Daqu fermentation control method and device based on fermentation stage and quality grade

Abstract

The present invention relates to the field of winemaking technology, and discloses a method and device for regulating Daqu fermentation based on fermentation stages and quality grades, aiming to solve the problems of delays and poor consistency and accuracy in existing regulation methods. The scheme mainly includes: obtaining metabolic compound content prediction results for future time periods based on a metabolic compound prediction model; obtaining first microbial abundance prediction results for future time periods based on a microbial abundance prediction model; obtaining first fermentation stage prediction results and first quality grade prediction results for future time periods based on a Daqu prediction model; regulating future environmental time series data based on a first difference between the first fermentation stage prediction result and future fermentation stage time series data, and a second difference between the first quality grade prediction result and the corresponding expected quality grade. The present invention avoids regulation delays, improves consistency and accuracy, and is suitable for monitoring Daqu fermentation processes.

Description

Technical Field

[0001] The present invention relates to the technical field of winemaking, and in particular to a method and device for regulating Daqu fermentation based on fermentation stages and quality grades. Background Art

[0002] Daqu (literally, "daqu") is a type of koji that undergoes a long fermentation process. It is typically made from wheat, millet, peas, and other raw materials through a series of fermentation, koji cultivation, and steaming processes. Daqu contains a variety of microorganisms, including molds, yeasts, and bacteria. Daqu is one of the primary fermentation agents used in traditional Chinese liquor production and is widely used in the brewing of Luzhou-flavor and Maotai-flavor liquors. Daqu fermentation refers to the process of using Daqu as a saccharifying and fermenting agent in the production of baijiu.

[0003] The fermentation process of Daqu is affected by many factors, including changes in the fermentation environment and changes in the chemical composition and microbial community of Daqu itself. In order to monitor, control and optimize the fermentation process of Daqu, the solution commonly adopted in the prior art is to conduct regular sampling and testing of Daqu, judge whether the fermentation of Daqu is abnormal based on the sampling and testing results, and regulate the fermentation environment based on experience by technical personnel. This method is labor-intensive and difficult to grasp the changes in the fermentation process in real time. In addition, the regulation of the fermentation environment relies on manual experience, and the consistency and accuracy are poor. In addition, if this method detects that the fermentation of Daqu is abnormal, a fermentation quality problem has already occurred, and the early prediction of the fermentation quality of Daqu cannot be achieved, resulting in a delay in environmental regulation. Summary of the Invention

[0004] The present invention aims to solve the problems of delay, poor consistency and accuracy in the existing Daqu fermentation environment control method, and proposes a Daqu fermentation control method and device based on fermentation stage and quality grade.

[0005] The technical solution adopted by the present invention to solve the above technical problems is:

[0006] In a first aspect, the present invention provides a method for regulating Daqu fermentation based on fermentation stage and quality grade, the method comprising:

[0007] Regularly obtain metabolic compound data and environmental data during the Daqu fermentation process, and record the actual fermentation stage data of the Daqu, generate metabolic compound historical time series data based on the metabolic compound data, generate environmental historical time series data based on the environmental data, and generate fermentation stage historical time series data based on the fermentation stage data;

[0008] Obtaining a prediction result of the metabolic compound content in a future time period based on the metabolic compound historical time series data, the environmental historical time series data, and the fermentation stage historical time series data and based on a pre-trained metabolic compound prediction model;

[0009] Determining future environmental time series data and future fermentation stage time series data for a future time period, and obtaining a first microbial abundance prediction result for the future time period based on the metabolic compound content prediction result for the future time period, the future environmental time series data, and the future fermentation stage time series data and a pre-trained microbial abundance prediction model;

[0010] Obtaining a first fermentation stage prediction result and a first quality grade prediction result for a future period according to the metabolic compound content prediction result and the first microbial abundance prediction result and based on a pre-trained Daqu prediction model;

[0011] Based on the first difference between the prediction result of the first fermentation stage and the future time series data of the fermentation stage, and the second difference between the prediction result of the first quality grade and the corresponding expected quality grade, it is judged whether the fermentation of Daqu is normal. If the fermentation of Daqu is abnormal, the future time series data of the environment in the future time period is adjusted according to the first difference and the second difference to minimize the first difference and the second difference.

[0012] Furthermore, the training methods of the metabolic compound prediction model, the microbial abundance prediction model and the Daqu prediction model include:

[0013] Obtaining fermentation data at various time points during the fermentation process of Daqu, wherein the fermentation data includes environmental data, fermentation stage data, metabolic compound data, microbial abundance data, and quality grade data;

[0014] The metabolic compound data at historical time points and their corresponding environmental data and fermentation stage data are used as input features, and the metabolic compound data at corresponding future time points are used as output labels to train a time series neural network to obtain a metabolic compound prediction model.

[0015] The environmental data, fermentation stage data, and metabolic compound data at each time point are used as input features, and the corresponding microbial abundance data are used as output labels to train a time series neural model to obtain a microbial abundance prediction model.

[0016] The metabolic compound data and microbial abundance data at each time point were used as input features, and the corresponding fermentation stage data and quality grade data were used as output labels. A composite neural network model was trained to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade.

