Method and device for predicting fermentation stage and quality grade of yeast for making hard liquor, and electronic equipment
By constructing a composite neural network model, using metabolic compounds and microbial abundance data to predict the Dako fermentation stage and quality level, the problems of poor accuracy and consistency in the existing technology are solved, and efficient fermentation stage and quality level judgment are achieved.
Patent Information
- Application Number
- CN202510364913.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-26
- Publication Date
- 2025-07-08
AI Technical Summary
The current judgment of the fermentation stage and quality level of Dakota mainly relies on manual experience, resulting in poor accuracy and consistency and low efficiency.
A composite neural network model is constructed, based on metabolic compound data and microbial abundance data, and through training the Daqu prediction model, it realizes simultaneous prediction of fermentation stage and quality level, including the combination of metabolic compound data sub-model, microbial abundance data sub-model, fusion layer, fermentation stage prediction layer and quality level prediction layer.
The accuracy and consistency of the Daqu fermentation stage and quality grade are improved, the dependence on manual experience is reduced, the prediction efficiency is improved, and the quality monitoring and optimization of the Daqu fermentation process is realized.
Smart Images
Figure CN120280040A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of brewing, and specifically relates to a method and device for predicting the fermentation stage and quality grade of Daqu, and an electronic device. Background Art
[0002] Daqu fermentation refers to the process of brewing Chinese liquor using Daqu as a saccharifying and fermenting agent. Its basic principle is that microorganisms in Daqu convert starch in raw materials into sugars, and then yeast ferments to produce alcohol. Daqu is a type of koji that has undergone long-term fermentation and is usually made from raw materials such as wheat, yellow rice, and peas through a series of processes including fermentation, koji mold cultivation, and steaming. Daqu contains various microorganisms, including molds, yeasts, and bacteria.
[0003] As an important matrix in traditional fermentation industry, the fermentation process of Daqu has a decisive impact on the flavor and quality of Chinese liquor. By determining the fermentation stage of Daqu during the fermentation process and the quality grades at different fermentation stages, quality monitoring of the Daqu fermentation process can be achieved, and a scientific basis can be provided for optimizing the Daqu fermentation process. Currently, the judgment of the fermentation stage and quality grade of Daqu mainly relies on manual experience. This method depends on manual experience and has strong subjectivity, resulting in poor accuracy and consistency. Summary of the Invention
[0004] The present invention aims to solve the problem of poor accuracy and consistency in the existing method for determining the fermentation stage and quality grade of Daqu, and proposes a method and device for predicting the fermentation stage and quality grade of Daqu, and an electronic device.
[0005] The technical solution adopted by the present invention to solve the above technical problems is as follows:
[0006] In a first aspect, the present invention provides a method for predicting the fermentation stage and quality grade of Daqu, the method comprising:
[0007] Obtaining fermentation data of the Daqu fermentation process, and establishing a Daqu fermentation data set according to the fermentation data, the fermentation data including metabolite data, microbial abundance data, fermentation stage data, and quality grade data;
[0008] Training a composite neural network model based on the Daqu fermentation data set to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade;
[0009] Obtaining the metabolite data and / or microbial abundance data of the Daqu to be predicted, and inputting the metabolite data and / or microbial abundance data of the Daqu to be predicted into the Daqu prediction model to obtain the fermentation stage and quality grade of the Daqu to be predicted.
[0010] Further, the Daqu prediction model includes:
[0011] A metabolic compound data sub-model for extracting metabolic compound features related to the fermentation stage and quality grade from metabolic compound data;
[0012] A microbial abundance data sub-model for extracting microbial abundance features related to the fermentation stage and quality grade from microbial abundance data;
[0013] A fusion layer for weighted fusion of metabolic compound features and microbial abundance features to obtain metabolic-microbial fusion features;
[0014] A Daqu fermentation stage prediction layer for predicting the fermentation stage based on metabolic compound features, microbial abundance features or metabolic-microbial fusion features;
[0015] A Daqu quality grade prediction layer for predicting the quality grade based on metabolic compound features, microbial abundance features or metabolic-microbial fusion features.
