Multi-stage fermentation method for improving activity and stability of thalli

Abstract

The invention discloses a multi-stage fermentation method for improving thallus activity and stability, and particularly relates to the technical field of microbial fermentation. Dynamically dividing a fermentation period into a plurality of metabolic staging sections based on the multi-source physiological characteristic parameter set; constructing a multi-stage environment regulation and control model, taking the optimal control parameter vector of each stage as a control target, combining a real-time feedback signal set, and carrying out environment factor regulation through a model prediction control algorithm; calculating a thallus activity improvement rate and a fermentation stress anomaly index in real time, performing cross-batch correction on the multi-stage environment regulation and control model after fermentation is finished, and outputting an optimized parameter template for next fermentation calling; the dynamic self-adaptive control of the fermentation system can be realized, the thallus metabolism efficiency, the process stability and the fermentation repeatability are improved, and the method is suitable for industrial large-scale microbial fermentation production scenes.

Description

Technical Field

[0001] This invention relates to the field of microbial fermentation technology, specifically to a multi-stage fermentation method for improving cell activity and stability. Background Technology

[0002] Microbial fermentation technology has wide applications in biopharmaceuticals, food industry, agriculture, and environmental engineering. The activity and stability of the microorganisms during fermentation directly affect the yield and quality of the fermentation products. Currently, traditional fermentation methods mostly employ single-stage or continuous fermentation processes with constant parameters, which are difficult to adapt to the dynamic needs of microorganisms at different growth stages for key environmental factors such as nutrients, dissolved oxygen, pH, and temperature. This leads to decreased microbial activity, metabolic disorders, and even premature death of the microorganisms.

[0003] Furthermore, during high-density cultivation and long-cycle fermentation, the activity and stability of the microorganisms are increasingly affected by adverse environmental stresses. Problems such as substrate accumulation, metabolite inhibition, and limited oxygen transfer further weaken the metabolic capacity of the microorganisms, leading to a decrease in the synthesis efficiency of the target product. Therefore, how to dynamically regulate fermentation environmental parameters based on the metabolic state of the microorganisms to improve their physiological activity and stability has become an urgent technical problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-stage fermentation method that improves cell activity and stability, thereby addressing the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a multi-stage fermentation method for improving cell activity and stability, comprising: Establish a multi-source physiological characteristic parameter set M, including metabolic flux, cell activity, conductivity and respiration rate, to obtain the dynamic metabolic state of the target microorganism at different time points. Based on the feature parameter set M, a clustering algorithm and phase space reconstruction method are used to automatically divide the entire fermentation cycle into n metabolic phase segments. Each phase segment defines a set of coupled control parameter vectors. Where: Ti is the temperature control value, DOi is the dissolved oxygen control value, Fi is the feed rate, Ni is the stirring rate, Pi is the pH control strategy, and i = 1, 2, ..., n; Construct a multi-stage environmental regulation model Q, using the parameter vector Wi as the control variable, and combine it with a real-time feedback signal set. k is the total number of feedback signals; key environmental factors in the fermentation process are dynamically adjusted to maintain them within the optimal range of the corresponding Wi. During each stage of the fermentation process, the improvement rate of cell activity ΔA and the fermentation stress anomaly index S are calculated in real time, and compared with the preset thresholds Ath and Sth respectively; if ΔA < Ath and S ≥ Sth, it is determined that the current stage fails, and the next stage is entered in advance, and at the same time, the updated parameter vector W(i + 1) is activated. After the fermentation is completed, according to the non-linear distribution of cell activity, product conversion rate and environmental regulation metabolic response in each stage, the correction data of the multi-stage environmental regulation model Q and the template of the optimal parameter vector Wi for the next batch of fermentation process are output.

[0006] Preferably, the entire fermentation cycle is automatically divided into n metabolic staging sections, including: Input the feature parameter set M into the clustering algorithm based on the dynamic time warping distance metric, and use the unsupervised clustering method to identify the similarity of the metabolic states in each time window, and obtain the candidate state segment set. Perform phase space reconstruction processing on the candidate state segment set, construct a multi-dimensional time-delay embedding vector sequence, and determine the phase transition inflection point of the metabolic behavior by calculating the degree of orbital convergence and the divergence interval. According to the positions of each phase transition inflection point, the entire fermentation cycle is automatically divided into n metabolic staging sections, and each section is labeled with the corresponding metabolic characteristic label.

[0007] Preferably, the determination of the phase transition inflection point includes: Embed the time series feature data in the feature parameter set M into the phase space, construct a time-delay embedding vector sequence, and each vector consists of the current time and its historical lag value to reconstruct the state orbit. Perform local neighborhood analysis on the state orbit, and calculate the convergence density value and the relative divergence speed between the orbit points in each time period. Based on the critical thresholds of the convergence density and the divergence speed, comprehensively judge the position of the phase transition inflection point of the phase space orbit, and mark the time series data before and after the inflection point as the boundary sections of different metabolic stages respectively.