[0017] Furthermore, the environmental data includes temperature data, humidity data and air oxygen content data; the fermentation stage data is the time node in the Daqu fermentation cycle, and the unit of the time node is day or hour.

[0018] Furthermore, the metabolic compound data includes content data of one or more compounds among esters, alcohols, aldehydes, acids, ketones, pyrazines, furans and aromatics, the esters include at least ethyl hexadecanoate and ethyl hexanoate, the alcohols include at least phenylethanol and 2,3-butanediol, the acids include at least acetic acid and hexanoic acid, and the pyrazines include at least 2,5-dimethylpyrazine, trimethylpyrazine and 2,6-dimethylpyrazine.

[0019] Furthermore, the microbial abundance data includes the types and quantities of microbial populations during the Daqu fermentation process, and the microbial populations include at least Bacillus, Aspergillus, Saccharomyces cerevisiae, Pediococcus acidilactici and Staphylococcus.

[0020] Furthermore, fermentation data at each time point during the fermentation process of Daqu is obtained by sampling according to a preset sampling frequency, and the preset sampling frequency is half a day, one day, two days or three days.

[0021] Furthermore, the method further comprises:

[0022] Obtain the current time series data of metabolic compounds, the current time series data of the environment, and the current time series data of the fermentation stage during the current period of the Daqu fermentation process;

[0023] Obtaining a second microbial abundance prediction result for the current time period according to the current time series data of the metabolic compound, the current time series data of the environment, and the current time series data of the fermentation stage and based on the microbial abundance prediction model;

[0024] Obtaining a second fermentation stage prediction result and a second quality grade prediction result for the current period based on the current time series data of the metabolic compound and the second microbial abundance prediction result and on the basis of the Daqu prediction model;

[0025] Whether the Daqu fermentation is normal is determined based on the third difference between the second fermentation stage prediction result and the current time series data of the fermentation stage, and the fourth difference between the second quality grade prediction result and the corresponding expected quality grade.

[0026] Furthermore, the method further comprises:

[0027] If it is determined that the Daqu fermentation is abnormal, an alarm will be issued for the Daqu fermentation abnormality.

[0028] Furthermore, regulating the future environmental time series data of the future period according to the first difference and the second difference includes:

[0029] Constructing a control model based on the microbial abundance prediction model and the Daqu prediction model and based on an evolutionary algorithm, wherein the evolutionary algorithm is a genetic algorithm, an ant colony algorithm, or a simulated annealing algorithm;

[0030] When the fermentation of Daqu is abnormal, the historical time series data of metabolic compounds, the historical time series data of the environment, the historical time series data of the fermentation stage, the future time series data of the fermentation stage and the expected quality level are input into the control model for optimization, and the target environmental data corresponding to the minimum of the first difference and the second difference are obtained, and the future time series data of the environment in the future time period are regulated according to the target environmental data.

[0031] In a second aspect, the present invention provides a device for regulating and controlling Daqu fermentation based on fermentation stage and quality grade, the device comprising:

[0032] An acquisition unit is used to regularly acquire metabolic compound data and environmental data during the fermentation process of Daqu, and record the actual fermentation stage data of Daqu, generate metabolic compound historical time series data based on the metabolic compound data, generate environmental historical time series data based on the environmental data, and generate fermentation stage historical time series data based on the fermentation stage data;

[0033] A first prediction unit is configured to obtain a prediction result of the content of the metabolic compound in a future time period based on the metabolic compound historical time series data, the environmental historical time series data, and the fermentation stage historical time series data and based on a pre-trained metabolic compound prediction model;

[0034] a second prediction unit, configured to determine future time series data of the environment and future time series data of the fermentation stage for a future time period, and obtain a first microbial abundance prediction result for the future time period based on the metabolic compound content prediction result for the future time period, the future time series data of the environment and the future time series data of the fermentation stage and a pre-trained microbial abundance prediction model;

[0035] a third prediction unit, configured to obtain a first fermentation stage prediction result and a first quality grade prediction result in a future time period based on the metabolic compound content prediction result and the first microbial abundance prediction result and a pre-trained Daqu prediction model;

[0036] The control unit is used to determine whether the fermentation of Daqu is normal based on a first difference between the prediction result of the first fermentation stage and the future time series data of the fermentation stage, and a second difference between the prediction result of the first quality grade and the corresponding expected quality grade. If the fermentation of Daqu is abnormal, the future time series data of the environment in the future time period is controlled based on the first difference and the second difference to minimize the first difference and the second difference.

[0037] The present invention has the following beneficial effects: the fermentation stage and quality grade-based daqu fermentation control method and device provided by the present invention predicts the fermentation stage and quality grade of daqu fermentation in the future using pre-trained metabolic compound prediction models, microbial abundance prediction models, and daqu prediction models. Daqu fermentation anomalies are then determined and environmental control is performed based on the difference between the predicted results and the actual fermentation stage and expected quality grade. This invention enables the early prediction of daqu fermentation quality, avoids delays in environmental control, and ensures the quality of daqu fermentation. Furthermore, the invention does not rely on manual experience, thereby improving the consistency and accuracy of environmental control. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic flow chart of a method for regulating Daqu fermentation based on fermentation stage and quality grade provided in an embodiment;

[0039] Figure 2 A schematic diagram of the structure of a Daqu fermentation control device based on fermentation stage and quality grade is provided in the embodiment. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solution of this embodiment will be clearly and completely described below in conjunction with the drawings in this embodiment.

