Fermentation optimization system and method for enhancing immune efficacy of Monascus based on data analysis

Abstract

The present invention belongs to the technical field of microbial fermentation engineering, and provides a fermentation optimization system and method for enhancing the immune efficacy of Monascus based on data analysis, comprising: a multi-source data edge preprocessing module deploys sensors to collect data such as dissolved oxygen and temperature, which is normalized through edge computing filtering; a bacterial oxygen consumption rate prediction model training module utilizes LSTM combined with an attention mechanism to establish an oxygen consumption rate prediction model based on historical data, with RMSE <5%; a dissolved oxygen dynamic control strategy module integrates MPC and fuzzy control, dynamically adjusts ventilation volume and stirring rate according to predicted oxygen consumption, with a control cycle of 30 seconds; an actuator and feedback correction module executes control through a mass flow controller, etc., and corrects model parameters according to DO deviation; a full-process closed-loop optimization module sets dual warnings, and updates control parameters based on a Bayesian optimization algorithm, so that the system realizes precise control of dissolved oxygen during the fermentation process and improves the biomass and active ingredient yield of Monascus.

Description

Technical Field

[0001] The present invention belongs to the technical field of microbial fermentation engineering, and in particular to a fermentation optimization system and method for enhancing the immune efficacy of Monascus based on data analysis. Background Art

[0002] During the fermentation process of Monascus, the oxygen consumption rate of the bacteria increases exponentially during the exponential growth period. The traditional dissolved oxygen control system relies on lagging feedback regulation and cannot match the oxygen consumption demand of the bacteria in real time, which often leads to the death of mycelium due to lack of oxygen, affecting the biomass of Monascus and the yield of active ingredients.

[0003] In the existing technology, the ventilation and stirring control strategy based on fixed parameters is difficult to adapt to the dynamically changing oxygen consumption demand, and there is an urgent need for a data-driven intelligent fermentation optimization method.

[0004] To this end, the present invention provides a fermentation optimization system and method for enhancing the immune efficacy of Monascus based on data analysis. Summary of the Invention

[0005] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0006] The technical solution adopted by the present invention to solve its technical problem is:

[0007] In a first aspect, the present invention provides a fermentation optimization system for enhancing the immune efficacy of Monascus based on data analysis, comprising:

[0008] Multi-source data edge pre-processing module: deploys multiple sensors to collect a variety of real-time data, and uses edge computing servers to pre-process the collected real-time data;

[0009] Bacteria oxygen consumption rate prediction model training module: Based on historical fermentation data, a bacteria oxygen consumption rate prediction model is established through machine learning algorithms;

[0010] Dissolved oxygen dynamic control strategy module: Establishes a dissolved oxygen dynamic control strategy and uses model predictive control to dynamically adjust the ventilation volume Q and stirring rate N. It also introduces fuzzy control rules to handle nonlinear disturbances, triggers fuzzy controller adjustments, and ensures smooth control actions.

[0011] Actuator and Feedback Correction Module: This module sets up actuator control and feedback correction mechanisms. The ventilation system uses a mass flow controller to regulate the flow of the air and oxygen mixture. The stirring system uses a variable frequency motor to drive a three-layer pitched-blade turbine impeller, dynamically adjusting the speed according to the fermentation stage. After each control action is completed, the feedback correction mechanism is activated to correct the prediction model parameters.

[0012] Full-process closed-loop optimization module: Establish a full-process optimization mechanism and ensure the stability of the fermentation process by setting up a dual early warning mechanism. After the fermentation batch is completed, analyze the correlation between fermentation indicators and model control parameters, and update the model prediction control parameters based on the Bayesian optimization algorithm.

[0013] As a further improvement of the present invention, the specific process of deploying multiple sensors is as follows:

[0014] Various sensors are deployed in a circular array at different heights in the fermentation tank, 5 cm, 15 cm, and 25 cm from the bottom of the tank. The sensors include: fluorescence quenching fiber optic dissolved oxygen sensor, thermocouple temperature sensor, glass electrode pH sensor, and optical density sensor.

[0015] As a further improvement of the present invention, the multiple real-time data include: real-time dissolved oxygen concentration DO data collected by a fluorescence quenching fiber optic dissolved oxygen sensor, real-time temperature T data collected by a thermocouple temperature sensor, real-time pH value data collected by a glass electrode pH sensor, and real-time bacterial concentration X data collected by an optical density sensor.

