Dynamic control method for fermentation process of Kaiqi koji based on microbial activity monitoring

Abstract

The invention provides a microbial activity monitoring-based Kaiqi yeast fermentation process dynamic control method, which comprises the following steps: extracting significant change trend characteristics according to a data set, and determining a time window and an amplitude range of pH value deviation and a corresponding product component deviation degree; according to the characteristics of the target flora combination, a corresponding pH value adjustment scheme is matched from a preset regulation and control strategy library, adaptive intervention measure parameters are obtained, and an execution instruction set is generated; according to the execution instruction set, an acid-base regulator is injected into the fermentation system, the environmental condition is adjusted in real time, pH value feedback data after intervention is obtained, and whether the target range interval is restored or not is judged; and carrying out component detection on the new fermentation product component data in the balance state to obtain a quantitative index of product stability and obtain an improved regulation and control strategy library.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a dynamic control method for the fermentation process of *Citrus aurantium* based on microbial activity monitoring. Background Technology

[0002] In the field of food fermentation and microbial regulation, studying the impact of environmental factors on microbial communities during fermentation is of great significance. This field is not only crucial for ensuring product quality and function but also central to modernizing traditional processes. However, current research and application methods often struggle to accurately grasp the deep interaction between environmental changes and microbial community dynamics when dealing with complex fermentation systems. In particular, insufficient understanding of the dynamic response mechanisms of key environmental factors in fermentation systems leads to a lack of real-time control measures for microbial community structure succession when environmental conditions fluctuate. This technical limitation is mainly reflected in the insufficient understanding of the dynamic response mechanisms of key environmental factors in fermentation systems, resulting in control measures often being lagging or unsuitable for specific changes, thus affecting the stability and functionality of fermentation products. Against this backdrop, changes in pH within the fermentation system have become a pressing challenge. pH shifts not only directly affect the composition of the microbial community but also further lead to an imbalance in the proportions of different functional microorganisms. For example, the activity of some beneficial bacteria may be inhibited, while other bacteria detrimental to the fermentation objective may become dominant. This imbalance causes the composition of fermentation products to deviate from expectations, affecting the quality and efficacy of the final product. A deeper issue lies in how to restore balance promptly through external intervention when pH levels shift, and how such intervention may have unpredictable impacts on the long-term succession of the microbial community. Therefore, how to dynamically regulate pH during fermentation to balance the proportions of different functional microorganisms and stabilize the composition of fermentation products through reasonable intervention strategies has become a critical problem that urgently needs to be solved. Summary of the Invention

[0003] This invention provides a dynamic control method for the fermentation process of Kaiyuqu based on microbial activity monitoring, mainly including:

[0004] In the fermentation system, pH change data and fermentation product component data are continuously collected and timestamped to generate a multidimensional dataset containing environmental state and product component information. Noise suppression and data smoothing are applied to the multidimensional dataset to extract the time window, amplitude range, and product component deviation degree of pH shift. Based on the time window and amplitude range of pH shift, a pre-established microbial community database is queried to obtain functional microbial activity data. Combined with the product component deviation degree, a risk index value for microbial community imbalance is calculated. The risk index value is classified using a support vector machine model to determine the target microbial community combination. Based on the target microbial community combination, a pH adjustment scheme is matched from a pre-set control strategy library to generate an execution instruction set. The execution instruction set is executed to obtain pH feedback data after intervention, determining whether it has recovered to the target range. Based on the pH feedback data after intervention, the dynamic recovery index of functional microbial activity is monitored to determine whether the fermentation system has reached a new equilibrium state. The fermentation product component data under the new equilibrium state is detected to obtain a quantitative index of product stability, and the control strategy library is optimized.

[0005] Furthermore, the continuous collection of pH change data and fermentation product component data in the fermentation system, embedding timestamp markers, generates a multidimensional dataset containing environmental state and product component information, including:

[0006] pH electrodes and conductivity sensors are deployed on the inner wall of the fermenter, at the inlet, outlet, and temperature control area to acquire raw acid-base voltage signals. Noise is processed using a Kalman filter algorithm to generate a pH value sequence. The concentrations of organic acids, alcohols, and sugar metabolites in the fermentation broth are detected using chromatographic analysis equipment to generate a component data matrix. A network time synchronization protocol is used to assign millisecond-level time stamps to the pH value sequence and the component data matrix. A support vector machine algorithm is used to identify key time nodes in the pH value sequence, dividing the fermentation stages and constructing a multidimensional dataset containing temporal relationships.

[0007] Furthermore, the step of performing noise suppression and data smoothing on the multidimensional dataset, and extracting the time window, amplitude range, and product component deviation of the pH shift, includes:

[0008] The moving average filtering method is used to suppress noise in the pH and product component data of the multidimensional dataset, generating first pH fluctuation information and first product component offset data. The first pH fluctuation information and first product component offset data are then smoothed using cubic spline interpolation to generate second pH fluctuation information and second product component offset data. The first derivative of the second pH fluctuation information is analyzed using gradient calculation to mark the time window and amplitude range of the pH offset. Based on the time window, the component concentration change value is extracted from the second product component offset data, the deviation percentage is calculated, and the degree of product component deviation is determined.

[0009] Furthermore, the step of querying a pre-established microbial community database based on the time window and amplitude range of the pH shift to obtain functional microbial community activity data, and calculating a risk index value for microbial community imbalance in conjunction with the degree of deviation of the product components, includes:

[0010] The system accesses the microbial community database through a query interface, retrieves historical microbial community response records that match the time window and amplitude range of the pH shift, and generates a first microbial community feature dataset. It then filters functional microbial communities in the first microbial community feature dataset whose acid tolerance is below the amplitude range, obtains metabolic activity decay coefficients, and generates functional microbial community activity data. Finally, it analyzes the degree of deviation between the functional microbial community activity data and the product components using a decision tree algorithm, constructs a correlation matrix between microbial community activity and product concentration, and calculates the risk index value for microbial community imbalance.

[0011] Furthermore, the step of classifying the risk indicator values ​​using a support vector machine model to determine the target microbial community combination includes:

[0012] The risk index values ​​are converted into feature vectors using a radial basis function kernel function, and an imbalance type identifier is generated using a support vector machine model. Based on the imbalance type identifier, a microbial community influence relationship lookup table is consulted to generate a list of candidate functional microbial communities. The list of candidate functional microbial communities is analyzed using a Bayesian probability calculation method to determine the list of functional microbial community categories. The interaction strength in the list of functional microbial community categories is analyzed using a correlation calculation method to construct a microbial community correlation matrix and generate the target microbial community combination.

[0013] Furthermore, the step of matching a pH adjustment scheme from a preset regulatory strategy library based on the target microbial community combination and generating an execution instruction set includes:

[0014] A target microbial community characteristic data table is generated based on the target microbial community combination; the target microbial community characteristic data table is compared with the regulation strategy library to generate a set of pH adjustment schemes; the set of pH adjustment schemes is evaluated by a similarity calculation method to generate a combination of intervention measures parameters; the combination of intervention measures parameters is converted into equipment control instructions to generate an execution instruction set.