[0041] In some of the processes described in the specification of the present invention and the figures above, multiple operations are included in a specific order. However, it should be understood that these operations may not be performed in the order in which they are presented herein or may be performed in parallel. The sequence numbers of the operations are merely used to distinguish between different operations and do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be performed sequentially or in parallel.

[0042] Currently, the most common approach for monitoring and controlling the fermentation process of Daqu is to periodically sample and test the Daqu. The results are used to determine if the fermentation is abnormal, and technicians then adjust the fermentation environment based on their experience. The inventors have discovered that this approach suffers from delays and poor consistency and accuracy.

[0043] Based on this, the technical solution of the present invention is proposed. In the present invention, metabolic compound data and environmental data in the fermentation process of Daqu are regularly obtained, and the actual fermentation stage data of Daqu is recorded, metabolic compound historical time series data is generated according to the metabolic compound data, environmental historical time series data is generated according to the environmental data, and fermentation stage historical time series data is generated according to the fermentation stage data; according to the metabolic compound historical time series data, the environmental historical time series data and the fermentation stage historical time series data and based on the pre-trained metabolic compound prediction model, the metabolic compound content prediction result of the future time period is obtained; the future environmental time series data and the future fermentation stage future time series data of the future time period are determined, and according to the metabolic compound content prediction result of the future time period, the environmental future time series data and the fermentation stage future time series data According to the future time series data of the fermentation stage and based on a pre-trained microbial abundance prediction model, a first microbial abundance prediction result for the future time period is obtained; according to the metabolic compound content prediction result and the first microbial abundance prediction result and based on a pre-trained Daqu prediction model, a first fermentation stage prediction result and a first quality grade prediction result for the future time period are obtained; according to a first difference between the first fermentation stage prediction result and the future time series data of the fermentation stage, and a second difference between the first quality grade prediction result and the corresponding expected quality grade, whether the Daqu fermentation is normal is judged; if the Daqu fermentation is abnormal, the environmental future time series data for the future time period is regulated according to the first difference and the second difference to minimize the first difference and the second difference.

[0044] Specifically, during the Daqu fermentation process, the present invention regularly acquires metabolic compound data, environmental data, and fermentation stage data, and generates corresponding historical time series data based on this data. Then, based on the pre-trained metabolic compound prediction model, microbial abundance prediction model, and Daqu prediction model, the fermentation stage and quality grade of Daqu fermentation in the future time period are predicted. Based on the difference between the predicted results and the actual fermentation stage and expected quality grade, abnormal Daqu fermentation is judged and the environment is regulated. The present invention realizes the early prediction of Daqu fermentation quality, avoids the delay of environmental regulation, and ensures the quality of Daqu fermentation. At the same time, the present invention does not rely on manual experience, and improves the consistency and accuracy of environmental regulation.

[0045] The technical solution of this embodiment will be clearly and completely described below in conjunction with the drawings in this embodiment. Obviously, the described embodiment is only a part of the embodiments of the present invention, rather than all the embodiments.

[0046] Figure 1 A schematic diagram of a method for regulating Daqu fermentation based on fermentation stage and quality grade is shown. Figure 1 , the method comprises the following steps:

[0047] Step 1: Regularly obtain metabolic compound data and environmental data during the fermentation process of Daqu, and record the actual fermentation stage data of Daqu, generate metabolic compound historical time series data based on the metabolic compound data, generate environmental historical time series data based on the environmental data, and generate fermentation stage historical time series data based on the fermentation stage data.

[0048] In this embodiment, the environmental data includes temperature, humidity, and air oxygen content data. After obtaining the environmental data, it can be normalized using the minimum-maximum normalization method. The maximum and minimum values ​​are obtained based on all the data in the data set. For example, if the temperature and humidity data at a certain point in time are [35, 32], and the maximum temperature data is 60°C and the minimum is 30°C, and the maximum humidity data is 36 and the minimum is 9, then the temperature and humidity data [35, 32] will be normalized to [0.14, 0.85].

