Construction method of Maotai-flavor liquor yield prediction model and liquor yield prediction method

Abstract

The invention belongs to the technical field of baijiu yield prediction, and relates to a construction method of a Maotai-flavor baijiu yield prediction model and a baijiu yield prediction method. The construction method comprises the following steps: 1, obtaining modeling data: detecting and recording stacking time, turning nodes, cellar entry top temperature, cellar entry medium temperature, cellar entry bottom temperature, cellar entry product temperature, cellar entry acidity, cellar entry moisture, cellar entry starch, cellar entry reducing sugar and base wine yield of a corresponding saccharification pile in a saccharification pile stacking fermentation stage to obtain the modeling data; and 2, constructing a prediction model: taking the modeling data obtained in the step 1 as modeling features, and constructing an XGBoost prediction model based on an XGBoost regression model. The invention establishes a wine yield prediction model capable of realizing manual regulation and control at the front end of a pit fermentation process, and further assists the improvement of the wine yield of the Maotai wine.

Description

Technical Field

[0001] The invention belongs to the technical field of liquor production prediction, and relates to a method for constructing a liquor production rate prediction model for a sauce-flavor liquor and a method for predicting the liquor production rate. Background Art

[0002] Baijiu is a unique type of liquor in China, and is listed as one of the world's seven major distilled liquors along with brandy, whiskey, vodka, rum, gin, and tequila. China's liquor industry has a long history of thousands of years and is one of China's traditional advantageous industries. Among them, sauce-flavored liquor is an important pillar of China's liquor industry. The development momentum of sauce-flavored liquor is strong, and the efficient improvement of its production and quality is of great significance to the development of the liquor industry.

[0003] The brewing process of Maotai-flavor liquor is complex, and it takes a year, two feedings, nine cookings, eight fermentations, and seven distillations. Each round is divided into two stages: pile fermentation and cellar fermentation. After the raw materials are cooked, spread out, and mixed with koji, they are piled up for pile fermentation. This process can collect, expand, and screen various microorganisms in the environment, enrich rich enzymes and flavor precursor compounds, and help produce ethanol and flavor compounds in the cellar fermentation stage; cellar fermentation is the subsequent fermentation of pile fermentation. Through anaerobic fermentation, the flavor characteristics of Maotai-flavor liquor are finally formed. Pile fermentation and cellar fermentation jointly affect the base wine yield of each round.

[0004] The prediction of wine yield plays an important role in promoting the efficient production of Maotai-flavor liquor. At present, there are technical reports that use the fermentation temperature of fermented mash corresponding to different depths of production pits, different fermentation times at the same depth of pits, and the accumulation time of fermented mash in the drying hall to establish a model to reasonably predict the output of liquor. However, this method mainly focuses on pit fermentation, which is a closed anaerobic fermentation without human intervention. The established model has little guidance for actual production. Therefore, it is necessary to establish a prediction model that can realize human control at the front end of the pit fermentation process to further help improve the yield of Maotai-flavor liquor. Summary of the invention

[0005] The present invention aims to provide a method for constructing a wine yield prediction model for a Maotai-flavor liquor and a method for predicting the wine yield.

[0006] According to a first aspect of the present invention, the present invention provides a method for constructing a Maotai-flavor liquor yield prediction model, comprising the following steps:

[0007] 1. Obtaining modeling data: Detect and record the stacking time, turning nodes, top temperature, middle temperature, bottom temperature, product temperature, acidity, moisture, starch, reducing sugar, and base wine yield of the corresponding saccharification pile during the fermentation stage to obtain modeling data;

[0008] 2. Build a prediction model: Use the modeling data obtained in step 1 as the modeling feature, and build an XGBoost prediction model based on the XGBoost regression model.

[0009] In some embodiments of the present invention, in step 2, the modeling data obtained in step 1 is used as a modeling feature, and an XGBoost prediction model is constructed based on the XGBoost regression model, including the following steps: dividing the modeling data into a training data set and a test data set, using the stacking time, turning nodes, top temperature of the cellar, middle temperature of the cellar, bottom temperature of the cellar, temperature of the cellar, acidity of the cellar, moisture of the cellar, starch of the cellar, and reducing sugar of the cellar in the training data set as independent variables, and using the base wine yield in the training data set as the dependent variable, importing the XGBoost regression model training in R language to construct the XGBoost prediction model.

[0010] In some embodiments of the present invention, in step 2, the parameter settings of the XGBoost regression model include: eta=0.04, gamma=0.03, max_depth=6, subsample=0.75, colsample_bytree=0.6, objective="reg:squarederror", nround=800, watchlist=watchlist, verbose=1, print_every_n=50, early_stopping_rounds=100.

[0011] In some embodiments of the present invention, the prediction method further comprises: evaluating the prediction accuracy of the XGBoost prediction model constructed in step 2 using at least one of the root mean square error, the coefficient of determination, and the mean absolute error as an evaluation index.

[0012] In some embodiments of the present invention, the prediction method further comprises:

[0013] In the step of building a prediction model, the modeling data obtained in step one is ranked according to the importance of independent variable indicators based on the random forest algorithm, and the modeling data is screened based on the importance ranking results. The indicators ranked first 6-8 in importance are used as modeling features to build an XGBoost prediction model.

[0014] In some embodiments of the present invention, the modeling features include: stacking time, turning pile node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, and top temperature in the cellar; or, stacking time, turning pile node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar, and bottom temperature in the cellar; or, stacking time, turning pile node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar, and bottom temperature in the cellar, and reducing sugar in the cellar.

[0015] According to a second aspect of the present invention, the present invention also provides a method for predicting the yield of Maotai-flavor liquor, the method comprising the following steps:

[0016] (1) Obtaining index data information of the saccharification pile fermentation stage;

[0017] (2) Inputting the index data obtained in step (1) into the XGBoost prediction model constructed by the construction method described in certain embodiments of the first aspect of the present invention, and predicting the base wine yield corresponding to the saccharification pile based on the output result of the XGBoost prediction model.

[0018] In some embodiments of the present invention, the indicator data information includes stacking time, turning node, top temperature entering the cellar, middle temperature entering the cellar, bottom temperature entering the cellar, product temperature entering the cellar, acidity entering the cellar, moisture entering the cellar, starch entering the cellar and reducing sugar entering the cellar.

[0019] According to a third aspect of the present invention, the present invention also provides a method for predicting the yield of Maotai-flavor liquor, the method comprising the following steps:

[0020] 1) Obtaining index data information of the saccharification pile fermentation stage;

[0021] 2) Inputting the data information obtained in step 1) into the XGBoost prediction model constructed by the construction method described in certain embodiments of the first aspect of the present invention, and predicting the base wine yield corresponding to the saccharification pile based on the output result of the XGBoost prediction model.

[0022] In some embodiments of the present invention, the indicator data information includes: stacking time, pile turning node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar and top temperature in the cellar data information; or, stacking time, pile turning node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar and bottom temperature in the cellar data information; or, stacking time, pile turning node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar, bottom temperature in the cellar and reducing sugar in the cellar.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] (1) In view of the fact that the wine yield prediction methods or prediction models disclosed in the prior art mainly focus on cellar fermentation, and the problem that cellar fermentation is a closed anaerobic fermentation without human intervention, the established models or prediction methods have little guidance for actual production, the present invention establishes a wine yield prediction model that can be manually controlled at the front end of the cellar fermentation process, further helping to improve the wine yield of sauce-flavored liquor.

[0025] (2) The liquor yield prediction method provided by the present invention takes the process parameters and physicochemical indicators of the mash in the stacking fermentation stage as independent variables, and the corresponding cellar yield as the dependent variable, and uses the XGBoost regression model to construct a liquor yield prediction model. Compared with the existing liquor fermentation yield prediction model established by using the temperature change of the mash during the closed fermentation stage in the cellar, the model provided by the present invention can predict the liquor yield according to the condition of the mash before entering the cellar, and has better operational guiding significance for actual production applications. In addition, the present invention screens the modeling indicators and the modeling model, so that the provided prediction method can realize the accurate prediction of the base liquor yield. DETAILED DESCRIPTION

[0026] The technical solution of the present invention is further described below by specific embodiments, which do not limit the protection scope of the present invention. Some non-essential modifications and adjustments made by others based on the concept of the present invention still fall within the protection scope of the present invention.

