White spirit quality evaluation and key flavor feature recognition method and system

Abstract

The invention provides a white spirit quality evaluation and key flavor feature recognition method and system, relates to the field of white spirit quality management, and obtains a white spirit quality evaluation model capable of predicting quality evaluation data through flavor feature data of a white spirit sample by establishing and training a neural network model. When the Baijiu quality evaluation model is used for prediction, the SHAP value of data of each flavor feature in a Baijiu sample to be detected is calculated based on the SHAP theory, the contribution degree of each flavor feature is evaluated according to the SHAP value obtained through calculation, and therefore the influence of each flavor feature on quality evaluation is obtained. Therefore, the flavor characteristics are selected as the key flavor characteristics of the to-be-detected white spirit sample according to the contribution degree, a scientific, objective and efficient guidance basis is provided for white spirit production and quality control, the problem that the influence of flavor compounds on quality evaluation is not clear in the prior art is solved, and the method is suitable for white spirit quality management.

Description

Technical Field

[0001] The present invention relates to the field of liquor quality management, and particularly to methods and systems for liquor quality evaluation and identification of key flavor characteristics. Background Art

[0002] Liquors have diverse flavor types and complex brewing processes. With the rapid development of modern analytical techniques and artificial intelligence, more and more research and production practices have begun to focus on the quantitative determination of flavor compounds in liquors and the correlation analysis between them and quality. By means of analytical methods such as chromatography and mass spectrometry, a series of key volatile or non-volatile components, including ethyl lactate, ethyl hexanoate, etc., can be objectively and accurately detected.

[0003] Generally, the evaluation of liquor quality mainly relies on senior liquor-tasting experts or professional liquor-tasters with rich experience to conduct sensory evaluations on taste, aroma, color, and style characteristics. However, although the sensory evaluation method has a certain degree of professionalism and authority, it also has defects such as inevitable subjective factors, limited expert resources, and difficulty in quantification and replication.

[0004] Therefore, in the prior art, the influence of flavor compounds on quality evaluation is not clear, and it is difficult to provide targeted guidance for the optimization of liquor quality. Summary of the Invention

[0005] The technical problem to be solved by the present invention: Provide a method and system for liquor quality evaluation and identification of key flavor characteristics, and solve the problem in the prior art that the influence of flavor compounds on quality evaluation is not clear.

[0006] The technical solution adopted by the present invention to solve the above technical problem: A method for liquor quality evaluation and identification of key flavor characteristics, comprising the following steps:

[0007] S1. Obtain the flavor characteristic data and corresponding quality evaluation data of liquor samples, construct a data set, and divide the data set into a training set and a validation set;

[0008] S2. Establish a neural network model, use the flavor characteristic data in the training set as input, use the quality evaluation data in the training set as output, train the neural network model, and use the validation set to verify the trained neural network model to obtain a liquor quality evaluation model;

[0009] S3. Use the liquor quality evaluation model to predict the flavor characteristic data of the liquor sample to be tested, obtain the corresponding quality evaluation data, and calculate the SHAP value of each flavor characteristic data in the liquor sample to be tested based on the SHAP theory;

[0010] S4. Based on the calculated SHAP values, evaluate the contribution degree of each flavor characteristic, and select flavor characteristics as the key flavor characteristics of the to-be-detected Baijiu samples according to the contribution degree.

[0011] Further, the flavor characteristic data includes data of multiple flavor compounds such as ethyl lactate, ethyl hexanoate, isoamyl alcohol, n-butanol, isobutanol, n-propanol, ethyl butyrate, methanol, acetal, and ethyl acetate, and the data is collected in the form of concentration.

[0012] Further, the quality evaluation data is obtained by averaging the scores given by experts or wine tasters to the Baijiu samples.

[0013] Further, the quality evaluation data is obtained by experts or wine tasters scoring the Baijiu samples, and the scoring results are weighted according to the qualifications of the experts or wine tasters.

[0014] Further, the neural network model includes one or a combination of more than one of MLP, LSTM, RNN, CNN, GRU, and Transformer.

[0015] Further, the Baijiu quality evaluation model uses the SmoothL1 loss function, and the expression of the SmoothL1 loss function is: where x is the predicted value of the Baijiu quality evaluation model, y is the true value of the quality evaluation data, β is the smoothing threshold, and SmoothL1(x, y) is the loss function value.

[0016] Further, the calculation formula of the SHAP value is: where j φ is the SHAP value of the j-th flavor characteristic data; N is the set of all flavor characteristic data; N\{j} is all subsets excluding the j-th flavor characteristic data, S is a subset in N\{j}; v(S) is the output value of the Baijiu quality evaluation model when only subset S is included, v(S∪{j}) - v(S) represents the change in the output value of the Baijiu quality evaluation model after adding the j-th flavor characteristic data; |S| is the size of subset S; |N| is the total number of flavor characteristic data.

