Prediction and evaluation method and system for fermentation quality of white spirit

Abstract

The invention discloses a baijiu fermentation quality prediction and evaluation method and system, and the method comprises the steps: collecting environmental parameters of a fermented grain fermentation process, and obtaining evaluation parameters of the fermentation quality; constructing a fermentation quality evaluation model, inputting the environmental parameters and the evaluation parameters as sample data into the fermentation quality evaluation model, and training a network model to obtain a trained evaluation model; inputting to-be-measured environmental parameters in the fermented grain fermentation process into the trained evaluation model, and outputting predicted evaluation parameters; and evaluating the fermentation quality by using the predicted evaluation parameters to obtain an evaluation result. According to the method, human interference can be remarkably reduced, evaluation parameters are quickly obtained by using the network model, and reliable and effective evaluation of the fermentation quality is realized.

Description

Technical Field

[0001] The invention relates to the field of liquor fermentation, and in particular to a method and system for predicting and evaluating liquor fermentation quality. Background Art

[0002] The most significant reflection of the quality of the mash fermentation process is the quality of the base wine, and the aroma of liquor mainly comes from the types and contents of chemical substances such as alcohols, esters, acids, aldehydes, phenols, etc. produced by the biochemical reactions in the fermentation process. These chemical substances have an important impact on the quality of mash fermentation.

[0003] At present, chemical analysis methods are usually used to predict and evaluate the fermentation quality of liquor. Traditional chemical analysis methods, such as gas chromatography, liquid chromatography and spectral analysis, have high requirements for equipment and are complex to operate. They require personnel with professional skills to operate and usually require complex sample pre-treatment steps, which not only increases the time and consumption of the experiment, but also may introduce human errors. In addition, these methods can often only analyze certain specific components in the sample, and it is difficult to fully reflect the changes in multiple substances during the fermentation process, which limits the reliable evaluation of the fermentation quality.

[0004] Therefore, a new prediction and evaluation technology for liquor fermentation quality is needed, which can significantly reduce human interference, use network models to quickly obtain evaluation parameters, and achieve reliable and effective evaluation of fermentation quality. Summary of the invention

[0005] In view of this, the purpose of the present invention is to overcome the defects in the prior art and provide a predictive evaluation method and system for liquor fermentation quality, which can significantly reduce human interference, use a network model to quickly obtain evaluation parameters, and achieve reliable and effective evaluation of fermentation quality.

[0006] The method for predicting and evaluating the fermentation quality of liquor of the present invention comprises:

[0007] Collect environmental parameters of the fermentation process of the mash and obtain evaluation parameters of the fermentation quality;

[0008] Constructing a fermentation quality evaluation model, inputting environmental parameters and evaluation parameters as sample data into the fermentation quality evaluation model, training the network model, and obtaining a trained evaluation model;

[0009] The environmental parameters to be measured during the fermentation of the mash are input into the trained evaluation model, and the predicted evaluation parameters are output;

[0010] The predicted evaluation parameters are used to evaluate the fermentation quality and obtain evaluation results.

[0011] Furthermore, the environmental parameters include temperature, humidity, carbon dioxide and pH value; and the evaluation parameters include total acid, total ester and total alcohol.

[0012] Furthermore, the fermentation quality evaluation model includes a graph convolutional network, a temporal convolutional network and an attention mechanism;

[0013] The graph convolutional network captures the complex relationship between nodes and extracts spatial features by defining convolution operations on the graph structure of non-Euclidean space; the graph convolutional network uses environmental parameters as nodes in the graph, and the interactions between environmental parameters are represented by edges, and the feature representation of each node is updated by aggregating the information of neighboring nodes;

[0014] The temporal convolutional network utilizes the parallel computing capability of convolutional neural networks and introduces convolution operations in the time dimension. The temporal convolutional network further analyzes the spatial features output by the graph convolutional network in the time dimension, extracts the temporal features, and combines them with the spatial features to provide insights into the time series for the final prediction.

