Prediction method for storage year of white spirit

Abstract

The invention discloses a baijiu storage year prediction method, and relates to the technical field of baijiu year analysis, the method comprises the following steps: sample data acquisition and pretreatment: extracting flavor substance information in a baijiu sample through GC-O-MS and GC-FID technologies; key flavor substances are screened, AEDA and OAV are adopted for analysis, and flavor substances with significant differences in baijiu of different years are recognized and screened out; feature data extraction and optimization: key flavor substance features related to storage years are determined through K-MEANS clustering and OPLS-DA analysis; prediction model construction and training: based on the screened key flavor substance data, adopting a machine learning algorithm to establish and optimize a prediction model of the storage years of the Baijiu; and year prediction and output: inputting the flavor substance data of the white spirit sample to be predicted into the trained prediction model, and outputting and storing a year prediction value. Therefore, the problems that in the prior art, storage year labeling lacks objective basis, the prediction technology is limited greatly, and it is difficult to accurately predict the storage year of Baijiu are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of baijiu vintage analysis, and more specifically, to a method for predicting the storage vintage of baijiu. Background Art

[0002] As a traditional Chinese distilled liquor, the quality of baijiu is closely related to its storage vintage. During the storage process, a series of complex chemical reactions such as esterification, oxidation, and polymerization occur inside the baijiu, which promote the change of flavor substances and thus improve the quality. However, the current situation of labeling the storage vintage of baijiu in the market is not optimistic. It mainly relies on the subjective records of enterprises and seriously lacks objective and scientific basis. This situation has led to many doubts from consumers about the product quality, making it difficult to judge the true quality of baijiu and is also not conducive to the healthy and orderly development of the baijiu market.

[0003] The existing vintage prediction technologies have obvious shortcomings:

[0004] On the one hand, for the method relying on a single chemical index, due to the complex composition of baijiu, relying solely on a certain chemical index simply cannot comprehensively reflect the complex changes of baijiu during storage, resulting in a large deviation in the prediction results.

[0005] On the other hand, although the sensory evaluation method can make judgments from aspects such as flavor, it is greatly affected by the subjective factors of the evaluators. The evaluations of the same baijiu by different evaluators may vary greatly, making it difficult to ensure the accuracy and reliability of the prediction.

[0006] Therefore, to meet the urgent market demand for accurately judging the storage vintage of baijiu and eliminate consumers' doubts, there is an urgent need for a scientific and effective prediction method that can overcome the defects of the existing technologies and achieve accurate and objective prediction of the storage vintage of baijiu. In view of this, a method for predicting the storage vintage of baijiu is proposed. Summary of the Invention

[0007] The purpose of the present invention is to provide a method for predicting the storage vintage of baijiu to solve the technical problems that the labeling of the storage vintage of baijiu in the market lacks objective basis and the existing prediction technologies have great limitations and are difficult to accurately and objectively predict the storage vintage of baijiu.

[0008] To solve the above technical problems, the present invention provides a method for predicting the storage vintage of baijiu, as Figure 1As shown, it includes the following steps: S1. Sample data collection and preprocessing: Obtain a white liquor sample, and extract the flavor substance information in the white liquor sample through gas chromatography-olfactometry-mass spectrometry, gas chromatography-flame ionization detection technology combined with liquid-liquid extraction and headspace solid-phase microextraction methods; S2. Screening of key flavor substances: Conduct odor activity evaluation and odor activity value analysis on the flavor substance information, and identify and screen flavor substances that have a significant impact on the senses and show significant differences in white liquors of different years; S3. Feature data extraction and optimization: Based on K-MEANS clustering and orthogonal partial least squares discriminant analysis, determine the key features related to the storage year from the flavor substances; S4. Prediction model construction and training: Based on the key feature data, establish and optimize a prediction model for the storage year of white liquor using machine learning algorithms; S5. Year prediction and output: Input the flavor substance data of the white liquor sample to be predicted into the prediction model to obtain the predicted storage year value.