[0017] Further, before training, it also includes preprocessing the flavor characteristic data and the corresponding quality evaluation data, and the preprocessing includes missing value processing, outlier processing, and normalization processing.

[0018] Further, after obtaining the normalized quality evaluation data using the Baijiu quality evaluation model, the inverse process corresponding to the normalization is adopted to restore the quality evaluation data to the true order of magnitude.

[0019] The present invention also provides a white liquor quality evaluation and key flavor characteristic identification system for implementing the white liquor quality evaluation and key flavor characteristic identification method as described above. The system includes a data acquisition module, a model training module, a model verification module, a model prediction module, a flavor characteristic data contribution module, and a key flavor characteristic identification module. The data acquisition module is used to acquire the flavor characteristic data and corresponding quality evaluation data of white liquor samples, construct a data set, and divide the data set into a training set and a verification set. The model training module is used to establish a neural network model, use the flavor characteristic data in the training set as input, and use the quality evaluation data in the training set as output to train the neural network model. The model verification module is used to verify the trained neural network model using the verification set to obtain a white liquor quality evaluation model. The model prediction module is used to predict the flavor characteristic data of the white liquor sample to be tested using the white liquor quality evaluation model to obtain the corresponding quality evaluation data. The flavor characteristic data contribution module is used to calculate the SHAP value of each flavor characteristic data in the white liquor sample to be tested based on the SHAP theory, and evaluate the contribution degree of each flavor characteristic according to the calculated SHAP value. The key flavor characteristic identification module is used to select flavor characteristics as the key flavor characteristics of the white liquor sample to be tested based on the contribution degree.

[0020] Advantages of the present invention: The present invention provides a white liquor quality evaluation and key flavor characteristic identification method and system. By acquiring the flavor characteristic data and corresponding quality evaluation data of white liquor samples, constructing a data set, dividing the data set into a training set and a verification set, establishing a neural network model, using the flavor characteristic data in the training set as input, using the quality evaluation data in the training set as output to train the neural network model, verifying the trained neural network model using the verification set to obtain a white liquor quality evaluation model, predicting the flavor characteristic data of the white liquor sample to be tested using the white liquor quality evaluation model to obtain the corresponding quality evaluation data, calculating the SHAP value of each flavor characteristic data in the white liquor sample to be tested based on the SHAP theory, evaluating the contribution degree of each flavor characteristic according to the calculated SHAP value, and selecting flavor characteristics as the key flavor characteristics of the white liquor sample to be tested based on the contribution degree, the problem in the prior art that the influence of flavor compounds on quality evaluation is not clear is solved. Description of the Drawings

[0021] Figure 1 is a schematic flowchart of a white liquor quality evaluation and key flavor characteristic identification method provided by the present invention;

[0022] Figure 2 is a schematic diagram of the quality evaluation result in a white liquor quality evaluation and key flavor characteristic identification method provided by the present invention;

[0023] Figure 3It is a descending order diagram of SHAP values of 10 flavor characteristics in a method for evaluating the quality of Chinese liquor and identifying key flavor characteristics provided by the present invention. Detailed implementation manners

[0024] Aiming at the problem in the prior art that the influence of flavor compounds on quality evaluation is not clear, the present invention provides a method and system for evaluating the quality of Chinese liquor and identifying key flavor characteristics. By establishing and training a neural network model, a Chinese liquor quality evaluation model capable of predicting quality evaluation data through flavor characteristic data of Chinese liquor samples is obtained. When using this Chinese liquor quality evaluation model for prediction, based on the SHAP theory, the SHAP value of each flavor characteristic data in the Chinese liquor sample to be measured is calculated, and according to the calculated SHAP value, the contribution degree of each flavor characteristic is evaluated. In this way, the influence of each flavor characteristic on quality evaluation is obtained, and thus flavor characteristics are selected as the key flavor characteristics of the Chinese liquor sample to be measured according to the contribution degree, providing a scientific, objective and efficient guiding basis for Chinese liquor production and quality control.

[0025] As Figure 1 shown, a method for evaluating the quality of Chinese liquor and identifying key flavor characteristics provided by the present invention includes the following steps:

[0026] S1. Obtain the flavor characteristic data and corresponding quality evaluation data of Chinese liquor samples, construct a data set, and divide the data set into a training set and a validation set.