[0015] The attention mechanism calculates weights for features after fusion of spatial features and temporal features, and comprehensively considers weighted information to enhance the sensitivity of the fermentation quality evaluation model to important features.

[0016] Furthermore, the graph convolution network takes the graph G(X,A) as input, updates the node feature information matrix X through the graph convolution operation, and obtains the new graph G(X k ,A), the final output is graph G(X',A); where X and X k The features of the middle nodes have different dimensions, and the dimension of the hidden layer node features is d.

[0017] Furthermore, the temporal convolutional network includes one-dimensional full convolution, causal convolution, dilated convolution and residual connection;

[0018] The temporal convolutional network generates an output sequence of the same length as the input sequence through one-dimensional full convolution to retain the information of the input data; the temporal convolutional network sets causal convolution so that the current convolution result only depends on the current and previous inputs;

[0019] The dilated convolution increases the number of intervals between the convolution kernel points so that the temporal convolution network obtains a larger receptive field, allowing the input information to be extracted faster.

[0020] The residual connection is used to train a deep network so that the network can transfer information across layers; the output of the residual connection includes input information and output information of the convolution operation.

[0021] A prediction and evaluation system for liquor fermentation quality, comprising a parameter acquisition unit, a model training unit and a quality evaluation unit;

[0022] The parameter collection unit is used to collect environmental parameters of the fermentation process of the fermented grains and obtain evaluation parameters of the fermentation quality;

[0023] The model training unit is used to construct a fermentation quality evaluation model, input environmental parameters and evaluation parameters as sample data into the fermentation quality evaluation model, train the network model, and obtain a trained evaluation model;

[0024] The quality evaluation unit is used to input the environmental parameters to be measured during the fermentation of the mash into the trained evaluation model, output the predicted evaluation parameters, and evaluate the fermentation quality using the predicted evaluation parameters to obtain the evaluation results.

[0025] Furthermore, the environmental parameters include temperature, humidity, carbon dioxide and pH value; and the evaluation parameters include total acid, total ester and total alcohol.

[0026] Furthermore, the fermentation quality evaluation model includes a graph convolutional network, a temporal convolutional network and an attention mechanism;

[0027] The graph convolutional network captures the complex relationship between nodes and extracts spatial features by defining convolution operations on the graph structure of non-Euclidean space; the graph convolutional network uses environmental parameters as nodes in the graph, and the interactions between environmental parameters are represented by edges, and the feature representation of each node is updated by aggregating the information of neighboring nodes;

[0028] The temporal convolutional network utilizes the parallel computing capability of convolutional neural networks and introduces convolution operations in the time dimension. The temporal convolutional network further analyzes the spatial features output by the graph convolutional network in the time dimension, extracts the temporal features, and combines them with the spatial features to provide insights into the time series for the final prediction.

[0029] The attention mechanism calculates weights for features after fusion of spatial features and temporal features, and comprehensively considers weighted information to enhance the sensitivity of the fermentation quality evaluation model to important features.

[0030] Furthermore, the graph convolution network takes the graph G(X,A) as input, updates the node feature information matrix X through the graph convolution operation, and obtains the new graph G(X k ,A), the final output is graph G(X',A); where X and X k The features of the middle nodes have different dimensions, and the dimension of the hidden layer node features is d.

[0031] Furthermore, the temporal convolutional network includes one-dimensional full convolution, causal convolution, dilated convolution and residual connection;

[0032] The temporal convolutional network generates an output sequence of the same length as the input sequence through one-dimensional full convolution to retain the information of the input data; the temporal convolutional network sets causal convolution so that the current convolution result only depends on the current and previous inputs;

[0033] The dilated convolution increases the number of intervals between the convolution kernel points so that the temporal convolution network obtains a larger receptive field, allowing the input information to be extracted faster.

[0034] The residual connection is used to train a deep network so that the network can transfer information across layers; the output of the residual connection includes input information and output information of the convolution operation.