[0009] Preferably, in the liquid-liquid extraction in step S1, it includes: Dilute the white liquor sample to 15% ethanol, add cinnamyl acetate as an internal standard to a final concentration of 40 mg / L, saturate it with NaCl, and extract it 3 times with freshly distilled dichloromethane, with the oscillation frequency of 300 r / min and the extraction time of 5 min each time; After combining the extracts, extract them 3 times with Na2CO3 solution, and collect the lower layer solution to obtain sample NBF; Acidify the upper layer solution to pH 2.0 with HCl, and then extract it 3 times with dichloromethane to obtain sample AF; Saturate the sample NBF and the sample AF with NaCl respectively, dry them with anhydrous Na2SO4, concentrate them by nitrogen blowing to 500 μL, and store them at -20 °C.

[0010] Preferably, in the headspace solid-phase microextraction method in step S1, it includes: Take 8 mL of a white liquor sample with a concentration of 15% vol, add 3 g of sodium chloride and cinnamyl acetate with a concentration of 40 mg / L, and -1 equilibrate at 45 °C and 400 r·min

[0011] rotation speed for 5 min, and adsorb the high-volatile components in the sample headspace with an SPME fiber head for 45 min.

[0012] Preferably, the gas chromatography-flame ionization detection conditions in step S1 are as follows: the flow rate of high-purity nitrogen is 40 mL / min, the flow rate of hydrogen is 40 mL / min, the flow rate of air is 500 mL / min, the injection volume is 1 μL, the injection port temperature is 250 °C, the detector temperature is 250 °C, and the split ratio is 40:1; the column oven temperature program is to maintain at 35 °C for 1 min initially, then rise to 180 °C at a rate of 3.5 °C / min, and then rise to 210 °C at a rate of 15 °C / min and hold for 6 min.

[0013] Preferably, the process of odor activity evaluation in step S2 includes: gradually diluting the sample after headspace solid-phase microextraction by setting the gas chromatography split ratio, where the split ratio can be 1:1, 3:1, 9:1, etc., until the odor cannot be perceived at the olfactory detection port, and calculating the flavor dilution value according to the split multiple; gradually diluting the sample NBF or the sample AF with dichloromethane in a volume ratio of 3:1, and performing gas chromatography-olfactometry-mass spectrometry detection in the order of decreasing concentration; screening the flavor substances with the flavor dilution value exceeding the flavor threshold for quantification, and analyzing the odor activity value to screen the substances with the odor activity value exceeding the odor threshold and being significantly different in different years, and obtaining 22 flavor substances that are significantly different in different years.

[0014] Preferably, in step S3, based on K-MEANS clustering and orthogonal partial least squares discriminant analysis, the key features related to the storage year are determined from the flavor substances, including: determining 11 key features related to the storage year among the 22 flavor substances that are significantly different in different years, where the 11 key features related to the storage year include 2-heptanol, ethyl 3-methylbutyrate, methyl hexanoate, furfuryl ethyl ether, ethyl 2-methylpropionate, dimethyl trisulfide, 1-butanol, butyl hexanoate, ethyl octanoate, 1,1-diethoxy-3-methylbutane, and ethyl phenylpropionate.

[0015] Preferably, the machine learning algorithm in step S4 includes support vector machine, random forest or neural network.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0017] 1. By comprehensively applying multi-dimensional information and neural network models, the present invention effectively solves the problems that the current labeling of liquor storage years lacks objective and scientific basis and the existing prediction technologies have limitations. This method no longer simply relies on a single chemical index or sensory evaluation, but combines the image, initial parameters and storage environment information of the liquor body to be stored, inputs them into the neural network model for analysis and prediction, comprehensively and objectively reflects the true situation of the liquor, and obtains a more accurate and reliable prediction result of the storage year, providing a scientific basis for the judgment of the liquor storage year and eliminating consumers' doubts about product quality;

[0018] 2. The present invention further optimizes the prediction accuracy. By obtaining the body liquor images in different year stages to form a basic image sample, denoising, normalizing, etc. are performed on it to generate a more representative sample image for training the neural network model. Moreover, the basic loss function, cross-entropy loss function, and a modified loss function selected according to the body liquor characteristics are used for multiple trainings to continuously optimize the model performance. At the same time, the alcohol content threshold is generated according to different storage years to accurately obtain the alcohol content information, enabling the model to more precisely capture the change characteristics of the white liquor during storage, further improving the prediction accuracy, and solving the problem of more accurately predicting the storage year of white liquor under complex components and changing conditions.