[0008] Preferably, if the convergence density is less than the convergence density critical threshold and the divergence speed is greater than the divergence speed critical threshold, the current time point is a metabolic phase transition inflection point.

[0009] Preferably, the construction of the multi-stage environmental regulation model Q includes: Take the optimal control parameter vector Wi corresponding to each metabolic staging section as the control target input, and construct a multi-variable control constraint set including temperature, dissolved oxygen, feeding rate, stirring rate and pH value adjustment. Collect and construct a real-time feedback signal set R, and the feedback signals include cell activity, product concentration, substrate residual amount, dissolved oxygen response delay and pH adjustment offset. Wi and the real-time feedback signal R are input into the optimizer under the model predictive control framework to calculate the dynamic deviation between the current environmental factors and the target value, and to predict the optimal adjustment path in the future control cycle. Based on the prediction results, adjustment instructions for each control factor are output, and the control strategy is continuously optimized to keep the key environmental parameters within the optimal range corresponding to Wi.

[0010] Preferably, the calculation of the cell activity enhancement rate ΔA includes: Set the start time within the current fermentation stage. With end time The specific growth rate μ(t) and unit substrate conversion efficiency η(t) of the bacteria were collected during the time period. Calculate the change in specific growth rate Δμ and the change in substrate conversion efficiency Δη during this time period, which are expressed as the endpoint value minus the starting value, respectively. The cell activity enhancement rate ΔA is calculated by weighting Δμ and Δη.

[0011] Preferably, the calculation of the fermentation stress abnormality index S includes the following steps: During the fermentation stage, real-time data on fluctuations in key environmental variables, including dissolved oxygen (DO), pH, temperature (T), and stirring rate (N), are collected and controlled. Calculate the standard deviations σ_DO, σ_pH, σ_T and σ_N of each variable within a set time window Δt, and divide them by the expected control mean of the corresponding variable to obtain their relative volatility. The stress anomaly index S is formed by comprehensively weighting the relative volatility.

[0012] Preferably, the output of the calibration data and optimal parameter vector Wi template for the multi-stage environmental control model Q in the next batch fermentation process includes: After the fermentation batch was completed, the cell activity indicators, product conversion rate and environmental parameter response data collected at each metabolic stage were statistically processed to construct a multidimensional dataset with stage labels. A nonlinear regression modeling method was used to fit the dataset and extract the coupling relationship between changes in environmental variables and metabolic response, which was used to identify bias and overcorrection problems in the staged control strategy. Based on the analysis results, the prediction function and control strategy weights in the original multi-stage regulation model Q are dynamically adjusted to generate model calibration parameters and optimized optimal parameter vector Wi templates for the next batch of fermentation processes, and then archived and stored according to stages.

[0013] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention constructs a multi-stage environmental regulation model Q and combines it with real-time feedback signals and a rolling optimization control mechanism to achieve dynamic identification and refined regulation of different metabolic stages during microbial fermentation. By introducing the cell activity enhancement rate ΔA and the fermentation stress anomaly index S, the physiological state and system stability of the current stage can be accurately assessed, enabling timely judgment of stage failures and automatic switching of control strategies, significantly improving the system's responsiveness, control precision, and product synthesis efficiency.

[0014] 2. This invention possesses cross-batch self-learning and control template optimization capabilities, enabling it to correct the multi-stage control model Q based on historical fermentation data and dynamically update the optimal parameter vector Wi template for the next batch. This mechanism effectively enhances the adaptability, robustness, and industrial repeatability of the fermentation control system, and is particularly suitable for medium- and high-density fermentation processes with complex metabolism and significant nonlinear responses, providing reliable support for efficient, stable, and intelligent fermentation process control. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0018] For examples, please refer to Figure 1 As shown in this embodiment, a multi-stage fermentation method for improving cell activity and stability includes: Establish a multi-source physiological characteristic parameter set M, including metabolic flux, cell activity, conductivity and respiration rate, to obtain the dynamic metabolic state of the target microorganism at different time points. Based on the feature parameter set M, a clustering algorithm and phase space reconstruction method are used to automatically divide the entire fermentation cycle into n metabolic phase segments. Each phase segment defines a set of coupled control parameter vectors. , where: Ti is the temperature control value, DOi is the dissolved oxygen control value, Fi is the feed rate, Ni is the stirring rate, Pi is the pH control strategy, and i = 1, 2, ..., n; Construct a multi-stage environmental regulation model Q, with the parameter vector Wi as the control variable, combined with the real-time feedback signal set , k is the total number of feedback signals; dynamically adjust the key environmental factors during the fermentation process to maintain them within the optimal range corresponding to Wi; During each stage of the fermentation process, calculate the cell activity improvement rate ΔA and the fermentation stress abnormal index S in real time, and compare them with the preset thresholds Ath and Sth respectively; if ΔA < Ath and S ≥ Sth, it is determined that the current stage fails, and the next stage is entered in advance, and at the same time, the updated parameter vector W(i + 1) is activated; After the fermentation is completed, according to the non-linear distribution of cell activity, product conversion rate and environmental regulation metabolic response in each stage, output the calibration data of the multi-stage environmental regulation model Q and the optimal parameter vector Wi template for the next batch of fermentation process.