[0035] The beneficial effects of the present invention are as follows: the present invention discloses a method and system for predicting and evaluating the fermentation quality of liquor, constructs an improved GCN-TCN-Attention model that integrates a graph convolutional network (GCN), a temporal convolutional network (TCN), and an attention mechanism. This combination enables the model to more comprehensively understand and predict the complexity of spatiotemporal data, while capturing the complex spatial interactions and temporal dynamic changes in the fermentation process; the introduction of a soft Attention mechanism enables the improved model to more accurately identify and emphasize features that have a greater impact on the prediction results, thereby improving the prediction accuracy and robustness of the model. The present invention significantly reduces human interference, and the above-mentioned improved model can quickly obtain evaluation parameters, thereby achieving a reliable and effective evaluation of the fermentation quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0037] Figure 1 This is a schematic diagram of the GCN-TCN-Attention improved model structure of the present invention;

[0038] Figure 2 Schematic diagram of the graph convolutional network structure of the present invention;

[0039] Figure 3 Schematic diagram of the temporal convolutional network structure of the present invention. DETAILED DESCRIPTION

[0040] The present invention is further described below in conjunction with the accompanying drawings, as shown in the drawings:

[0041] This embodiment discloses a method for predicting and evaluating the fermentation quality of liquor, comprising the following steps:

[0042] Collect environmental parameters of the fermentation process of the mash and obtain evaluation parameters of the fermentation quality;

[0043] Constructing a fermentation quality evaluation model, inputting environmental parameters and evaluation parameters as sample data into the fermentation quality evaluation model, training the network model, and obtaining a trained evaluation model;

[0044] The environmental parameters to be measured during the fermentation of the mash are input into the trained evaluation model, and the predicted evaluation parameters are output;

[0045] The predicted evaluation parameters are used to evaluate the fermentation quality and obtain evaluation results.

[0046] In this embodiment, the parameters such as temperature, humidity, carbon dioxide, pH value and the like during the fermentation of the fermented grains are important characteristic indicators during the fermentation of the fermented grains, and are also environmental parameters during the fermentation of the fermented grains; the environmental parameters include temperature, humidity, carbon dioxide and pH value;

[0047] The most significant manifestation of the quality of the fermentation process of the fermented grains is the quality of the base wine, and the aroma of the liquor mainly comes from the types and contents of alcohols, esters, acids, aldehydes, phenols and other substances produced by the biochemical reactions in the fermentation process. Among them, acids affect the taste balance and flavor characteristics of the wine, and have antibacterial and quality preservation effects. The main representatives are organic acids such as caproic acid, acetic acid, lactic acid, butyric acid, etc., which provide a material basis for the synthesis of the flavor and esters of the liquor. Esters are the main components of the aroma of liquor and have a great influence on the flavor. The main representatives are ethyl acetate, ethyl butyrate, ethyl caproate and ethyl lactate, which are the four major esters in liquor. Alcohols have the greatest influence and contribute to the overall aroma of the wine. The main representatives are ethanol and isopropanol. Therefore, the main components and contents of total acid, total ester and total alcohol in the fermented grains can be used as important indicators for evaluating the fermentation quality of the fermented grains; the evaluation parameters include total acid, total ester and total alcohol.

[0048] In this embodiment, a GCN-TCN-Attention improved model integrating a graph convolutional network (GCN), a temporal convolutional network (TCN), and an attention mechanism is constructed as a fermentation quality evaluation model. Figure 1 shown.

[0049] Graph Convolutional Network (GCN) is a deep learning method based on graph structured data. Its core idea is to learn a mapping function so that the nodes in the graph can aggregate the features of themselves and their neighboring nodes, thereby generating new representations of the nodes. Traditional convolutional neural networks (CNN) are good at processing spatial features in Euclidean spatial data, while GCN captures the complex relationships between nodes and extracts spatial features by defining convolution operations on graph structures in non-Euclidean space.