[0015] Furthermore, the execution of the execution instruction set, obtaining the pH feedback data after intervention, and determining whether it has recovered to the target range include:

[0016] According to the execution instruction set, an acid-base regulator is injected through an automatic feeding device, generating an injection operation record; environmental conditions are adjusted through a temperature regulation device and a ventilation regulation device, generating real-time environmental condition adjustment data; acid-base values ​​in the real-time environmental condition adjustment data are collected through a pH sensor network, generating post-intervention acid-base feedback data; and it is determined whether the post-intervention acid-base feedback data is within the target range.

[0017] Furthermore, the step of monitoring the dynamic recovery indicators of functional microbial community activity based on the pH feedback data after the intervention to determine whether the fermentation system has reached a new equilibrium state includes:

[0018] Collect the functional microbial community quantity distribution and metabolic activity indicators corresponding to the pH feedback data after the intervention, and generate a dynamic recovery index of functional microbial community activity; evaluate the change rate of the dynamic recovery index of functional microbial community activity through time series analysis, determine whether the microbial community ratio is within the stable threshold range, and determine whether the fermentation system has reached a new equilibrium state.

[0019] Furthermore, the detection of fermentation product component data under the new equilibrium state, obtaining quantitative indicators of product stability, and optimizing the regulatory strategy library includes:

[0020] Multi-band scanning is performed on the fermentation product component data under the new equilibrium state to generate spectral data; the spectral data is converted into quantitative concentration values ​​through peak identification and intensity analysis to generate product stability quantification index, and the pH adjustment threshold and environmental control parameters in the regulation strategy library are updated.

[0021] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0022] This invention discloses a dynamic control method for the fermentation process of *Aristolochia debilis* based on microbial activity monitoring. Addressing the business scenario where pH fluctuations during fermentation lead to microbial community imbalance and product component deviation, the method collects pH and product component data in real time, generates a preliminary dataset using timestamps, and extracts significant trends after denoising and smoothing to determine the time window and magnitude of pH shift. Through comparison with a microbial community database and classification using a support vector machine model, it accurately identifies suppressed functional microorganisms and the risk of imbalance. If the risk exceeds a threshold, the invention matches a pH adjustment scheme from a control strategy library, generates an execution instruction set, injects a regulator through an automated control system, and provides real-time feedback and dynamic optimization of intervention measures. If the target range is not reached, a multi-level decision algorithm iteratively enhances the intervention until an emergency mechanism is triggered. This invention verifies product stability through spectral analysis, optimizes the response mechanism model, and continuously improves the control strategy, ultimately achieving dynamic balance and efficient, stable operation of the fermentation system. Attached Figure Description

[0023] Figure 1 This is a flowchart of a dynamic control method for the fermentation process of *Cynanchum paniculatum* based on microbial activity monitoring, according to the present invention. Detailed Implementation

[0024] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] like Figure 1 This embodiment of a dynamic control method for the fermentation process of Kaiyuqu based on microbial activity monitoring may specifically include:

[0026] S101. Continuously collect pH change data in the fermentation system, obtain real-time pH fluctuation information, and simultaneously monitor the real-time changes of fermentation product components. Embed timestamp markers in the data stream to obtain a dataset.

[0027] Based on the pre-designed fermentation reactor layout, distributed sensor nodes are deployed at the fermenter's inner wall, inlet, outlet, and temperature control area, using pH electrodes and conductivity sensors. The raw pH voltage signals at each monitoring point are acquired via the ModBus communication protocol. Kalman filtering is used to process sensor noise interference, resulting in a stable pH value sequence and corresponding environmental parameter status values. If the variation amplitude of multiple consecutive sampling points in the pH value sequence exceeds a preset threshold, a product component detection program is triggered. Pre-configured chromatographic analysis equipment automatically analyzes the concentration changes of organic acids, alcohols, and sugar metabolites in the fermentation broth, obtaining a component data matrix containing product type identifiers and concentration values. A network time synchronization protocol is used to unify the clock reference of each sensor node. Millisecond-level time stamps are assigned to each set of pH value sequences and component data matrices. A support vector machine algorithm is used to identify key time nodes and abnormal fluctuation patterns in the pH value sequences, determining the time boundary points for fermentation stage transitions. Based on the fermentation stage information divided by the time boundary points, a data fusion method is used to correlate and map the pH value sequence, the component data matrix, and the environmental parameter state values ​​within the same time window. A multidimensional dataset containing time-series relationships is constructed through a structured storage format. The dataset can be used to characterize the environmental product correlation feature patterns of each fermentation stage.

[0028] For example, the deployment of distributed sensor nodes adopts a spatial hierarchical strategy. When setting pH electrodes on the inner wall of the fermenter, the flow characteristics of the fermentation broth and the temperature gradient distribution need to be considered.

[0029] For example, in a 2000-liter fermenter, sensors are typically deployed at three heights: 0.5 meters, 1.5 meters, and 2.5 meters from the bottom of the tank. Each height has four monitoring points, forming a monitoring network of 12 nodes. This layout can capture vertical and radial pH variations during fermentation, as microbial activity and nutrient concentrations differ significantly at different locations. The ModBus communication protocol plays a crucial role in sensor data acquisition, enabling unified management of multiple sensor nodes through a master-slave architecture.

[0030] In one embodiment, the main controller sends a query command to each sensor node every second to acquire a 4-20mA standard current signal, which is then converted into the corresponding pH value. The Kalman filter algorithm is applied based on the continuous nature of pH changes during fermentation. It predicts the pH value at the next moment by establishing a state transition model and compares and corrects it with the actual measured value, thereby eliminating instantaneous fluctuations caused by stirring, feeding, and other operations. Based on the aforementioned stable pH value sequence, the triggering mechanism for product component detection reflects the intelligent characteristics of fermentation process monitoring.

[0031] It should be noted that when the pH value changes significantly within a short period of time, it often indicates a shift in the fermentation metabolic pathway. Therefore, timely detection of changes in product components is crucial. Chromatographic analysis equipment uses an autosampler to extract samples from the fermenter and utilizes the differences in retention times of different components within the chromatographic column to achieve separation and detection.

[0032] For example, the retention time of lactic acid is approximately 3.2 minutes, ethanol is 5.8 minutes, and glucose is 8.1 minutes. The accurate concentration values ​​of each component can be calculated by integrating the peak area. Time synchronization plays a fundamental role in multi-sensor data fusion; network time synchronization protocols ensure that the timestamp accuracy of each node reaches the millisecond level.

[0033] One possible implementation employs the IEEE 1588 precise time protocol, using a network latency compensation mechanism to achieve high-precision synchronization between each sensor node and the master clock. The application of the Support Vector Machine (SVM) algorithm in time series analysis leverages its excellent nonlinear classification capabilities. By training on pH change patterns in historical fermentation data, it can identify characteristic signals of fermentation stage transitions; for example, pH values ​​typically exhibit a specific decreasing trend during the transition from the exponential growth phase to the stationary phase. The core of the data fusion method lies in establishing correlations between multidimensional data, using the time dimension as the primary index, and aligning pH value sequences, product component concentrations, and environmental parameters according to timestamps.