[0016] Furthermore, obtaining the fermentation stage and quality grade of the Daqu to be predicted includes:
[0017] If the Daqu prediction model only receives the metabolic compound data of the Daqu to be predicted, activate the metabolic compound data sub-model, the Daqu fermentation stage prediction layer and the Daqu quality grade prediction layer, and predict the fermentation stage and quality grade of the Daqu to be predicted based on the metabolic compound features;
[0018] If the Daqu prediction model only receives the microbial abundance data of the Daqu to be predicted, activate the microbial abundance data sub-model, the Daqu fermentation stage prediction layer and the Daqu quality grade prediction layer, and predict the fermentation stage and quality grade of the Daqu to be predicted based on the microbial abundance features;
[0019] If the Daqu prediction model receives both the metabolic compound data and the microbial abundance data of the Daqu to be predicted, activate the metabolic compound data sub-model, the microbial abundance data sub-model, the fusion layer, the Daqu fermentation stage prediction layer and the Daqu quality grade prediction layer, and predict the fermentation stage and quality grade of the Daqu to be predicted based on the metabolic-microbial fusion features.
[0020] Furthermore, the metabolic compound data includes concentration data of one or more compounds among esters, alcohols, aldehydes, acids, ketones, pyrazines, furans and aromatics. The esters at least include ethyl hexadecanoate and ethyl caproate. The alcohols at least include phenethyl alcohol and 2,3-butanediol. The acids at least include acetic acid and caproic acid. The pyrazines at least include 2,5-dimethylpyrazine, trimethylpyrazine and 2,6-dimethylpyrazine.
[0021] Further, the microbial abundance data includes the types and quantities of microbial populations during the Daqu fermentation process, and the microbial populations at least include Bacillus, Aspergillus, Saccharomyces cerevisiae, Pediococcus acidilactici, and Staphylococcus.
[0022] Further, the fermentation stage data is the time node in the Daqu fermentation cycle, and the unit of the time node is days or hours;
[0023] The quality grade data includes first-grade Daqu, second-grade Daqu, and third-grade Daqu.
[0024] Further, the composite neural network model is trained using a supervised learning method. The accuracy of the composite neural network model is determined through cross-validation, and based on the Dropout layer with regularization added, the cross-entropy loss function is used as the loss function of the network, and the model is optimized through the backpropagation of the neural network. When the loss function is less than the loss function threshold or the accuracy is greater than the accuracy threshold, the corresponding composite neural network model is used as the Daqu prediction model.
[0025] Further, the method further includes:
[0026] Data augmentation is performed on the metabolite data and microbial abundance data in the Daqu fermentation dataset, and the data augmentation method includes random noise.
[0027] In a second aspect, the present invention provides a device for predicting the fermentation stage and quality grade of Daqu, and the device includes:
[0028] An acquisition unit, configured to acquire fermentation data during the Daqu fermentation process, and establish a Daqu fermentation dataset according to the fermentation data. The fermentation data includes metabolite data, microbial abundance data, fermentation stage data, and quality grade data;
[0029] A training unit, configured to train a composite neural network model based on the Daqu fermentation dataset to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade;
[0030] A prediction unit, configured to acquire the metabolite data and / or microbial abundance data of the Daqu to be predicted, and input the metabolite data and / or microbial abundance data of the Daqu to be predicted into the Daqu prediction model to obtain the fermentation stage and quality grade of the Daqu to be predicted.
[0031] In a third aspect, the present invention provides an electronic device, and the electronic device includes a processor, a memory, and a communication bus;
[0032] The communication bus is used to realize the connection communication between the processor and the memory;
[0033] The processor is configured to execute one or more programs in the memory to implement the steps of the method for predicting the large - scale liquor fermentation stage and quality grade as described in the first aspect.
[0034] The beneficial effects of the present invention are as follows: The method, device, and electronic device for predicting the large - scale liquor fermentation stage and quality grade provided by the present invention can simultaneously predict the fermentation stage and quality grade of large - scale liquor by constructing a large - scale liquor prediction model that predicts both the fermentation stage and quality grade. Based on metabolite compound data and / or microbial abundance data, the simultaneous prediction of the large - scale liquor fermentation stage and quality grade can be achieved, thereby improving the accuracy and consistency of determining the large - scale liquor fermentation stage and quality grade. Moreover, with one prediction of the present invention, the large - scale liquor fermentation stage and its corresponding quality grade can be obtained simultaneously, without relying on manually determining the correspondence between the large - scale liquor fermentation stage and quality grade, further improving the efficiency. By determining the fermentation stage of large - scale liquor during the fermentation process and the quality grade at different fermentation stages, quality monitoring of the large - scale liquor fermentation process can be realized, and a scientific basis can be provided for optimizing the large - scale liquor fermentation process. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 A flowchart of a method for predicting the large - scale liquor fermentation stage and quality grade provided for an embodiment;
[0036] Figure 2 A schematic diagram showing the change of the loss function during the training process of a composite neural network model provided for an embodiment;
[0037] Figure 3 A schematic diagram showing the change of the accuracy rate during the training process of a composite neural network model provided for an embodiment;
[0038] Figure 4 A schematic structural diagram of a device for predicting the large - scale liquor fermentation stage and quality grade provided for an embodiment;
[0039] Figure 5 A schematic structural diagram of an electronic device provided for an embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0040] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the present embodiment will be clearly and completely described below in conjunction with the accompanying drawings in the present embodiment.