[0019] To achieve refined regulation and dynamic stage identification in the microbial fermentation process, the present invention first proposes a method for characterizing the metabolic state based on the fusion analysis of multi-source physiological characteristic parameters. By systematically collecting and constructing a multi-dimensional physiological characteristic parameter set M including metabolic flux, cell activity, conductivity and respiration rate, high-resolution characterization and quantitative analysis of the metabolic behavior of the target microorganism at different time sequence nodes can be realized.

[0020] Specifically, the parameter set M is mainly composed of the following four core indicators: Metabolic flux parameters: Metabolic flux refers to the conversion rate of substances in each main metabolic pathway in microbial cells per unit time. Common indicators include glycolysis flux, tricarboxylic acid cycle flux, and amino acid synthesis flux, etc. By offline sampling combined with mass spectrometry analysis (such as GC-MS or LC-MS) and metabolic network modeling techniques (such as FBA or 13C-isotope tracing), the metabolic flux level of a specific path in the cell can be indirectly measured. This parameter reflects the essential state of the overall metabolic activity and product synthesis efficiency of the cell, and is one of the important bases for metabolic state determination.

[0021] Cell activity parameters: Cell activity is the ability of microorganisms to maintain normal proliferation and metabolism under given environmental conditions. Common parameters include specific growth rate (μ) per unit time, unit substrate consumption rate, cell membrane integrity, and the ability to reduce reducing dyes (such as detected by TTC and MTT methods). In the present invention, the parameters with the most dynamic response characteristics in the cell activity index, that is, the specific growth rate of the cell and the membrane integrity index, are selected as characteristic factors and recorded in the form of a time series during the fermentation process.

[0022] Conductivity parameter: The conductivity of the culture medium mainly reflects changes in ion concentration within the system and is closely related to substrate consumption, metabolite accumulation, and cell lysis. Real-time monitoring of the conductivity changes in the fermentation broth using an online conductivity probe can help identify metabolic transition points, especially during cell rupture or when metabolites are released rapidly, resulting in a significant jump in the conductivity curve. As a non-invasive monitoring method, this parameter can provide real-time evidence of system stability or fluctuation trends.

[0023] Respiration rate parameter: Respiration rate is a comprehensive reflection of the oxygen uptake rate (OUR) and carbon dioxide release rate (CER) during bacterial metabolism, and is a key indicator of cellular energy metabolism. This invention uses an integrated gas sensor to perform online sampling and analysis of the intake and exhaust gases, calculates the flow difference of oxygen and carbon dioxide per unit time in real time to obtain OUR and CER curves, and further analyzes their slope changes, fluctuations, and peak periods as a dynamic benchmark for determining changes in activity at different stages.

[0024] The above four types of parameters together constitute the multi-source physiological characteristic parameter set M defined in this invention. During implementation, the above indicators are first collected in parallel according to the time intervals set for each fermentation batch (e.g., every 10 minutes, every 30 minutes, etc.), constructing the parameter vector m(t) at the corresponding time node, where t is the current time point. As the fermentation process progresses, the parameter sequence is continuously recorded. s represents the total number of data collection points, thus forming a multi-source dynamic metabolic state dataset covering the entire process.

[0025] To ensure comparability and compatibility among parameters with different dimensions, all original parameters were normalized and optimized for dimensionality reduction using principal component analysis (PCA) or feature selection algorithms. Subsequently, time window moving average and differential analysis methods were introduced to perform stationarity analysis on the parameter sequences, capturing potential trends and metabolic fluctuations.

[0026] This invention introduces a "metabolic behavior similarity matrix" to measure the evolutionary distance of metabolic states between different time points. This matrix is ​​obtained by calculating the Euclidean distance or dynamic time warping distance (DTW) between parameter vectors at any two time points, and can be used for subsequent stage identification, cluster analysis and state prediction.

[0027] Furthermore, the parameter set M can be used as an input variable in the multi-stage control model to support the decision-making of environmental control parameters for each fermentation stage, including real-time optimization of temperature, dissolved oxygen level, feeding strategy and stirring rate.