[0034] Specifically, each time point corresponds to a data vector containing 12 pH measurements, concentrations of 8 major metabolites, and 6 environmental parameters. Correlation analysis revealed a significant negative correlation between pH and lactic acid concentration, and a positive correlation with amino acid consumption rate. These correlation patterns provide important evidence for the precise control of the fermentation process.

[0035] S102. Based on the dataset, extract significant trend features, determine the time window and range of pH shift, and the corresponding degree of deviation of product components.

[0036] Based on the pH fluctuation and product component data in the dataset, a moving average filtering method is used to suppress noise in the raw data collected by each sensor node. High-frequency random interference and equipment jitter signals are removed by setting a filter window length, resulting in noise-suppressed first pH fluctuation information and first product component offset data. If the difference between adjacent data points in the first pH fluctuation information exceeds a preset fluctuation threshold, a smoothing process is triggered based on the first pH fluctuation information and the first product component offset data. A cubic spline interpolation method is used to perform curve fitting on the sequence, obtaining continuous and smooth second pH fluctuation information and second product component offset data. The gradient calculation method is used to analyze the first derivative change pattern of the second pH fluctuation information, identifying data segments where the absolute gradient value exceeds a preset significance standard. A sliding time window method is used to mark the start and end times of significant pH changes, determining the specific time window and corresponding amplitude range of the pH offset. Based on the time window information, component concentration changes within the same time period are extracted from the second product component offset data. By calculating the percentage deviation of each component concentration relative to a preset benchmark concentration, the specific degree of deviation and direction of change of the product components within the amplitude range is determined.

[0037] For example, the application of the moving average filtering method in fermentation data processing is based on the basic principles of signal processing, which eliminates random noise by averaging data points within a continuous time window.

[0038] For example, when the pH sensor is affected by mechanical vibrations such as the start / stop of a stirrer or the switching of a feed pump, the raw data may experience momentary jumps. In such cases, using a 5-point moving average can effectively smooth out these interference signals. The choice of filter window length directly affects the noise suppression effect; if the window is too small, the filtering effect will be insignificant, while if the window is too large, true information about biological process changes will be lost.

[0039] In one embodiment, noise suppression employs differentiated strategies for different types of interference. For high-frequency electromagnetic interference generated by the pH sensor, which typically manifests as random fluctuations with small amplitude but high frequency, moving average filtering can significantly reduce the impact of this type of noise. However, for deviations in product component detection caused by uneven sample pretreatment, outlier identification and removal are necessary by combining the statistical characteristics of the data. This stratified processing strategy ensures data quality while preserving the true dynamic characteristics of the fermentation process. Based on the noise-suppressed data, the introduction of cubic spline interpolation solves the problems of data discontinuity and uneven sampling intervals.

[0040] It should be noted that during fermentation, equipment maintenance, sensor calibration, and other operations may cause interruptions in data acquisition, resulting in missing data. Cubic spline interpolation constructs a piecewise cubic polynomial function to reasonably estimate the missing data points while ensuring the continuity of the function and its first and second derivatives.

[0041] For example, when pH data is missing for a certain 15-minute period, the interpolation algorithm uses the values ​​and trends from before and after the missing time points to generate a smooth transition curve. The setting of trigger conditions reflects the intelligent characteristics of adaptive data processing.

[0042] In one possible implementation, the preset fluctuation threshold is dynamically adjusted according to different fermentation stages, as the pH change amplitude differs significantly between the exponential growth phase and the stationary phase. When data fluctuations are detected to exceed the normal range, a smoothing process is automatically initiated, avoiding unnecessary processing of all data, thus improving computational efficiency while preserving the original characteristics of the data. Gradient calculation methods play a central role in identifying trends, quantifying the rate of change by calculating the ratio of differences between adjacent data points.

[0043] Specifically, when the pH value decreases by more than 0.5 units within 30 minutes, the absolute value of the first derivative increases significantly, indicating that the fermentation process may have entered a new metabolic phase. The sliding time window technique employs an overlapping window strategy, with each window covering 60 minutes of data and 50% overlap between windows. This design accurately captures the start and end points of changing trends, avoiding the omission of critical process transition points. The determination of the significance criteria combines prior knowledge of the fermentation process with statistical methods.

[0044] For example, based on statistical analysis of historical fermentation data, a significant change can be considered to have occurred when the absolute value of the gradient exceeds twice the standard deviation of the mean. A dual-threshold strategy is used to mark time boundaries: a high threshold is used to detect the onset of a change, and a low threshold is used to determine the end of the change. This method effectively avoids misjudgments caused by noise. The calculation of the degree of deviation of product components is based on a preset baseline concentration, which is usually derived from the target setting of the fermentation process or data from historical best batches.

[0045] In one embodiment, when a significant pH shift occurs, the system simultaneously calculates the concentration changes of key metabolites such as lactic acid, ethanol, and glucose. Percentage calculations visually reflect the deviations of each component; for example, a 15% increase in lactic acid concentration relative to the baseline and an 8% decrease in glucose concentration provide quantitative data for the diagnosis and optimized control of fermentation process anomalies.

[0046] S103. For the time window and range of pH shift, a comparative analysis is performed using a pre-established microbial community database to obtain data on the activity of functional microbial communities that may be inhibited under the shift conditions. The risk index value of microbial community ratio imbalance is calculated in combination with the degree of deviation of product components.

[0047] Based on the time window and amplitude range parameters of the pH offset, a pre-established microbial community database is accessed through a query interface. Historical microbial response records corresponding to the current offset conditions are retrieved using conditional matching, resulting in a first microbial community feature dataset containing microbial species identifiers, acid tolerance parameters, and optimal pH ranges. If any microbial records in the first microbial community feature dataset have acid tolerance parameters lower than the current amplitude range, functional microbial communities whose activity is inhibited under the current offset conditions are selected based on the first microbial community feature dataset. The metabolic activity attenuation coefficient of each functional microbial community under the inhibited state is obtained by combining this with a pre-defined microbial community metabolite comparison table, thus obtaining functional microbial community activity data. A weighted calculation method is used to process the functional microbial community activity data and the product component deviation data. By analyzing the correlation between microbial community activity and product concentration, the contribution changes of each functional microbial community to specific metabolites are analyzed. A decision tree algorithm is used to identify key microbial community species that cause product component deviations, determining the quantitative value of the imbalance degree of the microbial community structure. Based on the quantitative value of the degree of imbalance, the impact of the imbalance of the microbial community ratio on the stability of the fermentation process is calculated using preset risk assessment rules. By setting multi-level risk threshold intervals, the severity level of the current community imbalance is determined, and a risk index value of the microbial community ratio imbalance containing risk level code and score value is obtained.

[0048] For example, the construction of a microbial community database is based on the accumulation and classification of a large amount of historical fermentation data, which includes the physiological response characteristics of different strains under various environmental conditions.

[0049] For example, when the pH value drops from 6.8 to 5.2, the activity of *Lactobacillus acidophilus* remains above 85%, while the activity of common lactic acid bacteria decreases to around 40%. Each bacterial community record in the database contains key information such as acid tolerance parameters, optimal pH range, and metabolite profiles. These parameters were obtained through long-term experimental accumulation and statistical analysis. Conditional matching queries employ a multi-dimensional retrieval strategy, using the current time window and range as query conditions to search for similar historical cases in the database.