[0041] In some processes described in the specification of the present invention and the above - mentioned drawings, there are multiple operations that appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel.
[0042] Currently, the way to determine the fermentation stage and quality grade of Daqu during the fermentation process usually relies on manual experience to estimate the fermentation stage and quality grade of Daqu during the fermentation process. The inventor has found through research that the method of manual experience estimation relies on manual experience and has strong subjectivity, with problems of poor accuracy and consistency. In addition, it is also necessary to manually determine the corresponding relationship between the fermentation stage and quality grade of Daqu, and the efficiency is poor.
[0043] Based on this, the technical solution of the present invention is proposed. In the present invention, fermentation data of the Daqu fermentation process is obtained, and a Daqu fermentation data set is established according to the fermentation data. The fermentation data includes metabolite compound data, microbial abundance data, fermentation stage data, and quality grade data; a composite neural network model is trained based on the Daqu fermentation data set to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade; the metabolite compound data and / or microbial abundance data of the Daqu to be predicted are obtained, and the metabolite compound data and / or microbial abundance data of the Daqu to be predicted are input into the Daqu prediction model to obtain the fermentation stage and quality grade of the Daqu to be predicted.
[0044] Specifically, the present invention first constructs a Daqu fermentation data set including metabolite compound data, microbial abundance data, fermentation stage data, and quality grade data, then trains a Daqu prediction model based on the Daqu fermentation data set, and finally inputs the metabolite compound data and / or microbial abundance data of the Daqu to be predicted into the Daqu prediction model to obtain the fermentation stage of the Daqu to be predicted and its corresponding quality grade. The present invention does not rely on manual experience, improves the accuracy and consistency of determining the fermentation stage and quality grade of Daqu. And there is no need to rely on manual determination of the corresponding relationship between the fermentation stage and quality grade of Daqu, further improving the efficiency.
[0045] Figure 1 The flowchart of a method for predicting the fermentation stage and quality grade of Daqu is shown. Please refer to Figure 1 and this method includes the following steps:
[0046] Step 1, obtain the fermentation data of the Daqu fermentation process, and establish a Daqu fermentation data set according to the fermentation data. The fermentation data includes metabolite compound data, microbial abundance data, fermentation stage data, and quality grade data.
[0047] In this embodiment, the metabolic compound data represents compound data related to Daqu fermentation, and the compound data can be the concentration data of compounds. After obtaining the metabolic compound data, it is necessary to standardize the metabolic compound data by the minimum-maximum normalization method. For example, the metabolic compound data includes the concentration data of one or more compounds in esters, alcohols, aldehydes, acids, ketones, pyrazines, furans, and aromatics. The esters at least include ethyl hexadecanoate and ethyl caproate. The alcohols at least include phenethyl alcohol and 2,3-butanediol. The acids at least include acetic acid and caproic acid. The pyrazines at least include 2,5-dimethylpyrazine, trimethylpyrazine, and 2,6-dimethylpyrazine. The unit of the compound concentration data is ppm. For example, the metabolic compound data at a certain moment is shown as: [1.017500007, 0.360192202, 0.28742875, 0.282392042, 0, 0.071789635,...]. If the maximum value of the metabolic compound data is 4 ppm and the minimum value is 0 ppm, the metabolic compound data after normalization is shown as: [0.254375002, 0.090048051, 0.071857188, 0.070598011, 0, 0.017947409,...].