[0016] As a further improvement of the present invention, the specific process of establishing a bacterial oxygen consumption rate prediction model using a machine learning algorithm is as follows:

[0017] Collect data from at least 50 batches of Monascus fermentation process, with a fermentation period of ≥48 hours per batch and a sampling frequency of 10 minutes per batch. Ensure that each batch contains ≥1000 data points, including dissolved oxygen concentration (DO), temperature (T), pH value, bacterial concentration (X), and oxygen consumption rate (OUR).

[0018] Construct feature engineering to extract DO, T, pH, X, and historical oxygen consumption rate OUR data at the current time and in the previous 30 minutes as input features. The output structure of the bacterial oxygen consumption rate prediction model is output: the predicted value of oxygen consumption rate OUR in the next 30 minutes, and construct a time series dataset;

[0019] The prediction model is constructed by combining the long short-term memory network (LSTM) with the attention mechanism (Attention). The root mean square error (RMSE) is used as the loss function, and the model is trained using the Adam optimizer until the RMSE is less than 5%.

[0020] As a further improvement of the present invention, the specific process of dynamically adjusting the ventilation volume Q and the stirring rate N is as follows:

[0021] Set the dissolved oxygen target value, Set the dissolved oxygen target value 20%~30% air saturation is used as the control benchmark;

[0022] The model predictive control (MPC) optimization problem is solved by using the predicted oxygen consumption rate (OUR) for the next 30 minutes as input and the model predictive control (MPC) algorithm to solve the optimization problem, that is, to determine the regulation strategy of the ventilation volume (Q) and the stirring rate (N).

[0023] Based on the predicted value of oxygen consumption rate OUR output by the bacterial oxygen consumption rate prediction model, the optimal control quantity is solved through rolling optimization to balance the dissolved oxygen deviation and the change range of the control quantity.

[0024] As a further improvement of the present invention, the specific process of dynamically adjusting the ventilation volume Q and the stirring rate N is as follows:

[0025] The specific process of introducing fuzzy control rules to handle nonlinear disturbances, triggering fuzzy controller adjustments, and ensuring the smoothness of control actions is as follows:

[0026] When the deviation between the measured DO and the target value exceeds ±10% of the target value, the fuzzy controller is triggered to dynamically adjust the ventilation volume Q and stirring rate N according to the change trend of the collected real-time dissolved oxygen concentration DO data;

[0027] Control cycle setting The control cycle is set to 30 seconds, that is, adjustment is performed every 30 seconds to ensure smooth control action and avoid drastic fluctuations in fermentation parameters.

[0028] As a further improvement of the present invention, the specific process of the ventilation system using a mass flow controller to regulate the flow of the air and oxygen mixed gas is as follows:

[0029] A mass flow controller (MFC) with an accuracy of ±1% FS was used to adjust the flow of the mixed gas of air and oxygen. A microporous distributor was deployed at the gas inlet. The pore diameter of the microporous distributor was 50 , ensure that the average diameter of bubbles is less than 2mm, and improve the dissolved oxygen transfer efficiency.

[0030] As a further improvement of the present invention, the stirring system uses a variable frequency motor to drive a three-layer pitched blade turbine stirring paddle, and the specific process of dynamically adjusting the speed according to the fermentation stage is as follows:

[0031] A variable frequency motor is used to drive a three-layer inclined blade turbine agitator. The speed is dynamically adjusted according to the fermentation stage. If the fermentation stage is in the exponential growth phase, the speed is set to 300-500 rpm; if the fermentation stage is in the stable phase, the speed is set to 200-300 rpm.

[0032] As a further improvement of the present invention, the specific process of correcting the prediction model parameters is as follows:

[0033] By adjusting the weight layer parameters of the long short-term memory network (LSTM) and the attention mechanism (Attention) network, the deviation between the subsequent predicted values ​​and the measured values ​​is reduced, thereby achieving the purpose of correcting the parameters of the bacterial oxygen consumption rate prediction model.

[0034] In a second aspect, the present invention provides a fermentation optimization method for enhancing the immune efficacy of Monascus based on data analysis, comprising:

[0035] S1: Deploy multiple sensors to collect multiple real-time data, and use edge computing servers to pre-process the collected real-time data;

[0036] S2: Based on historical fermentation data, a prediction model for bacterial oxygen consumption rate is established using a machine learning algorithm.