[0027] The "stacking time" in the embodiment of the present invention specifically refers to: the stacking fermentation time experienced by the fermented grains before entering the cellar for fermentation in the cellar;

[0028] The "pile turning node" in the embodiment of the present invention specifically refers to: the stacking time from the completion of the pile of mash to the turning of the mash during the stacking fermentation process (the saccharification pile in the embodiment of the present invention only involves one pile turning). Turning the mash is a common method to assist the fermentation of mash during the stacking process of the saccharification pile. The mash at the top of the original saccharification pile is transferred to the target area as the new saccharification pile center; the mash remaining around the initial saccharification pile is separated by peeling and surrounded at the foot of the pile; the mash at the foot of the pile is transferred in batches from the outside to the inside to the center of the saccharification pile covering the target area. After multiple cycles, a new saccharification pile is formed and the pile turning is completed.

[0029] The "top temperature entering the cellar" in the embodiment of the present invention specifically refers to: the top temperature of the saccharification pile when the fermented grains after pile fermentation are about to enter the cellar for fermentation.

[0030] The "medium temperature entering the cellar" in the embodiment of the present invention specifically refers to: the center temperature of the saccharification pile when the mash after pile fermentation is about to enter the cellar for fermentation.

[0031] The "bottom temperature entering the cellar" in the embodiment of the present invention specifically refers to: the bottom temperature of the saccharification pile when the mash after pile fermentation is about to enter the cellar for fermentation.

[0032] The "cellar temperature" in the embodiment of the present invention specifically refers to: the comprehensive temperature of the mash after pile fermentation when it enters the cellar for fermentation.

[0033] The "acidity when entering the cellar" in the embodiment of the present invention specifically refers to the acidity of the fermented grains when they enter the cellar for fermentation after pile fermentation.

[0034] The "moisture content entering the cellar" in the embodiment of the present invention specifically refers to the moisture content of the fermented grains after pile fermentation when entering the cellar for fermentation.

[0035] The "cellar starch" in the embodiment of the present invention specifically refers to: the starch when the fermented grains enter the cellar for fermentation after pile fermentation.

[0036] The "reducing sugar entering the cellar" in the embodiment of the present invention specifically refers to the acidity of the fermented grains after pile fermentation when entering the cellar for fermentation.

[0037] The "base wine production rate" in the embodiment of the present invention specifically refers to: the ratio of the base wine production of each saccharification pile corresponding to the cellar in each round to the amount of grain input. The wine production rate in the embodiment of the present invention is the wine production rate of three rounds.

[0038] Example 1

[0039] Based on the invention content provided by the present invention, this embodiment provides a method for predicting the yield rate of Maotai-flavor liquor, comprising the following steps:

[0040] 1. Obtaining modeling data

[0041] Obtain data information on the stacking time (DJSJ) and turning node (FDJD) of the saccharification pile during the fermentation stage, as well as the top temperature (RJDW), middle temperature (RJZW), bottom temperature (RJDW.1), tasting temperature (RJPW), acidity (RJSD), moisture (RJSF), starch (RJDF), reducing sugar (RJHYT), and base wine yield (CJL) of the cellar where the corresponding batch of saccharification piles are located. , the data information of the top temperature of the fermented grains entering the cellar, the middle temperature of the fermented grains entering the cellar, the bottom temperature of the fermented grains entering the cellar, the tasting temperature of the fermented grains entering the cellar, and the wine yield rate are recorded manually, and the acidity, moisture content, starch content, and reducing sugar content of the fermented grains entering the cellar are detected and obtained by the FOSS near-infrared multifunctional quality analyzer (the specific detection steps are: fill the sample to be tested into the sample cup of the device to ensure that the bottom of the cup is not transparent, place it on the measuring pool after filling, cover it with a lid, select the measurement template and click OK to perform the detection). Finally, this embodiment obtains the above-mentioned indicator data corresponding to 115 saccharification piles;

[0042] 2. Building a Prediction Model

[0043] The modeling data obtained in step 1 is divided into a training data set and a test data set; the training data set accounts for 75% of the total data, and the test data set accounts for 25%; the modeling data corresponding to the training data set is input into the XGBoost algorithm to train and establish a prediction model, as follows:

[0044] The obtained wine yield CJL was taken as the dependent variable, the stacking time (DJSJ), the turning node (FDJD), and the top temperature of the mash entering the cellar (RJDW), the middle temperature entering the cellar (RJZW), the bottom temperature entering the cellar (RJDW.1), the tasting temperature entering the cellar (RJPW), the acidity entering the cellar (RJSD), the moisture entering the cellar (RJSF), the starch entering the cellar (RJDF), and the reducing sugar entering the cellar (RJHYT) were taken as the independent variables, and the obtained wine yield CJL was taken as the dependent variable. The XGBoost regression model in R language was imported to construct the corresponding wine yield prediction model; among them, XGBoos The optimal parameter settings for the regression model are as follows: eta = 0.04, gamma = 0.03, max_depth = 6, subsample = 0.75, colsample_bytree = 0.6, objective = "reg:squarederror", nround = 800, watchlist = watchlist, verbose = 1, print_every_n = 50, early_stopping_rounds = 100;

[0045] 3. Evaluation of the accuracy of the prediction model

[0046] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) are used as evaluation indicators to evaluate the accuracy of the prediction model established in step 3: the root mean square error (RMSE) represents the square root of the average of the squares of the differences between the predicted value and the actual value. The smaller the value, the better the model fit. The calculation method is shown in Formula I: Coefficient of determination (R 2 ), which indicates the degree of fit of the regression model to the data set, that is, the percentage of explanation of the regression model. The closer its value is to 1, the better the degree of fit of the model. The calculation method is shown in Formula II: Mean absolute error (MAE) represents the average value of the absolute error between the predicted value and the actual value. The smaller the value, the better the model fit. The calculation method is as described in Formula III: In the calculation formulas shown in the above formulas Ⅰ-Ⅲ, n is the number of samples, y is the sample label value, is the sample prediction value, is the sample average. The calculation results are shown in Table 1:

[0047] Table 1 Evaluation results of XGBoost model

[0048]

[0049] According to the results presented in Table 1, the XGBoost regression model constructed in this comparison has a training set RMSE value of 0.6782 and R2 The value is 0.8450 and the MAE value is 0.5433. The XGBoost regression model has good fitting ability.

[0050] 4. Verification of Prediction Model Accuracy

[0051] Based on the above method, the data information of the stacking time (DJSJ), turning node (FDJD) of 10 saccharification piles in the stacking fermentation stage, as well as the top temperature (RJDW), middle temperature (RJZW), bottom temperature (RJDW.1), tasting temperature (RJPW), acidity (RJSD), moisture (RJSF), starch (RJDF) and reducing sugar (RJHYT) of the fermented mash were collected and imported into the XGBoost regression model constructed in step 3 to obtain the predicted wine yield. At the same time, the wine yield was compared with the actual wine yield to verify the accuracy of the model, and the error rate between the predicted wine yield and the actual wine yield was calculated. The results are shown in Table 2:

[0052] The error rate calculation formula is:

[0053] Table 2 Comparison of the predicted wine production rate and the actual wine production rate under the XGBoost prediction model

[0054] Saccharification pile Actual wine production rate / % Example 1 Predicted wine production rate / % Error rate / % 1 8.50 8.54 0.47％ 2 5.56 5.54 0.36％ 3 8.14 8.06 0.98％ 4 6.03 6.03 0.00％ 5 10.21 10.27 0.59％ 6 10.11 10.16 0.49％ 7 10.56 10.48 0.76％ 8 9.25 9.28 0.32％ 9 9.45 9.47 0.21％ 10 9.83 9.89 0.61％

[0055] According to the results presented in Table 2, the XGBoost prediction model constructed by this embodiment has a high prediction accuracy for the wine yield, which is relatively close to the actual wine yield, and the overall error rate is less than 1%, so it can be used to predict the wine yield.