[0019] 3. In addition to accurately predicting the year, the present invention also considers the subsequent links of white liquor storage. According to the predicted storage year, initial parameters, and storage environment information, appropriate storage conditions are determined, and corresponding storage plans are generated. At the same time, in-depth analysis is carried out on the storage environment parameters and storage plans of the body liquor in different storage years, and the best storage plan for each storage year is obtained through evaluation indicators. This not only solves the problem of how to formulate a scientific storage plan according to the characteristics of white liquor, ensuring the stability or improvement of the quality of white liquor during subsequent storage, but also further improves the entire system of predicting the storage year of white liquor, providing a comprehensive and scientific solution for the white liquor industry from year judgment to storage management.

[0020] The additional aspects and advantages of this application will be partially given in the following description, partially will become obvious from the following description, or be understood through the practice of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above-mentioned and / or additional aspects and advantages of this application will become obvious and easy to understand from the following description of the embodiments in conjunction with the drawings, where:

[0022] Figure 1 is a flowchart of a method for predicting the storage year of white liquor provided by the present invention;

[0023] Figure 2 is a flowchart of a method for predicting the storage year of white liquor provided by an embodiment of the present invention;

[0024] Figure 3 is a schematic diagram of multi-dimensional feature analysis visualization provided by the present invention;

[0025] Figure 4 is a schematic diagram of a heat map of the concentration of key flavor substances in white liquor of different storage years provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] Embodiments of the present application will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where like or similar reference numerals denote like or similar elements or elements having like or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and should not be construed as limiting the present application.

[0027] A method for predicting the storage year of liquor according to an embodiment of the present application will be described below with reference to the accompanying drawings. In view of the problem in the above-mentioned background art that it is difficult to accurately and objectively predict the storage year of liquor, the present application provides a method for predicting the storage year of liquor. In this method, by adopting advanced technologies such as gas chromatography-olfaction-mass spectrometry and gas chromatography-flame ionization detection, the flavor substance information in the liquor sample is comprehensively and accurately extracted, and then the key flavor substances that have a significant impact on the senses and have significant differences in liquors of different years are identified. Based on K-MEANS clustering and orthogonal partial least squares discriminant analysis, the characteristic data closely related to the storage year is optimized, and a prediction model for the storage year of liquor based on a machine learning algorithm is established. This model can accurately predict the storage year of liquor, thereby improving the transparency of the liquor market, enhancing consumers' trust, and promoting the healthy development of the liquor industry. At the same time, this method also provides a more accurate reference for liquor tasting, promotes the progress of liquor detection technology, and better meets consumers' personalized and diversified needs for liquor. Thus, the technical problem that the labeling of the storage year of liquor in the market lacks an objective basis and the existing prediction technologies have great limitations and it is difficult to accurately and objectively predict the storage year of liquor is solved.

[0028] A method for predicting the storage year of liquor according to an embodiment of the present application will be described below with reference to the accompanying drawings.

[0029] As Figure 2 shown, a method for predicting the storage year of liquor according to the present invention includes the following steps:

[0030] S1. Input the image of the liquor body to be stored for a certain year into a neural network model, obtain its predicted year value and output it;

[0031] S101. Obtain a basic image sample, which is composed of liquor body information;

[0032] S102. Perform denoising processing and normalization processing on the basic image sample to generate a sample image;

[0033] In an embodiment of the present invention, step S102 performs denoising processing and normalization processing on the basic image sample to generate a sample image, which is carried out in the following manner: Take a part of the pictures of the basic image sample in proportion and shuffle the order, use Gaussian filtering to denoise the basic image sample and convert it into a grayscale image, perform standardization processing on the basic image sample, and use data augmentation methods such as random cropping and random flipping to perform normalization processing on the data to obtain the sample image;