[0028] By constructing a multi-source feature parameter set M, we can not only comprehensively characterize the dynamic response characteristics of the microbial metabolic system, but also provide high-dimensional support for subsequent metabolic homeostasis judgment, stage switching condition triggering and control model correction. This is one of the fundamental links to realize intelligent multi-stage fermentation control.

[0029] To achieve phased division of microbial metabolic behavior throughout the entire fermentation cycle, this invention proposes a method combining dynamic clustering based on a feature parameter set M with orbital phase transition determination, which automatically divides the fermentation cycle into n metabolic phase segments. This method integrates data mining, nonlinear system modeling, and time series analysis techniques, enabling intelligent identification and segmentation of the metabolic dynamics of the fermentation system without the need for manually setting thresholds or subjective time breakpoints.

[0030] In this invention, the complete fermentation stage division process includes: First, based on the acquisition of a multi-source physiological characteristic parameter set M, the changing trends of key physiological variables within it are extracted and standardized. The parameter set M includes physical and biological indicators closely related to microbial metabolism, such as metabolic flux, cell activity, conductivity, and respiration rate. This parameter set is represented at the time series level as follows: ,in Indicates at a point in time The parameter vector contains synchronous observations of multiple physiological variables.

[0031] Because different parameters have different dimensions, scales, and sampling errors, directly using the raw data for cluster analysis may cause dimensionality bias. Therefore, this invention employs zero-mean normalization, which involves subtracting the mean from each dimension's data and then dividing by its standard deviation to ensure that the contributions of each variable to subsequent analysis are on the same order of magnitude. This process not only enhances the numerical stability of the algorithm but also improves the accuracy of state difference discrimination.

[0032] To overcome the limitations of traditional clustering methods in handling asynchronous, variable-rate time series data, this invention introduces an unsupervised clustering algorithm based on Dynamic Time Warping (DTW) distance metric. This method can identify the morphological and trend similarities of metabolic states within different time windows, thereby obtaining a set of candidate state segments.

[0033] Specifically, the normalized parameter set M is split into multiple overlapping or non-overlapping time window sequences. Here, c represents the total number of windows. For any two windows Wc and Wj, the dynamic time warping algorithm is used to calculate their morphological matching distance. This involves dynamically pairing the two time series using dynamic programming to minimize their overall nonlinear offset error. The smaller the DTW distance, the more similar the metabolic behaviors within the two windows.

[0034] Based on the DTW distance matrix between all time windows, hierarchical clustering or density clustering methods are used for classification analysis to obtain several candidate state fragment sets with similar metabolic trends. , where m is the total number of segments. This set of segments provides the initial distribution basis for subsequent phase space orbit modeling and inflection point detection.

[0035] After obtaining the set of candidate state segments, this invention further introduces a phase space reconstruction method to model the nonlinear dynamic trajectory of the system in order to identify key metabolic stage transition points during fermentation. This method, derived from dynamical system theory, can embed a one-dimensional time series into a multi-dimensional state space to reconstruct the phase trajectory structure of the original system.

[0036] In practice, a representative component x(t) of the feature parameter vector m(t) is selected, and an appropriate time delay τ and embedding dimension d are set to construct a time delay vector sequence: This process can be extended to multidimensional parameter vectors. By constructing the state vector Y(t) at each time point, a multidimensional trajectory sequence of the entire fermentation process can be formed. .

[0037] Next, local neighborhood analysis is performed. At each time point... The corresponding state vector The orbital density within a certain neighborhood radius ε is calculated to evaluate the convergence density of the system at that moment. Simultaneously, the average Euclidean distance growth rate of state changes within a short time window is defined as the orbital divergence rate.

[0038] If the convergence density increases continuously over a period of time while the divergence rate decreases, it indicates that the system is in a metabolically stable phase. If the convergence density drops sharply while the divergence rate increases dramatically, it indicates that the metabolic behavior has entered a region of drastic change, which may be accompanied by regulatory switching, substrate depletion, or environmental stress.

[0039] Based on the above two orbital characteristic indicators, this invention sets a convergence density critical threshold and a divergence velocity critical threshold. When the following combined condition is met, the system is determined to be at a critical point in time. Phase transition inflection point where metabolic behavior occurs: If the convergence density is less than the convergence density critical threshold and the divergence velocity is greater than the divergence velocity critical threshold, then This represents a metabolic phase transition inflection point.

[0040] To improve the accuracy of judgment, a sliding window mechanism can be introduced to jointly judge multiple consecutive time points and eliminate misjudgments due to isolated fluctuations.

[0041] After completing the inflection point identification, each phase transition inflection point... Using the boundary as the boundary, the complete fermentation cycle is divided into n metabolic phase segments. Each phase segment defines a set of coupled control parameter vectors. Where: Ti is the temperature control value, DOi is the dissolved oxygen control value, Fi is the feed rate, Ni is the stirring rate, Pi is the pH control strategy, and i = 1, 2, ..., n.