[0027] Specifically, the flavor characteristic data includes data of various flavor compounds such as ethyl lactate, ethyl caproate, isoamyl alcohol, n-butanol, isobutanol, n-propanol, ethyl butyrate, methanol, acetal and ethyl acetate, and the data is collected in the form of concentration. The quality evaluation data is obtained by taking the average value after experts or wine tasters score the Chinese liquor samples. For example, 12 experts are used to score the Chinese liquor samples, using a 10-point system, that is, 10 points is considered the best and 0 points is the worst, and the average value of the scores of 12 experts is taken as the quality evaluation data. In addition, the quality evaluation data can also be obtained by experts or wine tasters scoring the Chinese liquor samples and weighting the scoring results according to the qualifications of the experts or wine tasters.

[0028] S2. Establish a neural network model, use the flavor characteristic data in the training set as the input and the quality evaluation data in the training set as the output, train the neural network model, and use the validation set to verify the trained neural network model to obtain a Chinese liquor quality evaluation model.

[0029] Specifically, the neural network model includes one or a combination of MLP, LSTM, RNN, CNN, GRU and Transformer. The Chinese liquor quality evaluation model uses the SmoothL1 loss function, and the expression of the SmoothL1 loss function is: Among them, x is the predicted value of the Baijiu quality evaluation model, y is the true value of the quality evaluation data, β is the smoothing threshold, and SmoothL1(x, y) is the loss function value. Before training, it also includes preprocessing the flavor feature data and the corresponding quality evaluation data. The preprocessing includes missing value processing, outlier processing, and normalization processing. After obtaining the normalized quality evaluation data using the Baijiu quality evaluation model, the inverse processing corresponding to the normalization is adopted to restore the quality evaluation data to the true order of magnitude. The schematic diagram of the quality evaluation result is as Figure 2 shown, where the coefficient of determination R 2 between the predicted value and the true value of the test set is 0.81, which is greater than 0.8, indicating that the model effect is good.

[0030] S3. Use the Baijiu quality evaluation model to predict the flavor feature data of the Baijiu sample to be tested, obtain the corresponding quality evaluation data, and calculate the SHAP value of each flavor feature data in the Baijiu sample to be tested based on the SHAP theory.

[0031] Specifically, the calculation formula of the SHAP value is: where, φ j is the SHAP value of the j-th flavor feature data; N is the set of all flavor feature data; N\{j} is the set of all subsets that do not include the j-th flavor feature data, S is a subset in N\{j}; v(S) is the output value of the Baijiu quality evaluation model when only the subset S is included, and v(S∪{j}) - v(S) represents the change in the output value of the Baijiu quality evaluation model brought about by adding the j-th flavor feature data; |S| is the size of the subset S; |N| is the total number of flavor feature data.

[0032] S4. According to the calculated SHAP values, evaluate the contribution degree of each flavor feature, and select the flavor features as the key flavor features of the Baijiu sample to be tested based on the contribution degree.

[0033] Specifically, the higher the SHAP value, the higher the contribution degree of the flavor feature. For example, taking 10 flavor features included in a certain Baijiu as an example, the calculated SHAP values corresponding to the 10 flavor features are arranged in descending order, as Figure 3 shown. Since the higher the SHAP value, the higher the contribution degree of the flavor feature, it is obtained that flavor feature 10, flavor feature 4, flavor feature 9, flavor feature 8, and flavor feature 2 are the top five flavor features in terms of contribution degree, and they are considered relatively important and used as the key flavor features of this Baijiu.

[0034] Specifically, by comparing the contribution degrees of the same flavor characteristics in different baijiu samples, the positive and negative effects of each flavor characteristic on quality evaluation are obtained. For example, flavor characteristic 10 and flavor characteristic 2 basically conform to the rule that the larger the characteristic value, the larger the SHAP value, which has a positive effect; flavor characteristic 4, flavor characteristic 9, and flavor characteristic 8 basically conform to the rule that the larger the characteristic value, the smaller the SHAP value, which has a negative effect. Therefore, it can provide a scientific, objective, and efficient guiding basis for baijiu production and quality control.

[0035] The present invention also provides a baijiu quality evaluation and key flavor characteristic identification system to implement the above-mentioned baijiu quality evaluation and key flavor characteristic identification method. The system includes a data acquisition module, a model training module, a model verification module, a model prediction module, a flavor characteristic data contribution module, and a key flavor characteristic identification module. The data acquisition module is used to acquire the flavor characteristic data and corresponding quality evaluation data of baijiu samples, construct a data set, and divide the data set into a training set and a verification set. The model training module is used to establish a neural network model, use the flavor characteristic data in the training set as input, and use the quality evaluation data in the training set as output to train the neural network model. The model verification module is used to verify the trained neural network model using the verification set to obtain a baijiu quality evaluation model. The model prediction module is used to use the baijiu quality evaluation model to predict the flavor characteristic data of the baijiu sample to be tested and obtain the corresponding quality evaluation data. The flavor characteristic data contribution module is used to calculate the SHAP value of each flavor characteristic data in the baijiu sample to be tested based on the SHAP theory, and evaluate the contribution degree of each flavor characteristic according to the calculated SHAP value. The key flavor characteristic identification module is used to select flavor characteristics as the key flavor characteristics of the baijiu sample to be tested based on the contribution degree.