[0050] During the fermentation process, parameters such as temperature, humidity, pH value, and carbon dioxide concentration can be regarded as nodes in the graph, and the interactions between them are represented by edges. The GCN model updates the feature representation of each node by aggregating the information of neighboring nodes, thereby learning the role and influence of each parameter in the fermentation process.

[0051] The GCN structure is as follows Figure 2 As shown in the figure, GCN takes the graph G(X,A) as input, updates the node feature information matrix X through the graph convolution operation, and obtains the new graph G(X k ,A), the final output is graph G(X',A). X and X k The features of the middle nodes have different dimensions, and the dimension of the hidden layer node features is d.

[0052] Among them, the node feature matrix of the k+1th layer can be obtained by the following formula:

[0053]

[0054] In the formula I is the identity matrix; D is the degree matrix of A, W k is the weight matrix of the kth layer; X k is the feature matrix of the k-th layer node; σ(·) is the nonlinear activation function.

[0055] Temporal Convolutional Network (TCN) is a deep learning framework for sequence modeling and prediction. Compared with traditional recurrent neural networks (RNN), TCN utilizes the parallel computing power of convolutional neural networks (CNN) and shows better performance in dealing with long-term dependency problems by introducing convolution operations in the time dimension. TCN is chosen for time series data processing mainly because of its efficiency and flexibility in capturing time dependencies and dynamic changes.

[0056] The data in the fermentation process is usually collected in chronological order. By using causal convolution and dilated convolution layers, TCN can flexibly adjust the perception range of historical information while maintaining efficient calculation. The structure of TCN allows the model to not only consider the parameter value at the current time point, but also combine the trend and pattern of changes over a period of time to provide more accurate predictions. In practical applications, TCN further analyzes the spatial features output by GCN in the time dimension, extracts the temporal features, and combines them with the spatial features to provide time series insights for the final prediction.

[0057] The architecture of TCN includes one-dimensional full convolution, causal convolution, dilated convolution and residual connection. It has the advantages of parallel computing capability, flexible receptive field, stable gradient and low memory usage. TCN is mainly composed of multiple residual modules stacked together, each of which contains two layers of dilated causal convolution to increase the direct mapping of input and speed up the calculation. The structure of TCN is as follows: Figure 3 shown.

[0058] Among them, TCN generates an output sequence of the same length as the input sequence through one-dimensional full convolution to retain the information of the input data;

[0059] Causal convolution makes convolution causal, and the current convolution result depends only on the current and previous inputs, thereby improving the prediction accuracy. t , its input value can only be the time t and the time before t, that is, x0…x t , not affected by x t+1 …x t The impact of the moment.

[0060] Dilated convolution increases the number of intervals between the convolution kernel points so that TCN can obtain a larger receptive field and extract input information faster. Assuming the one-dimensional input sequence is x, the one-dimensional convolution kernel f:{0,1,…,k-1}→R, the output value of the dilated sequence of the sth element is:

[0061]

[0062] Where: * is the dilation operation; d is the dilation coefficient; k is the convolution kernel size; x s-d·i is the sequence element corresponding to the element in the convolution kernel. d increases with the increase of network layers, and the receptive field is expanded by increasing the size of d or k.

[0063] Residual connection is an effective method for training deep networks, which can enable the network to transfer information across layers. Assume that the input of the residual module is x and the output is y:

[0064] y=H(x,P)+x;

[0065] Where: H is a series of convolution function operations; P is the weight function of the convolution value.

[0066] It can be seen from the formula that the output of the residual connection contains the input information and the output information of the convolution operation, which can effectively ensure the accuracy of the TCN network.

[0067] That is, the TCN module combines the design of residual connection and dilated convolution. Dilated convolution expands the receptive field of the model by increasing the interval of the convolution kernel, and can more effectively extract long-term dependent features; while residual connection ensures the stability of gradients in deep networks and avoids the problem of gradient vanishing, so that the model can maintain stable training results while increasing in complexity, further improving the model's expressiveness when processing long sequence data.