[0050] In one embodiment, when a pH decrease of 1.2 units is detected within 2 hours, the system automatically retrieves all historical records from the database showing pH decreases of 1.0-1.5 units over a duration of 1.5-2.5 hours. This fuzzy matching mechanism effectively handles individual differences and environmental changes during fermentation, improving the accuracy and applicability of the matching results. The microbial community screening process based on the retrieval results embodies the combination of biological principles and data analysis.

[0051] It should be noted that different bacterial groups have significantly different tolerances to acidic environments. When the pH of the environment is lower than the tolerance threshold of a certain bacterial group, the metabolic activity of that bacterial group will decrease sharply or even stop completely.

[0052] For example, the optimal pH range for butyric acid-producing bacteria is typically between 6.5 and 7.2. When the pH drops below 5.5, their butyric acid production capacity significantly weakens, thus affecting the metabolic balance of the entire fermentation system. The calculation of the metabolic activity attenuation coefficient is based on studies of bacterial community physiology, quantifying the changes in bacterial community activity under different degrees of inhibition.

[0053] In one possible implementation, when a functional microbial community is mildly inhibited, its metabolic activity attenuation coefficient might be 0.8, indicating a 20% reduction in activity; when severely inhibited, the attenuation coefficient might drop to 0.3, indicating that the activity is only 30% of the normal level. This quantitative characterization method allows complex biological processes to be described and predicted through mathematical models. Weighted calculation methods play a crucial role in data fusion, reflecting the importance of different microbial communities in the fermentation process by assigning them appropriate weight coefficients.

[0054] For example, acid-producing bacteria play a dominant role in pH control, and their weight coefficients are usually set high, while some minor bacterial groups have relatively low weights. The establishment of the correlation matrix reveals the quantitative relationship between bacterial activity and product concentration through statistical analysis; when the activity of a certain bacterial group decreases, the concentration trend of the corresponding metabolite can be predicted. The application of decision tree algorithms in key bacterial group identification is based on their excellent classification and regression capabilities.

[0055] Specifically, the algorithm uses microbial community activity data and product deviation data as input variables, and identifies the microbial community species that have the greatest impact on product component deviation by constructing a decision rule tree.

[0056] For example, when lactic acid concentration deviates by more than 15%, the decision tree may identify decreased activity of Streptococcus thermophilus as the main cause, while abnormal ethanol concentration may point to metabolic abnormalities in Saccharomyces cerevisiae. Quantifying the degree of imbalance assesses the stability of the entire microbial community by comprehensively considering the activity changes and interactions of multiple microbial groups.

[0057] In one embodiment, when the activity of the main acid-producing bacteria decreases by 30% while the activity of the gas-producing bacteria increases by 20%, the system calculates a high degree of imbalance, indicating a significant shift in the community structure.

[0058] S104. If the risk index value exceeds the preset threshold parameter, the prediction results of the imbalance of the microbial community ratio are classified by the support vector machine model to obtain the functional microbial community category most likely to be affected and determine the target microbial community combination that needs to be prioritized for intervention.

[0059] If the risk index value of the microbial community proportion imbalance exceeds a preset threshold parameter, a support vector machine (SVM) classification procedure is triggered. A radial basis function (RBF) kernel is used to convert the risk index value and the quantified imbalance degree value into a feature vector. A pre-trained SVM model is then used to perform pattern recognition on the current imbalance state, obtaining a predicted classification result of the microbial community proportion imbalance containing an imbalance type identifier. Based on the imbalance type identifier in the predicted classification result, a list of candidate functional microbial communities associated with the current imbalance pattern is obtained by searching a preset microbial community influence relationship lookup table. A Bayesian probability calculation method is used to analyze the probability of each candidate functional microbial community being affected, determining a list of the most likely affected functional microbial community categories ranked by influence probability. Based on this list of functional microbial community categories, a correlation calculation method is used to analyze the interaction strength and dependency relationships between each microbial community category. By constructing a microbial community correlation matrix, the key microbial communities with the highest degree of influence and related supporting microbial communities are identified, obtaining a priority sequence of target microbial communities ranked by the urgency of intervention. Based on the target microbial community priority sequence, the microbial community pair with the best synergistic regulatory effect is selected from the high-priority microbial communities. By comparing the expected repair effects of different microbial community combinations on the imbalance state, the target microbial community combination that needs to be prioritized for intervention, including the main intervention microbial community and the auxiliary regulatory microbial community, is determined.

[0060] For example, the triggering mechanism of the support vector machine (SVM) classification program embodies the core concept of intelligent monitoring. When the risk indicator value reaches a critical state, the deep analysis process is automatically initiated. The radial basis function (RBF) kernel function plays a key role in feature mapping, transforming the original risk values ​​and imbalance data into feature vectors in a high-dimensional space, making complex imbalance patterns that were originally linearly inseparable into accurately classifiable patterns.

[0061] For example, when the risk index is 0.75 and the imbalance level is 0.82, the radial basis function maps this set of two-dimensional data to a higher-dimensional feature space, in which different types of imbalance states can be clearly distinguished. The pre-trained support vector machine model is based on the accumulation and annotation of a large amount of historical fermentation data, which contains feature samples of various typical imbalance patterns.

[0062] In one embodiment, the model training data covers various imbalance types, including acid-producing, gas-producing, and alcohol-producing imbalances, each with corresponding feature labels and classification boundaries. When a new feature vector is input, the model can determine its imbalance type based on its position in the feature space, thus generating a classification result with a clear type identifier. Microbial community impact analysis based on the classification results reflects a logical progression from macroscopic imbalances to microscopic mechanisms. A microbial community impact relationship table records the correspondence between different imbalance types and related microbial communities; these relationships have been validated through long-term microbiological research and fermentation practice.

[0063] For example, when an acid-producing imbalance is identified, the reference table will point to candidate functional bacteria such as Lactobacillus acidophilus, Bifidobacterium, and Streptococcus acidophilus, because the metabolic activities of these bacteria directly affect the acidity balance of the fermentation system. The application of Bayesian probability calculation methods in impact probability assessment is based on conditional probability theory. It calculates the impact probability under the current situation by analyzing the frequency of each bacterial community's impact under specific imbalance types in historical data.

[0064] Specifically, when a bacterial community has an 85% probability of being affected in historical cases of acid-producing imbalances, the system assigns it a corresponding high-probability weight. This quantitative assessment method shifts the community selection process from empirical judgment to data-driven, precise decision-making. The construction of the community association matrix reveals the complex interaction network within the microbial community.

[0065] It should be noted that the microbial community in the fermentation system does not exist independently, but forms a complex ecological relationship through metabolic product exchange, nutrient competition, and symbiotic cooperation.

[0066] For example, organic acids produced by acid-producing bacteria can inhibit the activity of certain gas-producing bacteria, while carbon dioxide produced by gas-producing bacteria can affect the pH buffering capacity of the entire system. By calculating the correlation coefficients and influence strengths among various bacterial communities, the system can identify core bacterial communities that play a crucial role in the imbalance repair process. The priority of intervention is ranked based on the assessment of the influence of the bacterial communities on the overall balance, with key bacterial communities that directly affect metabolic balance being listed as the highest priority.