[0056] Example 2

[0057] Based on the invention content provided by the present invention, this embodiment provides a method for predicting the yield rate of Maotai-flavor liquor, comprising the following steps:

[0058] 1. Obtaining modeling data

[0059] Obtain the data information of indicators such as the stacking time (DJSJ), the turning node (FDJD) of the saccharification pile stacking and fermentation stage, the top temperature (RJDW), the middle temperature (RJZW), the bottom temperature (RJDW.1), the tasting temperature (RJPW), the acidity (RJSD), the moisture (RJSF), the starch (RJDF), the reducing sugar (RJHYT), and the base wine yield (CJL) of the cellar where the corresponding batch of wine mash is located; wherein, the data information of the stacking time, the turning node, the top temperature, the middle temperature, the bottom temperature, the tasting temperature, and the yield of the wine mash are recorded manually, and the acidity, moisture, starch, and reducing sugar content of the wine mash are obtained by detecting the acidity, moisture, and reducing sugar content of the wine mash. Finally, the embodiment obtains the above-mentioned indicator data corresponding to 115 saccharification piles;

[0060] 2. Screening Modeling Features

[0061] Based on the modeling data obtained in step 1, with wine production rate as the dependent variable, and data indicators such as stacking time, turning pile nodes, top temperature, middle temperature, bottom temperature, tasting temperature, acidity, moisture, starch and reducing sugar content as independent variables, the random forest algorithm in R language was imported to rank the importance of independent variable indicators. The specific parameter settings of the random forest algorithm are as follows: predictors<-colnames(dat)[2:11], model<-train(x=dat[,predictors], y=dat$CJL, method="rf", importance=TRUE, tuneGrid=expand.grid(mtry=c(2:4)), #length(predictors)or 2:6, trControl = trainControl(method = "cv", number = 20, p = 0.75, savePredictions = TRUE)); The ranking results of the importance of the modeling data features are: stacking time (DJSJ) > turning node (FDJD) > moisture in the pit (RJSF) > middle temperature in the pit (RJZW) > temperature in the pit (RJPW) > top temperature in the pit (RJDW) > bottom temperature in the pit (RJDW.1) > reducing sugar in the pit (RJHYT) > acidity in the pit (RJSD) > starch in the pit (RJDF);

[0062] The top 6 indicators in terms of importance were used as modeling indicators, i.e., stacking time (DJSJ), turning node (FDJD), moisture in the pit (RJSF), middle temperature in the pit (RJZW), product temperature in the pit (RJPW), and top temperature in the pit (RJDW) were used as modeling features for subsequent modeling.

[0063] 3. Building a Prediction Model

[0064] The modeling data obtained in step 1 is divided into a training data set and a test data set; the training data set accounts for 75% of the total data, and the test data set accounts for 25%; the modeling features corresponding to the training data set are input into the XGBoost algorithm to train and establish a prediction model, as follows:

[0065] With the pile time DJSJ, the turning node FDJD, the water content in the cellar RJSF, the middle temperature in the cellar RJZW, the tasting temperature in the cellar RJPW, and the top temperature in the cellar RJDW as independent variables, and the obtained wine yield CJL as the dependent variable, the XGBoost regression model in R language was imported to build the corresponding wine yield prediction model; among them, 75% of the data in the modeling features were divided into the experimental set, and 25% of the data were divided into the test set for modeling;

[0066] The optimal parameter settings for the XGBoos regression model are as follows: eta = 0.04, gamma = 0.03, max_depth = 6, subsample = 0.75, colsample_bytree = 0.6, objective = "reg:squarederror", nround = 800, watchlist = watchlist, verbose = 1, print_every_n = 50, early_stopping_rounds = 100;

[0067] 4. Evaluation of the accuracy of the prediction model

[0068] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) are used as evaluation indicators to evaluate the accuracy of the prediction model established in step 3: the root mean square error (RMSE) represents the square root of the average of the squares of the differences between the predicted value and the actual value. The smaller the value, the better the model fit. The calculation method is shown in Formula I: Coefficient of determination (R 2 ), which indicates the degree of fit of the regression model to the data set, that is, the percentage of explanation of the regression model. The closer its value is to 1, the better the degree of fit of the model. The calculation method is shown in Formula II: Mean absolute error (MAE) represents the average value of the absolute error between the predicted value and the actual value. The smaller the value, the better the model fit. The calculation method is as described in Formula III: (Formula III); In the calculation formulas shown in the above formulas I-III, n is the number of samples, y is the sample label value, is the sample prediction value, is the sample average. The calculation results are shown in Table 3:

[0069] Table 3 Evaluation results of different models

[0070]

[0071] According to the results presented in Table 3, the RMSE of the XGBoost regression model constructed in step 3 for the test set is 0.0920, R 2 The value is 0.9973 and the MAE is 0.0750, which shows an excellent fitting state.

[0072] 5. Verification of Prediction Model Accuracy

[0073] Based on the above method, the data information of the stacking time, turning node, moisture content, middle temperature, product temperature and top temperature of 10 saccharification piles in the stacking and fermentation stage were collected and imported into the XGBoost regression model constructed in step 3 to obtain the predicted wine yield. At the same time, the wine yield was compared with the actual wine yield to verify the accuracy of the model, and the error rate between the predicted wine yield and the actual wine yield was calculated. The results are shown in Table 4:

[0074] The error rate calculation formula is:

[0075] Table 4 Comparison of the predicted wine production rate and the actual wine production rate under the XGBoost prediction model

[0076]

[0077]

[0078] According to the results presented in Table 4, the error between the predicted result of the wine yield predicted by the XGBoost prediction model constructed by this embodiment and the actual wine yield is small, both less than 1%, which can achieve accurate prediction of the wine yield.

[0079] Example 3

[0080] This embodiment is based on Embodiment 2, and differs from Embodiment 2 in that the indicators corresponding to the top 7 ranking results of the modeling data feature importance are used as modeling indicators to construct an XGBoost prediction model, and a method for predicting the yield of Maotai-flavor liquor based on the constructed XGBoost prediction model is provided, comprising the following steps:

[0081] 1. Obtaining modeling data: same as in Example 2;

[0082] 2. Screening modeling features: The top 7 indicators ranked by importance are used as modeling indicators, that is, the pile time (DJSJ), the turning node (FDJD), the moisture content in the cellar (RJSF), the middle temperature in the cellar (RJZW), the product temperature in the cellar (RJPW), the top temperature in the cellar (RJDW), and the bottom temperature in the cellar (RJDW.1) are used as modeling features for subsequent modeling; the rest is the same as in Example 2;

[0083] 3. Building a Prediction Model

[0084] The stacking time (DJSJ), the turning node (FDJD), the moisture content in the cellar (RJSF), the middle temperature in the cellar (RJZW), the tasting temperature in the cellar (RJPW), the top temperature in the cellar (RJDW), and the bottom temperature in the cellar (RJDW.1) were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to construct the corresponding wine yield prediction model; the rest was the same as in Example 2;

[0085] 4. Evaluation of the accuracy of the prediction model

[0086] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction model established in step 3. The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were calculated, and the error rate between the predicted wine yield and the actual wine yield was calculated based on the method in Example 2. The calculation results are shown in Table 5:

[0087] Table 5 Evaluation results of different models

[0088]

[0089] According to the results presented in Table 5, the RMSE of the XGBoost regression model constructed in step 3 for the test set is 0.1170, R 2 The value is 0.9959 and the MAE is 0.0874, which shows a good fitting state.

[0090] 5. Verification of Prediction Model Accuracy

[0091] Based on the above method, data information of the stacking time (DJSJ), turning node (FDJD), water content in the cellar (RJSF), middle temperature in the cellar (RJZW), temperature in the cellar (RJPW), top temperature in the cellar (RJDW), and bottom temperature in the cellar (RJDW.1) of 10 saccharification piles in the stacking and fermentation stage were collected and imported into the XGBoost regression model constructed in step 3 to obtain the predicted wine yield. At the same time, the wine yield was compared with the actual wine yield to verify the accuracy of the model. The error rate between the predicted wine yield and the actual wine yield was calculated based on the method in Example 2. The calculation results are shown in Figure 6:

[0092] Table 6 Comparison of the predicted wine production rate and the actual wine production rate under the XGBoost prediction model

[0093] Saccharification pile Actual wine production rate / % Example 3 Predicted wine production rate / % Error rate 1 8.50 8.48 0.24 2 5.56 5.52 0.72 3 8.14 8.18 0.49 4 6.03 6.02 0.17 5 10.21 10.25 0.39 6 10.11 10.15 0.40 7 10.56 10.53 0.28 8 9.25 9.25 0.00 9 9.45 9.46 0.11 10 9.83 9.91 0.81

[0094] According to the results presented in Table 6, the error between the predicted result of the wine yield obtained by the XGBoost prediction model constructed by this embodiment and the actual wine yield is small, and the overall error rate is less than 1%, which can achieve accurate prediction of the wine yield.