[0034] Let the basic image sample be I base ={I base1 ,I base2 ,...,I baseN}, where N is the number of basic image samples. The image after Gaussian filtering denoising is I denoise ={I denoise1 ,I denoise2 ,...,I denoiseN}, the image after conversion to a grayscale image is I gray ={I gray1 ,I gray2 ,...,I grayN}, the image after standardization processing is I std ={I std1 ,I std2 ,...,I stdN}, and its calculation formula is:

[0035]

[0036] In the formula, μ is the mean of I gray , and σ is the standard deviation of I gray ;

[0037] S103. Establish a neural network model;

[0038] In an embodiment of the present invention, in step S103, establishing a neural network model specifically includes: Inputting the sample image into the neural network model, performing feature extraction on the sample image through multiple convolutional layers, and performing one-dimensional processing on the features obtained by the last convolutional layer and then inputting them into the regression output layer. The output of the regression output layer of the neural network model is:

[0039] O={o1,o2,...,o i ,...,o k};

[0040] In the formula, k is the regression input size of the neural network model, O is the regression output value of the neural network model, o i is the regression output value of the neural network model, and i is the serial number;

[0041] By performing one-dimensional processing on the features obtained from the convolutional layer, a regression feature channel is formed, and the regression output size m of the neural network model is generated according to O and the number n of wine body images;

[0042] S104. Input and output the training neural network model with the sample images;

[0043] In an embodiment of the present invention, the sample image I sample is used as the input, and the corresponding storage year label Y = {y1, y2,..., y n} is used as the output to train the neural network model;

[0044] S105. Train the neural network model using the basic loss function and the cross-entropy loss function;

[0045] In an embodiment of the present invention, in step S105, training the neural network model using the basic loss function and the cross-entropy loss function specifically includes the following steps:

[0046] S105a. Define the basic loss function and the cross-entropy loss function, obtain the basic loss value and the cross-entropy loss function value, define the weights of the basic loss value and the cross-entropy loss function value, and obtain the basic total loss value;

[0047] In an embodiment of the present invention, the basic total loss value is obtained in the following manner in step S105a. Let the basic loss function be L base , and the cross-entropy loss function be L ce . The weights of the basic loss value and the cross-entropy loss function value are α and β respectively, and α + β = 1. Then the basic total loss value L total is:

[0048] L total = αL base + βL ce ;

[0049] S105b. Train the neural network model using the basic loss value, the cross-entropy loss function value, and the basic total loss value;

[0050] Among them, through the backpropagation algorithm, the weight parameters and bias parameters of the neural network model are updated according to the basic total loss value L total ;

[0051] S106. Input the image of the wine body to be stored in the year into the neural network model, select the top three values of the output vector of the neural network model in descending order, obtain the storage year through mapping, and output it;

[0052] In an embodiment of the present invention, step S106 specifically includes: Let the output vector of the neural network model be O out={o out1 ,o out2 ,...,o outk}, sort it in descending order to obtain O sorted ={o orted1 ,o orted2 ,...,o ortedk}, select the first three values o orted1 ,o orted2 ,o orted3 , and obtain the corresponding storage years y pred1 ,y pred2 ,y pred3 through the predefined mapping relationship M and output them;

[0053] S107. According to the information of the wine body for the storage year, select a correction loss function that matches the wine body;

[0054] In the embodiment of the present invention, in step S107, according to the information of the wine body for the storage year, select a correction loss function that matches the wine body. Specifically, it is carried out in the following way. According to the color and alcohol content information of the wine body for the storage year, select a suitable correction loss function from the predefined correction loss function set L mod ={L mod1 ,L mod2 ,...,L modp};

[0055] S108. Obtain a corrected sample image, where the corrected sample image is a sample image obtained by blurring the basic image sample;

[0056] Blur the basic image sample I base to obtain the corrected sample image I mod1 ={I mod11 ,I mod12 ,...,I mod1N};

[0057] S109. Input and output the corrected sample image to the training neural network model;

[0058] Use the corrected sample image I mod1 as the input and the corresponding storage year label Y as the output to train the neural network model;

[0059] S110. Train the neural network model using the correction loss function and the cross-entropy loss function, and input the image I test of the wine body for the storage year into the trained neural network model to obtain the year prediction value y pred of the wine body for the storage year and output it;

