Method for predicting the quality grade of mellow liquor base wine

Through GC-MS detection and K nearest neighbor model combined with CART algorithm to screen characteristic substances, the problem of professional equipment and long time required for prediction of liquor quality grade in the existing technology is solved, and the rapid and accurate liquor grade prediction is achieved, suitable for large-scale Sample treatment.

CN115078573BActive Publication Date: 2024-11-05JIANGSU YANGHE BREWERY JOINT STOCK +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210650567.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-09
Publication Date
2024-11-05
Estimated Expiration
2042-06-09

AI Technical Summary

Technical Problem

The prior art requires professional large-scale instruments and equipment when predicting the quality level of liquor, or has not formed a stable and mature mathematical model, or the analysis steps are cumbersome and time is long, making it difficult to meet the needs of fast and accurate detection.

Method used

GC-MS was used to detect the fingerprint of volatile flavor components of soft liquor base wine, build a K nearest neighbor model, and use CART algorithm to screen characteristic flavor substances, and establish a model for predicting the quality level of liquor.

Benefits of technology

It achieves rapid and accurate prediction of liquor quality grades, is suitable for large-scale sample processing, improves the accuracy and reliability of the model, and reduces the requirements for the professional capabilities of equipment and operators.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115078573B_ABST
    Figure CN115078573B_ABST
Patent Text Reader

Abstract

The invention discloses a method for predicting the quality grade of a mellow-type liquor base wine, and belongs to the field of liquor brewing. The invention uses the fingerprint spectrum of the volatile flavor components of the mellow-type liquor base wine to establish and train and optimize the K nearest neighbor model, and then uses the CART algorithm to screen the volatile flavor components, removes substances that have no obvious effect on the prediction model or even have a negative effect, obtains characteristic substances, successfully realizes the compression of the characteristic space dimension, effectively and reliably improves the modeling quality, and can achieve a high accuracy prediction with less substances. In addition, the method of the invention is simple and fast to operate, and is suitable for high-throughput analysis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to a method for predicting the quality grade of a mellow-type liquor base wine, and belongs to the field of liquor brewing. Background Art

[0002] At present, the quality grade identification of liquor mainly relies on sensory evaluation, and the results of sensory evaluation are greatly affected by the subjective influence of tasters. When identifying large quantities of samples, it is time-consuming and laborious, and the results of tasting are difficult to guarantee. Therefore, it is of great significance to improve the scientificity and effectiveness of liquor quality grade evaluation and establish a rapid grading method. The non-sensory evaluation methods mainly include: Zhu Cheng et al. combined system cluster analysis to process gas chromatography fingerprints to identify liquor grades; Yin Yong et al. combined radial basis neural network with sensor array technology to establish a liquor quality identification model; Zhang Liang et al. used PCA+SVM cluster analysis of Fourier near-infrared spectroscopy fingerprints to identify liquor grades; Cheng Pingyan et al. established a liquor quality grade identification model based on empty solid phase microextraction mass spectrometry and chemometrics; Chen Mingju et al. established a Luzhou-flavor liquor evaluation model based on SPCA and GC-MS. The above research methods provide a variety of identification schemes for liquor grade identification, but these methods require more professional large-scale instruments and equipment, or have not formed a stable and mature mathematical model, or the analysis steps are more cumbersome and the analysis time is longer. At present, the market demand for liquor is growing, and the number of samples to be tested is increasing. How to develop simple, fast and accurate detection and identification technology has become a new urgent need. Summary of the invention

[0003] [Technical issues]

[0004] The technical problem to be solved by the present invention is that when the existing technology predicts the quality grade of liquor, professional large-scale instruments and equipment are needed, or a stable and mature mathematical model has not been formed, or the analysis steps are relatively cumbersome and the analysis time is long.

[0005] [Technical solution]

[0006] The present invention provides a method for predicting the quality grade of a mellow-flavored liquor base wine, comprising the following steps:

[0007] Step (1): using GC-MS to detect mellow-flavored liquor base liquors of known quality grades and different quality grades to obtain fingerprints of volatile flavor components;

[0008] Step (2): construct a K-Nearest Neighbor (KNN) model and use the fingerprint obtained in step (1) to train and optimize the K-Nearest Neighbor model to obtain a preliminary discrimination model;

[0009] Step (3): the fingerprint obtained in step (1) is calculated using the CART algorithm to screen out characteristic flavor substances, and the preliminary discrimination model is optimized using the obtained characteristic flavor substances to obtain a prediction model for predicting the quality grade of the mellow-flavor liquor base wine;

[0010] Step (4): Use the prediction model obtained in step (3) to predict the quality grade of the mellow-flavored liquor base wine.