[0095] Example 4

[0096] This embodiment is based on Embodiment 2, and differs from Embodiment 2 in that the indicators corresponding to the top 8 modeling data feature importance ranking results are used as modeling indicators to construct an XGBoost prediction model, and a method for predicting the yield of Maotai-flavor liquor based on the constructed XGBoost prediction model is provided, comprising the following steps:

[0097] 1. Obtaining modeling data: same as in Example 2;

[0098] 2. Screening modeling features: The top 8 indicators ranked by importance are used as modeling indicators, that is, the pile time (DJSJ), turning node (FDJD), moisture in the pit (RJSF), middle temperature in the pit (RJZW), product temperature in the pit (RJPW), top temperature in the pit (RJDW), bottom temperature in the pit (RJDW.1), and reducing sugar in the pit (RJHYT) are used as modeling features for subsequent modeling; the rest is the same as in Example 2;

[0099] 3. Building a Prediction Model

[0100] The obtained wine yield CJL is used as the dependent variable, and the stacking time (DJSJ), turning node (FDJD), water content in the cellar (RJSF), middle temperature in the cellar (RJZW), tasting temperature in the cellar (RJPW), top temperature in the cellar (RJDW), bottom temperature in the cellar (RJDW.1), and reducing sugar in the cellar (RJHYT) are used as independent variables, and the XGBoost regression model in the R language is imported to construct the corresponding wine yield prediction model; the rest is the same as in Example 2;

[0101] 4. Evaluation of the accuracy of the prediction model

[0102] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction model established in step 3. The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) are calculated, and the results are shown in Table 7:

[0103] Table 7 Evaluation results of different models

[0104]

[0105] According to the results presented in Table 7, the RMSE of the XGBoost regression model constructed in step 3 for the test set is 0.1934, R 2 The value is 0.9910 and the MAE is 0.1437, which shows a good fitting state.

[0106] 5. Verification of Prediction Model Accuracy

[0107] Based on the above method, data information of the stacking time (DJSJ), turning node (FDJD), water content in the cellar (RJSF), middle temperature in the cellar (RJZW), product temperature in the cellar (RJPW), top temperature in the cellar (RJDW), bottom temperature in the cellar (RJDW.1), and reducing sugar in the cellar (RJHYT) of 10 saccharification piles in the stacking and fermentation stage were collected and imported into the XGBoost regression model constructed in step 3 to obtain the predicted wine yield. At the same time, the wine yield was compared with the actual wine yield to verify the accuracy of the model. The error rate was calculated based on the same method as in Example 2. The calculation results are shown in Table 8:

[0108] Table 8 Comparison of the predicted wine production rate and the actual wine production rate under the XGBoost prediction model

[0109]

[0110]

[0111] According to the results presented in Table 8, the error between the predicted result of the wine yield obtained by the XGBoost prediction model constructed by this embodiment and the actual wine yield is small, and the overall error rate is less than 1%, which can achieve accurate prediction of the wine yield.

[0112] Comparative Example 1

[0113] This comparative example is based on Example 2 and provides a method for predicting the yield of Maotai-flavor liquor. The difference from Example 2 is that the top five indicators ranked by importance are used as modeling indicators, that is, the pile time, pile turning node, cellar moisture, cellar medium temperature, and cellar tasting temperature are used as modeling features for subsequent modeling; the details are as follows:

[0114] 1. Obtaining modeling data: same as in Example 2;

[0115] 2. Screening modeling features: taking the top 5 indicators in importance as modeling indicators, that is, taking the stacking time, turning node, moisture content in the cellar, medium temperature in the cellar, and product temperature in the cellar as modeling features for subsequent modeling; the rest is the same as in Example 2;

[0116] 3. Constructing a prediction model: taking the obtained wine yield CJL as the dependent variable, and taking the stacking time, turning pile node, water content in the cellar, medium temperature in the cellar, and tasting temperature in the cellar as independent variables, the XGBoost regression model in the R language is imported to construct a corresponding wine yield prediction model; the rest is the same as in Example 2;

[0117] 4. Evaluation of the accuracy of the prediction model

[0118] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction model established in step 3. The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) are calculated, and the results are shown in Table 9:

[0119] Table 9 Evaluation results of the model of Example 1

[0120]

[0121]

[0122] According to the results presented in Table 9, the RMSE of the XGBoost prediction model constructed in step 3 for the test set is 0.3576, R 2 The value is 0.9673 and the MAE is 0.2817, which means that the fitting state of the XGBoost prediction model is significantly reduced.

[0123] 5. Verification of Prediction Model Accuracy

[0124] Based on the above method, the data information of the stacking time, turning node, moisture content, medium temperature and tasting temperature of 10 saccharification piles in the stacking and fermentation stage were collected and imported into the XGBoost regression model constructed in step 3 to obtain the predicted wine yield. At the same time, the wine yield was compared with the actual wine yield to verify the accuracy of the model. The results are shown in Figure 10:

[0125] Table 10 Comparison of the predicted wine production rate and the actual wine production rate under the XGBoost prediction model

[0126] Saccharification pile Actual wine production rate / % Comparative Example 1 Predicted wine production rate / % Error rate / % 1 8.50 8.33 2.00％ 2 5.56 5.64 1.44％ 3 8.14 7.67 5.77％ 4 6.03 5.76 4.48％ 5 10.21 9.98 2.25％ 6 10.11 10.01 0.99％ 7 10.56 10.12 4.17％ 8 9.25 9.26 0.11％ 9 9.45 9.42 0.32％ 10 9.83 9.64 1.93％

[0127] According to the results presented in Table 10, the XGBoost prediction model constructed by this comparative example has significantly reduced prediction accuracy for wine yield, with an error rate as high as 5.77%, and cannot be used to predict wine yield.

[0128] Comparative Example 2

[0129] This embodiment is based on Embodiment 2 and provides a method for predicting the yield of Maotai-flavor liquor. The difference from Embodiment 2 is that the top four indicators in importance are used as modeling indicators, that is, the pile time, pile turning node, pit moisture, and pit temperature are used as modeling features for subsequent modeling; the details are as follows:

[0130] 1. Obtaining modeling data: same as in Example 2;

[0131] 2. Screening modeling features: taking the top 4 indicators in importance as modeling indicators, that is, taking the stacking time, turning node, moisture content in the cellar, and medium temperature in the cellar as modeling features for subsequent modeling; the rest is the same as in Example 2;

[0132] 3. Building a Prediction Model

[0133] The obtained wine yield CJL is used as the dependent variable, and the stacking time, turning node, moisture content in the cellar, and medium temperature in the cellar are used as independent variables. The XGBoost regression model in R language is imported to construct the corresponding wine yield prediction model; the rest is the same as in Example 2;

[0134] 4. Evaluation of the accuracy of the prediction model

[0135] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction model established in step 3. The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) are calculated, and the results are shown in Table 11:

[0136] Table 11 Evaluation results of the model of Example 2

[0137]

[0138] According to the results presented in Table 11, the fitting performance of the XGBoost regression model constructed by this comparative example deteriorates.

[0139] 5. Verification of Prediction Model Accuracy

[0140] Based on the above method, the data information of the stacking time, turning node, moisture content in the cellar, and medium temperature in the cellar of 10 saccharification piles in the stacking and fermentation stage were collected and imported into the XGBoost regression model constructed in step 3 to obtain the predicted wine yield. At the same time, the wine yield was compared with the actual wine yield to verify the accuracy of the model. The error rate was calculated based on the same method as in Example 2. The calculation results are shown in Figure 12:

[0141] Table 12 Comparison of the predicted wine production rate and the actual wine production rate under the XGBoost prediction model

[0142]

[0143]

[0144] According to the results presented in Table 12, the XGBoost prediction model constructed by this comparative example has significantly reduced prediction accuracy for wine yield, with an error rate as high as 4.32%, and cannot be used to predict wine yield.

