Coarse cereal staple food fermentation quality online monitoring method and system based on electronic nose

Abstract

The invention discloses a coarse cereal staple food fermentation quality online monitoring method and system based on an electronic nose, relates to the technical field of food fermentation monitoring, and discloses the coarse cereal staple food fermentation quality online monitoring method and system based on the electronic nose. An original gas response signal of an electronic nose sensor array is obtained and preprocessed to generate standardized gas response data, a dominant metabolic pathway and contribution weight thereof are matched based on the data, a current fermentation stage is diagnosed and final quality is predicted, and a regulation and control instruction is generated by comparing with a standard template to regulate fermentation parameters. A dominant metabolic pathway in the fermentation process can be monitored in real time, accurate stage diagnosis and quality prediction are realized, the fermentation process is regulated in time, and the fermentation quality stability and the production efficiency are improved.

Description

Technical Field

[0001] This application relates to the field of food fermentation monitoring technology, and in particular to a method and system for online monitoring of the fermentation quality of staple grains based on electronic nose. Background Technology

[0002] In the industrial fermentation production of staple grains, quality monitoring has long relied on manual experience or crude settings of macroscopic parameters such as temperature, humidity, and time, failing to achieve direct, online sensing and precise control of microbial metabolic activities. Grain raw materials possess multi-matrix characteristics and complex microbial community structures, involving dynamic interactions and metabolic pathway competition among various microbial groups such as yeast and lactic acid bacteria, resulting in a highly nonlinear fermentation process. While electronic nose technology has been introduced into the field of food fermentation monitoring due to its speed and non-destructive nature, it faces limitations when dealing with complex systems like grains with multiple matrices and microbial communities. Existing methods generally perform simple correlation modeling between the mixed gas signals collected by the electronic nose and macroscopic quality indicators such as the final product volume and acidity, forming a "black box" experience-based mapping. This method ignores the dynamic evolution of key metabolic pathways during fermentation, such as the activity changes and interactions of pathways like glycolysis and lactic acid fermentation, causing the model to heavily rely on training data from specific formulations and exhibiting weak generalization ability. When abnormal gas signals appear, harmful metabolites have often accumulated to an irreversible level, resulting in a severely delayed early warning mechanism that cannot promptly halt the quality deterioration process. Meanwhile, due to a lack of in-depth analysis of metabolic mechanisms, operators find it difficult to identify the specific metabolic pathway deviations behind abnormal signals, resulting in a lack of targeted control measures. This often requires repeated trial and error to adjust environmental parameters or material composition, leading to decreased production efficiency and waste of resources.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a method and system for online monitoring of the fermentation quality of staple grains based on an electronic nose, which aims to improve the stability of fermentation quality and production efficiency.

[0005] To achieve the above objectives, this application proposes an online monitoring method for the fermentation quality of coarse grain staple foods based on an electronic nose. The method includes: The original gas response signal generated by the electronic nose sensor array during the fermentation of the target grain staple food is obtained in real time, and the original gas response signal is preprocessed to generate standardized gas response data. Based on the standardized gas response data, at least one dominant metabolic pathway and its corresponding contribution weight data are matched from a pre-stored metabolic pathway feature database; wherein, the metabolic pathway feature database pre-stores standard gas response feature data corresponding to the metabolic pathways of various fermenting microorganisms. Based on the dominant metabolic pathway and the contribution weight data, the current fermentation stage is diagnosed to obtain stage diagnosis result data. Based on the dominant metabolic pathway, the contribution weight data and the stage diagnosis result data, a prediction of the final fermentation quality is generated to obtain quality prediction result data. The stage diagnosis result data and the quality prediction result data are compared with the pre-stored standard fermentation process template data, and process control instruction data for regulating the fermentation process is generated based on the comparison results. Based on the process control instruction data, the fermentation environment parameters or material composition of the target coarse grain staple food are adjusted.

[0006] In one embodiment, the step of matching at least one dominant metabolic pathway and its corresponding contribution weight data from a pre-stored metabolic pathway feature database based on the standardized gas response data includes: The standardized gas response data is correlated with each group of standard gas response feature data in the metabolic pathway feature database to obtain multiple correlation metrics. Based on the correlation metric, metabolic pathways corresponding to standard gas response feature data whose correlation metric exceeds a preset threshold are selected from the metabolic pathway feature database to obtain a candidate metabolic pathway set. The standardized gas response data is represented as a linear combination of the standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set. The linear coefficients of each candidate metabolic pathway are calculated by solving the linear equations corresponding to the linear combination, and are used as the contribution weight data corresponding to the candidate metabolic pathway. Based on the contribution weight data, the top N candidate metabolic pathways with the largest contribution weight data are determined from the candidate metabolic pathway set to serve as the dominant metabolic pathways, where N is a positive integer.

[0007] In one embodiment, the step of representing the standardized gas response data as a linear combination of the standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set, and calculating the linear coefficient of each candidate metabolic pathway by solving the linear equation system corresponding to the linear combination, and using it as the contribution weight data corresponding to the candidate metabolic pathway, includes: A linear combination equation is constructed, wherein the standardized gas response data is used as a known vector, the standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set is used as the column vector of the coefficient matrix, and the contribution weight data of each candidate metabolic pathway is used as the unknown vector to be solved. The linear combination equation is solved using a constrained optimization algorithm, wherein the constraint condition imposed by the constrained optimization algorithm is that the sum of all contribution weight data is a fixed constant and each contribution weight data is non-negative; The error between the standardized gas response data and the estimated response data reconstructed from the coefficient matrix and the unknown vector is minimized through iterative calculations to obtain the contribution weight data of each candidate metabolic pathway that satisfies the constraints.

[0008] In one embodiment, the steps of diagnosing the current fermentation stage based on the dominant metabolic pathway and the contribution weight data to obtain stage diagnosis result data, and generating a prediction of the final fermentation quality based on the dominant metabolic pathway, the contribution weight data, and the stage diagnosis result data to obtain quality prediction result data include: The dominant metabolic pathway and its corresponding contribution weight data are matched with pre-stored fermentation stage discrimination rules, and the current fermentation stage identifier is determined based on the matching results, which is used as the stage diagnosis result data; wherein, the fermentation stage discrimination rules define predefined dominant metabolic pathways and their preset weight ranges corresponding to different fermentation stages. Multidimensional time series data is constructed based on the set of dominant metabolic pathways corresponding to each sampling time point from the start of fermentation to the present moment, the contribution weight data of each dominant metabolic pathway, and the corresponding stage diagnostic results data. The multidimensional time series data is input into a pre-trained fermentation quality prediction model to output a predicted value of the evaluation index for the final fermentation quality, and the predicted value of the evaluation index is used as the quality prediction result data.

[0009] In one embodiment, the step of inputting the multidimensional time series data into a pre-trained fermentation quality prediction model to output a predicted value of the evaluation index for the final fermentation quality, and using the predicted value of the evaluation index as the quality prediction result data, includes: Trend analysis is performed on the contribution weight data of each dominant metabolic pathway in the multidimensional time series data to calculate the rate of change of the contribution weight of each dominant metabolic pathway within a preset time window. Based on the rate of change, a set of metabolic activity intensity change indicators are obtained by weighting the pre-stored weighting coefficients of each metabolic pathway. The metabolic activity intensity change index is matched with a pre-stored quality influence relationship table, and the predicted value of the final fermentation quality evaluation index is output as the quality prediction result data based on the matching result; wherein, the quality influence relationship table records the correspondence between the metabolic activity intensity change index and the final fermentation quality evaluation index.

[0010] In one embodiment, the step of comparing the stage diagnostic result data and the quality prediction result data with pre-stored standard fermentation process template data, and generating process control instruction data for regulating the fermentation process based on the comparison results, includes: The stage diagnosis result data is compared with the predefined dominant metabolic pathway and preset weight range of the corresponding stage in the standard fermentation process template data. If the dominant metabolic pathway does not belong to the predefined dominant metabolic pathway, or its corresponding contribution weight data exceeds the preset weight range, then a first type of deviation alarm data is generated. The predicted values ​​of the evaluation indicators in the quality prediction results data are compared with the final quality target range defined in the standard fermentation process template data. If the predicted values ​​of the evaluation indicators exceed the target range, a second type of deviation alarm data is generated. Based on at least one of the first type of deviation alarm data and the second type of deviation alarm data, a pre-stored control strategy mapping table is queried to generate control action instructions including the control object and control amount, which are used as the process control instruction data; wherein, the control strategy mapping table defines the correspondence between deviation type and control action.