[0037] S3: Establish a dynamic control strategy for dissolved oxygen, use model predictive control to achieve dynamic adjustment of ventilation volume Q and stirring rate N, and introduce fuzzy control rules to handle nonlinear disturbances, trigger fuzzy controller adjustment, and ensure smooth control action;

[0038] S4: Set up actuator control and feedback correction mechanisms. The ventilation system uses a mass flow controller to regulate the flow of the air and oxygen mixture. The stirring system uses a variable frequency motor to drive a three-layer pitched-blade turbine stirrer. The speed is dynamically adjusted according to the fermentation stage. After each control action is completed, the feedback correction mechanism is activated to correct the prediction model parameters.

[0039] S5: Establish a full-process optimization mechanism to ensure the stability of the fermentation process by setting up a dual early warning mechanism. After the fermentation batch is completed, analyze the correlation between the fermentation indicators and the model control parameters, and update the model prediction control parameters based on the Bayesian optimization algorithm.

[0040] The beneficial effects of the present invention are as follows:

[0041] 1. A prediction model is constructed using LSTM combined with an attention mechanism. Training based on more than 50 batches of historical data achieves a root mean square error (RMSE) of less than 5% in oxygen consumption rate (OUR) prediction, reducing the error rate of traditional ARIMA models. Peak bacterial oxygen consumption is captured 30 minutes in advance, avoiding hysteresis fluctuations in dissolved oxygen. The model is fine-tuned every two hours using the current batch's nearly two hours of real-time data to adapt to bacterial strain variation and environmental fluctuations, improve the dissolved oxygen matching degree during the exponential growth period, and address the problem that traditional fixed models cannot cope with dynamic changes during the fermentation stage.

[0042] 2. With an air saturation of 20% to 30% as the target, the ventilation volume Q and stirring rate N are continuously optimized through model predictive control (MPC). Fuzzy control is combined to handle nonlinear disturbances with DO deviation exceeding ±10%. A control cycle of 30 seconds ensures smooth operation, stabilizes the dissolved oxygen time ratio, and reduces the risk of mycelium death due to lack of oxygen. The mass flow controller is combined with a 50μm microporous distributor to make the bubble diameter less than 2mm, improving the dissolved oxygen transfer efficiency. The three-layer pitched blade turbine agitator dynamically adjusts the speed according to the fermentation stage to reduce stirring energy consumption. When DO is less than 15% air saturation or OUR is greater than 20% of the predicted value, an early warning is triggered. Combined with a 3-minute feedback correction mechanism, the response time to fermentation abnormalities is shortened to ensure the stability of the fermentation process.

[0043] 3. After each control action, the LSTM and attention network weights are adjusted to continuously reduce the prediction deviation. After 10 batches of iterations, the model prediction error converges, significantly enhancing the system's adaptability to fermentation parameter fluctuations. After the end of the batch, the Bayesian optimization algorithm is used to update the control parameters with Monacolin K yield and energy consumption as the objective functions, breaking through the yield bottleneck of traditional fixed parameter control. Combined with dry weight method, HPLC, ELISA and other detection methods, the correlation between fermentation indicators and control parameters is analyzed to achieve a multi-objective balance of biomass, active ingredient yield, and energy consumption, providing data support for industrial production. The edge computing server realizes real-time data filtering and normalization, reducing cloud transmission volume, lowering system latency, and adapting to the high real-time requirements of industrial sites. The attention mechanism quantifies the contribution of each parameter to oxygen consumption prediction, and the model decision-making process is transparent. The modular design supports the flexible expansion of sensor arrays and control algorithms, suitable for Monascus fermentation scenarios of different scales. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The present invention will be further described below with reference to the accompanying drawings.

[0045] Figure 1 This is a system module diagram of a fermentation optimization system for enhancing the immune efficacy of Monascus based on data analysis of the present invention;

[0046] Figure 2 The present invention is a flowchart of the steps of the fermentation optimization method for enhancing the immune effect of Monascus based on data analysis. DETAILED DESCRIPTION

[0047] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0048] Example 1

[0049] like Figure 1 As shown, the fermentation optimization system for enhancing the immune efficacy of Monascus based on data analysis according to an embodiment of the present invention includes:

[0050] Multi-source data edge pre-processing module: deploys multiple sensors to collect a variety of real-time data, and uses edge computing servers to pre-process the collected real-time data;

[0051] The specific process of deploying multiple sensors is as follows:

[0052] A variety of sensors were deployed in a circular array at different heights inside the fermentation tank, 5 cm, 15 cm, and 25 cm from the bottom of the tank. These sensors included: fluorescence quenching fiber optic dissolved oxygen sensor, thermocouple temperature sensor, glass electrode pH sensor, and optical density sensor.