[0060] In an embodiment of the present invention, in step S110, a neural network model is trained using a modified loss function and a cross-entropy loss function, and specifically, the following method is adopted: Let the modified loss value be Then the total modified loss value L at this time total1 is:

[0061]

[0062] In the formula, γ and δ are weights, and γ + δ = 1;

[0063] The present invention further optimizes the prediction accuracy. By obtaining the body liquor images in different year stages to form a basic image sample, performing denoising, normalization, etc. on it, generating a more representative sample image for training the neural network model, and using the basic loss function, cross-entropy loss function, and a modified loss function selected according to the body liquor characteristics for multiple trainings to continuously optimize the model performance. At the same time, an alcohol content threshold is generated according to different storage years to accurately obtain the alcohol content information, enabling the model to more finely capture the change characteristics of the white liquor during storage, further improving the prediction accuracy, and solving the problem of more accurately predicting the storage year of white liquor under complex components and changing conditions.

[0064] S2. Obtain the initial parameters and storage environment information of the body liquor to be stored for the year;

[0065] In an embodiment of the present invention, the initial parameter P init includes the color information P of the body liquor to be stored for the year color , alcohol information P alcohol , blending ratio information P blend , white liquor grade information P grade and body liquor information P body ;

[0066] The body liquor information includes the body liquor images at different time periods. According to the number n of the body liquor images, the regression input size k and the regression output size m of the neural network model are determined;

[0067] The storage environment information includes storage environment temperature information, humidity information, and storage condition information;

[0068] S3. Determine the storage year of the body liquor to be stored for the year according to the year prediction value and the initial parameters;

[0069] In an embodiment of the present invention, in S3, determining the storage year of the body liquor to be stored for the year according to the year prediction value and the initial parameters specifically includes the following steps:

[0070] S301. Input the year prediction value, the initial parameters, and the storage environment temperature information into the neural network model, and obtain the storage year through the neural network model;

[0071] Input the predicted value y of the year pred , the initial parameter P init = {P color , P alcohol , P blend , P grade , P body} and the storage environment temperature information T env into the neural network model to obtain the storage year y final ;

[0072] S302. Measure the storage year wine body environment temperature T through the temperature sensor S temo ; measure ;

[0073] S303. Obtain the alcohol content information of the wine body to be stored;

[0074] S303a. Generate the alcohol content threshold of the storage year wine body image I i according to different storage years y test ;

[0075] S303b. Obtain the image I test in the storage year wine body image I that exceeds the alcohol content threshold ; alcohol_above ;

[0076] S303c. Input the image I alcohol_above into the neural network model, and obtain the alcohol content information P test of the storage year wine body image I through the neural network model; alcohol_infol ;

[0077] S304. Input the storage year wine body environment temperature T measure , the alcohol content information P alcohol_infol , and the initial parameter P init obtained in steps S302 and S303 into the neural network model to obtain the storage year y final of the wine body to be stored;

[0078] By comprehensively applying multi-dimensional information and neural network models, the present invention effectively solves the problems that the current labeling of the storage year of Baijiu lacks objective and scientific basis and the existing prediction technologies have limitations. This method no longer simply relies on a single chemical index or sensory evaluation, but combines the image, initial parameters, and storage environment information of the wine body to be stored, inputs them into the neural network model for analysis and prediction, comprehensively and objectively reflects the true situation of Baijiu, and obtains a more accurate and reliable storage year prediction result, providing a scientific basis for the judgment of the storage year of Baijiu and eliminating consumers' doubts about product quality.