[0145] Comparative Example 3

[0146] This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on the above embodiments, which is as follows:

[0147] Group 1: This group 1 is based on Example 1, and the difference from Example 1 is that the wine yield prediction model is established based on the random forest regression model, the lasso regression model, and the decision tree regression model, respectively, and the rest is the same as Example 1; wherein the parameters of the random forest regression model used are as follows: data=traindata, ntree=500, mtry=6, importance=T; the parameters of the decision tree regression model are as follows: type=4, extra=100, fallen.leaves=TRUE, main="Decision Tree"; the parameters of the lasso regression model are as follows: set.seed(70), family="gaussian", alpha=1, type.measure="mse", type="coef", s="lambda.min";

[0148] Group 2: Based on Example 2, the difference from Example 2 is that the wine yield prediction model is established based on the random forest regression model, and the rest is the same as Example 1;

[0149] Group 3: Based on Example 3, the difference from Example 3 is that the wine yield prediction model is established based on the random forest regression model, and the rest is the same as Example 1;

[0150] Group 4: Based on Example 4, the difference from Example 4 is that the wine yield prediction model is established based on the random forest regression model, and the rest is the same as Example 1;

[0151] The accuracy of the prediction models established for the above groups was evaluated based on the root mean square error (RMSE), determination coefficient (R2) and mean absolute error (MAE) as evaluation indicators. The results are shown in Table 13:

[0152] Table 13 Evaluation results of different models

[0153]

[0154]

[0155] According to the results presented in Table 13, the degree of fit of the prediction model constructed for each group in this comparison ratio has decreased significantly.

[0156] The random forest regression model constructed based on the above groups 1, 2, 3, and 4 is used to predict the wine production rate. At the same time, the wine production rate is compared with the actual wine production rate to verify the accuracy of the model and calculate the error rate. The calculation results are shown in Figure 14:

[0157] Table 14 Comparison of predicted wine yield and actual wine yield of random forest regression model in different groups

[0158]

[0159] According to the results presented in Table 14, the prediction accuracy of the wine yield rate constructed by the prediction model of each group in this comparison ratio has obviously decreased, the error rate is high, and it cannot be used to predict the wine yield rate.

[0160] Comparative Example 4

[0161] This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 1, which is as follows:

[0162] Group 4-1: The difference from Example 1 is that the XGBoost regression model is built with the stacking time, the middle temperature of the cellar, the tasting temperature of the cellar, the top temperature of the cellar and the bottom temperature of the cellar as the modeling features, and the rest is the same as Example 2, that is, the obtained wine yield CJL is used as the dependent variable, and the stacking time, the middle temperature of the cellar, the tasting temperature of the cellar, the top temperature of the cellar and the bottom temperature of the cellar are used as the independent variables, and the XGBoost regression model is imported into the R language to obtain the corresponding wine yield prediction model;

[0163] Group 4-2: This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 1. The difference from Example 1 is that an XGBoost regression model is built with middle temperature entering the cellar, tasting temperature entering the cellar, top temperature entering the cellar, bottom temperature entering the cellar, moisture entering the cellar, acidity entering the cellar, and starch entering the cellar as modeling features, and the rest is the same as Example 1, that is, the obtained liquor yield CJL is used as the dependent variable, and middle temperature entering the cellar, tasting temperature entering the cellar, top temperature entering the cellar, bottom temperature entering the cellar, moisture entering the cellar, acidity entering the cellar, and starch entering the cellar are used as independent variables, and the XGBoost regression model in the R language is imported to obtain the corresponding liquor yield prediction model;

[0164] Group 4-3: This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 1. The difference from Example 1 is that the stacking time, cellar top temperature, cellar middle temperature, cellar bottom temperature, cellar tasting temperature, cellar acidity, cellar moisture, and cellar starch are used as modeling features for XGBoost regression model building, and the rest are the same as Example 1, that is, the obtained liquor yield CJL is used as the dependent variable, and the stacking time, cellar middle temperature, cellar tasting temperature, cellar top temperature, cellar bottom temperature, cellar moisture, cellar acidity, and cellar starch are used as independent variables, and the XGBoost regression model in R language is imported to obtain the corresponding liquor yield prediction model;

[0165] Group 4-4: This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 1. The difference from Example 1 is that the stacking time, top temperature, middle temperature, bottom temperature, acidity, moisture, starch and reducing sugar are used as modeling features to perform XGBoost regression model building, and the rest are the same as Example 1, that is, the obtained liquor yield CJL is used as the dependent variable, and the stacking time, top temperature, middle temperature, bottom temperature, acidity, moisture, starch and reducing sugar are used as independent variables, and the XGBoost regression model in R language is imported to obtain the corresponding liquor yield prediction model;

[0166] Group 4-5: This comparative example is based on Example 1 and provides a method for predicting the yield of Maotai-flavor liquor. The difference from Example 1 is that the stacking time, middle temperature into the cellar, bottom temperature into the cellar, tasting temperature into the cellar, acidity into the cellar, moisture into the cellar, starch into the cellar, and reducing sugar into the cellar are used as modeling features to perform XGBoost regression model modeling, and the rest are the same as Example 1, that is, the obtained liquor yield CJL is used as the dependent variable, and the stacking time, middle temperature into the cellar, bottom temperature into the cellar, tasting temperature into the cellar, acidity into the cellar, moisture into the cellar, starch into the cellar, and reducing sugar into the cellar are used as independent variables, and the XGBoost regression model in the R language is imported to obtain the corresponding liquor yield prediction model;

[0167] Group 4-6 This comparative example is based on Example 1, and provides a method for predicting the yield of Maotai-flavor liquor. The difference from Example 1 is that the stacking time, the middle temperature of entering the cellar, the bottom temperature of entering the cellar, the tasting temperature of entering the cellar, the acidity of entering the cellar, the moisture of entering the cellar, the starch of entering the cellar, and the reducing sugar of entering the cellar are used as modeling features to perform XGBoost regression model modeling, and the rest are the same as Example 1, that is, the obtained liquor yield CJL is used as the dependent variable, and the stacking time, the middle temperature of entering the cellar, the bottom temperature of entering the cellar, the tasting temperature of entering the cellar, the acidity of entering the cellar, the moisture of entering the cellar, the starch of entering the cellar, and the reducing sugar of entering the cellar are used as independent variables, and the XGBoost regression model in the R language is imported to obtain the corresponding liquor yield prediction model;

[0168] Group 4-7: This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 1. The difference from Example 1 is that XGBoost regression model is built with stacking time, water content in the cellar, top temperature in the cellar, bottom temperature in the cellar, reducing sugar in the cellar, starch in the cellar, and acidity in the cellar as modeling features, and the rest is the same as Example 1, that is, the obtained wine yield CJL is used as the dependent variable, and stacking time, water content in the cellar, top temperature in the cellar, bottom temperature in the cellar, reducing sugar in the cellar, starch in the cellar, and acidity in the cellar are used as independent variables, and the XGBoost regression model in R language is imported to obtain the corresponding wine yield prediction model;

[0169] Group 4-8: This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 1. The difference from Example 1 is that XGBoost regression model is built with stacking time, cellar temperature, cellar moisture, cellar bottom temperature, cellar reducing sugar, cellar starch and cellar acidity as modeling features, and the rest is the same as Example 1, that is, the obtained liquor yield CJL is used as the dependent variable, and stacking time, cellar temperature, cellar moisture, cellar bottom temperature, cellar reducing sugar, cellar starch and cellar acidity are used as independent variables, and the XGBoost regression model in R language is imported to obtain the corresponding liquor yield prediction model;

[0170] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction models constructed for the above groups. The results are shown in Table 15:

[0171] Table 15 Evaluation results of prediction models

[0172]

[0173] As shown in Table 15, the fitting performance of the XGBoost regression model constructed by each group in this comparative example is poor, significantly lower than that in Example 1.