[0068] Attention Mechanism is an advanced computing method that effectively improves the prediction accuracy of the model by prioritizing features with higher priority in situations where the amount of data is limited or information is overloaded. This mechanism is particularly suitable for predicting the fermentation process, because in the fermentation process, the parameters under different time periods or different environmental conditions have different effects on the final product.

[0069] The attention mechanism gives the model the ability to identify and emphasize factors that have a significant impact on the predicted target by self-learning the weights of input features. This process not only enhances the model's sensitivity to key information, but also significantly improves the accuracy and robustness of the prediction. The attention mechanism is divided into two forms: hard attention and soft attention. The hard attention mechanism only focuses on one type of input information at any time, while the soft attention mechanism calculates weights for all input information, takes these weighted information into comprehensive consideration, and inputs them into the neural network for training. In the present invention, soft attention is used because it can capture the importance differences of input information more carefully. The relevant formula is as follows:

[0070]

[0071] Among them, s(h i ,h t ) is the score of the feature vector. The larger the score, the greater the attention. i is the weight value of the i-th input feature in the attention mechanism, that is, the ratio of the score of the feature vector to the total, o is the final vector obtained by summing and averaging all vectors.

[0072] The present invention innovatively combines the graph convolutional network (GCN) and the temporal convolutional network (TCN), so that the model can simultaneously capture the complex spatial interaction relationship and temporal dynamic changes in the fermentation process. GCN is responsible for extracting the spatial interaction characteristics between various parameters in the fermentation process, and can effectively handle the complex relationship between parameters such as temperature, humidity, pH value and CO2. TCN accurately captures the dynamic changes in the time dimension and extracts time features through causal convolution and dilated convolution. This combination enables the model to more comprehensively understand and predict the complexity of spatiotemporal data. In addition, the Attention mechanism is introduced to calculate and enhance the weights of spatial features and temporal features respectively before feature fusion. Through the soft Attention mechanism, the model can more accurately identify and emphasize the features that have a greater impact on the prediction results, thereby improving the prediction accuracy and robustness of the model. This multi-level Attention application enables the model to more effectively capture key information when processing data with high-dimensional complex relationships.

[0073] In this embodiment, the corresponding environmental parameters such as temperature, humidity, carbon dioxide, pH value, etc. can be obtained from the historical data records of several fermented mash fermentations; accordingly, the evaluation parameters such as total acid, total ester, total alcohol, etc. can be obtained from the historical data records of several fermented mash fermentations; the obtained environmental parameters and evaluation parameters are used as sample data for model training, that is, the environmental parameters are used as the input of the GCN-TCN-Attention improved model of the present invention, and the evaluation parameters are used as the output of the GCN-TCN-Attention improved model of the present invention, and the improved model is trained. Among them, GCN is used to extract spatial features to process the complex relationship between parameters such as temperature, humidity, PH, CO2, etc. in the fermentation process; TCN extracts time features in the time dimension; the Attention mechanism is applied in multiple levels in the model to enhance the ability to capture key information; the Adam algorithm is used as the optimizer of the model, the number of cyclic training is 1500 times, the initial learning rate is 0.01, the gradient threshold is 1, and the learning rate is adjusted after 750 training times, and the learning rate reduction factor is 0.4.

[0074] Through the above training process of the GCN-TCN-Attention improved model, a trained evaluation model is obtained; the environmental parameters to be measured during the actual fermentation of the mash are input into the trained evaluation model, and the predicted evaluation parameters are output; according to the existing evaluation standards of the winery, the fermentation quality is evaluated using the predicted evaluation parameters to obtain the evaluation results. Among them, the winery can set the predicted evaluation parameters according to its own production process and product quality requirements: the standard value range of indicators such as total acid, total ester, and total alcohol. By comparing the predicted evaluation parameters with the standard values, it can be judged whether the fermentation quality meets the standards, thereby obtaining the final evaluation results.