[0011] In one embodiment, the step of querying a pre-stored control strategy mapping table based on at least one of the first type of deviation alarm data and the second type of deviation alarm data to generate a control action instruction including the control object and control amount, as the process control instruction data, includes: If the first type of deviation alarm data exists, the current fermentation stage and the dominant metabolic pathway type of the deviation are determined based on the data; If the second type of deviation alarm data exists, the type of final quality evaluation index that the prediction fails to meet the standard is determined based on the data; The current fermentation stage, the type of the deviated dominant metabolic pathway, and the type of the predicted substandard final quality evaluation index are used as composite query conditions. These conditions are then jointly queried in the pre-stored regulation strategy mapping table, and the process regulation instruction data is generated based on the query results. The dimensions of the regulation strategy mapping table correspond to the fermentation stage, the type of metabolic pathway deviation, and the type of quality index deviation. The regulation strategy mapping table stores regulation action instructions for different composite query conditions.

[0012] In one embodiment, the method further includes: In a controlled fermentation simulation environment, various pure culture fermentation microorganisms are induced to enter and maintain a preset metabolic pathway activity state. During the period when the preset metabolic pathway activity state is stable, the response signal of the electronic nose sensor array is collected. The collected response signal is processed to obtain the standard gas response signal corresponding to the metabolic pathway. Feature extraction is performed on the standard gas response signal corresponding to each metabolic pathway to obtain the standard gas response feature data of that metabolic pathway; The identification information of each metabolic pathway is associated with and stored with its corresponding standard gas response characteristic data to form the metabolic pathway characteristic database.

[0013] In one embodiment, the step of extracting features from the standard gas response signal corresponding to each metabolic pathway to obtain the standard gas response feature data of that metabolic pathway includes: The standard gas response signal is segmented along the time dimension to obtain multiple time segments; Calculate at least two characteristic parameters among the average value, maximum value, rising slope, and stable value of the response signals of each sensor in the electronic nose sensor array within each time segment; The characteristic parameters calculated from all time segments are combined and dimensionality reduced to form standard gas response characteristic data that can characterize the gas response pattern of this metabolic pathway.

[0014] Furthermore, to achieve the above objectives, this application also proposes an online monitoring system for the fermentation quality of coarse grain staple foods based on an electronic nose. The online monitoring system for the fermentation quality of coarse grain staple foods based on an electronic nose includes: a memory, a processor, and an online monitoring program for the fermentation quality of coarse grain staple foods based on an electronic nose stored in the memory and executable on the processor. The online monitoring program for the fermentation quality of coarse grain staple foods based on an electronic nose is configured to implement the steps of the online monitoring method for the fermentation quality of coarse grain staple foods based on an electronic nose.

[0015] The method and system for online monitoring of fermentation quality of staple grains based on electronic nose proposed in this application acquire the raw gas response signal of the electronic nose sensor array and preprocess it to generate standardized gas response data. Based on this data, the dominant metabolic pathway and its contribution weight are matched to diagnose the current fermentation stage and predict the final quality. The control instructions are generated by comparing with the standard template to adjust the fermentation parameters. It can monitor the dominant metabolic pathway in the fermentation process in real time, realize accurate stage diagnosis and quality prediction, and timely control the fermentation process to improve the stability of fermentation quality and production efficiency. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating an embodiment of the online monitoring method for fermentation quality of staple grains based on an electronic nose, as provided in this application. Figure 2 This is a schematic diagram of a structural embodiment of the online monitoring system for the fermentation quality of staple grains based on an electronic nose, as provided in this application.

[0019] Explanation of icon numbers: 10. Memory; 20. Processor.

[0020] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0021] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0022] It should be understood that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] In existing technologies, quality monitoring in the fermentation production of coarse grain staple foods mainly relies on manual experience or the setting of macroscopic parameters such as temperature, humidity, and time, lacking direct, online sensing and control of microbial metabolic activities. Current electronic nose technology interprets mixed gas signals based on superficial empirical correlations, resulting in poor model universality, delayed early warnings, and a lack of targeted control, failing to achieve real-time analysis and precise control of the activity of key microbial metabolic pathways during coarse grain fermentation.

[0024] Based on this, the embodiments of this application provide an online monitoring method for the fermentation quality of coarse grain staple foods based on an electronic nose, referring to... Figure 1 The online monitoring method for fermentation quality of coarse grain staple foods based on electronic nose includes steps S100 to S500, wherein: Step S100: Obtain the original gas response signal generated by the electronic nose sensor array in real time monitoring the gas produced during the fermentation of the target grain staple food, and preprocess the original gas response signal to generate standardized gas response data. Step S200: Based on the standardized gas response data, at least one dominant metabolic pathway and its corresponding contribution weight data are matched from the pre-stored metabolic pathway feature database; wherein, the metabolic pathway feature database pre-stores standard gas response feature data corresponding to the metabolic pathways of various fermentation microorganisms. Step S300: Based on the dominant metabolic pathway and the contribution weight data, diagnose the current fermentation stage to obtain stage diagnosis result data, and based on the dominant metabolic pathway, the contribution weight data and the stage diagnosis result data, generate a prediction of the final fermentation quality to obtain quality prediction result data. Step S400: Compare the stage diagnosis result data and the quality prediction result data with the pre-stored standard fermentation process template data, and generate process control instruction data for regulating the fermentation process based on the comparison results. Step S500: Adjust the fermentation environment parameters or material composition of the target grain staple food according to the process control instruction data.

[0025] In this embodiment, the electronic nose sensor array refers to an integrated module composed of multiple gas sensors with different sensitivity characteristics. It identifies and analyzes complex gas mixtures by simulating a biological olfactory system. Each sensor generates an electrical signal response to a specific gas component or gas mixture, and the combination patterns of these response signals can be used to characterize the overall characteristics of the gas. The raw gas response signal refers to the unprocessed electrical signal data directly output by each sensor when the electronic nose sensor array monitors fermentation gases in real time. These signals typically contain noise, drift, and the nonlinear response of the sensors themselves. Standardized gas response data refers to data obtained after a series of preprocessing steps (e.g., baseline correction, normalization, denoising, etc.) on the raw gas response signal. This data eliminates interference from non-biological factors, making signals collected from different batches or at different time points comparable and more accurately reflecting changes in volatile substances during fermentation. The metabolic pathway characteristic database refers to a pre-established collection storing standard gas response characteristic data produced by various fermenting microorganisms under specific metabolic pathways. This database is obtained through experiments or simulations and is used to characterize the unique volatile substance patterns produced by the metabolic activities of different microorganisms. The dominant metabolic pathway refers to the microbial metabolic pathway that has a major influence on the formation of fermentation products and quality at a specific stage of the fermentation of coarse grain staple foods. In complex microbial fermentation systems, multiple metabolic pathways may occur simultaneously, but a few of them play a decisive role in the overall fermentation process and quality.

[0026] In this embodiment, contribution weight data refers to the numerical value quantifying the contribution of each dominant metabolic pathway to the overall gas response signal at the current fermentation stage. This data reflects the activity level of different metabolic pathways or the relative abundance of their products in the mixed gas. Stage diagnosis result data refers to the identification information of the specific stage of fermentation of the coarse grain staple food by analyzing the dominant metabolic pathways and their contribution weight data. For example, it can be diagnosed as "start-up stage", "vigorous fermentation stage", "maturation stage", etc. Quality prediction result data refers to the evaluation index value that predicts the final quality (such as flavor, taste, nutritional components, etc.) of the coarse grain staple food product based on real-time data during the fermentation process. This data aims to provide a prediction of product quality before the end of fermentation. Standard fermentation process template data refers to a pre-set reference model of the ideal fermentation process of coarse grain staple food. This template defines key parameters such as the dominant metabolic pathways that should exist at different fermentation stages, their contribution weight range, and the final quality target range, which are used for comparison with actual fermentation data. Process control instruction data refers to the specific operation instructions automatically generated by the system based on the comparison results of actual fermentation data and standard template, used to adjust the fermentation environment or material composition. For example, increasing or decreasing temperature, humidity, aeration, or adjusting the ratio of auxiliary materials. Fermentation environmental parameters refer to external conditions that affect the growth and metabolic activities of microorganisms, such as temperature, humidity, pH value, oxygen concentration, and stirring speed. Material composition refers to the raw materials used in the fermentation process and their ratios, such as the types of grains, the amount of water added, primer strains, and nutritional supplements.