[0079] S4. Determine the storage conditions for the wine body of the year to be stored according to the initial parameters and storage environment information;

[0080] In an embodiment of the present invention, in S4, determining the storage conditions for the wine body of the year to be stored according to the initial parameters and storage environment information specifically includes the following steps:

[0081] S401. According to the storage year y final and the initial parameter P init , determine the storage environment information E env ={T env_new , H env , C env}; formulate a storage plan S plan ={T env_new , H env , t store , l store}, where T env_new is the storage environment temperature, H env is the storage environment humidity, C env is the storage condition, t store is the storage time, and l store is the storage location;

[0082] S402. Modify the weight parameters of the neural network model according to the storage environment temperature information T env_new ;

[0083] In an embodiment of the present invention, in step S402, modifying the weight parameters of the neural network model according to the storage environment temperature information T env_new , the modification formula is:

[0084] W env =W old +η(T env_new -T env_old )W old ;

[0085] In the formula, W old is the weight parameter before modification, W env is the weight parameter after modification, η is the modification coefficient, and T env_old is the storage environment temperature before modification;

[0086] S5. Generate a storage plan for the wine body of the year to be stored according to the storage year and storage conditions;

[0087] In an embodiment of the present invention, in step S5, generating a storage plan for the wine body of the year to be stored according to the storage year and storage conditions specifically includes the following steps:

[0088] S501. Generate a storage plan S based on the storage environment information E env and the storage year y final ; plan

[0089] S502. Analyze the storage environment parameters E of the liquor bodies with different storage years env and the storage plan S plan to obtain the optimal storage plan S for each storage year through evaluation indicators best and output it.

[0090] It should be noted that the concentrations and OAVs of the key flavor substances in the liquor with different storage years are shown in Table 1 below.

[0091] Table 1. Concentrations and OAVs of the key flavor substances in the liquor with different storage years

[0092]

[0093]

[0094] As shown in Table 1 above, a concentration heat map can be obtained based on the concentrations of the substances, as Figure 4 shown, which can quickly compare the contents or activity differences of various chemical substances in different samples, facilitating the exploration of data characteristics and laws.

[0095] In addition to accurately predicting the year, the present invention also takes into account the subsequent links of liquor storage. According to the predicted storage year, initial parameters and storage environment information, appropriate storage conditions are determined and corresponding storage plans are generated. At the same time, in-depth analysis is carried out on the storage environment parameters and storage plans of the liquor bodies with different storage years, and the optimal storage plan for each storage year is obtained through evaluation indicators. This not only solves the problem of how to formulate a scientific storage plan according to the characteristics of liquor, ensures the stability or improvement of the quality of liquor during subsequent storage, but also further improves the entire system of liquor storage year prediction, providing a comprehensive and scientific solution for the liquor industry from year judgment to storage management.

[0096] ​The present application provides a method for predicting the storage year of liquor. In this method, by adopting advanced technologies such as gas chromatography-olfactometry-mass spectrometry and gas chromatography-flame ionization detection, the flavor substance information in the liquor sample is comprehensively and accurately extracted, and then the key flavor substances that have a significant impact on the senses and show significant differences in liquors of different years are identified. Based on K-MEANS clustering and orthogonal partial least squares discriminant analysis, the characteristic data closely related to the storage year is optimized, and a prediction model for the storage year of liquor based on a machine learning algorithm is established. This model can accurately predict the storage year of liquor, thereby improving the transparency of the liquor market, enhancing consumers' trust, and promoting the healthy development of the liquor industry. At the same time, this method also provides a more accurate reference for liquor tasting, promotes the progress of liquor detection technology, and better meets consumers' personalized and diversified needs for liquor. Thus, the technical problem that the labeling of the storage year of liquor in the market lacks an objective basis and the existing prediction technologies have great limitations and it is difficult to accurately and objectively predict the storage year of liquor is solved.

[0097] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A method for predicting the storage years of Chinese liquor, characterized in that, The method includes: S1. Sample data collection and preprocessing: Obtain a white liquor sample, and extract the flavor substance information in the white liquor sample by using gas chromatography-olfactometry-mass spectrometry, gas chromatography-flame ionization detection technology in combination with liquid-liquid extraction and headspace solid-phase microextraction methods; S2. Screening of key flavor substances: Conduct odor activity evaluation and odor activity value analysis on the flavor substance information, and identify and screen flavor substances that have a significant impact on the senses and show significant differences in white liquors of different years; S3. Feature data extraction and optimization: Determine the key features related to the storage year from the flavor substances based on K-MEANS clustering and orthogonal partial least squares discriminant analysis; S4. Prediction model construction and training: Based on the key feature data, establish and optimize a prediction model for the storage year of white liquor by using a machine learning algorithm; S5. Year prediction and output: Input the flavor substance data of the white liquor sample to be predicted into the prediction model to obtain the predicted value of the storage year.