[0075] The present invention also relates to a prediction and evaluation system for liquor fermentation quality. The evaluation system corresponds to the evaluation method of the above embodiment and can be understood as a system for implementing the above evaluation method. The evaluation system includes a parameter acquisition unit, a model training unit and a quality evaluation unit.

[0076] The parameter collection unit is used to collect environmental parameters of the fermentation process of the fermented grains and obtain evaluation parameters of the fermentation quality;

[0077] The model training unit is used to construct a fermentation quality evaluation model, input environmental parameters and evaluation parameters as sample data into the fermentation quality evaluation model, train the network model, and obtain a trained evaluation model;

[0078] The quality evaluation unit is used to input the environmental parameters to be measured during the fermentation of the mash into the trained evaluation model, output the predicted evaluation parameters, and evaluate the fermentation quality using the predicted evaluation parameters to obtain the evaluation results.

[0079] In order to better understand the GCN-TCN-Attention improved model used by the fermentation quality prediction and evaluation method and system of the present invention, the following description is further made:

[0080] The data of temperature, humidity, carbon dioxide and pH value of a certain mash fermentation process were collected as environmental parameters and input into the BP model, LSTM model and GCN-TCN-Attention improved model respectively to obtain the predicted output of total acid, total ester and total alcohol, and the model regression prediction results shown in Table 1 were obtained.

[0081] Table 1

[0082]

[0083] As can be seen from Table 1, the GCN-TCN-Attention improved model of the present invention has a higher fitting degree than the BP neural network regression model and the LSTM long short-term memory neural network regression model in the prediction performance of acid ester alcohol, indicating that the regression accuracy of the GCN-TCN-Attention improved model is better than that of the BP neural network regression model and the LSTM long short-term memory neural network regression model. Since the mean absolute error (MAE) and the root mean square error (RMSE) of the GCN-TCN-Attention improved model are substantially smaller than those of the BP neural network regression model and the LSTM long short-term memory neural network regression model, it indicates that the stability of the GCN-TCN-Attention improved model is better than that of the BP neural network regression model and the LSTM long short-term memory neural network regression model.

[0084] Therefore, from the comprehensive data, the GCN-TCN-Attention improved model of the present invention has obvious advantages.

[0085] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting and evaluating liquor fermentation quality, characterized in that: include: Collect environmental parameters of the fermentation process of the mash and obtain evaluation parameters of the fermentation quality; Constructing a fermentation quality evaluation model, inputting environmental parameters and evaluation parameters as sample data into the fermentation quality evaluation model, training the network model, and obtaining a trained evaluation model; The environmental parameters to be measured during the fermentation of the mash are input into the trained evaluation model, and the predicted evaluation parameters are output; The predicted evaluation parameters are used to evaluate the fermentation quality and obtain evaluation results.

2. The method for predicting and evaluating liquor fermentation quality according to claim 1, characterized in that: The environmental parameters include temperature, humidity, carbon dioxide and pH value; the evaluation parameters include total acid, total ester and total alcohol.

3. The method for predicting and evaluating liquor fermentation quality according to claim 1, characterized in that: The fermentation quality evaluation model includes a graph convolutional network, a temporal convolutional network and an attention mechanism; The graph convolutional network captures the complex relationship between nodes and extracts spatial features by defining convolution operations on the graph structure of non-Euclidean space; the graph convolutional network uses environmental parameters as nodes in the graph, and the interactions between environmental parameters are represented by edges, and the feature representation of each node is updated by aggregating the information of neighboring nodes; The temporal convolutional network utilizes the parallel computing capability of convolutional neural networks and introduces convolution operations in the time dimension. The temporal convolutional network further analyzes the spatial features output by the graph convolutional network in the time dimension, extracts the temporal features, and combines them with the spatial features to provide insights into the time series for the final prediction. The attention mechanism calculates weights for features after fusion of spatial features and temporal features, and comprehensively considers weighted information to enhance the sensitivity of the fermentation quality evaluation model to important features.