[0048] In this embodiment, the microbial abundance data includes the types and quantities of microbial populations during the Daqu fermentation process. The microbial populations at least include Bacillus, Aspergillus, Saccharomyces cerevisiae, Pediococcus acidilactici, and Staphylococcus. For example, the microbial abundance data at a certain moment is shown as: [0.01207594, 0.017445474, 0.022036947,...]; the microbial abundance data itself is normalized, so there is no need for normalization. In each group of microbial abundance data, the sum of the abundances of various microorganisms is 1.
[0049] In this embodiment, the fermentation stage data is the time node in the Daqu fermentation cycle, and the unit of the time node is days or hours. That is, the fermentation stage data is used to identify the specific moment in the current Daqu fermentation cycle and can be initially represented in days or hours. For example, if the fermentation time is the 1st day, it is initially represented as 1, and if the fermentation time is the 16th day, it is initially represented as 16. In this embodiment, the Daqu fermentation cycle is taken as 30 days. Therefore, the initial data of the Daqu fermentation stage only includes the data from 1 to 30 days; in this embodiment, the initial data of 30 days can be divided into 3 groups according to the time development as the Daqu fermentation stage data, with 0 representing the 1st stage, 1 representing the 2nd stage, and 2 representing the 3rd stage; therefore, there are 3 types of Daqu fermentation stages, namely the 1st stage, the 2nd stage, and the 3rd stage.
[0050] In this embodiment, the quality grade data represents the quality grade of Daqu, which may include first-grade Daqu, second-grade Daqu, and third-grade Daqu. That is, there are 3 types of quality grade data. 0 represents first-grade Daqu, 1 represents second-grade Daqu, and 2 represents third-grade Daqu. Among them, the quality grade data can be divided according to actual production requirements or according to the grading standards used by experts in the field.
[0051] After obtaining a large amount of fermentation data during the Daqu fermentation process, a Daqu fermentation dataset can be established. In this embodiment, data augmentation methods such as random noise can also be used to augment the metabolite data and microbial abundance data in the Daqu fermentation dataset to improve the training efficiency and accuracy of the Daqu detection model. In the Daqu fermentation dataset, the metabolite data, microbial abundance data, fermentation stage data, and quality grade data correspond one by one. Each group of data in the Daqu fermentation dataset represents the metabolite data, microbial abundance data, fermentation stage data, and quality grade data of Daqu at the corresponding moment.
[0052] Step 2: Train a composite neural network model based on the Daqu fermentation dataset to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade.
[0053] In this embodiment, the Daqu prediction model includes:
[0054] A metabolite data sub-model for extracting metabolite features related to the fermentation stage and quality grade from the metabolite data;
[0055] A microbial abundance data sub-model for extracting microbial abundance features related to the fermentation stage and quality grade from the microbial abundance data;
[0056] A fusion layer for weighted fusion of the metabolite features and microbial abundance features to obtain metabolite-microbe fusion features;
[0057] A Daqu fermentation stage prediction layer for predicting the fermentation stage based on the metabolite features, microbial abundance features, or metabolite-microbe fusion features;
[0058] A Daqu quality grade prediction layer for predicting the quality grade based on the metabolite features, microbial abundance features, or metabolite-microbe fusion features.
[0059] Among them, the metabolite data sub-model and the microbial abundance data sub-model can be neural network models of types such as convolutional neural network (CNN), multi-layer perceptron (MLP), long short-term memory network (LSTM), etc. The Daqu fermentation stage prediction layer and the Daqu quality grade prediction layer can be neural network models of types such as convolutional neural network (CNN), multi-layer perceptron (MLP), etc.
[0060] During the training process of the Daqu prediction model, a metabolite data sub-model is trained based on the metabolite data in the Daqu fermentation dataset and its corresponding metabolite characteristics, so that the metabolite data sub-model can extract metabolite characteristics related to the fermentation stage and quality grade from the input metabolite data.
[0061] During the training process of the Daqu prediction model, a microbial abundance data sub-model is trained based on the microbial abundance data in the Daqu fermentation dataset and its corresponding microbial abundance characteristics, so that the microbial abundance data sub-model can extract microbial abundance characteristics related to the fermentation stage and quality grade from the input microbial abundance data.