[0042] To achieve precise regulation of the microbial growth environment at multiple metabolic stages of the fermentation process, this invention constructs a multi-stage environmental regulation model Q. This model uses the control parameter vector Wi corresponding to each metabolic stage as the target input, combined with the multi-source feedback signal set R collected in real time during fermentation. Based on the principle of model predictive control, it dynamically adjusts key environmental factors and performs rolling optimization adjustments during control execution, thereby enabling the system to adapt to fluctuations in metabolic state and continuously maintain the optimal stable state of the fermentation environment.

[0043] The construction and application of the regulation model Q described in this invention include: First, after the metabolic stages are automatically divided, the system assigns the optimal control parameter vector to each metabolic stage segment Pi. As input targets. Where Ti represents the target temperature for this stage, DOi represents the target dissolved oxygen level, Fi is the bottom flow acceleration rate, Ni is the stirring speed, and Pi is the target pH control value.

[0044] The system constructs a multivariate control constraint set based on the upper and lower limits, physiologically suitable ranges, and stage response characteristics of each variable in Wi, to constrain the adjustment space of each control variable. For example, the temperature constraint is... Dissolved oxygen is The remaining variables are set accordingly. The control model must satisfy this set of constraints when performing regulation calculations to avoid overshoot, environmental stress, and other situations that could affect the stability of the bacteria.

[0045] To achieve dynamic closed-loop regulation, this invention designs a real-time feedback signal set. This is used to reflect the current operating status of the system in real time. The feedback signal includes, but is not limited to, the following variables: Indicators of bacterial cell activity, such as respiration rate per unit time and specific growth rate; Product concentration: The concentration of the target fermentation product determined online or offline; Substrate residue: Concentration of unconsumed carbon or nitrogen source; Dissolved oxygen response delay: the time difference between changes in oxygen input and the system's DO response; pH adjustment offset: Sets the continuous deviation between the pH value and the measured value.

[0046] The feedback signal enters the control model at a certain sampling frequency (such as every 5 minutes or 10 minutes) and is used as the current state vector input. It is compared with the set target Wi in real time, forming the basis of model predictive control.

[0047] Based on this, the present invention employs a model predictive control method to analyze the difference between the current feedback state and the control objective, and predict the evolution trend of environmental variables over the next L control cycles.

[0048] Specifically, the system, based on a fermentation kinetics model (which can be a simplified mechanism model or a data-driven model), performs the following operations within each control cycle: Predict the changing trends of current environmental factors (such as temperature, dissolved oxygen, etc.) over the next L steps; construct an error function between the target and the predicted values, for example, the total error E is the sum of weighted deviations at each time step; Introducing an optimizer to solve a set of control action sequences This minimizes the error function E and ensures that all control actions satisfy the constraint set.

[0049] The sequence of control actions determines the adjustment range and speed in each future small cycle, such as whether the stirring rate should be gradually increased to improve dissolved oxygen, or whether the substrate feeding should be appropriately delayed to alleviate stress.

[0050] To ensure the control strategy has continuous responsiveness, this invention employs a rolling optimization control mechanism. That is, after each control cycle, the system uses the latest feedback state as the starting point to re-execute the prediction and optimization process, forming a closed-loop self-adjusting process of "perception-prediction-execution-update".

[0051] The mechanism mainly includes the following: The system re-predicts the future state in each control cycle to improve its adaptability to environmental fluctuations and disturbances. The weight of the recent prediction error is large, while the weight of the long-term prediction error is small, in order to balance prediction accuracy and response robustness. If the execution error or disturbance causes the actual state to deviate from the predicted trajectory, the control model will automatically adjust the control quantity to compensate for the deviation. By fine-tuning the control variables, the stability of parameter changes is maintained, and the stress of mutation regulation on the bacteria is avoided.

[0052] For example, if the target dissolved oxygen (DOi) is 30% at a certain stage, and the DO is detected to drop below 25% for two consecutive cycles, the system determines that the current oxygen supply is insufficient. Through rolling optimization and predictive analysis, it determines whether the stirring speed (Ni) or the ventilation volume should be increased immediately, and calculates the DO trend for the next three cycles under its influence, and provides an adjustment strategy within the constraints.

[0053] To improve the accuracy and timeliness of environmental control during multi-stage fermentation, this invention designs a mechanism for dynamically monitoring fermentation status and determining whether a stage has failed. This mechanism comprehensively considers the physiological and metabolic performance of microorganisms and the stability of the fermentation system, calculating two key indicators in real time: cell activity enhancement rate ΔA and fermentation stress anomaly index S, and comparing them with preset thresholds. This enables intelligent determination of the effectiveness of each fermentation stage and, when necessary, triggers the next stage control strategy in advance.