[0049] In this embodiment, the fermentation stage data is a time node in the Daqu fermentation cycle, and the unit of the time node is day or hour. That is, the fermentation stage data is used to identify a specific moment in the current Daqu fermentation cycle, and can be preliminarily represented by days or hours. For example, if the fermentation time is the 1st day, it is preliminarily represented as 1, and if the fermentation time is the 16th day, it is preliminarily represented as 16. In this embodiment, the Daqu fermentation cycle is 30 days, so the preliminary data of the Daqu fermentation stage only includes data from 1 to 30 days; this embodiment can divide the 30-day preliminary data into 3 groups according to the time development as 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 metabolite compound data represents compound data related to Daqu fermentation, and the compound data can be compound content data. After obtaining the metabolite compound data, the metabolite compound data can be normalized using a minimum-maximum normalization method. For example, the metabolite compound data includes one or more compound content data from esters, alcohols, aldehydes, acids, pyrazines, furans, and aromatics. The esters include at least ethyl hexadecanoate and ethyl hexanoate, the alcohols include at least 2,3-butanediol, and the pyrazines include at least 2,5-dimethylpyrazine, trimethylpyrazine, and 2,6-dimethylpyrazine. The unit of compound content data is ppm. For example, the metabolite 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 metabolite compound data is 4 ppm and the minimum value is 0 ppm, the metabolite compound data after normalization processing is shown as: [0.254375002, 0.090048051, 0.071857188, 0.070598011, 0, 0.017947409, …].

[0051] In practical applications, metabolic compound data from the Daqu fermentation process can be collected regularly at a preset frequency to generate corresponding historical time series data for the metabolic compounds. Environmental data can also be monitored regularly at a preset frequency to generate corresponding historical time series data for the environmental environment. Furthermore, data from the actual Daqu fermentation stages can be recorded to generate corresponding historical time series data for the fermentation stages. The preset sampling frequency can be half a day, one day, two days, three days, or other sampling frequencies determined by experts in the field.

[0052] Step 2: Obtain a prediction result of the metabolic compound content in a future time period based on the metabolic compound historical time series data, the environmental historical time series data, and the fermentation stage historical time series data and a pre-trained metabolic compound prediction model.

[0053] In this embodiment, the training method of the metabolic compound prediction model includes:

[0054] Obtain fermentation data at each time point during the Daqu fermentation process, including environmental data, fermentation stage data, and metabolic compound data; use the metabolic compound data at historical time points and their corresponding environmental data and fermentation stage data as input features, and the metabolic compound data at corresponding future time points as output labels, train a time series neural network, and obtain a metabolic compound prediction model;

[0055] In practical applications, first, the metabolic compound data at historical time points and their corresponding environmental data and fermentation stage data are used as input features, and the metabolic compound data at corresponding future time points are used as output labels of the time series neural network model to construct a training set and a validation set. Then, based on the training set and using a supervised learning method to train the time series neural network model, the output features of the time series neural network model and the corresponding output labels are compared, the corresponding loss function is calculated, and the model is optimized by back propagation. Finally, the time series neural network model is verified based on the validation set and the cross-validation test set. When the loss function of the time series neural network model is minimized and the prediction error of the validation set or the cross-validation test set is less than the error threshold, the corresponding time series neural network model is used as the metabolic compound prediction model.

[0056] After obtaining the historical time series data of metabolic compounds, the historical time series data of the environment and the historical time series data of the fermentation stage, the historical time series data of metabolic compounds, the historical time series data of the environment and the historical time series data of the fermentation stage are input into the trained metabolic compound prediction model to obtain the metabolic compound content of Daqu in the future period.

[0057] The future time period refers to a preset time period after the current time point, for example, one day, two days, three days, etc. after the current time point. The specific time length can be set according to actual needs.

[0058] Step 3: Determine the future time series data of the environment and the future time series data of the fermentation stage for the future time period, and obtain the first microbial abundance prediction result for the future time period based on the metabolic compound content prediction result of the future time period, the future time series data of the environment and the future time series data of the fermentation stage and based on the pre-trained microbial abundance prediction model.

[0059] It can be understood that the environmental future time series data is the environmental time series data of a preset future period, and the fermentation stage future time series data is the fermentation stage time series data of an actual future period.

[0060] In this embodiment, the training method of the microbial abundance prediction model includes:

[0061] Fermentation data at each time point during the Daqu fermentation process were obtained. The fermentation data included environmental data, fermentation stage data, metabolic compound data, and microbial abundance data. The environmental data, fermentation stage data, and metabolic compound data at each time point were used as input features, and the corresponding microbial abundance data were used as output labels to train a time series neural model and obtain a microbial abundance prediction model.

[0062] In practical applications, first, environmental data, fermentation stage data, and metabolic compound data at each time point are used as input features, and the corresponding microbial abundance data is used as the output label of the time series neural network model to construct a training set and a validation set. Then, based on the training set, the time series neural network model is trained using a supervised learning method. The output features of the time series neural network model are compared with the corresponding output labels, the corresponding loss function is calculated, and the model is optimized through backpropagation. Finally, the time series neural network model is verified based on the validation set. When the loss function of the time series neural network model is minimized and the prediction error is less than the error threshold, the corresponding time series neural network model is used as the microbial abundance prediction model.

[0063] After obtaining the predicted results of the metabolic compound content in the future time period, the future time series data of the environment, and the future time series data of the fermentation stage, the predicted results of the metabolic compound content in the future time period, the future time series data of the environment, and the future time series data of the fermentation stage are input into the trained microbial abundance prediction model to obtain the first microbial abundance prediction result in the future time period.