[0174] Comparative Example 5

[0175] This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 2, which is as follows:

[0176] First, in step 2 of screening modeling features in Example 2, the stacking time, cellar top temperature, cellar middle temperature, cellar bottom temperature, cellar product temperature, cellar acidity, cellar moisture, cellar starch, and cellar reducing sugar content data indicators are used as independent variables, and the random forest algorithm in the R language is imported to rank the importance of the independent variable indicators, and the importance ranking result is: stacking time>cellar moisture>cellar product temperature>cellar top temperature>cellar bottom temperature>cellar middle temperature>cellar reducing sugar>cellar acidity>cellar starch; based on the above importance ranking results, the following group experiments are designed:

[0177] Group 5-1: Based on Example 2, the top 8 indicators in the importance ranking were selected as modeling features from the above importance ranking results, i.e., stacking time, cellar moisture, cellar temperature, cellar top temperature, cellar bottom temperature, cellar middle temperature, cellar reducing sugar, and cellar acidity were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in R language was imported to construct the corresponding wine yield prediction model;

[0178] Group 5-2: Based on Example 2, the top 7 indicators in the importance ranking were selected as modeling features from the above importance ranking results, i.e., stacking time, moisture content in the cellar, product temperature in the cellar, top temperature in the cellar, bottom temperature in the cellar, middle temperature in the cellar, and reducing sugar in the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to construct the corresponding wine yield prediction model;

[0179] Group 5-3: Based on Example 2, the top 6 indicators in the importance ranking were selected as modeling features from the above importance ranking results, i.e., stacking time, cellar moisture, cellar temperature, cellar top temperature, cellar bottom temperature, and cellar middle temperature were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in R language was imported to construct the corresponding wine yield prediction model;

[0180] Group 5-4: Based on Example 2, the top 5 indicators in the importance ranking were selected as modeling features from the above importance ranking results, i.e., stacking time, moisture content in the cellar, temperature of the wine in the cellar, top temperature in the cellar, and bottom temperature in the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to construct the corresponding wine yield prediction model;

[0181] Group 5-5: Based on Example 2, the top 4 indicators in the importance ranking were selected as modeling features, i.e., stacking time, moisture content in the cellar, temperature in the cellar, and top temperature in the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to construct the corresponding wine yield prediction model;

[0182] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction models constructed for the above groups. The results are shown in Table 16:

[0183] Table 16 Evaluation results of prediction models

[0184]

[0185]

[0186] As shown in the results in Table 16, in step 2 of screening modeling features, importance screening based on different indicators or selecting indicators with different importance rankings as modeling features to establish a prediction model will greatly affect the prediction accuracy of the prediction model.

[0187] Comparative Example 6

[0188] This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 2, which is as follows:

[0189] Group 6-1: In step 2 of screening modeling features in Example 2, 9 data indicators including stacking time, turning pile nodes, and top temperature, middle temperature, bottom temperature, acidity, moisture, starch, and reducing sugar of fermented grains are used as independent variables, and the random forest algorithm in R language is imported to rank the independent variable indicators by importance, and the importance ranking result is: stacking time>turning pile node>moisture>bottom temperature>middle temperature>top temperature>reducing sugar>acidity>starch. Based on the above importance ranking result, the top 6 indicators in importance ranking are selected as modeling features for modeling, that is, stacking time, turning pile node, moisture, bottom temperature, middle temperature, and top temperature of fermented grains are used as independent variables, and the obtained wine yield CJL is used as the dependent variable. The XGBoost regression model in R language is imported to construct a corresponding wine yield prediction model.

[0190] Group 6-2: In step 2 of screening modeling features in Example 2, 9 data indicators including stacking time, turning pile node, middle temperature of entering cellar, bottom temperature of entering cellar, product temperature of entering cellar, acidity of entering cellar, moisture of entering cellar, starch of entering cellar and reducing sugar of entering cellar are used as independent variables, and the random forest algorithm in R language is imported to rank the importance of the independent variable indicators, and the importance ranking result is obtained as follows: stacking time> turning pile node> product temperature of entering cellar> middle temperature of entering cellar> moisture of entering cellar> bottom temperature of entering cellar> acidity of entering cellar> reducing sugar of entering cellar> starch of entering cellar. Based on the above importance ranking result, the top 6 indicators in importance ranking are selected as modeling features for modeling, namely, stacking time, turning pile node, product temperature of entering cellar, middle temperature of entering cellar, moisture of entering cellar and bottom temperature of entering cellar are used as independent variables, and the obtained wine yield CJL is used as the dependent variable, and the XGBoost regression model in R language is imported to construct the corresponding wine yield prediction model;

[0191] Group 6-3: In step 2 of screening modeling features in Example 2, 9 data indicators including stacking time, turning pile node, cellar top temperature, cellar bottom temperature, cellar odor temperature, cellar acidity, cellar moisture, cellar starch and cellar reducing sugar are used as independent variables, and the random forest algorithm in R language is imported to rank the importance of the independent variable indicators, and the importance ranking result is: stacking time> turning pile node> cellar odor temperature> cellar moisture> cellar bottom temperature> cellar top temperature> cellar reducing sugar> cellar acidity> cellar starch. Based on the above importance ranking result, the top 6 indicators in the importance ranking are selected as modeling features for modeling, that is, stacking time, turning pile node, cellar odor temperature, cellar moisture, cellar bottom temperature and cellar top temperature are used as independent variables, and the obtained wine yield CJL is used as the dependent variable, and the XGBoost regression model in R language is imported to construct the corresponding wine yield prediction model;

[0192] Group 6-4: In step 2 of screening modeling features in Example 2, eight data indicators, namely, stacking time, cellar top temperature, cellar middle temperature, cellar bottom temperature, cellar acidity, cellar moisture, cellar starch, and cellar reducing sugar, were used as independent variables. The random forest algorithm in R language was imported to rank the importance of the independent variable indicators, and the importance ranking result was: stacking time > cellar moisture > cellar top temperature > cellar bottom temperature > cellar middle temperature > cellar reducing sugar > cellar starch > cellar acidity. Based on the above importance ranking result, the top six indicators in the importance ranking were selected as modeling features for modeling, namely, stacking time, cellar moisture, cellar top temperature, cellar bottom temperature, cellar middle temperature, and cellar reducing sugar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in R language was imported to construct a corresponding wine yield prediction model.

[0193] Group 6-5: In step 2 of screening modeling features in Example 2, eight data indicators, including stacking time (DJSJ), middle temperature in the cellar (RJZW), bottom temperature in the cellar (RJDW.1), product temperature in the cellar (RJPW), acidity in the cellar (RJSD), moisture in the cellar (RJSF), starch in the cellar (RJDF), and reducing sugar in the cellar (RJHYT), were used as independent variables. The random forest algorithm in R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ)> product temperature in the cellar (RJPW)> Moisture content in the cellar (RJSF) > medium temperature in the cellar (RJZW) > bottom temperature in the cellar (RJDW.1) > reducing sugar in the cellar (RJHYT) > starch in the cellar (RJDF) > acidity in the cellar (RJSD). Based on the above importance ranking results, the top 6 indicators in the importance ranking were selected as modeling features for modeling, namely, stacking time, cellar temperature, moisture content in the cellar, medium temperature in the cellar, bottom temperature in the cellar, and reducing sugar in the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to construct the corresponding wine yield prediction model.

[0194] Group 6-6: In step 2 of screening modeling features in Example 2, eight data indicators, including stacking time (DJSJ), top temperature entering the cellar (RJDW), bottom temperature entering the cellar (RJDW.1), product temperature entering the cellar (RJPW), acidity entering the cellar (RJSD), moisture entering the cellar (RJSF), starch entering the cellar (RJDF), and reducing sugar entering the cellar (RJHYT), were used as independent variables. The random forest algorithm in R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ)> moisture entering the cellar (RJSF)> Wine entering cellar temperature (RJPW) > wine entering cellar bottom temperature (RJDW.1) > wine entering cellar top temperature (RJDW) > wine entering cellar reducing sugar (RJHYT) > wine entering cellar acidity (RJSD) > wine entering cellar starch (RJDF). Based on the above importance ranking results, the top 6 indicators in importance ranking were selected as modeling features for modeling, namely, stacking time, wine entering cellar temperature, wine entering cellar moisture, wine entering middle temperature, wine entering cellar bottom temperature, and wine entering reducing sugar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in R language was imported to construct the corresponding wine yield prediction model.