[0062] During the training process of the Daqu prediction model, feature fusion is performed on the metabolite characteristics corresponding to the metabolite data and the microbial abundance characteristics corresponding to the microbial abundance data in the Daqu fermentation dataset to obtain the metabolite-microbe fusion characteristics of each group of data in the Daqu fermentation dataset. Then, a Daqu fermentation stage prediction layer is trained based on the metabolite-microbe fusion characteristics and their corresponding fermentation stage data, and a Daqu quality grade prediction layer is trained based on the metabolite-microbe fusion characteristics and their corresponding quality grade data, so that the Daqu fermentation stage prediction layer can predict the fermentation stage of Daqu based on the input metabolite characteristics, microbial abundance characteristics, or metabolite-microbe fusion characteristics, and the Daqu quality grade prediction layer can predict the quality grade of Daqu based on the input metabolite characteristics, microbial abundance characteristics, or metabolite-microbe fusion characteristics.
[0063] In this embodiment, the composite neural network model is trained using a supervised learning method. The accuracy of the composite neural network model is determined through cross-validation, and based on the Dropout layer with regularization added, the cross-entropy loss function is used as the loss function of the network, and the model is optimized through the backpropagation of the neural network. When the loss function is less than the loss function threshold or the accuracy is greater than the accuracy threshold, the corresponding composite neural network model is used as the Daqu prediction model.
[0064] Specifically, the composite neural network model can be trained using a supervised learning method, and the stability and accuracy of the model can be evaluated through cross-validation, and an appropriate loss function, such as mean squared error (MSE), mean absolute error (MAE), or cross-entropy loss function, is used for model optimization to ensure the wide applicability and high accuracy of the model during the Daqu fermentation process. For example, the cross-entropy loss function is used as the loss function to optimize the model, and this loss function is defined based on the nn.CrossEntropyLoss() function of the PyTorch library. For the results of the change of the loss function during the training process, please refer to Figure 2 in Figure 2Among them, the abscissa represents the number of Epochs, that is, the number of times the entire training set is processed by the composite neural network model, and the ordinate represents the loss function. Meanwhile, during the training process, the generalization ability of the model can be improved and overfitting can be prevented by adding a regularized Dropout layer.
[0065] In practical applications, the Daqu fermentation dataset can be divided into a training set and a validation set in a ratio of 4:1. The composite neural network model is trained using the training set, and the accuracy of the composite neural network model is verified using the validation set. When the loss function is less than the loss function threshold or the accuracy is greater than the accuracy threshold, the Daqu prediction model can be obtained. For the results of the change in the accuracy of the validation set during the training process, please refer to Figure 3 , in Figure 3 Among them, the abscissa represents the number of Epochs, that is, the number of times the entire training set is processed by the composite neural network model, and the ordinate represents the accuracy.
[0066] Step 3: Obtain the metabolite data and / or microbial abundance data of the Daqu to be predicted, and input the metabolite data and / or microbial abundance data of the Daqu to be predicted into the Daqu prediction model to obtain the fermentation stage and quality grade of the Daqu to be predicted.
[0067] In practical applications, if it is necessary to determine the fermentation stage and quality grade of the Daqu to be predicted, first obtain the metabolite data and / or microbial abundance data of the Daqu to be predicted, and then input it into the Daqu prediction model, and the fermentation stage and quality grade of the Daqu to be predicted can be obtained. Among them, the metabolite data of the Daqu to be predicted can be obtained by gas or liquid chromatography-mass spectrometry detection, and the microbial abundance data can be obtained by genomic sequencing, PLFA (phospholipid fatty acid analysis), high-throughput sequencing, etc.
[0068] Specifically, in the Daqu prediction model, if the Daqu prediction model only receives the metabolite data of the Daqu to be predicted, the metabolite data sub-model, the Daqu fermentation stage prediction layer, and the Daqu quality grade prediction layer are activated, and the fermentation stage and quality grade of the Daqu to be predicted are predicted based on the metabolite characteristics.
[0069] If the Daqu prediction model only receives the microbial abundance data of the Daqu to be predicted, the microbial abundance data sub-model, the Daqu fermentation stage prediction layer, and the Daqu quality grade prediction layer are activated, and the fermentation stage and quality grade of the Daqu to be predicted are predicted based on the microbial abundance characteristics.
[0070] If the Daqu prediction model receives both the metabolic compound data and the microbial abundance data of the Daqu to be predicted simultaneously, it activates the metabolic compound data sub-model, the microbial abundance data sub-model, the fusion layer, the Daqu fermentation stage prediction layer, and the Daqu quality grade prediction layer, and predicts the fermentation stage and quality grade of the Daqu to be predicted based on the metabolic-microbial fusion features.