[0064] Step 4: Obtain a first fermentation stage prediction result and a first quality grade prediction result for a future period based on the metabolic compound content prediction result and the first microbial abundance prediction result and a pre-trained Daqu prediction model.

[0065] In this embodiment, the Daqu prediction model includes a metabolic compound data sub-model, a microbial abundance data sub-model, a fusion layer, a Daqu fermentation stage prediction layer, and a Daqu quality grade prediction layer. The training method of the Daqu prediction model includes:

[0066] First, fermentation data at each time point during the Daqu fermentation process is obtained. The fermentation data includes fermentation stage data, metabolite compound data, microbial abundance data, and quality grade data. Then, a metabolite compound data sub-model is trained based on the metabolite compound data and their corresponding metabolite compound features in the Daqu fermentation dataset, so that the metabolite compound data sub-model can extract metabolite compound features related to the fermentation stage and quality grade from the input metabolite compound data. Then, a microbial abundance data sub-model is trained based on the microbial abundance data and their corresponding microbial abundance features in the Daqu fermentation dataset, so that the microbial abundance data sub-model can extract microbial abundance features related to the fermentation stage and quality grade from the input microbial abundance data. Then, feature fusion is performed on the metabolic compound features corresponding to the metabolic compound data in the Daqu fermentation dataset and the microbial abundance features corresponding to the microbial abundance data to obtain the metabolic-microbial fusion features of each group of data in the Daqu fermentation dataset. Finally, the Daqu fermentation stage prediction layer is trained based on the metabolic-microbial fusion features and its corresponding fermentation stage data, and the Daqu quality grade prediction layer is trained based on the metabolic-microbial fusion features and its corresponding quality grade data, so that the Daqu fermentation stage prediction layer can predict the fermentation stage of Daqu according to the input metabolic-microbial fusion features, and the Daqu quality grade prediction layer can predict the quality grade of Daqu according to the input metabolic-microbial fusion features.

[0067] After obtaining the metabolic compound content prediction results and the first microbial abundance prediction results, the metabolic compound content prediction results and the first microbial abundance prediction results are input into the trained Daqu prediction model to obtain the first fermentation stage prediction results and the first quality grade prediction results for the future period.

[0068] In this example, the microbial abundance data includes the species and abundance of microbial populations during the Daqu fermentation process, including at least Bacillus, Aspergillus, Saccharomyces cerevisiae, Pediococcus acidilactici, and Staphylococcus. For example, the microbial abundance data at a specific moment is shown as: [0.01207594, 0.017445474, 0.022036947, ...]. The microbial abundance data is already normalized, so normalization is not required. In each set of microbial abundance data, the sum of the abundances of the various microorganisms is 1.

[0069] In this embodiment, the quality grade data represents the quality grade of Daqu, which can include first-grade, second-grade, and third-grade. That is, there are three types of quality grade data, with 0 representing first-grade, 1 representing second-grade, and 2 representing third-grade. The quality grade data can be divided based on actual production needs or according to grading standards used by experts in the field.

[0070] In this embodiment, the fermentation data at each time point used for model training is sampled according to a preset sampling frequency, which is half a day, one day, two days, or three days.

[0071] In this embodiment, the temporal neural network model may be, but is not limited to, a long short-term memory network (LSTM), a gated recurrent unit (GRU), or a temporal neural network model based on an attention mechanism.

[0072] Step 5: Based on the first difference between the prediction result of the first fermentation stage and the future time series data of the fermentation stage, and the second difference between the prediction result of the first quality grade and the corresponding expected quality grade, determine whether the fermentation of Daqu is normal. If the fermentation of Daqu is abnormal, the future time series data of the environment in the future time period is adjusted according to the first difference and the second difference to minimize the first difference and the second difference.

[0073] After the prediction results of the first fermentation stage and the first quality grade are obtained, if the first difference between the prediction results of the first fermentation stage in the future time period and the future time series data of the actual fermentation stage of Daqu is greater than the preset fermentation stage difference threshold, or the second difference between the prediction results of the first quality grade in the future time period and the corresponding expected Daqu quality grade is greater than the preset fermentation stage difference threshold, it indicates that the Daqu fermentation is abnormal. At this time, an abnormal alarm is issued for the fermentation situation of this batch of Daqu, and environmental control is performed.

[0074] In this embodiment, the preset fermentation stage difference threshold can be set according to factors such as the Daqu fermentation cycle and quality requirements in the actual production environment, or it can be set by experts in the field through statistical analysis of historical data.

[0075] In this embodiment, a warning threshold value that is one level different from the expectation can be set. For example, if the future time period should actually belong to the second stage of the Daqu fermentation stage, but the predicted result is the first stage, then a warning will be issued for the Daqu fermentation situation; or, if the expected Daqu level is the first level, but the predicted Daqu level is the second level, then a warning will be issued for the Daqu fermentation situation.