[0027] In this embodiment, the online monitoring method for the fermentation quality of coarse grain staple foods based on electronic nose first acquires the raw gas response signal generated by the electronic nose sensor array in real time monitoring the gases produced during the fermentation process of the target coarse grain staple food, and then preprocesses the raw gas response signal to generate standardized gas response data. Specifically, the electronic nose sensor array is deployed in the fermentation tank or fermentation chamber to continuously collect information on volatile organic compounds in the fermentation gases and generate raw electrical signal data. These raw signals may be affected by fluctuations in ambient temperature and humidity or sensor drift. Therefore, these raw signals need to be preprocessed. For example, simple moving average filtering or baseline correction methods can be used to remove noise and drift, thereby obtaining relatively stable standardized gas response data.

[0028] Furthermore, based on this standardized gas response data, at least one dominant metabolic pathway and its corresponding contribution weight data are matched from a pre-stored metabolic pathway feature database. This database stores standard gas response feature data corresponding to the metabolic pathways of various fermenting microorganisms. For example, a simple Euclidean distance calculation can be performed between the current standard gas response data and the standard feature data of each metabolic pathway stored in the database, selecting the closest metabolic pathway as the dominant metabolic pathway and assigning it a preset fixed weight value. Alternatively, a known metabolic pathway can be directly associated with the peak value of a specific sensor response in the standardized gas response data, and its contribution weight can be roughly estimated based on the magnitude of that peak value.

[0029] Based on this, the current fermentation stage is diagnosed using the dominant metabolic pathway and its contribution weight data to obtain stage diagnosis results. Then, based on the dominant metabolic pathway, its contribution weight data, and the stage diagnosis results, a prediction of the final fermentation quality is generated to obtain quality prediction results. Specifically, a series of simple rules can be set; for example, when the contribution weight of a specific metabolic pathway exceeds a certain threshold, the current fermentation is determined to be in the "early" or "middle" stage. For quality prediction, a simple linear model can be used, taking the current dominant metabolic pathway and its contribution weight as input, and directly outputting a predicted final quality score.

[0030] In this embodiment, the diagnostic results and quality prediction data for this stage are then compared with pre-stored standard fermentation process template data. Based on the comparison results, process control instructions for regulating the fermentation process are generated. For example, operators can manually view the diagnosed fermentation stage and predicted quality and compare them with pre-printed or displayed standard fermentation process charts. If the current stage does not match the standard template, or the predicted quality is lower than expected, the operator can choose a general control measure based on experience, such as "increasing the fermentation temperature" or "extending the fermentation time".

[0031] In this embodiment, finally, the fermentation environment parameters or material composition of the target grain staple food are adjusted according to the process control instruction data. For example, if the generated instruction is "increase fermentation temperature", the operator can manually adjust the heating device of the fermentation tank to raise its temperature. If the instruction is "add auxiliary materials", the operator can manually add a predetermined amount of auxiliary materials to the fermentation system.

[0032] In this embodiment, by acquiring and standardizing fermentation gas response signals in real time and analyzing the dominant metabolic pathways of microorganisms and their contribution weights, this application can achieve refined stage diagnosis of the fermentation process and early prediction of the final quality. Therefore, the system can promptly detect fermentation anomalies and generate targeted process control instructions, thereby precisely adjusting fermentation environmental parameters or material composition. This effectively avoids the shortcomings of traditional "black box" experience-based correlations and improves the stability and intelligent control level of fermented grain staple foods.

[0033] In one feasible implementation, the step of matching at least one dominant metabolic pathway and its corresponding contribution weight data from a pre-stored metabolic pathway feature database based on the standardized gas response data includes: performing correlation calculations on the standardized gas response data and each set of standard gas response feature data in the metabolic pathway feature database to obtain multiple correlation metrics; selecting metabolic pathways corresponding to standard gas response feature data whose correlation metrics exceed a preset threshold from the metabolic pathway feature database according to the correlation metrics to obtain a candidate metabolic pathway set; representing the standardized gas response data as a linear combination of standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set, and calculating the linear coefficient of each candidate metabolic pathway by solving the linear equation system corresponding to the linear combination, as the contribution weight data corresponding to the candidate metabolic pathway; and determining the top N candidate metabolic pathways with the largest contribution weight data from the candidate metabolic pathway set according to the contribution weight data, as the dominant metabolic pathway, where N is a positive integer.

[0034] In this embodiment, the standardized gas response data is correlated with each set of standard gas response feature data in the metabolic pathway feature database to obtain multiple correlation metrics. This step aims to preliminarily identify metabolic pathways that may be related to the current fermentation state, thereby effectively narrowing the scope of subsequent analysis. Correlation calculations can employ various statistical methods, such as Pearson correlation coefficient, cosine similarity, or the reciprocal of Euclidean distance. For example, the standardized gas response data and each set of standard gas response feature data in the metabolic pathway feature database can be treated as multidimensional vectors, and then their similarity or distance can be calculated. In this way, the degree of similarity between the currently observed gas response pattern and known metabolic pathway feature patterns can be quantified.

[0035] In this embodiment, based on the correlation metric, metabolic pathways corresponding to standard gas response feature data whose correlation metric values ​​exceed a preset threshold are selected from the metabolic pathway feature database to obtain a candidate metabolic pathway set. Setting the preset threshold is crucial; its value can be optimized based on actual application scenarios, historical data analysis, or expert experience. For example, a correlation coefficient threshold can be set, and only metabolic pathways with calculated correlation metric values ​​higher than this threshold are included in the candidate set. This screening process ensures that only metabolic pathways relevant to the current fermentation process are further considered, thereby improving the efficiency and accuracy of subsequent calculations.

[0036] Based on this, the standardized gas response data is represented as a linear combination of the standard gas response characteristic data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set. By solving the system of linear equations corresponding to this linear combination, the linear coefficients of each candidate metabolic pathway are calculated, serving as the contribution weight data for that pathway. This step establishes a mathematical model that decomposes the observed complex gas response data into the superposition of contributions from each candidate metabolic pathway. By solving the system of linear equations, the contribution of each candidate metabolic pathway to the overall gas response, i.e., its activity level, can be quantified.

[0037] In this embodiment, finally, based on the contribution weight data, the top N candidate metabolic pathways with the largest contribution weight data are determined from the candidate metabolic pathway set to serve as the dominant metabolic pathways, where N is a positive integer. After calculating the contribution weight data of all candidate metabolic pathways, these weights are sorted in descending order, and the top N metabolic pathways with the largest weight values ​​are selected as the dominant metabolic pathways. The value of N can be set according to actual needs and the complexity of the fermentation process. For example, N can be 1, 2, or 3, representing one to three metabolic pathways that have the greatest impact on the fermentation process. This helps to focus on the few pathways that have the greatest impact on the current fermentation process, simplifying subsequent diagnostic and predictive analysis.

[0038] In this embodiment, by calculating the correlation between standardized gas response data and standard data in the metabolic pathway feature database, candidate metabolic pathways highly correlated with the current fermentation state can be effectively screened. Furthermore, the actual gas response data is represented as a linear combination of these candidate pathways, and their linear coefficients are calculated as contribution weights, providing a clear quantitative basis for the activity level of each metabolic pathway. Finally, the top N dominant metabolic pathways are determined based on their contribution weights, ensuring accurate capture of the core microbial activities during the fermentation process. This refined matching mechanism improves the accuracy and reliability of fermentation stage diagnosis and final quality prediction, providing solid data support for subsequent fermentation process regulation. This effectively avoids misjudgments and inappropriate interventions caused by inaccurate pathway identification, ensuring the quality stability of coarse grain staple foods.

[0039] In one feasible implementation, representing the standardized gas response data as a linear combination of standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set, and calculating the linear coefficients of each candidate metabolic pathway by solving the linear equations corresponding to the linear combination to obtain the contribution weight data corresponding to the candidate metabolic pathway, includes: constructing a linear combination equation, wherein the standardized gas response data is used as a known vector, the standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set is used as the column vector of the coefficient matrix, and the contribution weight data of each candidate metabolic pathway is used as the unknown vector to be solved; solving the linear combination equation using a constrained optimization algorithm, wherein the constraint condition applied by the constrained optimization algorithm is that the sum of all contribution weight data is a fixed constant and each contribution weight data is non-negative; minimizing the error between the standardized gas response data and the estimated response data reconstructed from the coefficient matrix and the unknown vector through iterative calculation to obtain the contribution weight data of each candidate metabolic pathway that satisfies the constraint condition.