[0011] Further, step (1) comprises the following steps:

[0012] Step 101, using mellow-flavored liquor base liquors of known quality grades and different quality grades as test samples, using ultrapure water to reduce the alcohol content of the test samples to 8 degrees, and adding sodium chloride and an internal standard to obtain test samples; the quality grades are divided into two grades: superior-grade liquor and general-grade liquor;

[0013] Step 102, using a headspace solid phase microextraction method to obtain volatile compounds from the sample through an extraction head;

[0014] Step 103, after the extraction head is subjected to analytical adsorption at the GC-MS injection port, the fingerprint information of the volatile flavor components is collected by GC-MS to obtain the fingerprint information of the volatile flavor components of the mellow-type liquor base wine of different quality grades.

[0015] Furthermore, the step 101 is to use mellow-type liquor base liquors of different quality grades of known quality grades as samples to be tested, use ultrapure water to reduce the alcohol content of the liquor sample to 8% vol, then take 6 mL and put it in an injection bottle, add 0.25 g / mL sodium chloride to make the solution supersaturated, and then add 10 μL of internal standard to obtain the sample to be tested; the internal standard is tert-amyl alcohol, which is a solution with a concentration of 8.05 g / L obtained by diluting the volume with methanol.

[0016] Furthermore, in step 102, the parameters of headspace solid phase microextraction are: equilibration at 40-60° C. for 1-25 min, and extraction time of 40 min.

[0017] Furthermore, in the step 103, the GC analysis conditions are: using a 30m×0.25mm×0.50μm DB-WAX capillary gas chromatography column, the carrier gas is high-purity helium, the flow rate is 1.2mL / min, the split ratio is 5:1, the program temperature is: start at 50°C and maintain for 2min, increase the temperature to 145°C at 3°C / min, and then increase the temperature to 230°C at 15°C / min and maintain for 3min, and the injection port temperature is maintained at 250°C; the MS analysis conditions are: transmission line temperature 230°C, ion source temperature 260°C, scanning mass range m / z: 33~350amu, ionization mode: EI+; electron energy: 70eV.

[0018] Furthermore, step (2) includes the following steps:

[0019] Step 201, using the fingerprint data of the volatile flavor components obtained in step (1) as a test sample;

[0020] Step 202: Establish a K-nearest neighbor model, and train and optimize the K-nearest neighbor model using the test samples obtained in step (1) according to a 10-fold cross-validation method to obtain a preliminary discriminant model and the accuracy A1 of the preliminary discriminant model.

[0021] Furthermore, step (3) specifically includes the following steps:

[0022] Step 301, calculating the Gini index of the test samples obtained in step (1) based on the CART algorithm in the decision tree, and obtaining the contribution of each volatile flavor substance to the quality grade of the mellow-flavored liquor base wine according to the Gini index, sorting them in descending order, and screening out characteristic substances;

[0023] Step 302: respectively bring the top 10, top 20, top 30, and top 100 characteristic substances ranked in step 301 (10 substances are added each time) back to the preliminary discrimination model of step (2), and obtain the accuracy An of the preliminary discrimination model under different numbers of characteristic substances;

[0024] Step 303, compare the accuracy of the preliminary discrimination model under different numbers of characteristic substances. If Ax≥An and Ax≥A1, the preliminary discrimination model with the top x characteristic substances is used as a model for predicting the quality grade of the mellow-type liquor base wine.

[0025] Further, in certain embodiments of the present invention, the top x characteristic substances include ethyl lactate, isopentyl lactate, benzaldehyde, phenylethyl acetate, diethyl succinate, ethyl 3-phenylpropionate, hexyl lactate, n-hexanal, 4-ethyl-2-methoxyphenol, ethyl phenylacetate, phenylethyl butyrate, ethyl acetate, caproic anhydride, amyl acetate, (2,2-diethoxyethyl)-benzene, butyl phenylacetate, diethyl azelaic acid, ethyl 4-phenylbutyrate, n-octanol, 2-phenylethyl hexanoate, furfuryl hexanoate, isopentyl acetate, isobutanol, 2-pentadecanone, geranyl acetone, n-octanol, ethyl 9-tetradecenoate, butyl acetate, ethyl benzoate, and 2-undecenal.