[0076] In this embodiment, regulating the future environmental time series data in the future period according to the first difference and the second difference includes:

[0077] A control model is constructed based on the microbial abundance prediction model and the Daqu prediction model and based on an evolutionary algorithm, wherein the evolutionary algorithm is a genetic algorithm, an ant colony algorithm or a simulated annealing algorithm; when the Daqu fermentation is abnormal, the historical time series data of metabolic compounds, the historical time series data of the environment, the historical time series data of the fermentation stage, the future time series data of the fermentation stage and the expected quality grade are input into the control model for optimization, and the target environmental data corresponding to the minimum of the first difference and the second difference are obtained, and the future time series data of the environment in the future time period are regulated according to the target environmental data.

[0078] Among them, the optimization process of the regulation model is specifically as follows: all possible metabolic compound contents and microbial abundances are clearly input into the Daqu prediction model, and the fermentation stage data and quality grade are predicted. Then, the predicted fermentation stage data and quality grade are compared with the corresponding future time series data of the fermentation stage and the expected quality grade respectively, and the fitness is calculated according to the fitness function. According to the fitness, the optimal metabolic compound content and the optimal microbial abundance are determined, and the optimal metabolic compound content, future time series data of the fermentation stage and all possible environmental data are input into the microbial abundance prediction model to predict the microbial abundance. Then, the predicted microbial abundance is compared with the optimal microbial abundance, and the fitness is calculated according to the fitness function. According to the fitness, the optimal environmental data is determined and used as the target environmental data.

[0079] In this embodiment, a genetic algorithm is used to implement the optimization process of the above-mentioned control model. The corresponding Python code is shown as follows:

[0080] # Initialization phase

[0081] # Input the current metabolic compound data, environmental data and actual fermentation stage data to define the length of the future time period data_metabolite:np.ndarray#Current metabolic compound data

[0082] data_environment:np.ndarray#Current environment data

[0083] data_stage:np.ndarray#Current actual fermentation stage data

[0084] time_future:int#Future time length

[0085] data_stage_future_real:np.ndarray#actual fermentation stage of Daqu in the future

[0086] data_grade_future_target:np.ndarray#Expected quality level of Daqu in the future

[0087] #Define model

[0088] MetabolitePredictionModel:Callable#Metabolic compound prediction model

[0089] MicrobePredictionModel:Callable#Microbial abundance prediction model

[0090] StageGradePredictionModel:Callable#fermentation stage and quality grade prediction model

[0091] #Genetic algorithm parameters

[0092] population_size=50#population size

[0093] generations=100#maximum number of iterations

[0094] # Initialize the population

[0095] population = initialize the environment variable population (population_size, time_future)

[0096] #Fitness function

[0097] def fitness_function(candidate_environment):

[0098] data_metabolite_future=MetabolitePredictionModel(data_metabolite,data_environment,data_stage)

[0099] data_microbes_future=MicrobePredictionModel(data_metabolite_future, candidate_environment, data_stage_future_real)

[0100] data_stage_future_pred,data_grade_future_pred=StageGradePredictionModel(data_metabolite_future,data_microbes_future)

[0101] stage_error = Calculate the future stage prediction error (data_stage_future_pred, data_stage_future_real)

[0102] grade_error = Calculate Quality Grade Prediction Error (data_grade_future_pred, data_grade_future_target)

[0103] return comprehensive fitness (stage_error, grade_error)

[0104] #Genetic algorithm main loop

[0105] for generation in range(generations):

[0106] #Evaluate the current population fitness

[0107] fitness_scores=[fitness_function(candidate)for candidate inpopulation]

[0108] #Selection operation (e.g. tournament selection)

[0109] selected_population=select(population,fitness_scores)

[0110] #Crossover operation (such as single point crossover or uniform crossover)

[0111] offspring=cross(selected_population)

[0112] #Mutation operation (such as random perturbation)

[0113] Offspring = Variation (offspring)

[0114] #Update population

[0115] population = update population (offspring, selected_population)

[0116] # Output current generation information

[0117] Output the current generation's optimal fitness information (population, fitness_scores)

[0118] # Output the optimal solution

[0119] best_environment = Get the best environment (population, fitness_scores)

[0120] In this embodiment, the method further includes:

[0121] Obtain current time series data of metabolic compounds, current time series data of the environment, and current time series data of the fermentation stage during the current period of the Daqu fermentation process; obtain a second microbial abundance prediction result for the current period based on the current time series data of the metabolic compounds, current time series data of the environment, and current time series data of the fermentation stage and based on the microbial abundance prediction model; obtain a second fermentation stage prediction result and a second quality grade prediction result for the current period based on the current time series data of the metabolic compounds and the second microbial abundance prediction result and based on the Daqu prediction model; determine whether the Daqu fermentation is normal based on a third difference between the second fermentation stage prediction result and the current time series data of the fermentation stage, and a fourth difference between the second quality grade prediction result and the corresponding expected quality grade.

[0122] It will be understood that this embodiment also determines the second fermentation stage prediction result and the second quality grade prediction result for the current time period. Specifically, first, after obtaining the current time series data of the metabolic compounds, the current time series data of the environment, and the current time series data of the fermentation stage during the current time period of the Daqu fermentation process, the current time series data of the metabolic compounds, the current time series data of the environment, and the current time series data of the fermentation stage are input into the microbial abundance prediction model to obtain the second microbial abundance prediction result for the current time period. Then, the current time series data of the metabolic compounds and the second microbial abundance prediction result are input into the Daqu prediction model to obtain the second fermentation stage prediction result and the second quality grade prediction result for the current time period.