[0054] During each metabolic phase of operation, the system continuously calculates the current values ​​of ΔA and S, and compares them with the set critical thresholds Ath (lower limit of activity enhancement) and Sth (upper limit of stress fluctuation): If ΔA is less than Ath, it means that the bacterial activity has not been improved as expected during this stage. Meanwhile, if S is greater than or equal to Sth, it indicates that the system is in a state of strong fluctuation or stress. When the above two conditions are met, the system determines that the current fermentation stage has failed, immediately triggers a stage jump, enters the next stage ahead of schedule, and activates the parameter vector W(i+1) to execute the optimal combination of environmental control parameters for the next stage.

[0055] This strategy avoids decreased cell activity, metabolic disorders, or reduced product synthesis efficiency due to stage delays, exhibiting high responsiveness and system stability.

[0056] The cell activity enhancement rate ΔA is used to measure the degree of enhancement in microbial activity during the current stage, and its calculation method is as follows: Set a fixed time window during the current fermentation stage. Two key physiological parameters of the bacteria were collected during this time period: Specific growth rate μ(t): represents the growth rate of the bacterial cells per unit time; Substrate conversion efficiency η(t): represents the amount of target product converted from each unit of substrate, reflecting the metabolic efficiency of the bacteria.

[0057] Within the time window, calculate the changes in μ(t) and η(t) respectively: Specific growth rate change ; Change in conversion efficiency ; This change can be estimated by differentiation or obtained by directly interpolating the time endpoints.

[0058] We combine Δμ and Δη in a weighted manner to form a single index ΔA: ; where α and β are pre-set positive weighting coefficients, the values ​​of which can be set according to the metabolic characteristics of the target strain or application requirements. For example, when more emphasis is placed on product efficiency, β can be greater than α.

[0059] In this way, ΔA can comprehensively reflect the changes in cell activity level and resource utilization ability, and is an important basis for judging whether the bacterial cells are continuously in a good physiological state.

[0060] The fermentation stress anomaly index S is a composite index for evaluating the intensity and stability of the fermentation system environment, and its calculation method is as follows: During each fermentation stage, continuously collect the key environmental control variables of the system, including: dissolved oxygen (DO), pH value, temperature (T), and stirring rate (N); these parameters directly affect the physiological activities of the bacterial cells and are the core reflections of the system environmental stability.

[0061] Within the set time window Δt, calculate the standard deviations of the above four variables, denoted as: σ_DO represents the standard deviation of DO; σ_pH represents the standard deviation of pH; σ_T represents the standard deviation of temperature; σ_N represents the standard deviation of the stirring rate.

[0062] At the same time, calculate the expected control means of each variable, denoted as: DŌ, , , respectively represent the target stable levels of each variable.

[0063] Then, divide each standard deviation by its expected mean to obtain the relative volatility: relative volatility ; relative volatility ; relative fluctuation ; relative volatility .

[0064] Perform a weighted combination of the above relative volatilities to form a single index S: Among them, to are the weight coefficients of each variable, representing their influence degrees on the system stability, and satisfy . The weights can be set according to the sensitivity of the bacterial strain to a certain type of environmental variable. For example, if a certain bacterial strain is more sensitive to pH changes, then should be appropriately increased. The larger the value of the index S, the more intense the environmental fluctuation in the current stage and the worse the system stability, and it is an important parameter reflecting potential fermentation risks.

[0065] In practical applications, the system continuously monitors the current values of ΔA and S, and compares them with their preset threshold values Ath and Sth: When ΔA ≥ Ath and S < Sth: It indicates that the current stage is still valid, and the system maintains the current control strategy; When ΔA < Ath and S ≥ Sth: Determine that the current stage fails, and the system actively switches to the next stage in advance, activates the parameter vector W(i+1), and updates the control model; If only ΔA or S approaches the boundary, an early warning mechanism or a delay confirmation mechanism can be triggered to avoid misjudgment.

[0066] This judgment logic can be embedded in the feedback decision-making module of the model predictive controller to achieve dynamic, intelligent, and multi-dimensional management and control of the fermentation stage, effectively improving the stability, response ability, and product yield of the fermentation system.

[0067] To enhance the transferability, self-learning ability, and cross-batch adaptability of the fermentation process control system, after the fermentation is completed, based on the operation data of each stage of the previous batch, the multi-stage environmental regulation model Q is corrected, and the optimal parameter vector Wi template applicable to the next batch of fermentation tasks is output. By establishing a cross-batch correction mechanism driven by "process feedback data" as the core, not only can the model prediction performance be optimized, but the structure and parameters of the control strategy can also be dynamically adjusted to improve the starting accuracy and response efficiency of subsequent fermentation batches.