[0026] [Beneficial Effects]

[0027] Compared with other models, the K nearest neighbor model is more suitable for the present invention. The present invention uses the fingerprint spectrum of the volatile flavor components of the mellow-type liquor base wine to establish and train and optimize the K nearest neighbor model. Then, the volatile flavor components are screened using the CART algorithm to remove substances that have no obvious impact on the prediction model or even have a negative effect, thereby obtaining characteristic substances, successfully achieving the compression of the dimension of the feature space, effectively and reliably improving the modeling quality, and achieving a higher accuracy prediction with fewer substances.

[0028] The fingerprint spectrum input into the prediction model of the present invention can be obtained by means of GC-MS. The GC-MS technology is stable and mature, the instrumental analysis has high precision, small errors between samples, high repeatability, reliable results, and large analysis throughput. In addition, the GC-MS detection technology does not have excessive requirements on the professional ability of equipment and operators, and can quickly complete data collection, thereby realizing rapid and accurate prediction of liquor grades, and is suitable for large-scale sample processing and screening.

[0029] The prediction model obtained by the present invention has a high accuracy rate when predicting the quality grade of mellow and soft liquor, and has high applicability for mellow and soft liquor. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The present invention is a flow chart of a method for determining the grade of a mellow-type liquor base wine.

[0031] Figure 2 A curve chart showing the accuracy of the model for every ten characteristic substances. DETAILED DESCRIPTION

[0032] like Figure 1 As shown, the method for predicting the quality grade of a mellow-flavored liquor base wine of the present invention comprises the following steps:

[0033] S1: GC-MS is used to detect mellow-flavored liquor base liquors of known quality grades and different quality grades to obtain fingerprints of volatile flavor components;

[0034] S2: construct a K nearest neighbor model and use the fingerprint spectrum obtained in step S1 to train and optimize the K nearest neighbor model to obtain a preliminary discrimination model;

[0035] S3: The fingerprint spectrum obtained in S1 is calculated using the CART algorithm to screen out characteristic flavor substances, and the obtained characteristic flavor substances are used to optimize the preliminary discrimination model to obtain a prediction model for predicting the quality grade of the base wine of the mellow and soft liquor;

[0036] S4: Use the prediction model obtained in S3 to predict the quality grade of the mellow-type liquor base wine.

[0037] Example 1 Method for predicting the grade of mellow and soft liquor base wine

[0038] Step S1, using GC-MS to obtain fingerprints of volatile flavor components of mellow-flavored liquor base liquors of known quality grades and different quality grades.

[0039] Specifically:

[0040] Step 101, take 4 production batches of mellow liquor base liquor, a total of 58 samples to be tested, it is known that the quality grades of these 58 samples to be tested are: 29 premium liquor samples and 29 general liquor samples. Take 38 samples to establish the model, and the other 10 premium liquor samples and 10 general liquor samples are used to verify the accuracy of the model prediction results.

[0041] The alcohol content of the sample to be tested is reduced to 8% vol, 6 mL is taken and placed in a sample injection bottle, 0.25 g / mL sodium chloride is added until the solution is supersaturated, and 10 μL of internal standard is added to obtain the sample to be tested; wherein, the internal standard is tert-amyl alcohol; the concentration of the internal standard is 8.05 g / L; tert-amyl alcohol is stable in nature and is not prone to unnecessary reactions that affect the results. At the same time, it will not change due to storage, so it is a suitable internal standard choice, and its physical and chemical properties are similar to those of the volatile components in white wine, which can reduce errors.

[0042] Step 102, using a headspace solid phase microextraction method, extracting volatile compounds from the sample through an extraction head, the parameters of the headspace solid phase microextraction are: 40-60°C equilibration for 1-25 minutes, and extraction time of 40 minutes.

[0043] Step 103, after the extraction head is desorbed and adsorbed at the injection port, GC-MS is used to collect fingerprint information of volatile components; wherein, the GC analysis conditions are: using a 30m×0.25mm×0.50μm DB-WAX capillary gas chromatography column, the carrier gas is high-purity helium, the flow rate is 1.2mL / min, the split ratio is 5:1, the program temperature is: starting at 50°C and maintaining for 2min, heating to 145°C at 3°C / min, and then heating to 230°C at 15°C / min and maintaining for 3min, and the injection port temperature is maintained at 250°C; MS analysis conditions are: transmission line temperature 230°C, ion source temperature 260°C, scanning mass range m / z: 33-350amu, ionization mode: EI+; electron energy: 70eV;

[0044] Each sample was measured three times in parallel, and the corresponding data were counted to obtain the fingerprints of volatile flavor components of different grades of liquor.

