CN-122024880-A - Method for predicting liquor yield of liquor heap, method and system for constructing model

CN122024880ACN 122024880 ACN122024880 ACN 122024880ACN-122024880-A

Abstract

The application belongs to the technical field of liquor yield prediction, and relates to a liquor yield prediction method, a model construction method and a model construction system for liquor piles. A model construction method for predicting wine yield of a white spirit pile comprises the steps of collecting characteristic data of a plurality of piles and corresponding wine yield, preprocessing the characteristic data of the piles of the same production round to form an input matrix X and a wine yield vector Y, carrying out coding processing or continuous characterization processing on year and workshop information of the piles to obtain characteristic variables capable of representing systematic deviation of the year and the workshop relative to an overall average level, adding the characteristic variables into the input matrix X, and building a regression prediction model from the input matrix X to the wine yield vector Y based on a partial least square regression model according to the input matrix X and the wine yield vector Y. According to the application, the prediction of the output of the heap sub-level is realized in the early stage of the heap fermentation, and the coding variable which can represent the systematic deviation of the input matrix X relative to the overall average level is added into the input matrix X to realize early warning and accurate prediction.

Inventors

YAN SUI
Zhao Yigao
Tian Shaorun
LI XIAOBO
CHEN ZONGXIAO
TIAN YUAN
ZHANG XIAOMING
ZHANG PING

Assignees

贵州茅台酒股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260107

Claims (10)

1. The method for constructing the liquor yield prediction model of the liquor pile is characterized by comprising the following steps of: collecting characteristic data of a plurality of piles and the corresponding wine yield; The characteristic data of the piles of the same production round are preprocessed to form an input matrix X and a wine yield vector Y, the year and workshop information of the piles are subjected to coding processing or continuous characterization processing to obtain characteristic variables capable of representing systematic deviation of the year and the workshop relative to the overall average level, and the characteristic variables are added into the input matrix X; And establishing a regression prediction model from the input matrix X to the wine yield vector Y based on a partial least square regression model according to the input matrix X and the wine yield vector Y.
2. The model construction method according to claim 1, wherein the characteristic data includes process index data and physicochemical index data; preferably, the physical and chemical indexes comprise temperature, moisture, acidity, sugar, ethanol, starch and acetic acid data measured on a stack surface, a stack core and a stack bottom of the stack; preferably, the process index data comprise stacking time, spreading and airing time, starter propagation temperature, stacking temperature and airing room temperature data when stacking.
3. The model construction method according to claim 1, wherein the encoding process is a dummy variable process, comprising: selecting a reference year and a reference workshop; for the year, setting a 0-1 variable for each year except the reference year, and taking a value of 1 when the year of the heap belongs to the year, or else taking a value of 0; For workshops, setting a 0-1 variable for each workshop except the reference workshop, and taking a value of 1 when the workshop to which the pile belongs is the workshop, or taking a value of 0 when the pile belongs to the workshop; the reference year and reference plant are implicitly represented by the corresponding dummy variable being all 0.
4. The method according to claim 1, wherein the continuous characterization process includes constructing a continuous environmental index reflecting a year or a whole fermentation environment of the plant based on statistical features or principal component analysis scores of the historical process and the physicochemical index, and taking the continuous environmental index as a feature variable representing systematic deviation of the year or the plant.
5. The model construction method according to claim 1, wherein the encoding process includes effect encoding of setting a number of offset variables for the years or the workshops equal to the number of categories thereof minus one, and learning the offset intensity of each year or the workshops with respect to the overall mean value in model training.
6. The model building method according to any one of claims 1 to 3, wherein the preprocessing comprises cleaning, wherein the cleaning comprises identifying abnormal data of three times of standard deviation or missing of each index in each group from a mean value, eliminating a single pile if two or more indexes are abnormal, and replacing the corresponding index value of the pile with the smallest overall distance from the abnormal pile in the pile with all normal indexes in the same group if only the single index is abnormal.
7. A model building method according to any one of claims 1 to 3, wherein the preprocessing includes normalization processing including normalization of subtracting a mean value from a process index and a physicochemical index divided by a standard deviation, centering of subtracting a mean value from a dummy variable index generated by year and shop information conversion, and normalization of subtracting a mean value from a wine yield vector divided by a standard deviation.
8. A method for predicting the wine yield of a wine pile is characterized by comprising the following steps: Acquiring characteristic data of a pile to be predicted, wherein the characteristic data comprises process indexes, physicochemical indexes and year and workshop information; Carrying out coding processing or continuous characterization processing on the year and workshop information of the heap to obtain a characteristic variable capable of representing systematic deviation of the year and the workshop relative to the overall average level, and forming an input vector based on the characteristic data and the characteristic variable; inputting the input vector into a wine yield prediction model which is pre-built for the same production round, and outputting a wine yield predicted value of the pile by the wine yield prediction model, wherein the wine yield prediction model is obtained by the model building method according to any one of claims 1-7.
9. A predictive model construction system for liquor yield of a liquor pile, the system comprising: Comprising the following steps: The data acquisition module is used for acquiring characteristic data of a plurality of piles and the corresponding wine yield in the same production round; the data preprocessing module is used for preprocessing the characteristic data, converting year and workshop information of the piles into characteristic variables capable of representing systematic deviation of the characteristic variables relative to the overall average level, and forming an input matrix X and a wine yield vector Y; The model construction module is used for constructing a regression prediction model from X to Y by adopting a partial least square regression method based on the input matrix X and the wine yield vector Y, and the regression prediction model is used as a wine yield prediction model of the production turn.
10. A system for predicting wine yield of a wine pile, the system comprising: Comprising the following steps: the characteristic acquisition module is used for acquiring characteristic data of the pile to be predicted, wherein the characteristic data comprises process indexes, physicochemical indexes, the year and workshop information; The code conversion module is used for converting the year and workshop information into characteristic variables capable of representing systematic deviation of the year and the workshop information relative to the overall average level and generating an input vector based on the characteristic data and the characteristic variables; the prediction execution module is used for inputting the input vector into a wine yield prediction model which is pre-constructed for the same production round and outputting a wine yield prediction value of the pile; The wine yield prediction model is a partial least squares regression model obtained by the model construction method according to any one of claims 1-7.