Claims

1. A method for evaluating liquor quality and identifying key flavor characteristics, characterized in that: The following steps are involved: S1. Obtain flavor characteristic data and corresponding quality evaluation data of liquor samples, construct a data set, and divide the data set into a training set and a validation set; S2, establishing a neural network model, taking the flavor characteristic data in the training set as input, taking the quality evaluation data in the training set as output, training the neural network model, and using the validation set to validate the trained neural network model, to obtain a liquor quality evaluation model; S3, using the liquor quality evaluation model to predict the flavor characteristic data of the liquor sample to be tested, to obtain corresponding quality evaluation data, and calculating the SHAP value of each flavor characteristic data in the liquor sample to be tested based on the SHAP theory; S4. According to the calculated SHAP value, the contribution of each flavor characteristic is evaluated, and the flavor characteristic is selected as the key flavor characteristic of the liquor sample to be tested based on the contribution.

2. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 1, characterized in that: The flavor characteristic data include data of various flavor compounds in ethyl lactate, ethyl hexanoate, isopentanol, n-butanol, isobutanol, n-propanol, ethyl butyrate, methanol, acetal and ethyl acetate, and the data are collected in the form of concentration.

3. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 1, characterized in that: The quality evaluation data is obtained by averaging the scores given to the liquor samples by experts or wine appraisers.

4. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 1, characterized in that: The quality evaluation data is obtained by having experts or wine appraisers score the liquor samples and weighting the scoring results according to the qualifications of the experts or wine appraisers.

5. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 1, characterized in that: The neural network model includes one or more combinations of MLP, LSTM, RNN, CNN, GRU and Transformer.

6. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 1, characterized in that: The liquor quality evaluation model uses the SmoothL1 loss function, and the expression of the SmoothL1 loss function is: Among them, x is the predicted value of the liquor quality evaluation model, y is the true value of the quality evaluation data, β is the smoothing threshold, and SmoothL1(x,y) is the loss function value.

7. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 1, characterized in that: The calculation formula of SHAP value is: Among them, φ j is the SHAP value of the j-th flavor feature data; N is the set of all flavor feature data; N\{j} is all subsets that do not contain the j-th flavor feature data, S is a subset in N\{j}; v(S) is the output value of the liquor quality evaluation model when it only contains subset S, v(S∪{j})-v(S) represents the change in the output value of the liquor quality evaluation model after adding the j-th flavor feature data; |S| is the size of subset S; |N| is the total number of flavor feature data.

8. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 1, characterized in that: Before training, the flavor characteristic data and the corresponding quality evaluation data are preprocessed, and the preprocessing includes missing value processing, outlier processing and normalization processing.

9. The method for evaluating liquor quality and identifying key flavor characteristics according to claim 8, characterized in that: After using the liquor quality evaluation model to obtain normalized quality evaluation data, the inverse processing corresponding to the normalization is used to restore the quality evaluation data to the true order of magnitude.

10. Liquor quality evaluation and key flavor characteristics recognition system, characterized in that: The method for evaluating liquor quality and identifying key flavor characteristics as claimed in claim 1 is implemented, wherein the system comprises a data acquisition module, a model training module, a model verification module, a model prediction module, a flavor characteristic data contribution module and a key flavor characteristic identification module; the data acquisition module is used to acquire the flavor characteristic data and the corresponding quality evaluation data of the liquor sample, construct a data set, and divide the data set into a training set and a validation set; the model training module is used to establish a neural network model, take the flavor characteristic data in the training set as input, take the quality evaluation data in the training set as output, and train the neural network model; the model verification module is used to verify the trained neural network model using the validation set to obtain the liquor quality evaluation model; the model prediction module is used to predict the flavor characteristic data of the liquor sample to be tested using the liquor quality evaluation model to obtain the corresponding quality evaluation data; the flavor characteristic data contribution module is used to calculate the SHAP value of each flavor characteristic data in the liquor sample to be tested based on the SHAP theory, and evaluate the contribution of each flavor characteristic according to the calculated SHAP value; the key flavor characteristic identification module is used to select the flavor characteristic as the key flavor characteristic of the liquor sample to be tested according to the contribution.

Images