[0040] In this embodiment, when constructing the linear combination equation, the preprocessed and standardized gas response data acquired at the current moment is considered as a known vector. This vector reflects the comprehensive response of the electronic nose sensor array to the volatile substances produced during the fermentation of staple grains. Simultaneously, the standard gas response feature data corresponding to each candidate metabolic pathway selected from the metabolic pathway feature database is used as the column vector of the coefficient matrix. These column vectors represent the typical gas response patterns produced by a single metabolic pathway under ideal or standard conditions. The contribution weight data of each candidate metabolic pathway, i.e., the relative contribution of each metabolic pathway to the current total gas response, is considered as an unknown vector to be solved. The establishment of this linear combination equation aims to describe, through a mathematical model, how the observed complex gas response is composed of the superposition of contributions from multiple basic metabolic pathways.

[0041] In this embodiment, when solving the linear combination equation using a constrained optimization algorithm, specific constraints are introduced to ensure that the solved contribution weight data has practical physical and biological significance. First, the imposed constraints require that the sum of all contribution weight data be a fixed constant, such as 1 (representing 100% contribution), reflecting that the observed gas response is a collective contribution from all candidate metabolic pathways. Second, each contribution weight data must be non-negative, consistent with the biological reality that metabolic pathways can only produce or contribute substances, not make negative contributions. Commonly used constrained optimization algorithms include non-negative least squares (NNLS), quadratic programming, or more general convex optimization algorithms. These algorithms can find the optimal solution while satisfying the preset constraints.

[0042] In this embodiment, iterative calculations minimize the error between the standardized gas response data and the estimated response data reconstructed from the coefficient matrix and the unknown vectors, thereby obtaining contribution weight data for each candidate metabolic pathway that satisfies the constraints. This iterative process aims to continuously adjust the unknown vectors (i.e., the contribution weight data) so that the difference (error) between the estimated response data reconstructed from these weights and coefficient matrices (standard gas response characteristic data) and the actually observed standardized gas response data is minimized. Minimization of the error is typically measured using mean squared error (MSE) or other appropriate loss functions. Iterative calculations ensure that, while satisfying constraints such as non-negativity and a fixed sum, the model can fit the actual observed data to the maximum extent, thus obtaining contribution weight data for each candidate metabolic pathway that best explains the current gas response pattern.

[0043] In this embodiment, through the above technical solution, this application can ensure that the calculated contribution weight data is not only mathematically optimal but also biologically meaningful. Specifically, the contribution weight of each metabolic pathway is non-negative, and the sum of the contributions of all candidate metabolic pathways conforms to a preset physical constant. This method effectively avoids instability or negative values ​​in contribution weight calculations caused by data noise or collinearity of metabolic pathway features, thereby improving the accuracy of dominant metabolic pathway identification and the reliability of contribution weight evaluation. By minimizing the reconstruction error through iterative calculation, the model's ability to fit actual gas response data is further enhanced, making the quantitative analysis of microbial metabolic activities during the fermentation of coarse grain staple foods more accurate, and providing a more solid data foundation for subsequent fermentation stage diagnosis and quality prediction.

[0044] In one feasible implementation, the steps of diagnosing the current fermentation stage based on the dominant metabolic pathway and the contribution weight data to obtain stage diagnosis result data, and generating a prediction of the final fermentation quality based on the dominant metabolic pathway, the contribution weight data, and the stage diagnosis result data to obtain quality prediction result data include: matching the dominant metabolic pathway and its corresponding contribution weight data with pre-stored fermentation stage discrimination rules, and determining the stage identifier of the current fermentation based on the matching result, as the stage diagnosis result data; wherein, the fermentation stage discrimination rules define predefined dominant metabolic pathways and their preset weight ranges corresponding to different fermentation stages; constructing multidimensional time series data based on the set of dominant metabolic pathways corresponding to each sampling time point from the start of fermentation to the current time, the contribution weight data of each dominant metabolic pathway, and the corresponding stage diagnosis result data; inputting the multidimensional time series data into a pre-trained fermentation quality prediction model to output a predicted value of the evaluation index for the final fermentation quality, and using the predicted value of the evaluation index as the quality prediction result data.

[0045] In this embodiment, the dominant metabolic pathway and its corresponding contribution weight data are matched with pre-stored fermentation stage discrimination rules, and the current fermentation stage identifier is determined based on the matching results, serving as stage diagnosis result data. This step aims to accurately determine the specific stage of fermentation of the coarse grain staple food based on the real-time detected microbial metabolic activity characteristics. The fermentation process is usually divided into multiple stages such as the initiation stage, vigorous stage, and maturity stage, and the microbial community composition and metabolic activity patterns differ in different stages. By matching the real-time identified dominant metabolic pathway and its contribution weight with the preset stage discrimination rules, refined management of the fermentation process can be achieved. The fermentation stage discrimination rules can be established in advance using experimental data or expert knowledge. For example, in the early stage of fermentation, the metabolic pathway of acid-producing bacteria may be dominant, with a high contribution weight; while in the later stage of fermentation, it may shift to the metabolic pathway of aroma-producing substances. These rules can be stored as lookup tables, decision trees, or rule-based expert systems. The matching process can employ methods such as fuzzy logic matching, threshold judgment, or machine learning classifiers to determine the current stage identifier based on whether the type of dominant metabolic pathway and its contribution weight fall within a preset range. For example, if the contribution weight of the lactic acid fermentation pathway is detected to be in a specific high value range, it may be determined as the vigorous fermentation period.

[0046] Based on this, multidimensional time series data is constructed using the set of dominant metabolic pathways corresponding to each sampling time point from the start of fermentation to the current moment, the contribution weight data of each dominant metabolic pathway, and the corresponding stage diagnostic results. This step aims to integrate the dynamically changing metabolic and stage information during fermentation into a data structure with a time dimension. Fermentation is a dynamic process, and snapshot information at a single time point is insufficient to fully reflect its development trend and final quality. By constructing multidimensional time series data, the evolutionary patterns of the fermentation process can be captured, providing richer and more context-sensitive input for subsequent quality prediction. Multidimensional time series data can be a matrix or tensor, where each row or slice represents a sampling time point, and the columns or dimensions contain the dominant metabolic pathway identifier (which can be one-hot encoded or numerically mapped), the contribution weight of each dominant metabolic pathway (e.g., normalized values), and the fermentation stage identifier diagnosed at that time point. This data can be stored chronologically to form a historical trajectory. For example, a sliding window approach can be used to retain only the data from the most recent N sampling points to focus on recent trends, or to retain all data from the start of fermentation to the present moment for global analysis.

[0047] In this embodiment, multidimensional time-series data is input into a pre-trained fermentation quality prediction model to output predicted values ​​of evaluation indicators for the final fermentation quality, which are then used as the quality prediction result data. This step aims to utilize machine learning or deep learning models to predict the quality of the final product based on the dynamic metabolism and stage information of the fermentation process. By predicting the final quality, potential problems can be identified and intervened in a timely manner during fermentation, avoiding product defects and improving production efficiency and product quality stability. The pre-trained fermentation quality prediction model can be various machine learning models, such as deep learning models suitable for processing time-series data, such as recurrent neural networks (RNN), long short-term memory networks (LSTM), and gated recurrent units (GRU), or traditional machine learning models such as support vector machines (SVM) and random forests, but feature engineering processing of the time-series data is required. The model's training data comes from historical fermentation batch data, including multidimensional time-series data during the fermentation process and corresponding final product quality evaluation indicators (such as taste, flavor, and nutritional components). By learning the patterns and correlations in this historical data, the model can predict the final quality based on new real-time multidimensional time-series data. For example, the model can output a numerical value representing the acidity, alcohol content, or content of a certain flavor substance in the final product, or output a classification result indicating whether the product meets the quality level of "excellent", "good", or "qualified".

[0048] In this embodiment, through the above-described technical solution, this application can achieve accurate stage diagnosis and effective prediction of the final quality of the fermentation process of coarse grain staple foods. Specifically, by matching the real-time acquired dominant metabolic pathways and their contribution weights with preset fermentation stage discrimination rules, the specific stage of fermentation can be dynamically and accurately identified, thereby avoiding the lag and inaccuracy caused by traditional experience-based judgment or offline detection. Based on this, by constructing multi-dimensional time-series data containing the entire fermentation process or key stages and inputting it into a pre-trained fermentation quality prediction model, the model can fully learn the dynamic evolution of the fermentation process and capture the complex nonlinear relationship between microbial metabolic activity and the quality of the final product. This allows the system to reliably predict the quality of the final product even before fermentation is complete, thus providing a forward-looking basis for the generation of subsequent control instructions. This forward-looking quality prediction capability makes online monitoring of the fermentation process no longer just about monitoring the state, but also about predicting future quality and proactively intervening, improving the intelligence level of the fermentation process and the stability of product quality.