[0071] It can be understood that the Daqu prediction model can automatically adjust its calculation path according to different combinations of input data. If the input includes metabolic compound data and microbial abundance data, it activates the metabolic compound data sub-model, the microbial abundance data sub-model, and the fusion layer simultaneously. If the input only includes metabolic compound data or only includes microbial abundance data, it correspondingly activates only the corresponding sub-model for prediction, ensuring that the Daqu prediction model can flexibly adapt to different input conditions. In this way, when only the metabolic compound data or the microbial abundance data of the Daqu to be predicted can be obtained, the fermentation stage and quality grade can still be predicted, reducing the requirements for prediction. When both the metabolic compound data and the microbial abundance data of the Daqu to be predicted can be obtained simultaneously, prediction is carried out based on both types of data, thereby improving the accuracy of prediction.
[0072] In summary, the prediction method for the Daqu fermentation stage and quality grade provided in this embodiment does not rely on artificial experience. By constructing a Daqu prediction model that simultaneously predicts the fermentation stage and quality grade, based on the metabolic compound data and / or microbial abundance data, the simultaneous prediction of the Daqu fermentation stage and quality grade can be achieved, thereby improving the accuracy and consistency of determining the Daqu fermentation stage and quality grade. And in the present invention, the fermentation stage of the Daqu and its corresponding quality grade can be obtained simultaneously in one prediction, without relying on artificial determination of the correspondence between the Daqu fermentation stage and the quality grade, further improving the efficiency. By determining the fermentation stage of the Daqu during the fermentation process and the quality grade at different fermentation stages, the quality monitoring of the Daqu fermentation process is realized, and a scientific basis is provided for optimizing the Daqu fermentation process.
[0073] Figure 4 The structure diagram of a prediction device for the Daqu fermentation stage and quality grade is shown. Please refer to Figure 4 , the device includes:
[0074] An acquisition unit, configured to acquire the fermentation data of the Daqu fermentation process, establish a Daqu fermentation data set according to the fermentation data, and the fermentation data includes metabolic compound data, microbial abundance data, fermentation stage data, and quality grade data;
[0075] A training unit, configured to train a composite neural network model based on the Daqu fermentation data set to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade;
[0076] A prediction unit, configured to obtain metabolite data and / or microbial abundance data of the Daqu to be predicted, and input the metabolite data and / or microbial abundance data of the Daqu to be predicted into a Daqu prediction model to obtain the fermentation stage and quality grade of the Daqu to be predicted.
[0077] Figure 5 FIG. shows a schematic structural diagram of an electronic device. Please refer to Figure 5 , the electronic device includes a processor, a memory, and a communication bus;
[0078] The communication bus is used to implement connection communication between the processor and the memory;
[0079] The processor is configured to execute one or more programs in the memory to implement the steps of the method for predicting the fermentation stage and quality grade of Daqu as described in this embodiment.
[0080] It can be understood that since the device and electronic device for predicting the fermentation stage and quality grade of Daqu described in this embodiment are used to implement the device and electronic device for the method for predicting the fermentation stage and quality grade of Daqu described in the embodiment, for the device and electronic device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For the relevant parts, please refer to the partial description of the method, and details are not elaborated here.
Claims
1. A method for predicting the Daqu fermentation stage and quality grade, characterized in that, The method includes: Obtaining fermentation data of Daqu fermentation process, and establishing a Daqu fermentation data set according to the fermentation data, where the fermentation data includes metabolic compound data, microbial abundance data, fermentation stage data, and quality grade data; Training a composite neural network model based on the Daqu fermentation data set to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade; Obtaining the metabolic compound data and / or microbial abundance data of the Daqu to be predicted, and inputting the metabolic compound data and / or microbial abundance data of the Daqu to be predicted into the Daqu prediction model to obtain the fermentation stage and quality grade of the Daqu to be predicted.
2. The prediction method for the Daqu fermentation stage and quality grade according to claim 1, wherein The Daqu prediction model includes: A metabolic compound data sub-model for extracting metabolic compound features related to the fermentation stage and quality grade from the metabolic compound data; A microbial abundance data sub-model for extracting microbial abundance features related to the fermentation stage and quality grade from the microbial abundance data; A fusion layer for weighted fusion of the metabolic compound features and microbial abundance features to obtain metabolic-microbial fusion features; A Daqu fermentation stage prediction layer for predicting the fermentation stage based on the metabolic compound features, microbial abundance features, or metabolic-microbial fusion features; A Daqu quality grade prediction layer for predicting the quality grade based on the metabolic compound features, microbial abundance features, or metabolic-microbial fusion features.