[0045] Step S2, constructing a K-Nearest Neighbor (KNN) model and using the fingerprint spectrum obtained in step S1 to train and optimize the K-Nearest Neighbor model to obtain a preliminary discrimination model.

[0046] Specifically:

[0047] Step 201, taking the fingerprint data of the volatile flavor components of 38 samples to be tested as test samples.

[0048] Step 202, establish a K-nearest neighbor model, train and optimize the K-nearest neighbor model with 38 test samples according to the ten-fold cross validation method, and obtain a preliminary discriminant model and the accuracy A1 of the preliminary discriminant model in judging the quality grade of the base wine of mellow and soft liquor, A1 is 0.789474. As shown in Table 3, the AUC value is a standard for comprehensively measuring the quality of the classification model. The AUC value is the area covered by the ROC curve (Receiver Operating Characteristic Curve), and the value is 0.5-1.0. A larger AUC value represents that the model has better classification ability. The performance of the classifier is usually evaluated by classification accuracy (CA), precision (Precision) and recall (Recall). The accuracy (CA) represents the ratio of the number of samples correctly classified by the classifier to the total number of samples for a given test data set. Precision (Precision) usually represents the proportion of positive examples in the results predicted to be positive examples. Recall (Recall) represents the proportion of positive examples predicted to be positive examples in the samples that are actually positive examples. The F1 value (F1) represents the harmonic mean of precision and recall.

[0049] Step S3, the fingerprint spectrum obtained in step S1 is calculated using the CART algorithm to screen out characteristic flavor substances, and the obtained characteristic flavor substances are used to optimize the preliminary discrimination model to obtain a prediction model for predicting the quality grade of the base wine of mellow and soft liquor.

[0050] Step 301, importing the fingerprint data of 38 samples to be tested into the CART algorithm, calculating the Gini index, obtaining the contribution of each volatile flavor substance to the quality grade of the mellow-type liquor base wine according to the Gini index, sorting them in descending order, and screening out characteristic substances that can be used for modeling;

[0051] Step 302, respectively, bring the top 10, top 20, top 30, and finally top 100 characteristic substances (10 substances are added each time) back to the preliminary discrimination model to obtain the accuracy An under different numbers of characteristic substances, as shown in Table 2;

[0052] Step 303, compare the accuracy A of the preliminary discrimination model under different numbers of characteristic substances. If Ax≥An and Ax≥A1, the preliminary discrimination model after introducing the top x characteristic substances is used as the prediction model for predicting the quality grade of the mellow-type liquor base wine. In this embodiment, when the first 30 characteristic substances are used, the highest accuracy of the model is 0.968596, as shown in Table 3. Comparison of various aspects with the evaluation criteria shows that they are improved after optimization, indicating that the accuracy of this model is high. The top 30 characteristic substances are shown in Table 1, which are ethyl lactate, isopentyl lactate, benzaldehyde, phenylethyl acetate, diethyl succinate, ethyl 3-phenylpropionate, hexyl lactate, n-hexanal, 4-ethyl-2-methoxyphenol, ethyl phenylacetate, phenylethyl butyrate, ethyl acetate, caproic anhydride, amyl acetate, (2,2-diethoxyethyl)-benzene, butyl phenylacetate, diethyl azelaic acid, ethyl 4-phenylbutyrate, n-octanol, 2-phenylethyl hexanoate, furfuryl hexanoate, isopentyl acetate, isobutanol, 2-pentadecanone, geranyl acetone, n-octanol, ethyl 9-tetradecenoate, butyl acetate, ethyl benzoate, and 2-undecenal.

[0053] Step S4, predicting the grade of the mellow-flavored liquor base wine through the prediction model.

[0054] Step 401, 10 premium wine samples and 10 general wine samples are input into the prediction model as test samples to obtain the predicted base wine grade, as shown in Table 4, with an accuracy of 95%.

[0055] Table 1 The first 30 characteristic substances selected in step S3 for modeling

[0056]

[0057]

[0058]

[0059] Table 2 The accuracy of the preliminary discrimination model after different numbers of characteristic substances were brought back to the preliminary discrimination model

[0060] serial number Quantity of substances Accuracy 1 10 0.947368421052632 2 20 0.956140350877193 3 30 0.96859649122807 4 40 0.947368421052632 5 50 0.947368421052632 6 60 0.868421052631579 7 70 0.833333333333333 8 80 0.833333333333333 9 90 0.798245614035088 10 100 0.798245614035088

[0061] Table 3 Comparison of model accuracy before and after feature variable screening

[0062] AUC CA F1 Precision Recall Before screening 0.844568 0.789474 0.789409 0.789831 0.789474 After screening 0.984918 0.968596 0.968596 0.968596 0.968596

[0063] Table 4 Prediction results of 20 samples introduced into the prediction model

[0064]

[0065]

[0066] Comparative Example 1

[0067] Step S1 is the same as S1 in Example 1.