[0123] After predicting and obtaining the second fermentation stage prediction result and the second quality grade prediction result of the current time period, if the third difference between the second fermentation stage prediction result of the current time period and the current time series data of the actual fermentation stage of Daqu is greater than the preset fermentation stage difference threshold, or the fourth difference between the second quality grade prediction result of the current time period and the corresponding expected Daqu quality grade is greater than the preset fermentation stage difference threshold, it indicates that the Daqu fermentation is abnormal, and an abnormal alarm is issued for the fermentation situation of this batch of Daqu.

[0124] In summary, the fermentation stage and quality grade-based Daqu fermentation control method provided in this embodiment predicts the fermentation stage and quality grade of Daqu fermentation in future time periods using pre-trained metabolic compound prediction models, microbial abundance prediction models, and Daqu prediction models. It then determines Daqu fermentation anomalies and performs environmental control based on the difference between the predicted results and the actual fermentation stage and expected quality grade. This embodiment achieves early prediction of Daqu fermentation quality, avoids delays in environmental control, and ensures Daqu fermentation quality. Furthermore, this embodiment does not rely on manual experience, improving the consistency and accuracy of environmental control.

[0125] Figure 2 A schematic diagram of a Daqu fermentation control device based on fermentation stage and quality grade is shown. Figure 2 , the device comprises:

[0126] An acquisition unit is used to regularly acquire metabolic compound data and environmental data during the fermentation process of Daqu, and record the actual fermentation stage data of Daqu, generate metabolic compound historical time series data based on the metabolic compound data, generate environmental historical time series data based on the environmental data, and generate fermentation stage historical time series data based on the fermentation stage data;

[0127] A first prediction unit is configured to obtain a prediction result of the content of the metabolic compound in a future time period based on the metabolic compound historical time series data, the environmental historical time series data, and the fermentation stage historical time series data and based on a pre-trained metabolic compound prediction model;

[0128] a second prediction unit, configured to determine future time series data of the environment and future time series data of the fermentation stage for a future time period, and obtain a first microbial abundance prediction result for the future time period based on the metabolic compound content prediction result for the future time period, the future time series data of the environment and the future time series data of the fermentation stage and a pre-trained microbial abundance prediction model;

[0129] a third prediction unit, configured to obtain a first fermentation stage prediction result and a first quality grade prediction result in a future time period based on the metabolic compound content prediction result and the first microbial abundance prediction result and a pre-trained Daqu prediction model;

[0130] The control unit is used to determine whether the fermentation of Daqu is normal based on a first difference between the prediction result of the first fermentation stage and the future time series data of the fermentation stage, and a second difference between the prediction result of the first quality grade and the corresponding expected quality grade. If the fermentation of Daqu is abnormal, the future time series data of the environment in the future time period is controlled based on the first difference and the second difference to minimize the first difference and the second difference.

[0131] It can be understood that since the Daqu fermentation control device based on fermentation stage and quality grade described in this embodiment is a device for implementing the Daqu fermentation control method based on fermentation stage and quality grade described in the embodiment, for the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple. For relevant matters, please refer to the partial description of the method, which will not be repeated here.

Claims

1. A method for regulating and controlling Daqu fermentation based on fermentation stage and quality grade, characterized in that: The method comprises: Regularly obtain metabolic compound data and environmental data during the Daqu fermentation process, and record the actual fermentation stage data of the Daqu, generate metabolic compound historical time series data based on the metabolic compound data, generate environmental historical time series data based on the environmental data, and generate fermentation stage historical time series data based on the fermentation stage data; Obtaining a prediction result of the metabolic compound content in a future time period based on the metabolic compound historical time series data, the environmental historical time series data, and the fermentation stage historical time series data and based on a pre-trained metabolic compound prediction model; Determining future environmental time series data and future fermentation stage time series data for a future time period, and obtaining a first microbial abundance prediction result for the future time period based on the metabolic compound content prediction result for the future time period, the future environmental time series data, and the future fermentation stage time series data and a pre-trained microbial abundance prediction model; Obtaining a first fermentation stage prediction result and a first quality grade prediction result for a future period according to the metabolic compound content prediction result and the first microbial abundance prediction result and based on a pre-trained Daqu prediction model; Based on the first difference between the prediction result of the first fermentation stage and the future time series data of the fermentation stage, and the second difference between the prediction result of the first quality grade and the corresponding expected quality grade, it is judged whether the fermentation of Daqu is normal. If the fermentation of Daqu is abnormal, the future time series data of the environment in the future time period is adjusted according to the first difference and the second difference to minimize the first difference and the second difference.

2. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 1, characterized in that: The training methods of the metabolic compound prediction model, the microbial abundance prediction model and the Daqu prediction model include: Obtaining fermentation data at each time point during the Daqu fermentation process, wherein the fermentation data includes environmental data, fermentation stage data, metabolic compound data, microbial abundance data, and quality grade data; The metabolic compound data at historical time points and their corresponding environmental data and fermentation stage data are used as input features, and the metabolic compound data at corresponding future time points are used as output labels to train a time series neural network to obtain a metabolic compound prediction model. The environmental data, fermentation stage data, and metabolic compound data at each time point are used as input features, and the corresponding microbial abundance data are used as output labels to train a time series neural model to obtain a microbial abundance prediction model. The metabolic compound data and microbial abundance data at each time point were used as input features, and the corresponding fermentation stage data and quality grade data were used as output labels. A composite neural network model was trained to obtain a Daqu prediction model for simultaneously predicting the fermentation stage and quality grade.

3. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 2, characterized in that, The environmental data includes temperature data, humidity data and air oxygen content data; the fermentation stage data is the time node in the Daqu fermentation cycle, and the unit of the time node is day or hour.

4. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 2, characterized in that, The metabolic compound data includes content data of one or more compounds among esters, alcohols, aldehydes, acids, ketones, pyrazines, furans and aromatics, wherein the esters include at least ethyl hexadecanoate and ethyl hexanoate, the alcohols include at least phenylethanol and 2,3-butanediol, the acids include at least acetic acid and hexanoic acid, and the pyrazines include at least 2,5-dimethylpyrazine, trimethylpyrazine and 2,6-dimethylpyrazine.

5. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 2, characterized in that, The microbial abundance data includes the types and quantities of microbial populations during the Daqu fermentation process, and the microbial populations include at least Bacillus, Aspergillus, Saccharomyces cerevisiae, Pediococcus acidilactici and Staphylococcus.

6. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 2, characterized in that: The fermentation data at each time point during the fermentation of Daqu are obtained by sampling according to a preset sampling frequency, which is half a day, one day, two days or three days.

7. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 1, characterized in that: The method further comprises: Obtain the current time series data of metabolic compounds, the current time series data of the environment, and the current time series data of the fermentation stage during the current period of the Daqu fermentation process; Obtaining a second microbial abundance prediction result for the current time period according to the current time series data of the metabolic compound, the current time series data of the environment, and the current time series data of the fermentation stage and based on the microbial abundance prediction model; Obtaining a second fermentation stage prediction result and a second quality grade prediction result for the current period based on the current time series data of the metabolic compound and the second microbial abundance prediction result and on the basis of the Daqu prediction model; Whether the Daqu fermentation is normal is determined based on the third difference between the second fermentation stage prediction result and the current time series data of the fermentation stage, and the fourth difference between the second quality grade prediction result and the corresponding expected quality grade.

8. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 1 or 7, characterized in that: The method further comprises: If it is determined that the Daqu fermentation is abnormal, an alarm will be issued for the Daqu fermentation abnormality.

9. The Daqu fermentation control method based on fermentation stage and quality grade according to claim 1, characterized in that: Adjusting the future time series data of the environment in the future period according to the first difference and the second difference includes: Constructing a control model based on the microbial abundance prediction model and the Daqu prediction model and based on an evolutionary algorithm, wherein the evolutionary algorithm is a genetic algorithm, an ant colony algorithm, or a simulated annealing algorithm; When the fermentation of Daqu is abnormal, the historical time series data of metabolic compounds, the historical time series data of the environment, the historical time series data of the fermentation stage, the future time series data of the fermentation stage and the expected quality level are input into the control model for optimization, and the target environmental data corresponding to the minimum of the first difference and the second difference are obtained, and the future time series data of the environment in the future time period are regulated according to the target environmental data.

10. A Daqu fermentation control device based on fermentation stage and quality grade, characterized in that: The device comprises: An acquisition unit is used to regularly acquire metabolic compound data and environmental data during the fermentation process of Daqu, and record the actual fermentation stage data of Daqu, generate metabolic compound historical time series data based on the metabolic compound data, generate environmental historical time series data based on the environmental data, and generate fermentation stage historical time series data based on the fermentation stage data; A first prediction unit is configured to obtain a prediction result of the content of the metabolic compound in a future time period based on the metabolic compound historical time series data, the environmental historical time series data, and the fermentation stage historical time series data and based on a pre-trained metabolic compound prediction model; a second prediction unit, configured to determine future time series data of the environment and future time series data of the fermentation stage for a future time period, and obtain a first microbial abundance prediction result for the future time period based on the metabolic compound content prediction result for the future time period, the future time series data of the environment and the future time series data of the fermentation stage and a pre-trained microbial abundance prediction model; a third prediction unit, configured to obtain a first fermentation stage prediction result and a first quality grade prediction result in a future time period based on the metabolic compound content prediction result and the first microbial abundance prediction result and a pre-trained Daqu prediction model; The control unit is used to determine whether the fermentation of Daqu is normal based on a first difference between the prediction result of the first fermentation stage and the future time series data of the fermentation stage, and a second difference between the prediction result of the first quality grade and the corresponding expected quality grade. If the fermentation of Daqu is abnormal, the future time series data of the environment in the future time period is controlled based on the first difference and the second difference to minimize the first difference and the second difference.

Images