[0195] Group 6-7: In step 2 of screening modeling features in Example 2, eight data indicators, including stacking time (DJSJ), turning node (FDJD), middle temperature in the cellar (RJZW), bottom temperature in the cellar (RJDW.1), acidity in the cellar (RJSD), moisture in the cellar (RJSF), starch in the cellar (RJDF), and reducing sugar in the cellar (RJHYT), were used as independent variables. The random forest algorithm in R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ)> turning node (FDJD)> The middle temperature of the wine entering the cellar (RJZW)> the moisture content of the wine entering the cellar (RJSF)> the bottom temperature of the wine entering the cellar (RJDW.1)> the reducing sugar content of the wine entering the cellar (RJHYT)> the acid content of the wine entering the cellar (RJSD)> the starch content of the wine entering the cellar (RJDF). Based on the above importance ranking results, the top 6 indicators in the importance ranking were selected as modeling features for modeling, namely, stacking time, turning pile node, middle temperature of the wine entering the cellar, moisture content of the wine entering the cellar, bottom temperature of the wine entering the cellar, and reducing sugar of the wine entering the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to construct the corresponding wine yield prediction model.

[0196] Group 6-8: In step 2 of screening modeling features in Example 2, eight data indicators, including stacking time (DJSJ), turning node (FDJD), bottom temperature in the cellar (RJDW.1), product temperature in the cellar (RJPW), acidity in the cellar (RJSD), moisture in the cellar (RJSF), starch in the cellar (RJDF), and reducing sugar in the cellar (RJHYT), were used as independent variables. The random forest algorithm in R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ) > turning node (FDJD) >Pit temperature (RJPW) >Pit moisture (RJSF) >Pit bottom temperature (RJDW.1) >Pit acidity (RJSD) >Pit reducing sugar (RJHYT) >Pit starch (RJDF). Based on the above importance ranking results, the top 6 indicators in importance ranking are selected as modeling features for modeling, that is, stacking time, turning pile node, pit temperature, pit moisture, pit bottom temperature, and pit acidity are used as independent variables, and the obtained wine yield CJL is used as the dependent variable. The XGBoost regression model in R language is imported to build the corresponding wine yield prediction model;

[0197] Group 6-9: In step 2 of screening modeling features in Example 2, seven data indicators, including stacking time (DJSJ), middle temperature in the cellar (RJZW), bottom temperature in the cellar (RJDW.1), acidity in the cellar (RJSD), moisture in the cellar (RJSF), starch in the cellar (RJDF), and reducing sugar in the cellar (RJHYT), were used as independent variables. The random forest algorithm in R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ)> moisture in the cellar (RJSF)> Middle temperature in the cellar (RJZW) > bottom temperature in the cellar (RJDW.1) > reducing sugar in the cellar (RJHYT) > starch in the cellar (RJDF) > acidity in the cellar (RJSD). Based on the above importance ranking results, the top 6 indicators in the importance ranking were selected as modeling features for modeling, namely, stacking time, moisture in the cellar, middle temperature in the cellar, bottom temperature in the cellar, reducing sugar in the cellar, and starch in the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to construct the corresponding wine yield prediction model.

[0198] Group 6-10: In step 2 of screening modeling features in Example 2, seven data indicators, including stacking time (DJSJ), top temperature entering the cellar (RJDW), bottom temperature entering the cellar (RJDW.1), acidity entering the cellar (RJSD), moisture entering the cellar (RJSF), starch entering the cellar (RJDF), and reducing sugar entering the cellar (RJHYT), were used as independent variables. The random forest algorithm in R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ) > moisture entering the cellar (RJSF). >Top temperature of wine entering the cellar (RJDW) > Bottom temperature of wine entering the cellar (RJDW.1) > Reducing sugar of wine entering the cellar (RJHYT) > Starch of wine entering the cellar (RJDF) > Acidity of wine entering the cellar (RJSD). Based on the above importance ranking results, the top 6 indicators of importance ranking were selected as modeling features for modeling, namely, stacking time, moisture of wine entering the cellar, top temperature of wine entering the cellar, bottom temperature of wine entering the cellar, reducing sugar of wine entering the cellar, and starch of wine entering the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in R language was imported to construct the corresponding wine yield prediction model.

[0199] Group 6-11: In step 2 of screening modeling features in Example 2, seven data indicators, including stacking time (DJSJ), bottom temperature in the cellar (RJDW.1), product temperature in the cellar (RJPW), acidity in the cellar (RJSD), moisture in the cellar (RJSF), starch in the cellar (RJDF), and reducing sugar in the cellar (RJHYT), were used as independent variables. The random forest algorithm in the R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ) > product temperature in the cellar (RJPW). >Moisture content before cellaring (RJSF) > Bottom temperature before cellaring (RJDW.1) > Reducing sugar before cellaring (RJHYT) > Starch before cellaring (RJDF) > Acidity before cellaring (RJSD). Based on the above importance ranking results, the top 6 indicators in importance ranking were selected as modeling features for modeling, namely, stacking time, cellar temperature, moisture content before cellaring, bottom temperature before cellaring, reducing sugar before cellaring, and starch before cellaring were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in R language was imported to construct the corresponding wine yield prediction model.

[0200] Group 6-12: In step 2 of screening modeling features in Example 2, seven data indicators, including stacking time (DJSJ), turning node (FDJD), bottom temperature in the cellar (RJDW.1), acidity in the cellar (RJSD), moisture in the cellar (RJSF), starch in the cellar (RJDF), and reducing sugar in the cellar (RJHYT), were used as independent variables. The random forest algorithm in R language was introduced to rank the independent variable indicators by importance, and the importance ranking result was: stacking time (DJSJ) > turning node (FDJD) >Moisture content in the cellar (RJSF) > Bottom temperature in the cellar (RJDW.1) > Acidity in the cellar (RJSD) > Reducing sugar in the cellar (RJHYT) > Starch in the cellar (RJDF). Based on the above importance ranking results, the top 6 indicators in the importance ranking were selected as modeling features for modeling, namely, stacking time, turning pile node, moisture content in the cellar, bottom temperature in the cellar, acidity in the cellar, and reducing sugar in the cellar were used as independent variables, and the obtained wine yield CJL was used as the dependent variable. The XGBoost regression model in the R language was imported to build the corresponding wine yield prediction model;

[0201] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction models constructed for the above groups. The results are shown in Table 17:

[0202] Table 17 Evaluation results of prediction models

[0203]

[0204]

[0205] As shown in Table 17, in step 2 of screening modeling features, the prediction accuracy of the constructed prediction model is completely different when importance screening is performed based on different indicators. That is, screening the importance of independent variable indicators based on different indicators will also greatly affect the prediction accuracy of the prediction model.