[0049] In one feasible implementation, the step of inputting the multidimensional time series data into a pre-trained fermentation quality prediction model to output predicted values ​​of evaluation indicators for the final fermentation quality, and using these predicted values ​​as the quality prediction result data, includes: performing trend analysis on the contribution weight data of each dominant metabolic pathway in the multidimensional time series data, calculating the rate of change of the contribution weight of each dominant metabolic pathway within a preset time window; performing weighted calculation based on the rate of change and pre-stored influence weight coefficients of each metabolic pathway to obtain a set of metabolic activity intensity change indicators; matching the metabolic activity intensity change indicators with a pre-stored quality influence relationship table, and outputting the predicted values ​​of the evaluation indicators for the final fermentation quality as the quality prediction result data based on the matching results; wherein, the quality influence relationship table records the correspondence between the metabolic activity intensity change indicators and the final fermentation quality evaluation indicators.

[0050] In this embodiment, trend analysis is performed on the contribution weight data of each dominant metabolic pathway in the multidimensional time series data to reveal the dynamic pattern of these weights changing over time. Calculating the rate of change of the contribution weight of each dominant metabolic pathway within a preset time window is a key step in quantifying this dynamic pattern. This trend analysis and rate of change calculation can be implemented using various mathematical methods. For example, for the contribution weight data of each dominant metabolic pathway, the instantaneous rate of change can be obtained within the preset time window by calculating the ratio of the difference between adjacent sampling points to the time interval; alternatively, time series analysis methods such as linear regression and exponential smoothing can be used to fit a trend line, and the slope of the trend line can be calculated as the rate of change. The length of the preset time window can be flexibly set according to the characteristics of the fermentation process of the target grain staple food, the duration of the fermentation stage, and the sampling frequency of the electronic nose sensor array. For example, it can be the most recent 3, 5, or more sampling periods to ensure that meaningful changes in metabolic activity can be captured.

[0051] Based on this, a weighted calculation is performed according to the rate of change and the pre-stored weight coefficients of each metabolic pathway to obtain a set of indicators of metabolic activity intensity change. The weight coefficients of each metabolic pathway are predetermined and used to characterize the relative importance of the changes in the contribution weights of different dominant metabolic pathways to the final fermentation quality. These weight coefficients can be obtained in various ways, such as by setting them based on the experience of fermentation experts, by conducting statistical analysis (e.g., correlation analysis, regression analysis) on historical fermentation data, or by using machine learning models (e.g., feature importance assessment algorithms) for learning and optimization. The weighted calculation can be simply performed by multiplying the rate of change of each dominant metabolic pathway by its corresponding weight coefficient, and then summing all the weighted values ​​or performing other forms of combination (e.g., weighted average) to obtain a comprehensive indicator of metabolic activity intensity change.

[0052] Furthermore, the metabolic activity intensity change index is matched with a pre-stored quality influence relationship table to output the predicted value of the final fermentation quality evaluation index as the quality prediction result data. The quality influence relationship table is a pre-built knowledge base that establishes a mapping relationship between the metabolic activity intensity change index and the final fermentation quality evaluation index. This table can exist in various forms; for example, it can be a lookup table recording specific ranges of metabolic activity intensity change indices corresponding to specific quality grades (such as "Excellent," "Good," "Medium," and "Poor"), or corresponding to specific quality index numerical ranges (such as acidity, flavor compound content, and taste score). The matching process can be direct lookup, threshold-based range judgment, or more complex matching algorithms such as fuzzy logic. The construction of the quality influence relationship table is usually based on a large amount of experimental data, fermentation mechanism research, and the accumulation of expert knowledge.

[0053] In this embodiment, through the above technical solution, this application can conduct in-depth trend analysis on the contribution weight data of each dominant metabolic pathway in multidimensional time series data and quantify their rate of change, thereby capturing the dynamic evolution information of microbial metabolic activities during fermentation more precisely. By combining pre-stored metabolic pathway influence weight coefficients for weighted calculation, the comprehensive impact of changes in different metabolic pathways on the final quality can be assessed more accurately, resulting in a set of more representative and interpretable indicators of metabolic activity intensity changes. Furthermore, by matching these indicators with a pre-stored quality influence relationship table, the predicted values ​​of the final fermentation quality evaluation indicators can be output directly, efficiently, and with good interpretability. This method not only improves the accuracy and robustness of fermentation quality prediction, especially in complex and dynamically changing fermentation processes, but also enhances the interpretability of the prediction results through clear indicators and relationship tables, helping operators understand the reasons for quality formation and thus providing a more accurate basis for subsequent fermentation regulation.

[0054] In one feasible implementation, the step of comparing the stage diagnostic result data and the quality prediction result data with pre-stored standard fermentation process template data, and generating process control instruction data for regulating the fermentation process based on the comparison results, includes: comparing the stage diagnostic result data with the predefined dominant metabolic pathway and preset weight range of the corresponding stage in the standard fermentation process template data; if the dominant metabolic pathway does not belong to the predefined dominant metabolic pathway, or its corresponding contribution weight data exceeds the preset weight range, then generating first-type deviation alarm data; comparing the predicted value of the evaluation index in the quality prediction result data with the final quality target range defined in the standard fermentation process template data; if the predicted value of the evaluation index exceeds the target range, then generating second-type deviation alarm data; querying a pre-stored control strategy mapping table based on at least one of the first-type deviation alarm data and the second-type deviation alarm data to generate control action instructions including the control object and control amount, as the process control instruction data; wherein, the control strategy mapping table defines the correspondence between deviation types and control actions.

[0055] In this embodiment, during fermentation process monitoring, the stage diagnostic results data are first compared with the predefined dominant metabolic pathways and preset weight ranges for the corresponding stages in the standard fermentation process template data. This step aims to identify deviations from metabolic pathways during fermentation. The stage diagnostic results data reflect the main microbial metabolic activities of the current fermentation stage. The standard fermentation process template data predefines the dominant metabolic pathways that should appear in a specific fermentation stage and the preset range within which their corresponding contribution weights should fall. By comparing the actually detected dominant metabolic pathways with the predefined pathways in the template, it can be determined whether the current fermentation is proceeding as expected. If the actual dominant metabolic pathway is found to be inconsistent with the predefined pathway, for example, if an unexpected metabolic pathway appears, or if the contribution weight data of the identified dominant metabolic pathway exceeds the preset normal fluctuation range, it indicates that there is an abnormality at the metabolic level in the fermentation process. At this time, the system will generate a first type of deviation alarm data, which may include the type of deviation (such as metabolic pathway inconsistency or abnormal weight), the degree of deviation, and the specific metabolic pathway identifiers involved, providing a basis for subsequent precise control.

[0056] In this embodiment, the predicted values ​​of evaluation indicators in the quality prediction results data are simultaneously compared with the final quality target range defined in the standard fermentation process template data. This step is used to assess the potential impact of the fermentation process on the final product quality. The quality prediction results data provides predicted values ​​of evaluation indicators for the final quality of the whole grain staple food, such as taste, flavor substance content, and nutritional components. The standard fermentation process template data defines the final quality target range that these evaluation indicators should achieve under normal fermentation conditions. By comparing the predicted values ​​with the target range, it can be predicted whether the current fermentation process can achieve the expected product quality. If the predicted value exceeds the target range, whether too high or too low, it means that the final quality may not meet the standard. At this time, the system will generate a second type of deviation alarm data, which may include the type of quality indicator that is predicted to be substandard, the direction of deviation (higher or lower than the target), and the degree of deviation, to guide the control of quality problems.

[0057] Further, based on at least one of the first and second types of deviation alarm data, a pre-stored control strategy mapping table is queried to generate control action instructions, including the control object and control amount, as process control instruction data. This step intelligently generates specific control schemes based on the identified fermentation deviations. When at least one of the first type of deviation alarm data (abnormal metabolic pathway) or the second type of deviation alarm data (abnormal quality prediction) is detected, the system activates the control mechanism. The control strategy mapping table is a pre-established knowledge base that records in detail the correspondence between various deviation types (e.g., abnormality of a specific metabolic pathway, failure to meet the prediction of a certain quality indicator) and corresponding control actions. Each control action instruction explicitly specifies the "control object" (e.g., fermentation temperature, humidity, pH value, oxygen concentration, type or quantity of added microorganisms, material ratio, etc.) and the "control amount" (e.g., temperature increase of 2°C, humidity decrease of 5%, pH value adjustment to 4.5, addition of 10g of a specific strain, etc.). By querying this mapping table, the system can quickly and accurately retrieve the most appropriate control action command based on specific alarm information, thereby achieving precise intervention in the fermentation process. This mapping table can be constructed and optimized through expert experience, historical data analysis, or machine learning model training.

