represents the content of condensed tannins, HT represents the content of hydrolyzed tannins, TF3 represents the total flavanol content of polymeric tannins, TP3 represents the total phenol content of polymeric tannins, TF2 represents the total flavanol content of oligomeric tannins, TP2 represents the total phenol content of oligomeric tannins, mDP3 represents the degree of polymerization of polymeric tannins, mDP2 represents the degree of polymerization of oligomeric tannins, %P2 represents the proportion of protozoalcium nitrate in oligomeric tannins, PT3 represents the proportion of pigment tannins in polymeric tannins, and %G3 represents the proportion of gallic tannins in polymeric tannins.

[0099] The eight input variables are matrix data and tannin quantification parameters, and the output variable is the astringency quality grade (poor = 1, inferior = 2, medium = 3, good = 4, excellent = 5). The eight input variables and one output variable are organized into an Excel table as shown in Table 2 below.

[0100] Table 2. Matrix data and tannin quantification parameters of wine

[0101]

[0102]

[0103] (5) In data preprocessing, the data is first divided into training and test sets. Thirty random wine samples from the original data are selected as the training set, and ten wine samples are used as the test set. Normalization is then performed on the matrix data and tannin quantitative parameter data of the wines. Sensory quality grade data are not normalized. The corresponding calculation formulas are as follows:

[0104]

[0105] Where, x i Let x be any value in the sample data. max x is the maximum value in the sample data. min y is the minimum value in the sample data. i The value is the normalized value, and its range is [0, 1].

[0106] (6) Design neural networks and optimize using genetic algorithms. For example... Figure 2 As shown, it is a structural diagram of the BP neural network model in an embodiment of the present invention.

[0107] Step S601: The input layer of the BP neural network has 8 cells, the output layer has 5 categories, and the number of neurons in the hidden layer is... Where n is the number of nodes in the input layer, l is the number of nodes in the output layer, c is a constant between [1, 10], and j ranges between [3, 13]. The maximum number of iterations for the BP neural network is 1000, and the error threshold is 10. -6 The learning rate is 0.01.

[0108] Step S602, Genetic Algorithm Parameter Design and Optimization: The number of optimized parameters is S = n*j + j*l + j + l, where j is the number of hidden layer nodes, n is the number of output layer nodes, and l is the number of output layer nodes.

[0109] Initialize the genetic algorithm population by randomly generating an initial population containing N individuals.

[0110] Step S603: Determine the fitness function. SSE is the sum of squared errors between the predicted and actual values.

[0111] Step S604: Use the roulette wheel algorithm to select the chromosome with higher fitness.

[0112] Step S605, the crossover operation of the population genes is as follows:

[0113] C n1 =C n1 (1-h)+C m1 h

[0114] C m1 =C m1 (1-h)+C n1 h

[0115] Where h is a random number between [0,1], and C n1 It is the first gene on the nth chromosome, C m1 It is the first gene on the m-th chromosome.

[0116] Step S606, mutation operation: change the gene values ​​at certain loci of the individual strings in the population to generate new individuals, so that the genetic algorithm has local random search capability.

[0117] Step S607: When the fitness threshold is reached, the set number of iterations is reached, or the fitness of the best individual or group no longer increases, the algorithm can be terminated if any of these three conditions are met, and the optimal weight and threshold are output.

[0118] The parameters of the genetic algorithm described in this embodiment are set as follows: initial population size is 50, number of generations is 50, crossover probability is 0.5, and mutation probability is 0.03.

[0119] (7) Compare the actual sensory quality level data from the training set and the test set with the predicted sensory quality level data obtained by the GA-BP neural network model to determine the accuracy of the prediction results. The results are shown in […]. Figure 3 .from Figure 3 It can be seen that the accuracy rate of the training set is 83.3%, and the accuracy rate of the test set reaches 90.9%.