[0067] In one possible implementation, when decreased Lactobacillus acidophilus activity is identified as the primary cause of insufficient acid production, this microbial community is designated as the highest-priority intervention target. Simultaneously, synergistic auxiliary microbial communities are also prioritized, forming a systematic intervention strategy. This combined assessment approach embodies a systems optimization mindset in the final microbial community selection, considering not only the reparative effects of individual microbial communities but also evaluating the synergistic potential of different microbial community combinations.

[0068] For example, supplementing with Lactobacillus acidophilus alone may only partially restore acid-producing capacity, but if prebiotic-producing bacteria that provide nutritional support are also supplemented, the overall repair effect will be significantly improved. The assessment of expected repair effects involves simulating the theoretical degree of improvement of the imbalance state by different combinations of treatments, providing a scientific basis for practical biological interventions.

[0069] S105. Based on the characteristics of the target bacterial community combination, match the corresponding pH adjustment scheme from the preset regulation strategy library, obtain the appropriate intervention parameters, and generate an execution instruction set.

[0070] Based on the information of the primary interventional and auxiliary regulatory microorganisms in the target microbial community combination to be prioritized for intervention, the optimal pH range, nutrient requirements, and metabolite characteristics of each microbial community are analyzed. A target microbial community characteristic data table is constructed, including acid tolerance indicators, nutrient preferences, and synergistic relationships, in conjunction with microbial community growth environment preference parameters. This target microbial community characteristic data table is compared and queried with a pre-set regulatory strategy library. By searching for microbial community characteristic tags and environmental condition requirements, a list of candidate regulatory schemes matching the current target microbial community combination is obtained, determining a set of pH adjustment schemes including pH adjustment range, nutrient supplementation type, and environmental optimization measures. If multiple candidate schemes exist in the set of pH adjustment schemes, the matching degree between each adjustment scheme and the target microbial community characteristic data table is evaluated using a similarity calculation method. A weighted scoring method is used to comprehensively consider the expected value of the regulatory effect, operational complexity, and cost assessment value to obtain the optimal combination of intervention parameters. Based on the parameter combination of the intervention measures, the abstract regulation parameters are converted into specific equipment control instructions. By setting parameters for pH regulator addition, nutrient solution supply, and environmental control equipment operation, a set of targeted instructions for regulating the balance of the microbial community, including operation sequence and parameter settings, is obtained.

[0071] For example, the construction of a target microbial community characteristic data table embodies a key transformation process from biological principles to engineering applications. Data extraction methods, by analyzing the physiological and biochemical characteristics of the microbial community, convert complex biological information into quantifiable engineering parameters.

[0072] For example, *Lactobacillus acidophilus* has an optimal pH range of 5.5-6.2, prefers carbohydrates and organic nitrogen sources for nutrition, and its main metabolites are lactic acid and acetic acid. These characteristics are standardized to form structured data records. Acid tolerance is quantified by measuring the survival rate and activity retention rate of the bacterial community under different pH conditions, while nutritional preferences are assessed based on the community's utilization efficiency of different nutrient substrates. Pattern matching queries based on the characteristic data table embody the core mechanism of intelligent decision support. The regulation strategy base, serving as a knowledge base, stores a large number of validated regulation schemes, each labeled with its applicable bacterial community characteristic range and environmental condition requirements.

[0073] In one embodiment, when the system detects that the target bacterial community combination contains bacteria with strong acid-producing capacity, the pattern matching algorithm automatically searches the strategy library for all regulatory schemes applicable to high acid-producing bacteria. The matching process employs multi-dimensional similarity calculations, comprehensively considering the degree of matching across multiple dimensions such as pH tolerance range, nutrient requirement type, and metabolite profile. The screening process for candidate regulatory schemes demonstrates a decision-making optimization logic from multiple schemes to the optimal scheme.

[0074] It should be noted that different control schemes vary significantly in practical applications. Some schemes may have good regulatory effects but are complex to operate, while others, although simple, have limited effects. The similarity calculation method quantifies the degree of matching between each scheme and the target requirements, providing an objective basis for subsequent comprehensive evaluation.

[0075] For example, the similarity score for scheme A is 0.85, for scheme B it is 0.92, and for scheme C it is 0.78. These values ​​reflect the degree to which each scheme fits the current microbial community characteristics. The application of the weighted scoring method reflects the complexity of multi-objective optimization decision-making. The expected value of the adjustment effect predicts the degree of improvement after the scheme is implemented based on historical data and theoretical models. The operational complexity assesses the technical difficulty and time cost of implementing the scheme, and the cost assessment considers the required material input and equipment usage costs.

[0076] In one possible implementation, the system assigns 60% weight to the regulatory effect, 25% to operational complexity, and 15% to cost, calculating a comprehensive score for each scheme through weighted average. This multi-dimensional evaluation ensures that the final selected scheme achieves the expected results while also possessing good operability and economy. The parameter transformation method achieves the crucial transformation from abstract strategies to concrete operational instructions. The combination of intervention parameter combinations contains the core elements of the regulatory scheme, but these parameters are usually expressed in a relatively abstract form and need to be converted into concrete instructions that the equipment can recognize and execute.

[0077] For example, when the control scheme requires adjusting the pH to within the range of 6.0 ± 0.2, the parameter conversion method calculates the specific amount and frequency of regulator addition based on factors such as the current pH value, the buffering capacity of the fermentation system, and the concentration of the regulator. The generation of the execution instruction set reflects the precision requirements of automated control. The setting of the operation sequence considers the coordination and priority between different control measures, avoiding mutual interference from simultaneous operations.

[0078] Specifically, pH adjusters are typically added before nutrient solution supply to prevent nutrient precipitation or inactivation under unsuitable pH conditions. Environmental control equipment operating parameters include temperature controller setpoints, stirrer speed, and aeration rate adjustments. The coordinated configuration of these parameters ensures the microbial community regains equilibrium under optimal environmental conditions. The structured format of the instruction set facilitates parsing and execution by the automated control system. Each instruction contains a clearly defined target, parameter settings, and execution time, forming a complete control and execution plan.

[0079] S106. For the execution instruction set, inject acid-base regulators into the fermentation system, adjust environmental conditions in real time, obtain acid-base feedback data after intervention, and determine whether it has recovered to the target range.

[0080] Based on the operation sequence and parameter settings information in the instruction set for targeted regulation of microbial community balance, the abstract instructions are converted into equipment control signals by an automated control system. An automatic feeding device and an acid-base regulator supply device precisely inject alkaline or acidic regulators into the fermentation system, resulting in an acid-base regulator injection operation record containing the injection time and amount. Based on the time and dosage parameters in the acid-base regulator injection operation record, a temperature control device, a stirring control device, and aeration control device coordinate the control of temperature distribution, mixing intensity, and oxygen supply within the fermenter. An environmental monitoring device continuously tracks the changing trends of various environmental indicators, obtaining real-time environmental condition adjustment data including temperature changes, dissolved oxygen levels, and stirring status. If the real-time environmental condition adjustment data shows that all environmental parameters have reached a stable state, a distributed pH sensor network collects pH values ​​at each monitoring point in the fermentation system. A weighted average method is used to fuse the multi-point measurement results, determining the post-intervention pH feedback data representing the overall fermentation environmental state. The pH feedback data after the intervention is compared with the preset upper and lower limits of the target pH range to check whether the current pH falls within the target range, and to obtain a judgment result indicating whether the balance of the microbial community has been restored to the target range.