3. The prediction method for the Daqu fermentation stage and quality grade according to claim 2, wherein Obtaining the fermentation stage and quality grade of the Daqu to be predicted includes: If the Daqu prediction model only receives the metabolic compound data of the Daqu to be predicted, activate the metabolic compound data sub-model, the Daqu fermentation stage prediction layer, and the Daqu quality grade prediction layer, and predict the fermentation stage and quality grade of the Daqu to be predicted based on the metabolic compound features; If the Daqu prediction model only receives the microbial abundance data of the Daqu to be predicted, activate the microbial abundance data sub-model, the Daqu fermentation stage prediction layer, and the Daqu quality grade prediction layer, and predict the fermentation stage and quality grade of the Daqu to be predicted based on the microbial abundance features; If the Daqu prediction model receives both the metabolic compound data and the microbial abundance data of the Daqu to be predicted, activate the metabolic compound data sub-model, the microbial abundance data sub-model, the fusion layer, the Daqu fermentation stage prediction layer, and the Daqu quality grade prediction layer, and predict the fermentation stage and quality grade of the Daqu to be predicted based on the metabolic-microbial fusion features.
4. The prediction method for the Daqu fermentation stage and quality grade according to claim 1, characterized in that, The metabolic compound data includes concentration data of one or more compounds among esters, alcohols, aldehydes, acids, ketones, pyrazines, furans, and aromatics. The esters at least include ethyl hexadecanoate and ethyl caproate. The alcohols at least include phenethyl alcohol and 2,3-butanediol. The acids at least include acetic acid and caproic acid. The pyrazines at least include 2,5-dimethylpyrazine, trimethylpyrazine, and 2,6-dimethylpyrazine.
5. The prediction method for the Daqu fermentation stage and quality grade according to claim 1, characterized in that, The microbial abundance data includes the types and quantities of microbial populations during the Daqu fermentation process. The microbial populations at least include Bacillus, Aspergillus, Saccharomyces cerevisiae, Pediococcus acidilactici, and Staphylococcus.
6. The prediction method for the Daqu fermentation stage and quality grade according to claim 1, wherein, The fermentation stage data is the time node in the Daqu fermentation cycle, and the unit of the time node is days or hours; The quality grade data includes first-class koji, second-class koji, and third-class koji.
7. The prediction method for the Daqu fermentation stage and quality grade according to claim 1, wherein The composite neural network model is trained using a supervised learning method. The accuracy of the composite neural network model is determined through cross-validation. Based on the Dropout layer with regularization added, the cross-entropy loss function is used as the loss function of the network, and the model is optimized through the backpropagation of the neural network. When the loss function is less than the loss function threshold or the accuracy is greater than the accuracy threshold, the corresponding composite neural network model is used as the koji prediction model.
8. The prediction method for the Daqu fermentation stage and quality grade according to claim 1, characterized in that, The method further includes: Data augmentation is performed on the metabolite data and microbial abundance data in the koji fermentation dataset. The data augmentation method includes random noise.
9. Prediction device for the Daqu fermentation stage and quality grade, characterized in that, The device includes: An acquisition unit for acquiring fermentation data during the koji fermentation process, and establishing a koji fermentation dataset according to the fermentation data. The fermentation data includes metabolite data, microbial abundance data, fermentation stage data, and quality grade data; A training unit for training a composite neural network model based on the koji fermentation dataset to obtain a koji prediction model for simultaneously predicting the fermentation stage and quality grade; A prediction unit for acquiring the metabolite data and / or microbial abundance data of the koji to be predicted, and inputting the metabolite data and / or microbial abundance data of the koji to be predicted into the koji prediction model to obtain the fermentation stage and quality grade of the koji to be predicted.
10. An electronic device, characterized in that, The electronic device includes a processor, a memory, and a communication bus; The communication bus is used to realize the connection and communication between the processor and the memory; The processor is used to execute one or more programs in the memory to implement the steps of the method for predicting the fermentation stage and quality grade of koji according to any one of claims 1 to 8.