[0206] Comparative Example 7

[0207] This comparative example provides a method for predicting the yield of Maotai-flavor liquor based on Example 2, which is as follows:

[0208] Group 7-1: Based on Example 2, the difference from Example 2 is that in step three of constructing the prediction model in Example 2, the parameters of the XGBoost regression model used are different, and the rest are the same as Example 2; wherein, the XGBoost regression model parameters used are as follows: eta=0.08, gamma=0.01, max_depth=4, subsample=0.75, colsample_bytree=0.6, objective="reg:squarederror", nround=1000, watchlist=watchlist, verbose=1, print_every_n=50, early_stopping_rounds=100, that is, with stacking time DJSJ, turning node FDJD, cellar moisture RJSF, cellar middle temperature RJZW, cellar tasting temperature RJPW, cellar top temperature RJDW as independent variables, and the obtained wine yield CJL as the dependent variable, a corresponding wine yield prediction model is constructed based on the above XGBoost regression model parameters;

[0209] Group 7-2: Based on Example 2, the difference from Example 2 is that in step three of constructing the prediction model in Example 2, the parameters of the XGBoost regression model used are different, and the rest are the same as Example 2; wherein, the XGBoost regression model parameters used are as follows: eta=0.03, gamma=0.8, max_depth=7, subsample=0.75, colsample_bytree=0.06, objective="reg:squarederror", nround=800, watchlist=watchlist, verbose=1, print_every_n=60, early_stopping_rounds=100, that is, with stacking time DJSJ, turning node FDJD, cellar moisture RJSF, cellar middle temperature RJZW, cellar tasting temperature RJPW, cellar top temperature RJDW as independent variables, and the obtained wine yield CJL as the dependent variable, a corresponding wine yield prediction model is constructed based on the above XGBoost regression model parameters;

[0210] Group 7-3: Based on Example 2, the difference from Example 2 is that in step three of constructing the prediction model in Example 2, the parameters of the XGBoost regression model used are different, and the rest are the same as Example 2; wherein, the XGBoost regression model parameters used are as follows: eta=0.06, gamma=0.5, max_depth=6, subsample=0.75, colsample_bytree=0.6, objective="reg:squarederror", nround=1000, watchlist=watchlist, verbose=1, print_every_n=40, early_stopping_rounds=120, that is, with stacking time DJSJ, turning node FDJD, cellar moisture RJSF, cellar middle temperature RJZW, cellar tasting temperature RJPW, cellar top temperature RJDW as independent variables, and the obtained wine yield CJL as the dependent variable, a corresponding wine yield prediction model is constructed based on the above XGBoost regression model parameters;

[0211] Group 7-4: Based on Example 2, the difference from Example 2 is that in step three of constructing the prediction model in Example 2, the parameters of the XGBoost regression model used are different, and the rest are the same as Example 2; wherein, the XGBoost regression model parameters used are as follows: eta=0.08, gamma=0.02, max_depth=5, subsample=0.75, colsample_bytree=0.5, objective="reg:squarederror", nround=600, watchlist=watchlist, verbose=1, print_every_n=60, early_stopping_rounds=100, that is, with stacking time DJSJ, turning pile node FDJD, cellar moisture RJSF, cellar middle temperature RJZW, cellar tasting temperature RJPW, cellar top temperature RJDW as independent variables, and the obtained wine yield CJL as the dependent variable, a corresponding wine yield prediction model is constructed based on the above XGBoost regression model parameters;

[0212] Group 7-5: Based on Example 2, the difference from Example 2 is that in step three of constructing the prediction model in Example 2, the parameters of the XGBoost regression model used are different, and the rest are the same as Example 2; wherein, the XGBoost regression model parameters used are as follows: eta=0.3, gamma=0.08, max_depth=6, subsample=0.75, colsample_bytree=0.4, objective="reg:squarederror", nround=900, watchlist=watchlist, verbose=1, print_every_n=40, early_stopping_rounds=100, that is, with stacking time DJSJ, turning node FDJD, cellar moisture RJSF, cellar middle temperature RJZW, cellar tasting temperature RJPW, cellar top temperature RJDW as independent variables, and the obtained wine yield CJL as the dependent variable, a corresponding wine yield prediction model is constructed based on the above XGBoost regression model parameters;

[0213] Group 7-6: Based on Example 2, the difference from Example 2 is that in step three of constructing the prediction model in Example 2, the parameters of the XGBoost regression model used are different, and the rest are the same as Example 2; wherein the XGBoost regression model parameters used are as follows: eta=0.6, gamma=0.3, max_depth=8, subsample=0.75, colsample_bytree=0.6, objective="reg:squarederror", nround=1000, watchlist=watchlist, verbose=1, print_every_n=30, early_stopping_rounds=100, that is, with stacking time DJSJ, turning pile node FDJD, cellar moisture RJSF, cellar middle temperature RJZW, cellar tasting temperature RJPW, cellar top temperature RJDW as independent variables, and the obtained wine yield CJL as the dependent variable, a corresponding wine yield prediction model is constructed based on the above XGBoost regression model parameters;

[0214] The root mean square error (RMSE), determination coefficient (R 2 ) and mean absolute error (MAE) were used as evaluation indicators to evaluate the accuracy of the prediction models constructed for the above groups. The results are shown in Table 18:

[0215] Table 18 Evaluation results of prediction models

[0216]

[0217]

[0218] As shown in the results in Table 18, even if the prediction model is constructed based on the XGBoost regression model, the change of model parameters will greatly affect the prediction accuracy of the constructed model.

[0219] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of the present invention are within the scope of protection of the present invention.

Claims

1. A method for constructing a prediction model for the yield of Maotai-flavor liquor, characterized in that: The steps include:

1. Obtaining modeling data: Detect and record the stacking time, turning nodes, top temperature, middle temperature, bottom temperature, product temperature, acidity, moisture, starch, reducing sugar, and base wine yield of the corresponding saccharification pile during the fermentation stage to obtain modeling data; 2. Build a prediction model: Use the modeling data obtained in step 1 as the modeling feature, and build an XGBoost prediction model based on the XGBoost regression model.

2. The prediction method according to claim 1, characterized in that: In step 2, the modeling data obtained in step 1 is used as the modeling feature, and the XGBoost prediction model is constructed based on the XGBoost regression model, including the following steps: The modeling data is divided into a training data set and a test data set. The stacking time, turning nodes, top temperature, middle temperature, bottom temperature, taste temperature, acidity, moisture, starch and reducing sugar in the training data set are used as independent variables, and the base wine yield in the training data set is used as the dependent variable. The XGBoost regression model training in R language is imported to obtain the XGBoost prediction model.

3. The prediction method according to claim 1, characterized in that: In step 2, the parameter settings of the XGBoost regression model include: eta=0.04, gamma=0.03, max_depth=6, subsample=0.75, colsample_bytree=0.6, objective="reg:squarederror", nround=800, watchlist=watchlist, verbose=1, print_every_n=50, early_stopping_rounds=100.

4. The prediction method according to claim 1, characterized in that: The prediction method further comprises: The prediction accuracy of the XGBoost prediction model constructed in step 2 is evaluated using at least one of the root mean square error, determination coefficient, and mean absolute error as an evaluation index.

5. The prediction method according to claim 1, characterized in that: The prediction method further comprises: In the step of building a prediction model, the modeling data obtained in step one is ranked according to the importance of independent variable indicators based on the random forest algorithm, and the modeling data is screened based on the importance ranking results. The indicators ranked first 6-8 in importance are used as modeling features to build an XGBoost prediction model.

6. The prediction method according to claim 5, characterized in that: The modeling features include: stacking time, turning pile node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, and top temperature in the cellar; or, stacking time, turning pile node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar, and bottom temperature in the cellar; or, stacking time, turning pile node, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar, bottom temperature in the cellar, and reducing sugar in the cellar.

7. A method for predicting the yield of Maotai-flavor liquor, characterized in that: The prediction method comprises the following steps: (1) Obtaining index data information of the saccharification pile fermentation stage; (2) Inputting the index data obtained in step (1) into the XGBoost prediction model constructed by any construction method described in claims 1-4, and predicting the base wine yield corresponding to the saccharification pile based on the output result of the XGBoost prediction model.

8. The prediction method according to claim 7, characterized in that: The index data information includes stacking time, turning node, top temperature entering the cellar, middle temperature entering the cellar, bottom temperature entering the cellar, product temperature entering the cellar, acidity entering the cellar, moisture entering the cellar, starch entering the cellar and reducing sugar entering the cellar.

9. A method for predicting the yield of Maotai-flavor liquor, characterized in that: The prediction method comprises the following steps: 1) Obtaining index data information of the saccharification pile fermentation stage; 2) Inputting the data information obtained in step 1) into the XGBoost prediction model constructed by any of the construction methods described in claims 5-6, and predicting the base wine yield corresponding to the saccharification pile based on the output result of the XGBoost prediction model.

10. The prediction method according to claim 9, characterized in that: The indicator data information includes: Data information on the piling time, turning time, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, and top temperature in the cellar; or Data information on stacking time, turning pile nodes, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar, and bottom temperature in the cellar; or, stacking time, turning pile nodes, moisture content in the cellar, middle temperature in the cellar, product temperature in the cellar, top temperature in the cellar, bottom temperature in the cellar, and reducing sugar in the cellar.

Images