[0058] In this embodiment, by comparing the stage diagnostic results data with the predefined dominant metabolic pathways and preset weight ranges in the standard fermentation process template data, it is possible to promptly detect whether the metabolic activities of fermenting microorganisms deviate from the normal track, such as the presence of abnormal bacterial community activity or insufficient or excessive production of key metabolites, thereby generating the first type of deviation alarm data. Simultaneously, by comparing the predicted values ​​of evaluation indicators in the quality prediction results data with the final quality target range, it is possible to predict in advance whether the final product quality will fail to meet the standards, thereby generating the second type of deviation alarm data. These clearly categorized alarm data enable the system to accurately query the pre-stored control strategy mapping table based on the nature of the deviation and its impact on quality, quickly locating the most suitable control object and control amount, and generating targeted process control instruction data. This improves the intelligence level of fermentation process monitoring and the accuracy of control, avoids blind or delayed intervention, ensures the stability and consistency of fermented grain staple food quality, and effectively reduces production risks and costs.

[0059] In one feasible implementation, the step of querying a pre-stored control strategy mapping table based on at least one of the first type of deviation alarm data and the second type of deviation alarm data to generate control action instructions including the control object and control amount, as the process control instruction data, includes: if the first type of deviation alarm data exists, determining the current fermentation stage and the dominant metabolic pathway type of the deviation based on the data; if the second type of deviation alarm data exists, determining the predicted final quality evaluation index type that is not up to standard based on the data; constructing the current fermentation stage, the dominant metabolic pathway type of the deviation, and the predicted final quality evaluation index type that is not up to standard into a composite query condition, performing a joint query in the pre-stored control strategy mapping table, and generating the process control instruction data based on the query result; wherein, the dimensions of the control strategy mapping table correspond to the fermentation stage, metabolic pathway deviation type, and quality index deviation type, and the control strategy mapping table stores control action instructions for different composite query conditions.

[0060] In this embodiment, when the system detects the first type of deviation alarm data, it indicates that the dominant metabolic pathway or its contribution weight in the current fermentation process deviates from the standard fermentation process template. For precise control, it is necessary to parse the specific stage of the current fermentation and identify which one or more dominant metabolic pathways have deviated from the alarm data. This can be achieved by parsing the metadata contained in the first type of deviation alarm data. For example, the alarm data can be a structured data packet that explicitly includes the fermentation stage identifier (e.g., early, middle, late) when the alarm is triggered, as well as the ID or name of the deviated dominant metabolic pathway. Upon receiving the alarm, the system can directly read this information. Alternatively, if the alarm data is merely a simple identifier, the system can infer these details by combining the current timestamp, preset fermentation stage division rules, and deviation metabolic pathway information recorded when comparing with the standard template.

[0061] In this embodiment, similarly, when the system detects the second type of deviation alarm data, it indicates that, based on the current prediction, one or more evaluation indicators of the final fermentation quality will fail to meet the preset target. To effectively correct this, it is necessary to identify which one or more specific quality indicators (such as flavor compound content, acidity, taste, etc.) are at risk of failing to meet the standard. The second type of deviation alarm data should contain sufficient information to indicate the type of quality evaluation indicator that is not meeting the standard. For example, the alarm data could clearly indicate that "the predicted acidity value exceeds the target range" or "the predicted value of a specific flavor compound content is lower than the target lower limit." The system parses the alarm data and extracts the specific quality indicator name or type identifier. This helps in the formulation of subsequent control strategies, enabling targeted improvement of specific quality defects.

[0062] In this embodiment, a unique query key is formed by combining three types of information: the current fermentation stage, the type of the deviated dominant metabolic pathway, and the type of the predicted substandard final quality evaluation index. This constitutes a composite query condition. The composite query condition can be a tuple or a structure. In a pre-stored control strategy mapping table, each record's key corresponds to such a composite condition. After the system generates a composite query condition, it directly searches the mapping table for a record that perfectly matches the condition. If multiple deviations exist (e.g., both Type I and Type II alarms exist simultaneously), a composite query condition can be constructed based on preset priorities or combination rules, or multiple queries can be performed and the results merged.

[0063] To support such compound queries, the control strategy mapping table is categorized by dimensions corresponding to fermentation stage, metabolic pathway deviation type, and quality indicator deviation type. Furthermore, the table stores control action instructions for different compound query conditions. This means the mapping table is expanded into a multidimensional structure, capable of providing more targeted control instructions based on more detailed contextual information (fermentation stage, specific deviation type). The control strategy mapping table can be implemented as a multidimensional lookup table, hash table, or relational database. For example, it can be a three-dimensional array or nested dictionary, with indices corresponding to fermentation stage, metabolic pathway deviation type, and quality indicator deviation type, respectively. Each cell stores one or a set of predefined control action instructions, including the control object (e.g., temperature, humidity, pH, oxygen concentration, type and amount of additives) and the control amount (e.g., increase by 2°C, decrease by 0.5 pH, add 5g of yeast). These control action instructions are optimized results obtained through prior experiments, expert experience, or machine learning model training, aiming to achieve the best correction effect with minimal intervention. For example, in the middle of fermentation, if the lactic acid bacteria metabolic pathway deviates and the predicted acidity is insufficient, it might suggest "increasing the temperature by 2°C and adding a small amount of lactic acid bacteria activator."

[0064] In this embodiment, the above-mentioned technical solution refines the single deviation alarm data into three categories: the current fermentation stage, the dominant metabolic pathway type of the deviation, and the predicted type of final quality evaluation index that fails to meet the standards. These categories form a composite query condition. This multi-dimensional and refined query method allows for joint queries in a pre-stored control strategy mapping table, thereby obtaining control action instructions for specific fermentation scenarios. Compared to control based solely on a single deviation type, this solution provides more accurate and personalized process control instruction data. For example, when a specific metabolic pathway deviation occurs in the early stages of fermentation and the predicted final flavor is insufficient, the system can query the instruction "adjust fermentation temperature and supplement specific flavor precursors" based on composite conditions, rather than the general "adjust temperature." This improves the intelligence and refinement of fermentation process control, effectively avoids excessive or inappropriate intervention, ensures the stability and optimization of the fermentation quality of staple grains, and thus more effectively guarantees the quality of the final product and improves production efficiency.

[0065] In one feasible implementation, the method further includes: in a controlled fermentation simulation environment, inducing various pure cultured fermentation microorganisms to enter and maintain a preset metabolic pathway activity state, and collecting response signals from the electronic nose sensor array during the period when the preset metabolic pathway activity state remains stable; processing the collected response signals to obtain standard gas response signals corresponding to the metabolic pathway; extracting features from the standard gas response signals corresponding to each metabolic pathway to obtain standard gas response feature data for that metabolic pathway; and associating and storing the identification information of each metabolic pathway with its corresponding standard gas response feature data to form the metabolic pathway feature database.

[0066] In this embodiment, in a controlled fermentation simulation environment, multiple pure-cultured fermenting microorganisms are induced to enter and maintain a preset metabolic pathway activity state. The response signals of the electronic nose sensor array are collected while the preset metabolic pathway activity state remains stable. The collected response signals are processed to obtain the standard gas response signal corresponding to the metabolic pathway. This step aims to obtain the gas response characteristics of a specific metabolic pathway in a stable state. First, a controlled fermentation simulation environment needs to be constructed, which can precisely control key fermentation parameters such as temperature, humidity, pH, oxygen concentration, and nutrient composition to ensure that the growth and metabolic activities of microorganisms are under control. Second, a single pure-cultured fermenting microorganism is selected to avoid mutual interference between different microorganisms. By adjusting environmental parameters or adding specific metabolic substrates / inhibitors, the microorganism can be induced to enter and maintain a preset specific metabolic pathway activity state. For example, by changing the type and concentration of carbon or nitrogen sources, microorganisms can be guided to preferentially carry out alcoholic fermentation, lactic acid fermentation, or acetic acid fermentation. When the metabolic activity of the microorganism stabilizes in the target pathway, the volatile gas signals generated are collected in real time using the electronic nose sensor array. The acquired raw response signal needs to be preprocessed, such as baseline correction, noise filtering, and drift compensation, to eliminate environmental interference and errors caused by the sensor's own characteristics, thereby obtaining a pure and stable standard gas response signal that can accurately characterize the gas fingerprint of a specific metabolic pathway.