[0053] The multiple real-time data include: real-time dissolved oxygen concentration DO data collected by a fluorescence quenching optical fiber dissolved oxygen sensor, real-time temperature T data collected by a thermocouple temperature sensor, real-time pH value data collected by a glass electrode pH sensor, and real-time bacterial concentration X data collected by an optical density sensor;

[0054] Through the 5G wireless transmission module, the real-time data collected by various deployed sensors is transmitted to the edge computing server at a frequency of 10Hz;

[0055] The edge computing server performs sliding average filtering and normalization on the collected real-time data. By setting a sliding window size of 10 seconds, the collected real-time data is smoothed to reduce noise interference;

[0056] Using the normalization method, the collected data is mapped to the [0,1] interval to eliminate the influence of different parameter dimensions and facilitate subsequent model training;

[0057] Specifically, when using the sliding average filter, the real-time data collected in the window is , where n is the window size, n=10×10=100 data points. Since the sampling frequency is 10Hz, there are 100 points in 10s. After filtering, all the real-time data collected in the window are for: ;

[0058] Map the real-time data collected in the window to the interval [0,1]. Specifically, let the real-time data collected in the window be The minimum value of the real-time data collected in the window is , the maximum value of the real-time data collected in the window is , the normalized value is: ;

[0059] Through the above steps, methods and formulas, real-time and accurate collection and preprocessing of fermentation parameters were achieved, providing a reliable data basis for subsequent oxygen consumption prediction and control strategies;

[0060] Bacteria oxygen consumption rate prediction model training module: Based on historical fermentation data, a bacteria oxygen consumption rate prediction model is established through machine learning algorithms;

[0061] The specific process of establishing the bacterial oxygen consumption rate prediction model through the machine learning algorithm is as follows:

[0062] Collect data from at least 50 batches of Monascus fermentation process, with a fermentation period of ≥48 hours per batch and a sampling frequency of 10 minutes per batch. Ensure that each batch contains ≥1000 data points, including dissolved oxygen concentration (DO), temperature (T), pH value, bacterial concentration (X), and oxygen consumption rate (OUR).

[0063] Construct feature engineering to extract DO, T, pH, X, and historical oxygen consumption rate OUR data at the current time and in the previous 30 minutes as input features. The output structure of the bacterial oxygen consumption rate prediction model is output: the predicted value of oxygen consumption rate OUR in the next 30 minutes, and construct a time series dataset;

[0064] Time series feature extraction extracts dynamic features based on a sliding time window to capture the temporal dependencies of parameters during the fermentation process;

[0065] The prediction model is constructed by combining the long short-term memory network LSTM with the attention mechanism Attention, and the root mean square error RMSE is used as the loss function. Specifically, the prediction value is , the true value is , the number of samples n, then RMSE is defined as: , this formula is used to measure the degree of deviation between the model prediction value and the actual value. The training goal is to make RMSE < 5%;

[0066] The model is trained using the Adam optimizer, with steps including first-order moment estimation, second-order moment estimation, momentum update, adaptive learning rate update, bias correction, and parameter update, until RMSE is less than 5%.

[0067] LSTM is used to handle long-sequence temporal dependencies, and the attention mechanism enhances the Attention model's ability to focus on key time point features, improving prediction accuracy.

[0068] After each fermentation batch is completed, the new fermentation batch data is added to the training set, and the model is fine-tuned every 2 hours using the real-time data of the current batch fermentation for nearly 2 hours to adapt to bacterial strain variation and environmental fluctuations;

[0069] The model is trained with historical data and fine-tuned with online data to enable the model to adapt to dynamic changes in the fermentation process.

[0070] Through the above steps, methods and formulas, a prediction model adapted to the dynamic oxygen consumption characteristics of Monascus fermentation was constructed, providing an accurate prediction basis for subsequent dissolved oxygen control strategies;

[0071] Dissolved oxygen dynamic control strategy module: Establishes a dissolved oxygen dynamic control strategy and uses model predictive control to dynamically adjust the ventilation volume Q and stirring rate N. It also introduces fuzzy control rules to handle nonlinear disturbances, triggers fuzzy controller adjustments, and ensures smooth control actions.