[0081] For example, the instruction conversion mechanism of an automated control system embodies the crucial bridging role from digital instructions to physical operations. The operation timing and parameter setting information contained in the instruction set are transmitted to the control system through a standardized data format, and the system's built-in instruction parsing function converts abstract control parameters into specific equipment action signals.

[0082] For example, when the instruction set requires the addition of 0.1M sodium hydroxide solution at a specific time point, the control system automatically calculates the required pump operating time and flow rate setpoint to ensure precise dosage of the regulator. This automated conversion avoids errors and delays caused by manual operation, improving the accuracy and reproducibility of the control process. The automatic dosing device plays a crucial role in the injection of acid-base regulators, achieving quantitative dosing through precise flow control technology.

[0083] In one embodiment, the feeding device is driven by a peristaltic pump, enabling precise control at the micro-level and avoiding the flow fluctuation problems that may occur with traditional gravity feeding methods. The acid-base regulator supply device typically includes multiple storage tanks, storing acidic and alkaline solutions of different concentrations respectively, and automatically switching valves select the appropriate regulator type and concentration. The real-time generation of injection operation records provides an important data foundation for subsequent effect evaluation and process optimization. Environmental coordination control based on injection records reflects the complexity of multi-parameter coupled regulation during fermentation.

[0084] It should be noted that the addition of acid-base regulators not only affects the pH value but also has a cascading impact on the temperature distribution, viscosity characteristics, and mass transfer efficiency of the fermentation system. The temperature control device monitors temperature changes after regulator injection and automatically adjusts heating or cooling power to maintain fermentation temperature stability. The stirring control device adjusts the stirring speed based on changes in solution viscosity to ensure rapid and uniform distribution of the regulator in the fermentation broth. The aeration control device adjusts the aeration rate based on the effect of pH changes on gas-liquid mass transfer to maintain an appropriate dissolved oxygen level. The continuous tracking function of the environmental monitoring device provides dynamic feedback information for real-time control.

[0085] Specifically, a temperature sensor array monitors the temperature distribution at different locations within the fermenter, a dissolved oxygen probe measures changes in oxygen concentration in real time, and a stirring power sensor reflects the changing trend of mixing efficiency. Comprehensive analysis of this data can promptly detect abnormal fluctuations in environmental parameters, providing a basis for further control decisions. Acquiring real-time environmental condition adjustment data lays the foundation for determining whether control measures have achieved the expected results. The deployment of a distributed pH sensor network demonstrates the importance of spatially distributed monitoring.

[0086] In one possible implementation, a sensor network is deployed at multiple monitoring points at different heights and radial positions within the fermenter, with each sensor independently collecting local pH values. This distributed monitoring approach effectively captures potential pH gradients and localized inhomogeneities within the fermentation broth, avoiding representativeness errors that may arise from single-point measurements. The weighted averaging method considers the spatial representativeness and data reliability of different monitoring points, calculating a more accurate overall pH level by assigning appropriate weighting coefficients to measurements from different locations. The data fusion process involves not only simple numerical averaging but also outlier identification and data quality assessment.

[0087] For example, when a sensor's measurement deviates significantly from other sensors, the system automatically reduces the weight of that data point or marks it as an outlier. Determining the pH feedback data after intervention provides an objective standard for the final effect assessment. The combined application of numerical comparison and interval judgment methods reflects the requirements of precise control. The target pH range is set based on the optimal growth conditions of the microbial community and process requirements, typically including two boundary parameters: an upper limit and a lower limit. When the measured pH value falls within this range, it indicates that the control measures have achieved the expected effect, and the growth environment of the microbial community has been effectively improved.

[0088] S107. If the pH level returns to the target range, continuously monitor the changes in the proportion of the microbial community to determine whether the fermentation system has reached a new equilibrium state. If the pH level does not return to the target range, reassess the current pH deviation and the severity of the microbial community imbalance, implement enhanced intervention, and obtain pH feedback data after enhanced intervention. If the target range is still not reached after a preset number of adjustments, record the abnormal state parameters and enter the manual intervention process.

[0089] If the judgment result indicates successful regulation of the microbial community balance returning to the target range, a continuous monitoring program is established based on the pH feedback data after the intervention. The quantity distribution and metabolic activity indicators of each functional microbial community in the fermentation system are periodically collected using a pre-set microbial community detection device and activity measurement device to obtain dynamic recovery indicators of functional microbial community activity, including changes in microbial community quantity and activity indicators. A time-series analysis method is used to assess the changing trend of the dynamic recovery indicators of functional microbial community activity. The stability of each microbial community is determined by calculating the rate of change in the microbial community proportion and the fluctuation range of the activity indicators. If the microbial community proportion and activity indicators remain within a stable threshold range within a pre-set monitoring period, the fermentation system is determined to have reached a new equilibrium state. If the judgment result indicates that the balance has not returned to the target range, the current pH deviation and the severity of the microbial community imbalance are re-analyzed. A multi-level decision-making algorithm is used to extract enhanced intervention schemes from the regulation strategy library, adjusting the dosage and time interval of the pH regulator, and simultaneously activating the temperature and ventilation regulation devices to obtain enhanced pH regulation parameters. The enhanced pH control operation is performed according to the enhanced pH control parameters. The pH feedback data after the enhanced intervention is obtained through the monitoring device. If the pH still does not reach the target range after a preset number of enhanced control operations, the abnormal recording program is triggered to save the current state parameters of the fermentation system, start the alarm device, and generate an abnormal command that requires manual handling.

[0090] For example, the establishment of continuous monitoring procedures embodies the core concept of dynamic management of the fermentation process, shifting from verifying the results of pH control to deeply tracking the state of the microbial community. Microbial community detection devices typically employ flow cytometry or fluorescence in situ hybridization (FISH) to identify and count different types of functional microorganisms in real time. The advantage of this method is its ability to distinguish between live and dead cells, providing accurate information on microbial activity. Activity assay devices assess the metabolic state of the microbial community by detecting the activity of key enzymes; for example, measuring lactate dehydrogenase activity to assess the metabolic capacity of acid-producing bacteria, and measuring cellulase activity to assess the functional state of cellulose-decomposing bacteria.

[0091] In one embodiment, a multi-dimensional data acquisition strategy is employed to obtain dynamic recovery indicators of functional microbial community activity. Changes in microbial community quantity are obtained through periodic sampling and colony counting, typically data is collected every 2-4 hours to form a continuous time series. Changes in activity indicators are comprehensively assessed using multiple methods, including enzyme activity assays, metabolite concentration detection, and respiration rate measurements. This multi-dimensional monitoring method can comprehensively reflect the dynamic process of the microbial community transitioning from an imbalanced to a balanced state, providing a reliable basis for judging the recovery effect. The application of time-series analysis methods is based on the time-dependent characteristics of biological processes, identifying system state transitions by analyzing the time-varying patterns of data.