[0067] In this embodiment, feature extraction is then performed on the standard gas response signal corresponding to each metabolic pathway to obtain the standard gas response feature data for that metabolic pathway. After obtaining the standard gas response signal, feature extraction is required to transform it into structured data that can be used for database storage and subsequent matching. The purpose of feature extraction is to identify and quantify representative and discriminative information from the original signal. This may include, but is not limited to: signal peak value, peak area, response time, recovery time, signal integral value, ratio of responses from different sensors, eigenvectors after dimensionality reduction by principal component analysis (PCA) or independent component analysis (ICA), wavelet transform coefficients, Fourier transform spectral features, etc. Through these feature parameters, complex time-series response signals can be transformed into a set of values, thereby forming standard gas response feature data unique to that metabolic pathway. This feature data should be able to effectively distinguish different metabolic pathways and have a certain degree of robustness to environmental changes.

[0068] In this embodiment, the identification information of each metabolic pathway is finally associated and stored with its corresponding standard gas response feature data to form the metabolic pathway feature database. After feature extraction, the unique identification information of each metabolic pathway (e.g., metabolic pathway name, microbial strain information, induction conditions, etc.) is bound to its corresponding standard gas response feature data and stored in a structured database. This database can be in the form of a relational database, a non-relational database, or a file system. The purpose of the associated storage is to establish a mapping relationship between metabolic pathways and gas response features, so that in the subsequent online monitoring process, the corresponding metabolic pathway and its features can be quickly and accurately queried and matched using the gas response data collected in real time. The construction of this database is the foundation of the entire online monitoring method, and its completeness and accuracy directly determine the system's ability to identify fermentation status and predict quality.

[0069] In this embodiment, through the above-described technical solution, specific metabolic pathways of pure cultured microorganisms are systematically induced in a controlled fermentation simulation environment, and their stable electronic nose response signals are collected and processed. This application can accurately acquire the standard gas response signals unique to each metabolic pathway. Furthermore, feature extraction is performed on these standard signals, transforming them into structured standard gas response feature data, and they are associated and stored with metabolic pathway identification information, thereby constructing a comprehensive and accurate metabolic pathway feature database. The establishment of this database provides a solid data foundation for the above-described online monitoring method for the fermentation quality of coarse grain staple foods based on electronic noses. It solves the technical problem of accurately identifying the dominant metabolic activity due to the coexistence of multiple microorganisms and the complex intertwining of metabolic pathways during actual fermentation. By pre-acquiring pure and clear metabolic pathway features, the system can more accurately match real-time gas response data with standard features in the database during online monitoring, thereby accurately identifying the dominant metabolic pathway and its contribution weight in the current fermentation stage. This precise identification capability improves the accuracy of fermentation stage diagnosis and the reliability of final quality prediction, making the generation of process control instructions more targeted and effective, ultimately ensuring the stability and controllability of the fermentation quality of coarse grain staple foods.

[0070] In one feasible implementation, the step of extracting features from the standard gas response signal corresponding to each metabolic pathway to obtain the standard gas response feature data of that metabolic pathway includes: segmenting the standard gas response signal in the time dimension to obtain multiple time segments; calculating at least two feature parameters among the average value, maximum value, rising slope, and stable value of the response signals of each sensor in the electronic nose sensor array within each time segment; and combining and dimensionality-reducing the feature parameters calculated for all time segments to form standard gas response feature data that can characterize the gas response mode of that metabolic pathway.

[0071] In this embodiment, the standard gas response signal is first segmented along the time dimension to obtain multiple time segments. This step aims to decompose the continuous sensor response data stream into more easily analyzable discrete units, thereby enabling the capture of the dynamic changes in the gas response signal at different time stages. For example, the entire response period can be divided into several fixed-length time windows, or adaptive segmentation can be performed based on the signal change points to ensure that each time segment represents a specific state or stage in the gas response process.

[0072] In this embodiment, next, at least two characteristic parameters are calculated from the average value, maximum value, rising slope, and stable value of the response signals of each sensor in the electronic nose sensor array within each time segment. These characteristic parameters can quantify the sensor's response characteristics to the gas from different perspectives. For example, the average value can reflect the overall level of gas concentration within that time segment; the maximum value can indicate the peak value of gas concentration or response intensity; the rising slope can characterize the speed of sensor response, i.e., the drastic degree of change in gas composition; and the stable value represents the sustained response level of the sensor after reaching equilibrium. By selecting at least two parameters for calculation, it can be ensured that the extracted features include both static and dynamic information, thereby providing a more comprehensive description of the gas response pattern. For example, the average value and rising slope can be calculated simultaneously to take into account both response intensity and response speed.

[0073] In this embodiment, the feature parameters calculated from all time segments are finally combined and dimensionality-reduced to form standard gas response feature data that can characterize the gas response pattern of the metabolic pathway. During the combination stage, the feature parameters calculated by all sensors in all time segments are concatenated to form a high-dimensional feature vector. Subsequently, to remove data redundancy, reduce computational complexity, and improve the robustness of the features, principal component analysis (PCA), linear discriminant analysis (LDA), or other suitable dimensionality reduction algorithms can be used to process this high-dimensional feature vector. The dimensionality-reduced feature data will be a compact and highly discriminative vector that can effectively capture and represent the unique gas response patterns of volatile substances produced by a specific metabolic pathway.

[0074] In this embodiment, the standard gas response signal is segmented along the time dimension using the aforementioned technical solution, and multi-dimensional feature parameters are extracted from each time segment. This comprehensively captures the dynamic changes and key features of the gas response signal. Subsequently, these feature parameters are combined and dimensionality-reduced, effectively removing redundant information and noise, resulting in compact and highly discriminative standard gas response feature data. This refined feature extraction method enables the feature data stored in the metabolic pathway feature database to more accurately characterize the gas response patterns of different metabolic pathways, thereby improving the accuracy and robustness of matching standardized gas response data with pre-stored metabolic pathway feature data during subsequent online monitoring. Ultimately, this helps to more accurately identify dominant metabolic pathways and their contribution weights, providing more reliable basic data support for online monitoring of the fermentation quality of coarse grain staple foods.

[0075] In the embodiments of this application, the online monitoring method for fermentation quality of staple grains based on electronic nose acquires the raw gas response signal of the electronic nose sensor array and preprocesses it to generate standardized gas response data. Based on this data, it matches the dominant metabolic pathway and its contribution weight, diagnoses the current fermentation stage and predicts the final quality, and generates control instructions to adjust fermentation parameters by comparing with standard templates. This method can monitor the dominant metabolic pathway in the fermentation process in real time, achieve accurate stage diagnosis and quality prediction, and timely control the fermentation process to improve the stability of fermentation quality and production efficiency.

[0076] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the online monitoring method for the fermentation quality of staple grains based on electronic nose. Any simple modifications based on this technical concept are within the protection scope of this application.

[0077] This application also provides an online monitoring system for the fermentation quality of staple grains based on an electronic nose, referenced. Figure 2 The electronic nose-based online monitoring system for the fermentation quality of coarse grain staple foods includes: a memory 10, a processor 20, and an electronic nose-based online monitoring program for the fermentation quality of coarse grain staple foods stored on the memory 10 and executable on the processor 20. The electronic nose-based online monitoring program for the fermentation quality of coarse grain staple foods is configured to implement the steps of the electronic nose-based online monitoring method for the fermentation quality of coarse grain staple foods.

[0078] The electronic nose-based online monitoring system for the fermentation quality of staple grains provided in this application, employing the electronic nose-based online monitoring method for the fermentation quality of staple grains in the above embodiments, can improve the stability of fermentation quality and production efficiency. Compared with the prior art, the beneficial effects of the electronic nose-based online monitoring system for the fermentation quality of staple grains provided in this application are the same as those of the electronic nose-based online monitoring method for the fermentation quality of staple grains provided in the above embodiments, and other technical features of the electronic nose-based online monitoring system for the fermentation quality of staple grains are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0079] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0080] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. All equivalent structural transformations made under the technical concept of this application using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the scope of patent protection of this application.

Claims

1. A method for online monitoring of the fermentation quality of coarse grain staple foods based on an electronic nose, characterized in that, The method includes: The original gas response signal generated by the electronic nose sensor array during the fermentation of the target grain staple food is obtained in real time, and the original gas response signal is preprocessed to generate standardized gas response data. Based on the standardized gas response data, at least one dominant metabolic pathway and its corresponding contribution weight data are matched from a pre-stored metabolic pathway feature database; wherein, the metabolic pathway feature database pre-stores standard gas response feature data corresponding to the metabolic pathways of various fermenting microorganisms. Based on the dominant metabolic pathway and the contribution weight data, the current fermentation stage is diagnosed to obtain stage diagnosis result data. Based on the dominant metabolic pathway, the contribution weight data and the stage diagnosis result data, a prediction of the final fermentation quality is generated to obtain quality prediction result data. The stage diagnosis result data and the quality prediction result data are compared with the pre-stored standard fermentation process template data, and process control instruction data for regulating the fermentation process is generated based on the comparison results. Based on the process control instruction data, the fermentation environment parameters or material composition of the target coarse grain staple food are adjusted.