[0068] Step S2 is the same as S2 in Example 1.

[0069] Step S3, 10 premium wine samples and 10 general wine samples are input into the preliminary discrimination model as test samples to obtain the predicted base wine grade, as shown in Table 5, with an accuracy rate of 75%.

[0070] Table 5 Prediction results of 20 samples introduced into the prediction model

[0071]

[0072]

[0073] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Anyone familiar with this technology can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the definition of the claims.

Claims

1. A method for predicting the quality grade of a mellow-flavored liquor base wine, characterized in that: The following steps are involved: Step (1): Step 101, taking 19 premium wine samples and 19 general wine samples as mellow-type liquor base wine test samples for establishing a model, using ultrapure water to reduce the alcohol content of the test samples to 8 degrees, taking 6 mL and placing it in a sample injection bottle, adding 0.25 g / mL sodium chloride until the solution is supersaturated, and adding 10 μL of tert-amyl alcohol to obtain the test sample; Step 102, using a headspace solid phase microextraction method, extracting volatile compounds from the sample through an extraction head, the parameters of the headspace solid phase microextraction are: 40-60°C equilibration for 1-25 minutes, extraction time 40 minutes; Step 103, after the extraction head is subjected to desorption and adsorption at a GC-MS injection port, GC-MS is used to collect fingerprint information of volatile flavor components, wherein the GC analysis conditions are: using a 30m×0.25mm×0.50μm DB-WAX capillary gas chromatography column, the carrier gas is high-purity helium, the flow rate is 1.2mL / min, Split ratio: 5:1, heating program: start at 50℃ and maintain for 2min, then heat to 145℃ at 3℃ / min, then heat to 230℃ at 15℃ / min and maintain for 3min, injection port temperature is maintained at 250℃; MS analysis conditions: transmission line temperature 230℃, ion source temperature 260℃, scanning mass range m / z: 33~350amu, ionization mode: EI+; electron energy: 70eV, each sample is measured 3 times in parallel, corresponding data are counted, and fingerprints of volatile flavor components of mellow-type liquor base wines of different quality grades are obtained; Step (2): Step 201, using the fingerprint data of the volatile flavor components obtained in step (1) as a test sample; Step 202: Establish a K-nearest neighbor model, and train and optimize the K-nearest neighbor model using the test samples obtained in step (1) according to the ten-fold cross-validation method to obtain a preliminary discriminant model and the accuracy A1 of the preliminary discriminant model, where A1 is 0.789474 and the ACU value is 0.5 to 1.0; Step (3): Step 301, calculating the Gini index of the test samples obtained in step (1) based on the CART algorithm in the decision tree, and obtaining the contribution of each volatile flavor substance to the quality grade of the mellow-flavored liquor base wine according to the Gini index, sorting them in descending order, and screening out characteristic substances; Step 302: respectively bring the top 30 characteristic substances ranked in step 301 back to the preliminary discrimination model of step (2), and obtain the accuracy of the preliminary discrimination model An=0.968596, which is used as the prediction model for predicting the quality grade of the mellow-flavored liquor base wine; Step (4): 10 premium liquor samples and 10 general liquor samples are input as test samples into the prediction model, and the prediction model obtained in step (3) is used to predict the quality grade of the mellow-flavored liquor base liquor, with an accuracy rate of 95%; The quality grades are divided into two grades: superior wine and general wine.

2. The method according to claim 1, characterized in that The top 30 characteristic substances include ethyl lactate, isopentyl lactate, benzaldehyde, phenylethyl acetate, diethyl succinate, ethyl 3-phenylpropionate, hexyl lactate, n-hexanal, 4-ethyl-2-methoxyphenol, ethyl phenylacetate, phenylethyl butyrate, ethyl acetate, caproic anhydride, amyl acetate, (2,2-diethoxyethyl)-benzene, butyl phenylacetate, diethyl azelaic acid, ethyl 4-phenylbutyrate, n-octanedioic anhydride, 2-phenylethyl hexanoate, furfuryl hexanoate, isopentyl acetate, isobutanol, 2-pentadecanone, geranyl acetone, n-octanol, ethyl 9-tetradecenoate, butyl acetate, ethyl benzoate, and 2-undecenal.