[0092] It should be noted that the recovery process of microbial communities typically follows a specific time pattern, with potentially dramatic fluctuations in the initial stages followed by a gradual stabilization. The rate of change in microbial community proportions is calculated by comparing the differences in microbial abundance at consecutive time points to quantify the speed of change. When the rate of change continuously decreases and approaches zero, it indicates that the microbial community proportions are stabilizing. The assessment of the fluctuation range of activity indicators is achieved by calculating the standard deviation or coefficient of variation of activity values ​​within a certain time window. The setting of the stability threshold range is based on the quality control requirements of the fermentation process and the principles of microbial ecology.

[0093] For example, when the proportion of major functional microbiota changes by less than 5% and lasts for more than 6 hours, the microbiota proportion can be considered to have reached a stable state. The stability assessment of activity indicators requires that the fluctuation range of key enzyme activities be controlled within 10% of the normal level. This quantitative stability assessment standard provides a clear basis for automated judgment. Multi-level decision-making algorithms play a crucial role in the selection of reinforcement intervention programs, addressing complex regulatory problems by constructing a hierarchical decision structure.

[0094] Specifically, the first level of decision-making determines the intervention intensity level based on the current degree of deviation. Mild deviations are addressed with a mild control plan, moderate deviations with a standard enhanced plan, and severe deviations with maximum intensity control. The second level of decision-making selects appropriate combinations of control measures based on the specific characteristics of the imbalance. For acid-producing imbalances, pH and nutrient supply are adjusted primarily; for gas-producing imbalances, temperature and ventilation conditions are controlled. The main difference between the enhanced intervention plan and the conventional control plan lies in the setting of the control parameters.

[0095] In one possible implementation, the enhanced scheme might increase the dosage of the acid-base regulator to 1.5-2 times the conventional amount, and increase the dosing frequency from once per hour to once every 30 minutes, while simultaneously coordinating fine-tuning of temperature and optimization of ventilation parameters. The temperature control device may need to increase its control precision from...

[0096] Increasing the temperature from ±0.5℃ to ±0.2℃ may require adjusting the dissolved oxygen level from the standard range to the upper limit of the optimal range. The activation of the anomaly recording procedure indicates that conventional control methods are insufficient to resolve the current problem, necessitating higher-level intervention. The storage of status parameters includes all critical data from the fermentation process, such as historical pH curves, records of microbial community changes, environmental parameter fluctuations, and control measure execution logs. This data provides crucial reference for subsequent fault analysis and process improvement. The activation of the alarm device notifies operators via audible and visual signals and data communication, ensuring timely handling of abnormal situations and preventing further losses from fermentation failure.

[0097] S108. For the fermentation product component data under the new equilibrium state, perform component detection, obtain quantitative indicators of product stability, and obtain an improved regulatory strategy library.

[0098] Based on the confirmation that the fermentation system has reached a new equilibrium state, representative samples are extracted from the fermenter using a pre-set sampling device. Multi-band scanning detection of the samples is performed using a spectroscopic detection device to obtain spectral data of fermentation product components, including organic acid content, alcohol concentration, and sugar residue. Spectral data processing methods are used to identify peaks and analyze intensity in the spectral data of the fermentation product components. By establishing a pre-set standard control relationship, the spectral characteristic signals are converted into quantitative concentration values. Combined with product concentration stability assessment and component ratio consistency calculation, a stability quantification index reflecting the stability of product quality is obtained. If the stability quantification index reaches the pre-set stability standard, a neural network algorithm is used to update the pre-established response mechanism understanding model based on data from current successful control cases. By analyzing the correspondence between pH change patterns and microbial community succession patterns, the model parameters are retrained to determine optimized model parameters reflecting the latest correlation understanding. Based on the correlation information contained in the optimized model parameters, a parameter transformation method is used to convert the correlation data output by the model into specific control parameter suggestions. By updating the pH adjustment threshold, microbial intervention timing, and environmental control parameter settings in the control strategy library, an improved control strategy library containing improved control rules and optimized parameter configurations is obtained.

[0099] For example, the design of the sampling device reflects the precision requirements of quality control in the fermentation process, ensuring the representativeness and consistency of the samples through an automated sampling mechanism. The sampling device typically has multiple sampling points distributed at different heights and locations within the fermenter, comprehensively reflecting the product distribution of the entire fermentation system. The timing of sampling is based on the stabilization period after equilibrium confirmation, at which point the microbial community has reached a relatively stable state, and the product composition has also stabilized, providing a reliable basis for subsequent spectroscopic analysis. Spectroscopic detection equipment plays a central role in product composition analysis, rapidly obtaining molecular composition information of the sample through non-destructive optical detection methods.

[0100] In one embodiment, near-infrared spectroscopy is particularly suitable for the quantitative analysis of organic compounds, enabling the simultaneous detection of multiple organic acids, alcohols, and sugars. Different compounds exhibit characteristic absorption peaks at specific wavelengths; by scanning the 800-2500 nm wavelength range, a complete spectral fingerprint of the sample can be obtained. Raman spectroscopy provides molecular vibrational information, offering unique advantages for identifying chemical bond types and molecular structures. The combined use of these two techniques provides more comprehensive and accurate compositional information. Quantitative compositional analysis based on spectral data represents a crucial transition from qualitative identification to quantitative assessment. The spectral data processing method first improves data quality through preprocessing steps such as baseline correction, noise filtering, and peak identification, and then uses chemometric methods to establish a quantitative relationship between spectral characteristics and compound concentrations.

[0101] It should be noted that the relationship between the spectral response intensity and concentration of different compounds generally follows the Beer-Lambert law. Establishing a standard control relationship allows for an accurate conversion from spectral intensity to actual concentration. The calculation of stability quantification indicators comprehensively considers multiple dimensions of product composition, including not only the concentration stability of individual components but also the consistency of the proportional relationships between components. Product concentration stability is assessed by calculating the relative standard deviation of concentration changes between consecutive sampling points. When the concentration fluctuation of major products such as lactic acid and ethanol is less than 5%, the concentration stability is considered good. The calculation of component proportion consistency focuses on the relative proportions between different products, such as the ratio of lactic acid to acetic acid, and the ratio of sugar consumption to acid production. The stability of these proportions reflects the consistency and predictability of microbial metabolism. The application of neural network algorithms in updating response mechanism understanding models is based on their powerful nonlinear modeling capabilities.

[0102] Specifically, the network's input layer receives time-series data on pH changes, including characteristic parameters such as the magnitude, rate, and duration of pH changes, as well as succession data of the microbial community, such as changes in the proportion of major bacterial groups, community diversity indices, and functional gene expression levels. The hidden layers learn the complex relationships between the input variables through a multilayer perceptron structure, and the output layer predicts the response patterns and succession trends of the microbial community. The retraining process of the model parameters embodies the incremental learning concept in machine learning.