2. The online monitoring method for fermentation quality of staple grains based on electronic nose as described in claim 1, characterized in that, Based on the standardized gas response data, the step of matching at least one dominant metabolic pathway and its corresponding contribution weight data from a pre-stored metabolic pathway feature database includes: The standardized gas response data is correlated with each group of standard gas response feature data in the metabolic pathway feature database to obtain multiple correlation metrics. Based on the correlation metric, metabolic pathways corresponding to standard gas response feature data whose correlation metric exceeds a preset threshold are selected from the metabolic pathway feature database to obtain a candidate metabolic pathway set. The standardized gas response data is represented as a linear combination of the standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set. The linear coefficients of each candidate metabolic pathway are calculated by solving the linear equations corresponding to the linear combination, and are used as the contribution weight data corresponding to the candidate metabolic pathway. Based on the contribution weight data, the top N candidate metabolic pathways with the largest contribution weight data are determined from the candidate metabolic pathway set to serve as the dominant metabolic pathways, where N is a positive integer.

3. The online monitoring method for fermentation quality of staple grains based on an electronic nose as described in claim 2, characterized in that, The steps of representing the standardized gas response data as a linear combination of the standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set, and calculating the linear coefficients of each candidate metabolic pathway by solving the linear equations corresponding to the linear combination, to serve as the contribution weight data corresponding to that candidate metabolic pathway, include: A linear combination equation is constructed, wherein the standardized gas response data is used as a known vector, the standard gas response feature data corresponding to all candidate metabolic pathways in the candidate metabolic pathway set is used as the column vector of the coefficient matrix, and the contribution weight data of each candidate metabolic pathway is used as the unknown vector to be solved. The linear combination equation is solved using a constrained optimization algorithm, wherein the constraint condition imposed by the constrained optimization algorithm is that the sum of all contribution weight data is a fixed constant and each contribution weight data is non-negative; The error between the standardized gas response data and the estimated response data reconstructed from the coefficient matrix and the unknown vector is minimized through iterative calculations to obtain the contribution weight data of each candidate metabolic pathway that satisfies the constraints.

4. The online monitoring method for fermentation quality of staple grains based on an electronic nose as described in claim 1, characterized in that, The steps of diagnosing the current fermentation stage based on the dominant metabolic pathway and the contribution weight data to obtain stage diagnosis result data, and generating a prediction of the final fermentation quality based on the dominant metabolic pathway, the contribution weight data, and the stage diagnosis result data to obtain quality prediction result data include: The dominant metabolic pathway and its corresponding contribution weight data are matched with pre-stored fermentation stage discrimination rules, and the current fermentation stage identifier is determined based on the matching results, which is used as the stage diagnosis result data; wherein, the fermentation stage discrimination rules define predefined dominant metabolic pathways and their preset weight ranges corresponding to different fermentation stages. Multidimensional time series data is constructed based on the set of dominant metabolic pathways corresponding to each sampling time point from the start of fermentation to the present moment, the contribution weight data of each dominant metabolic pathway, and the corresponding stage diagnostic results data. The multidimensional time series data is input into a pre-trained fermentation quality prediction model to output a predicted value of the evaluation index for the final fermentation quality, and the predicted value of the evaluation index is used as the quality prediction result data.

5. The online monitoring method for fermentation quality of staple grains based on an electronic nose as described in claim 4, characterized in that, The steps of inputting the multidimensional time series data into a pre-trained fermentation quality prediction model to output predicted values ​​of evaluation indicators for the final fermentation quality, and using these predicted values ​​as the quality prediction result data, include: Trend analysis is performed on the contribution weight data of each dominant metabolic pathway in the multidimensional time series data to calculate the rate of change of the contribution weight of each dominant metabolic pathway within a preset time window. Based on the rate of change, a set of metabolic activity intensity change indicators are obtained by weighting the pre-stored weighting coefficients of each metabolic pathway. The metabolic activity intensity change index is matched with a pre-stored quality influence relationship table, and the predicted value of the final fermentation quality evaluation index is output as the quality prediction result data based on the matching result; wherein, the quality influence relationship table records the correspondence between the metabolic activity intensity change index and the final fermentation quality evaluation index.

6. The online monitoring method for fermentation quality of staple grains based on electronic nose as described in claim 1, characterized in that, The steps of comparing the stage diagnostic results data and the quality prediction results data with pre-stored standard fermentation process template data, and generating process control instruction data for regulating the fermentation process based on the comparison results, include: The stage diagnosis result data is compared with the predefined dominant metabolic pathway and preset weight range of the corresponding stage in the standard fermentation process template data. If the dominant metabolic pathway does not belong to the predefined dominant metabolic pathway, or its corresponding contribution weight data exceeds the preset weight range, then a first type of deviation alarm data is generated. The predicted values ​​of the evaluation indicators in the quality prediction results data are compared with the final quality target range defined in the standard fermentation process template data. If the predicted values ​​of the evaluation indicators exceed the target range, a second type of deviation alarm data is generated. Based on at least one of the first type of deviation alarm data and the second type of deviation alarm data, a pre-stored control strategy mapping table is queried to generate control action instructions including the control object and control amount, which are used as the process control instruction data; wherein, the control strategy mapping table defines the correspondence between deviation type and control action.

7. The online monitoring method for fermentation quality of staple grains based on an electronic nose as described in claim 6, characterized in that, The step of querying a pre-stored control strategy mapping table to generate a control action instruction including the control object and control amount, based on at least one of the first type of deviation alarm data and the second type of deviation alarm data, as the process control instruction data, includes: If the first type of deviation alarm data exists, the current fermentation stage and the type of dominant metabolic pathway deviation are determined based on the data. If the second type of deviation alarm data exists, the type of final quality evaluation index that is predicted to be unsatisfactory is determined based on the data. The current fermentation stage, the type of the deviated dominant metabolic pathway, and the type of the predicted substandard final quality evaluation index are used as composite query conditions. These conditions are then jointly queried in the pre-stored regulation strategy mapping table, and the process regulation instruction data is generated based on the query results. The dimensions of the regulation strategy mapping table correspond to the fermentation stage, the type of metabolic pathway deviation, and the type of quality index deviation. The regulation strategy mapping table stores regulation action instructions for different composite query conditions.

8. The method for online monitoring of fermentation quality of staple grains based on electronic nose as described in claim 1, characterized in that, The method further includes: In a controlled fermentation simulation environment, various pure culture fermentation microorganisms are induced to enter and maintain a preset metabolic pathway activity state. During the period when the preset metabolic pathway activity state is stable, the response signal of the electronic nose sensor array is collected. The collected response signal is processed to obtain the standard gas response signal corresponding to the metabolic pathway. Feature extraction is performed on the standard gas response signal corresponding to each metabolic pathway to obtain the standard gas response feature data of that metabolic pathway; The identification information of each metabolic pathway is associated with and stored with its corresponding standard gas response characteristic data to form the metabolic pathway characteristic database.

9. The online monitoring method for fermentation quality of staple grains based on an electronic nose as described in claim 8, characterized in that, The steps of extracting features from the standard gas response signal corresponding to each metabolic pathway to obtain the standard gas response feature data of that metabolic pathway include: The standard gas response signal is segmented along the time dimension to obtain multiple time segments; Calculate at least two characteristic parameters among the average value, maximum value, rising slope, and stable value of the response signals of each sensor in the electronic nose sensor array within each time segment; The characteristic parameters calculated from all time segments are combined and dimensionality reduced to form standard gas response characteristic data that can characterize the gas response pattern of this metabolic pathway.

10. An online monitoring system for the fermentation quality of coarse grain staple foods based on an electronic nose, characterized in that, The electronic nose-based online monitoring system for the fermentation quality of coarse grain staple foods includes: a memory, a processor, and an electronic nose-based online monitoring program for the fermentation quality of coarse grain staple foods stored in the memory and executable on the processor. The electronic nose-based online monitoring program for the fermentation quality of coarse grain staple foods is configured to implement the steps of the electronic nose-based online monitoring method for the fermentation quality of coarse grain staple foods as described in any one of claims 1 to 9.

Images