[0068] After the entire fermentation cycle is completed, the system automatically archives the key operation data of all metabolic stages and marks them by stage. Specifically, it includes: Cell activity data: such as specific growth rate, respiration rate, substrate conversion efficiency, etc.; Product conversion rate data: the change in the synthesis concentration of the target metabolite per unit time; Environmental regulation variables: including actual control parameters (temperature, dissolved oxygen, pH, stirring rate, feed flow rate) and system response delay, fluctuation characteristics, etc.

[0069] This data is divided into multiple time series segments according to each metabolic stage, and is structurally matched with the current control parameter vector Wi to form a stage operation data set with the triple attributes of "input-response-feedback".

[0070] For example, in the metabolic stage The corresponding parameter vector is , including temperature , dissolved oxygen , etc. The system will record Under the control, the cell activity promotion rate , stress index , product conversion rate , etc. of the feedback results, and compare and analyze them with the actual production target.

[0071] Because fermentation is a strongly coupled nonlinear system, the sensitivity and response patterns of microbial cells to control parameters may vary significantly at different stages. This invention introduces a data fitting and modeling mechanism to extract the mapping relationship between changes in environmental variables and the system's metabolic response. The steps are as follows: Normalize the process data within each stage to ensure that the dimensions of different variables are consistent. Nonlinear modeling methods, such as support vector regression (SVR), random forest regression, and neural network fitting, are used to establish a staged input-output prediction model by taking control parameters (such as temperature, DO, and pH) as input features and bacterial cell response parameters (such as ΔA or η) as output labels. Based on model evaluation metrics (such as goodness of fit R² and mean squared error MSE), determine whether there is a significant nonlinear relationship, and further explore the parameter factors that cause the deviation of the stage control strategy. If a small disturbance in a certain control parameter has a significant impact on the fluctuation of the system output, then the parameter can be identified as a "highly sensitive variable" and should be given higher weight in subsequent model calibration.

[0072] After completing the modeling of the input-output relationship, this invention enters the parameter correction and strategy update stage, which mainly includes the following: Based on the aforementioned fitting results, the system can update the structure of the multi-stage control model Q. If a certain stage exhibits significant prediction bias or adjustment delay, the control prediction window length, rolling optimization frequency, or objective function weight coefficients can be appropriately adjusted to enhance the model's real-time performance and robustness.

[0073] For each control variable in the original parameter vector Wi, template optimization is performed as follows: If a certain variable (such as stirring rate Ni) frequently reaches its upper or lower limit during the control period and the output deviates significantly from the target, the starting template for the next batch can be optimized by adjusting its setpoint or control range. If there is a lag response relationship between the change of a certain variable and the product synthesis, then increase its adjustment step size or prediction window period; The system can also construct parameter iteration curves and extract the optimal starting interval based on multiple batches of historical templates and feedback data.

[0074] The final Wi template structure is as follows: Each of these is an optimized target control value, which can be directly called in the initial stage of the next batch.

[0075] To support standardized management of large-scale fermentation tasks, this invention designs a template archiving mechanism, which stores the Wi(opt) template and Q model parameters output for each batch in a knowledge base and indexes them by strain, culture conditions, and process objectives.

[0076] In subsequent fermentation, the system can call the most suitable starting control strategy from historical templates according to the current task requirements, shortening the modeling time and improving the fermentation success rate.

[0077] Meanwhile, this invention supports a continuous iterative learning mechanism for templates. That is, after each batch, the historical best template is scored and replaced based on the model prediction accuracy and feedback deviation, so as to ensure that the control strategy is always updated.

[0078] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A multi-stage fermentation method for improving cell activity and stability, characterized in that: Comprising: Establish a multi-source physiological characteristic parameter set M including metabolic flux, cell activity, conductivity, and respiration rate, and obtain the dynamic metabolic state of the target microorganism at different time sequence nodes; Based on the feature parameter set M, a clustering algorithm and phase space reconstruction method are used to automatically divide the entire fermentation cycle into n metabolic phase segments. Each phase segment defines a set of coupled control parameter vectors. Where: Ti is the temperature control value, DOi is the dissolved oxygen control value, Fi is the feed rate, Ni is the stirring rate, Pi is the pH control strategy, and i = 1, 2, ..., n; Construct a multi-stage environmental regulation model Q, using the parameter vector Wi as the control variable, and combine it with a real-time feedback signal set. k is the total number of feedback signals; key environmental factors in the fermentation process are dynamically adjusted to maintain them within the optimal range of the corresponding Wi. During each stage of the fermentation process, calculate the cell activity increase rate ΔA and the fermentation stress anomaly index S in real time, and compare them with the preset thresholds Ath and Sth respectively; if ΔA < Ath and S ≥ Sth, it is determined that the current stage fails, and the next stage is entered in advance, and at the same time, the updated parameter vector W(i + 1) is activated; After the fermentation is completed, according to the non-linear distribution of cell activity, product conversion rate, and environmental regulation metabolic response in each stage, output the correction data of the multi-stage environmental regulation model Q and the optimal parameter vector Wi template for the next batch of fermentation processes.