[0072] The specific process of dynamically adjusting the ventilation volume Q and the stirring rate N is as follows:

[0073] Set the dissolved oxygen target value, Set the dissolved oxygen target value 20%~30% air saturation is used as the control benchmark;

[0074] The model predictive control (MPC) optimization problem is solved by taking the predicted oxygen consumption rate OUR in the next 30 minutes as input and solving the optimization problem by the model predictive control MPC algorithm, that is, determining the regulation strategy of ventilation volume Q and stirring rate N;

[0075] Based on the oxygen consumption rate OUR predicted value output by the bacterial oxygen consumption rate prediction model, the optimal control amount is solved through rolling optimization to balance the dissolved oxygen deviation and the change range of the control amount;

[0076] Model predictive control (MPC) is responsible for optimal control under normal operating conditions, while fuzzy control handles nonlinear deviations. The combination of the two enables dynamic adaptive control.

[0077] The specific process of introducing fuzzy control rules to handle nonlinear disturbances, triggering fuzzy controller adjustment, and ensuring the smoothness of control actions is as follows:

[0078] When the deviation between the measured DO and the target value exceeds ±10% of the target value, the fuzzy controller is triggered to dynamically adjust the ventilation volume Q and stirring rate N according to the change trend of the collected real-time dissolved oxygen concentration DO data;

[0079] The process of dynamically adjusting the ventilation volume Q and stirring rate N is:

[0080] The control cycle is set to 30 seconds, that is, the adjustment is performed every 30 seconds to ensure smooth control action and avoid drastic fluctuations in fermentation parameters;

[0081] Through the combined MPC and fuzzy control strategy described above, the system can respond in real time to the dynamic oxygen demand during Monascus fermentation, ensuring dissolved oxygen stability while reducing fluctuations in the controlled quantity and improving fermentation efficiency. It also uses nonlinear fuzzy rules to handle sudden disturbances and dynamically adjusts ventilation and agitation parameters based on the deviation between the measured DO and the target value, enhancing system robustness.

[0082] Actuator and Feedback Correction Module: This module sets up actuator control and feedback correction mechanisms. The ventilation system uses a mass flow controller to regulate the flow of the air and oxygen mixture. The stirring system uses a variable frequency motor to drive a three-layer pitched-blade turbine impeller, dynamically adjusting the speed according to the fermentation stage. After each control action is completed, the feedback correction mechanism is activated to correct the prediction model parameters.

[0083] The specific process of the ventilation system using a mass flow controller to regulate the flow of air and oxygen mixed gas is as follows:

[0084] A mass flow controller (MFC) with an accuracy of ±1% FS was used to adjust the flow of the mixed gas of air and oxygen. A microporous distributor was deployed at the gas inlet. The pore diameter of the microporous distributor was 50 , ensure that the average bubble diameter is less than 2mm, and improve the dissolved oxygen transfer efficiency;

[0085] The stirring system uses a variable frequency motor to drive a three-layer pitched blade turbine stirring paddle. The specific process of dynamically adjusting the speed according to the fermentation stage is as follows:

[0086] A variable frequency motor is used to drive a three-layer pitched blade turbine agitator. The speed is dynamically adjusted according to the fermentation stage. If the fermentation stage is in the exponential growth phase, the speed is set to 300-500 rpm; if the fermentation stage is in the stable phase, the speed is set to 200-300 rpm.

[0087] The startup feedback correction mechanism is specifically as follows:

[0088] After each control action is executed, a 3-minute timer monitoring is started. If the deviation between the measured DO and the target value is not reduced to within ±5% of the target value within 3 minutes, the model error compensation algorithm is triggered to correct the parameters of the bacterial oxygen consumption rate prediction model based on the DO deviation data and the actual oxygen consumption rate;

[0089] Specifically, the calculation process of the deviation between the measured DO and the target value is as follows: ,in, is the measured DO value, is the preset target dissolved oxygen concentration value;

[0090] When the deviation ratio is greater than 5% and lasts for 3 minutes, the error compensation and model error compensation mechanism are triggered;

[0091] The specific process of correcting the prediction model parameters is as follows:

[0092] By adjusting the weight layer parameters of the long short-term memory network (LSTM) and the attention mechanism (Attention) network, the deviation between the subsequent predicted values ​​and the measured values ​​was reduced, achieving the purpose of correcting the parameters of the bacterial oxygen consumption rate prediction model.

[0093] This step ensures the effective execution of control instructions and the continuous optimization of the bacterial oxygen consumption rate prediction model through hardware precision control and dynamic feedback mechanism, thereby achieving closed-loop control of the fermentation process.