2. The prediction method for the storage years of liquor according to claim 1, characterized in that The liquid-liquid extraction in step S1 includes: Dilute the white liquor sample to 15% ethanol, add cinnamyl acetate as an internal standard to a final concentration of 40 mg / L, saturate with NaCl, and extract 3 times with freshly distilled dichloromethane at an oscillation frequency of 300 r / min and an extraction time of 5 min each time; After combining the extracts, extract 3 times with Na2CO3 solution, and collect the lower layer solution to obtain sample NBF; Acidify the upper layer solution to pH 2.0 with HCl, and then extract 3 times with dichloromethane to obtain sample AF; Saturate the sample NBF and the sample AF with NaCl respectively, dry with anhydrous Na2SO4, concentrate by nitrogen blowing to 500 μL, and store at -20 °C.

3. A method for predicting the storage years of liquor according to claim 1, characterized in that, The headspace solid-phase microextraction method in step S1 includes: Take 8 mL of a white liquor sample with a concentration of 15% vol, add 3 g of sodium chloride and cinnamyl acetate with a concentration of 40 mg / L, balance at 45 °C and a rotation speed of 400 r·min-1 for 5 min, and adsorb the high-volatile components in the sample headspace with an SPME fiber head for 45 min.

4. A method for predicting the storage year of Chinese liquor according to claim 1, characterized in that, The temperature programming of the gas chromatography-olfactometry-mass spectrometry in step S1 is: The initial temperature is maintained at 45 °C for 3 min, and then increased to 230 °C at a rate of 5 °C / min and maintained for 10 min; The mass spectrometry conditions are EI mode, ionization energy 70 eV, full scan mode, and mass scan range 30-350 m / z.

5. A method for predicting the storage year of liquor according to claim 1, characterized in that, The gas chromatography-flame ionization detection conditions in step S1 are: The flow rate of high-purity nitrogen is 40 mL / min, the flow rate of hydrogen is 40 mL / min, the flow rate of air is 500 mL / min, the injection volume is 1 μL, the injection port temperature is 250 °C, the detector temperature is 250 °C, and the split ratio is 40:1; The column oven temperature programming is to maintain at 35 °C for 1 min initially, increase to 180 °C at a rate of 3.5 °C / min, and then increase to 210 °C at a rate of 15 °C / min and maintain for 6 min.

6. The prediction method of the storage year of white liquor according to claim 1, characterized in that, The process of odor activity evaluation in step S2 includes: The samples after headspace solid-phase microextraction are gradually diluted by setting the gas chromatography split ratio. Among them, the split ratio can be 1:1, 3:1, 9:1, etc., until the odor detection port can no longer perceive the odor, and the flavor dilution value is calculated according to the split multiple. The sample NBF or the sample AF is gradually diluted with dichloromethane at a volume ratio of 3:1, and gas chromatography-olfactometry-mass spectrometry detection is carried out in the order from high to low concentration. The flavor substances with flavor dilution values exceeding the flavor threshold are screened for quantification, and substances with odor activity values exceeding the odor threshold and being significantly different in different years are screened through odor activity value analysis, and 22 flavor substances that are significantly different in different years are obtained.

7. A method for predicting the storage years of Chinese liquor according to claim 1 or 6, characterized in that, In step S3, based on K-MEANS clustering and orthogonal partial least squares discriminant analysis, the key features related to the storage year are determined from the flavor substances, including: 11 key features related to the storage year are determined from the 22 flavor substances that are significantly different in different years. Among them, the 11 key features related to the storage year include 2-heptanol, ethyl 3-methylbutyrate, methyl hexanoate, furfuryl ethyl ether, ethyl 2-methylpropionate, dimethyl trisulfide, 1-butanol, butyl hexanoate, ethyl octanoate, 1,1-diethoxy-3-methylbutane, and ethyl phenylpropionate.

8. A method for predicting the storage years of liquor according to claim 1, characterized in that The machine learning algorithms in step S4 include support vector machines, random forests, or neural networks.

Images