[0120] The embodiments described above merely illustrate the implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A diagnostic method for the astringency quality of wine based on GA-BP neural network, characterized in that, include: S1, collect matrix parameters of the wine sample to be tested; S2, after the wine sample to be tested is de-alcoholized, the tannins in the wine sample to be tested are separated into three parts: tannin monomers, oligotannins and high-polymer tannins using solid phase extraction separation technology, and then redissolved into the model wine solution to prepare the corresponding model tannin solutions. Each model tannin solution was tested and analyzed to determine the structure and content of substances in the three parts of tannin monomer, oligotannin, and polymeric tannin. Specifically, the proportions of total phenols, total flavanols, and pigment tannins were determined by spectroscopic methods; and the contents of hydrolyzed tannins, condensed tannins, degree of polymerization, protochlor tannin percentage, and galloyl acylated tannins were determined by high performance liquid chromatography. S3, Calculate the tannin quantification parameters based on the structure and content of substances in the three parts of tannin monomers, oligotannins, and polymeric tannins; the tannin quantification parameters are one or more of τ1, τ2, τ3, τ4, and τ5: Wherein, τ1 represents the content parameters of condensed tannins and hydrolyzed tannins, τ2 represents the quantitative parameter of high-polymer tannin content, τ3 represents the quantitative parameter of oligo-polymer tannins, τ4 represents the quantitative parameter of tannin polymerization, τ5 represents the proportion of each tannin structure at the extended end, CT represents the content of condensed tannins, HT represents the content of hydrolyzed tannins, TF3 represents the total flavanol content of high-polymer tannins, TP3 represents the total phenol content of high-polymer tannins, TF2 represents the total flavanol content of oligo-polymer tannins, TP2 represents the total phenol content of oligo-polymer tannins, mDP3 represents the degree of polymerization of high-polymer tannins, mDP2 represents the degree of polymerization of oligo-polymer tannins, %P2 represents the proportion of protozoaltin in oligo-polymer tannins, PT3 represents the proportion of pigment tannins in high-polymer tannins, and %G3 represents the proportion of gallic tannins in high-polymer tannins. S4. Input the matrix parameters and tannin quantification parameters into the BP neural network model optimized by the genetic algorithm. The BP neural network model optimized by the genetic algorithm predicts and outputs the astringency quality grade of the wine sample to be tested.

2. The method for diagnosing the astringency quality of wine based on a GA-BP neural network according to claim 1, characterized in that, In S1, the matrix parameters are one or more of the following: the pH of the matrix, the ethanol content in the matrix, and the titratable acid content in the matrix.

3. The method for diagnosing the astringency quality of wine based on a GA-BP neural network according to claim 1, characterized in that, In S2, solid-phase extraction separation technology is used to separate the tannins in the wine sample into three parts: tannin monomers, oligotannins, and high-polymer tannins. Specifically: After injecting the wine sample into a C18 solid-phase extraction column, it was dried with nitrogen and washed with distilled water, ethyl acetate and methanol respectively. The fractions were collected and freeze-dried into powder to obtain three parts: tannin monomers, oligotannins and polytannins.

4. The method for diagnosing the astringency quality of wine based on a GA-BP neural network according to claim 1, characterized in that, In S3, the BP neural network model optimized by the genetic algorithm is constructed using the following method: (1) Collect sensory data of multiple wine samples and calculate the astringency quality grade of multiple wine samples based on the sensory data; (2) Collect matrix parameters from multiple wine samples; (3) After de-alcoholizing multiple wine samples, solid phase extraction separation technology was used to separate the tannins in the wine samples into three parts: tannin monomers, oligotannins and high-polymer tannins, and then redissolved them into the model wine solution to prepare the corresponding model tannin solutions. Each model tannin solution was tested and analyzed to determine the structure and content of tannin monomers, oligotannins, and polymeric tannins; (4) Calculate the tannin quantification parameters based on the structure and content of tannin monomers, oligotannins and polymeric tannins; (5) Divide the astringency quality grade, matrix parameters and tannin quantitative parameters of multiple wine samples into training set and test set; (6) Using the matrix parameters and tannin quantification parameters of the wine sample as inputs and the astringency quality grade of the wine sample as output, construct a BP neural network model optimized by genetic algorithm, train it using the training set, and obtain the BP neural network model optimized by genetic algorithm. (7) The reliability of the prediction results of the BP neural network model optimized by the genetic algorithm was tested using a test set.

5. The method for diagnosing the astringency quality of wine based on a GA-BP neural network according to claim 4, characterized in that, In step (1), the collection of sensory data specifically involves using the five-point oral sensory method and a linear scale to assess the intensity of astringency in the palate, cheeks, tongue, upper and lower lips, and periodontal areas.

6. The method for diagnosing the astringency quality of wine based on a GA-BP neural network according to claim 5, characterized in that, Astringency quality grade of wine samples Intensity of astringency on upper and lower lips The score is calculated and rounded down, ranging from 1 to 5. Y represents poor, average, good, and excellent, respectively, from 1 to 5.

7. The method for diagnosing the astringency quality of wine based on a GA-BP neural network according to claim 4, characterized in that, Step (5) also includes: normalizing the matrix data and tannin quantification parameters of the wine.

Images