Description

Method for predicting liquor yield of liquor heap, method and system for constructing model Technical Field The application belongs to the technical field of liquor yield prediction, and relates to a liquor yield prediction method, a model construction method and a model construction system for liquor piles. Background The white spirit is a Chinese traditional distilled spirit, is prepared by taking grains such as sorghum, wheat and the like as raw materials through complex processes such as solid state fermentation, distillation, ageing, blending and the like, and has long history and deep cultural background. According to different flavors, the white spirit is mainly classified into a Maotai-flavor type, a strong flavor type, a faint scent type and the like, wherein the Maotai-flavor type white spirit is favored by virtue of unique brewing process and flavor characteristics. Maotai-flavor liquor is represented by Guizhou Maotai liquor, adopts a '12987' process (namely annual production period, twice feeding, nine times of cooking, eight times of fermentation and seven times of liquor taking), relies on natural microbial communities to perform open solid state fermentation, and is highly sensitive to climate, water quality, pit environment and operation experience. The wine yield is not only affected by the proportion of raw materials, but also closely related to dynamic factors such as temperature, humidity, succession of microorganisms and the like in the fermentation process. At present, white spirit enterprises rely on manual experience or a simple statistical model to estimate the wine yield, and the prior art lacks the actual requirement for fine management in the complex brewing process. Disclosure of Invention The invention aims to provide a method for predicting liquor yield of a liquor pile, a method for constructing a model and a system. The first aspect of the application provides a model construction method for predicting liquor yield of a liquor pile, which comprises the following steps: collecting characteristic data of a plurality of piles and the corresponding wine yield; The characteristic data of the piles of the same production round are preprocessed to form an input matrix X and a wine yield vector Y, the year and workshop information of the piles are subjected to coding processing or continuous characterization processing to obtain characteristic variables capable of representing systematic deviation of the year and the workshop relative to the overall average level, and the characteristic variables are added into the input matrix X; And establishing a regression prediction model from the input matrix X to the wine yield vector Y based on a partial least square regression model according to the input matrix X and the wine yield vector Y. In some embodiments of the application, the characteristic data includes process index data and physical and chemical index data; In some embodiments of the application, the physical and chemical indicators include measured temperature, moisture, acidity, sugar, ethanol, starch, and acetic acid data at the stack table, core, bottom of the stack; In some embodiments of the application, the process index data includes stacking time, spreading and airing time, starter propagation temperature, stacking temperature and airing hall room temperature data when stacking. In some embodiments of the application, the encoding process is a dummy variable process comprising: selecting a reference year and a reference workshop; for the year, setting a 0-1 variable for each year except the reference year, and taking a value of 1 when the year of the heap belongs to the year, or else taking a value of 0; For workshops, setting a 0-1 variable for each workshop except the reference workshop, and taking a value of 1 when the workshop to which the pile belongs is the workshop, or taking a value of 0 when the pile belongs to the workshop; the reference year and reference plant are implicitly represented by the corresponding dummy variable being all 0. In some embodiments of the application, the continuous characterization process includes constructing a continuous environmental index reflecting a year or a plant overall fermentation environment based on statistical features or principal component analysis scores of historical process and physicochemical metrics, and taking the continuous environmental index as a characteristic variable characterizing the year or plant systematic deviation. In some embodiments of the application, the encoding process includes effect encoding by setting a number of offset variables for a year or plant equal to its number of categories minus one and learning the offset strength of each year or plant relative to the overall mean in model training. In some embodiments of the application, the preprocessing comprises cleaning, wherein the cleaning comprises identifying abnormal data of which each index deviates from a mean value by three times of