[0103] In one possible implementation, each successful regulation case is added to the training dataset, and the network weights are updated through backpropagation, enabling the model to continuously learn new regulation experiences and response patterns. This continuous learning mechanism allows the model's predictive accuracy to gradually improve with the accumulation of regulation experience, while also adapting to the behavioral characteristics of microbial communities under different fermentation conditions. The parameter transformation method serves as a crucial bridge from abstract model output to concrete operational guidance. The correlation data output by the model typically expresses the correlation coefficients and influence weights between different variables in numerical form. The parameter transformation method converts these values ​​into practical regulation parameter suggestions through predefined mapping rules.

[0104] For example, when the model shows that a 0.3 unit decrease in pH leads to a 15% increase in the proportion of acid-producing bacteria, the conversion method will adjust the pH adjustment threshold setting accordingly, tightening the original ±0.2 adjustment range to...

[0105] ±0.15 was used to achieve more precise community control. The update of the regulation strategy library reflects a systematic approach to knowledge management and experience accumulation. The updated strategy library not only includes new parameter settings but also incorporates the latest discoveries of regulation patterns and response mechanisms, forming a more complete regulation knowledge system and providing more accurate and effective guidance for subsequent fermentation process optimization.

[0106] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for dynamic control of the fermentation process of Kaiyuqu based on microbial activity monitoring, characterized in that, The method includes: In the fermentation system, pH change data and fermentation product component data are continuously collected and timestamped to generate a multidimensional dataset containing environmental state and product component information. Noise suppression and data smoothing are applied to the multidimensional dataset to extract the time window, amplitude range, and product component deviation degree of pH shift. Based on the time window and amplitude range of pH shift, a pre-established microbial community database is queried to obtain functional microbial activity data. Combined with the product component deviation degree, a risk index value for microbial community imbalance is calculated. The risk index value is classified using a support vector machine model to determine the target microbial community combination. Based on the target microbial community combination, a pH adjustment scheme is matched from a pre-set control strategy library to generate an execution instruction set. The execution instruction set is executed to obtain pH feedback data after intervention, determining whether it has recovered to the target range. Based on the pH feedback data after intervention, the dynamic recovery index of functional microbial activity is monitored to determine whether the fermentation system has reached a new equilibrium state. The fermentation product component data under the new equilibrium state is detected to obtain a quantitative index of product stability, and the control strategy library is optimized.

2. The dynamic control method for the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The process involves continuously collecting pH change data and fermentation product component data within the fermentation system, embedding timestamps, and generating a multidimensional dataset containing environmental state and product component information, including: pH electrodes and conductivity sensors are deployed on the inner wall of the fermenter, at the inlet, outlet, and temperature control area to acquire raw acid-base voltage signals. Noise is processed using a Kalman filter algorithm to generate a pH value sequence. The concentrations of organic acids, alcohols, and sugar metabolites in the fermentation broth are detected using chromatographic analysis equipment to generate a component data matrix. A network time synchronization protocol is used to assign millisecond-level time stamps to the pH value sequence and the component data matrix. A support vector machine algorithm is used to identify key time nodes in the pH value sequence, dividing the fermentation stages and constructing a multidimensional dataset containing temporal relationships.

3. The dynamic control method for the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The process of noise suppression and data smoothing of the multidimensional dataset, and extraction of the time window, amplitude range, and product component deviation of the pH shift, includes: The moving average filtering method is used to suppress noise in the pH and product component data of the multidimensional dataset, generating first pH fluctuation information and first product component offset data. The first pH fluctuation information and first product component offset data are then smoothed using cubic spline interpolation to generate second pH fluctuation information and second product component offset data. The first derivative of the second pH fluctuation information is analyzed using gradient calculation to mark the time window and amplitude range of the pH offset. Based on the time window, the component concentration change value is extracted from the second product component offset data, the deviation percentage is calculated, and the degree of product component deviation is determined.

4. The dynamic control method for the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The process involves querying a pre-established microbial community database based on the time window and amplitude range of the pH shift to obtain functional microbial community activity data, and calculating a risk index value for microbial community imbalance based on the degree of deviation of the product components, including: The system accesses the microbial community database through a query interface, retrieves historical microbial community response records that match the time window and amplitude range of the pH shift, and generates a first microbial community feature dataset. It then filters functional microbial communities in the first microbial community feature dataset whose acid tolerance is below the amplitude range, obtains metabolic activity decay coefficients, and generates functional microbial community activity data. Finally, it analyzes the degree of deviation between the functional microbial community activity data and the product components using a decision tree algorithm, constructs a correlation matrix between microbial community activity and product concentration, and calculates the risk index value for microbial community imbalance.

5. The dynamic control method for the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The step of classifying the risk indicator values ​​using a support vector machine model to determine the target microbial community combination includes: The risk index values ​​are converted into feature vectors using a radial basis function kernel function, and an imbalance type identifier is generated using a support vector machine model. Based on the imbalance type identifier, a microbial community influence relationship lookup table is consulted to generate a list of candidate functional microbial communities. The list of candidate functional microbial communities is analyzed using a Bayesian probability calculation method to determine the list of functional microbial community categories. The interaction strength in the list of functional microbial community categories is analyzed using a correlation calculation method to construct a microbial community correlation matrix and generate the target microbial community combination.

6. The method for dynamic control of the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The step of matching a pH adjustment scheme from a preset regulatory strategy library based on the target microbial community combination and generating an execution instruction set includes: A target microbial community characteristic data table is generated based on the target microbial community combination; the target microbial community characteristic data table is compared with the regulation strategy library to generate a set of pH adjustment schemes; the set of pH adjustment schemes is evaluated by a similarity calculation method to generate a combination of intervention measures parameters; the combination of intervention measures parameters is converted into equipment control instructions to generate an execution instruction set.

7. The method for dynamic control of the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The execution of the instruction set, obtaining post-intervention pH feedback data, and determining whether the pH has recovered to the target range include: According to the execution instruction set, an acid-base regulator is injected through an automatic feeding device, generating an injection operation record; environmental conditions are adjusted through a temperature regulation device and a ventilation regulation device, generating real-time environmental condition adjustment data; acid-base values ​​in the real-time environmental condition adjustment data are collected through a pH sensor network, generating post-intervention acid-base feedback data; and it is determined whether the post-intervention acid-base feedback data is within the target range.

8. The dynamic control method for the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The step of monitoring the dynamic recovery indicators of functional microbial community activity based on the pH feedback data after the intervention to determine whether the fermentation system has reached a new equilibrium state includes: Collect the functional microbial community quantity distribution and metabolic activity indicators corresponding to the pH feedback data after the intervention, and generate a dynamic recovery index of functional microbial community activity; evaluate the change rate of the dynamic recovery index of functional microbial community activity through time series analysis, determine whether the microbial community ratio is within the stable threshold range, and determine whether the fermentation system has reached a new equilibrium state.

9. The method for dynamic control of the fermentation process of Kaiyuqu based on microbial activity monitoring according to claim 1, characterized in that, The process of detecting fermentation product component data under the new equilibrium state, obtaining quantitative indicators of product stability, and optimizing the regulatory strategy library includes: Multi-band scanning is performed on the fermentation product component data under the new equilibrium state to generate spectral data; the spectral data is converted into quantitative concentration values ​​through peak identification and intensity analysis to generate product stability quantification index, and the pH adjustment threshold and environmental control parameters in the regulation strategy library are updated.

Images