2. The multi-stage fermentation method for improving cell activity and stability according to claim 1, characterized in that: Automatically divide the entire fermentation cycle into n metabolic staging sections, including: Input the characteristic parameter set M into a clustering algorithm based on dynamic time warping distance metric, and use unsupervised clustering to identify the similarity of metabolic states within each time window, obtaining a candidate state segment set; Perform phase space reconstruction processing on the candidate state segment set, construct a multi-dimensional time delay embedding vector sequence, and determine the phase transition inflection point of the metabolic behavior by calculating the degree of orbit convergence and the divergence interval; According to the positions of each phase transition inflection point, automatically divide the entire fermentation cycle into n metabolic staging sections, and label corresponding metabolic characteristic labels for each section.

3. The multi-stage fermentation method for improving cell activity and stability according to claim 2, characterized in that: The determination of the phase transition inflection point includes: Embed the time series feature data in the characteristic parameter set M into the phase space, construct a time delay embedding vector sequence, and each vector consists of the current moment and its historical lag value to reconstruct the state orbit; Perform local neighborhood analysis on the state orbit, and calculate the convergence density value and relative divergence speed between orbit points within each time period; Based on the critical thresholds of convergence density and divergence speed, comprehensively judge the position of the phase transition inflection point of the phase space orbit, and label the time series data before and after the inflection point as the boundary sections of different metabolic stages respectively.

4. The multi-stage fermentation method for improving cell activity and stability according to claim 3, characterized in that: If the convergence density is less than the convergence density critical threshold and the divergence speed is greater than the divergence speed critical threshold, the current time point is a metabolic phase transition inflection point.

5. The multi-stage fermentation method for improving cell activity and stability according to claim 1, characterized in that: The construction of the multi-stage environmental regulation model Q includes: Take the optimal control parameter vector Wi corresponding to each metabolic staging section as the control target input, and construct a multi-variable control constraint set including temperature, dissolved oxygen, feeding rate, stirring rate, and pH value adjustment; Collect and construct a real-time feedback signal set R, and the feedback signals include cell activity, product concentration, substrate residue, dissolved oxygen response delay, and pH adjustment offset; Input Wi and the real-time feedback signal R into an optimizer under the model predictive control framework, calculate the dynamic deviation between the current environmental factor and the target value, and predict the optimal adjustment path within a future control period; Based on the prediction results, output the adjustment instructions for each control factor, and perform rolling optimization on the regulation strategy to continuously maintain the key environmental parameters within the optimal range corresponding to Wi.

6. The multi-stage fermentation method for improving cell activity and stability according to claim 1, characterized in that: The calculation of the cell activity increase rate ΔA includes: Set the start time within the current fermentation stage. With end time The specific growth rate μ(t) and unit substrate conversion efficiency η(t) of the bacteria were collected during the time period. Calculate the change amount Δμ of the specific growth rate and the change amount Δη of the substrate conversion efficiency within this time period, which are respectively expressed as the end value minus the start value; The cell activity enhancement rate ΔA is calculated by weighting Δμ and Δη.

7. The multi-stage fermentation method for improving cell activity and stability according to claim 6, characterized in that: The calculation of the fermentation stress abnormality index S includes the following steps: During the fermentation stage, real-time data on fluctuations in key environmental variables, including dissolved oxygen (DO), pH, temperature (T), and stirring rate (N), are collected and controlled. Calculate the standard deviations σ_DO, σ_pH, σ_T and σ_N of each variable within a set time window Δt, and divide them by the expected control mean of the corresponding variable to obtain their relative volatility. The stress anomaly index S is formed by comprehensively weighting the relative volatility.

8. The multi-stage fermentation method for improving cell activity and stability according to claim 1, characterized in that: The output, used for the calibration data and optimal parameter vector Wi template of the multi-stage environmental control model Q in the next batch fermentation process, includes: After the fermentation batch was completed, the cell activity indicators, product conversion rate and environmental parameter response data collected at each metabolic stage were statistically processed to construct a multidimensional dataset with stage labels. A nonlinear regression modeling method was used to fit the dataset and extract the coupling relationship between changes in environmental variables and metabolic response, which was used to identify bias and overcorrection problems in the staged control strategy. Based on the analysis results, the prediction function and control strategy weights in the original multi-stage regulation model Q are dynamically adjusted to generate model calibration parameters and optimized optimal parameter vector Wi templates for the next batch of fermentation processes, and then archived and stored according to stages.

Images