4. The method for predicting and evaluating the fermentation quality of liquor according to claim 3, characterized in that: The graph convolution network takes the graph G(X,A) as input, updates the node feature information matrix X through the graph convolution operation, and obtains the new graph G(X k ,A), the final output is graph G(X',A); where X and X k The features of the middle nodes have different dimensions, and the dimension of the hidden layer node features is d.

5. The method for predicting and evaluating the fermentation quality of liquor according to claim 3, characterized in that: The temporal convolutional network includes one-dimensional full convolution, causal convolution, dilated convolution and residual connection; The temporal convolutional network generates an output sequence of the same length as the input sequence through one-dimensional full convolution to retain the information of the input data; the temporal convolutional network sets causal convolution so that the current convolution result only depends on the current and previous inputs; The dilated convolution increases the number of intervals between the convolution kernel points so that the temporal convolution network obtains a larger receptive field, allowing the input information to be extracted faster. The residual connection is used to train a deep network so that the network can transfer information across layers; the output of the residual connection includes input information and output information of the convolution operation.

6. A prediction and evaluation system for liquor fermentation quality, characterized in that: It includes parameter acquisition unit, model training unit and quality assessment unit; The parameter collection unit is used to collect environmental parameters of the fermentation process of the fermented grains and obtain evaluation parameters of the fermentation quality; The model training unit is used to construct a fermentation quality evaluation model, input environmental parameters and evaluation parameters as sample data into the fermentation quality evaluation model, train the network model, and obtain a trained evaluation model; The quality evaluation unit is used to input the environmental parameters to be measured during the fermentation of the mash into the trained evaluation model, output the predicted evaluation parameters, and evaluate the fermentation quality using the predicted evaluation parameters to obtain the evaluation results.

7. The prediction and evaluation system for liquor fermentation quality according to claim 6, characterized in that: The environmental parameters include temperature, humidity, carbon dioxide and pH value; the evaluation parameters include total acid, total ester and total alcohol.

8. The prediction and evaluation system for liquor fermentation quality according to claim 6, characterized in that: The fermentation quality evaluation model includes a graph convolutional network, a temporal convolutional network and an attention mechanism; The graph convolutional network captures the complex relationship between nodes and extracts spatial features by defining convolution operations on the graph structure of non-Euclidean space; the graph convolutional network uses environmental parameters as nodes in the graph, and the interactions between environmental parameters are represented by edges, and the feature representation of each node is updated by aggregating the information of neighboring nodes; The temporal convolutional network utilizes the parallel computing capability of convolutional neural networks and introduces convolution operations in the time dimension. The temporal convolutional network further analyzes the spatial features output by the graph convolutional network in the time dimension, extracts the temporal features, and combines them with the spatial features to provide insights into the time series for the final prediction. The attention mechanism calculates weights for features after fusion of spatial features and temporal features, and comprehensively considers weighted information to enhance the sensitivity of the fermentation quality evaluation model to important features.

9. The prediction and evaluation system for liquor fermentation quality according to claim 8, characterized in that: The graph convolution network takes the graph G(X,A) as input, updates the node feature information matrix X through the graph convolution operation, and obtains the new graph G(X k ,A), the final output is graph G(X',A); where X and X k The features of the middle nodes have different dimensions, and the dimension of the hidden layer node features is d.

10. The prediction and evaluation system for liquor fermentation quality according to claim 8, characterized in that: The temporal convolutional network includes one-dimensional full convolution, causal convolution, dilated convolution and residual connection; The temporal convolutional network generates an output sequence of the same length as the input sequence through one-dimensional full convolution to retain the information of the input data; the temporal convolutional network sets causal convolution so that the current convolution result only depends on the current and previous inputs; The dilated convolution increases the number of intervals between the convolution kernel points so that the temporal convolution network obtains a larger receptive field, allowing the input information to be extracted faster. The residual connection is used to train a deep network so that the network can transfer information across layers; the output of the residual connection includes input information and output information of the convolution operation.

Images