[0094] Full-process closed-loop optimization module: Establishes a full-process optimization mechanism and ensures the stability of the fermentation process by setting up a dual early warning mechanism. After the fermentation batch is completed, the correlation between fermentation indicators and model control parameters is analyzed, and the model prediction control parameters are updated based on the Bayesian optimization algorithm;

[0095] The specific process of setting up the dual early warning mechanism is as follows:

[0096] Real-time display of dissolved oxygen concentration DO data, oxygen consumption rate OUR data, ventilation volume Q, and stirring rate N parameter curves through the human-machine interface HMI;

[0097] A dual early warning mechanism is set up: Early warning mechanism 1: dissolved oxygen concentration DO data is less than 15% of air saturation; early warning mechanism 2: oxygen consumption rate OUR data is greater than 20% of the predicted value; if the dissolved oxygen concentration DO or oxygen consumption rate OUR data triggers any of the above early warning conditions, an early warning signal will be generated immediately;

[0098] After the fermentation batch is completed, the specific process of analyzing the correlation between the fermentation index and the model control parameters is as follows:

[0099] After the fermentation batch is completed, a fermentation data report is automatically generated and analyzed:

[0100] The biomass was determined by dry weight method, the Monacolin K content was determined by HPLC, and the GABA concentration was determined by ELISA. The correlation between fermentation indices and model control parameters, namely, the aeration volume Q and the stirring rate N, was calculated.

[0101] Analyze the correlation between fermentation product indicators and model control parameters based on statistical methods to provide data support for optimization;

[0102] The Bayesian optimization algorithm searches for the optimal model control parameters, updates the bacterial oxygen consumption rate prediction model, fits the objective function (i.e., the relationship between the fermentation product yield and the model control parameters) through a probabilistic model, and uses the acquisition function to efficiently search for the optimal parameter combination.

[0103] After every 10 batches of fermentation, the comprehensive index of Monacolin K yield and energy consumption is used as the objective function, and the model prediction control parameters are updated based on the Bayesian optimization algorithm, forming a closed-loop iterative process of data collection-model training-control optimization.

[0104] The technical solution of the embodiment of the present invention is to achieve intelligent control of the fermentation process through five modules. The system deploys multiple types of sensors, such as fluorescence quenching fiber optic dissolved oxygen sensors, which are distributed in a circular array at different heights of the fermentation tank to collect real-time data such as DO, temperature, pH, and bacterial concentration. The data is then subjected to sliding average filtering and normalization preprocessing by the edge computing server. Using at least 50 batches of historical fermentation data, an oxygen consumption rate prediction model combining LSTM and attention mechanism is constructed. The model is trained with a target of RMSE < 5%, and fine-tuned every 2 hours using nearly 2 hours of real-time data. The dissolved oxygen control integrates model predictive control (MPC) and fuzzy control, with an air saturation of 20% to 30%. The ventilation volume Q and stirring rate N are optimized based on the predicted oxygen consumption value for the next 30 minutes. The fuzzy controller is triggered when the DO deviation exceeds the target value by ±10%. The control cycle is 30 seconds to ensure smooth operation. The execution layer uses a mass flow controller with an accuracy of ±1%FS to adjust the air flow rate, combined with a microporous distributor with a pore size of 50μm. A variable frequency motor drives a three-layer pitched-blade turbine agitator, with a speed of 300-500rpm during the exponential growth phase and 200-300rpm during the stable phase. If the DO deviation does not shrink to ±5% within 3 minutes after each control, the model parameters are corrected by adjusting the LSTM and attention network weights. A double warning is set for the entire process. After the batch is completed, the correlation between indicators such as biomass and Monacolin K content and the control parameters is analyzed. With the comprehensive yield and energy consumption as the goal, the MPC parameters are updated through Bayesian optimization every 10 batches, forming a data collection-model training-dynamic control-feedback optimization closed loop, thereby achieving increased production of immune active ingredients in Monascus and optimized energy consumption.

[0105] Example 2

[0106] like Figure 2 As shown in Example 1, the present invention provides a fermentation optimization method for enhancing the immune efficacy of Monascus based on data analysis, comprising:

[0107] S1: Deploy multiple sensors to collect multiple real-time data, and use edge computing servers to pre-process the collected real-time data;

[0108] S2: Based on historical fermentation data, a prediction model for bacterial oxygen consumption rate is established using a machine learning algorithm.

[0109] S3: Establish a dynamic control strategy for dissolved oxygen, use model predictive control to achieve dynamic adjustment of ventilation volume Q and stirring rate N, and introduce fuzzy control rules to handle nonlinear disturbances, trigger fuzzy controller adjustment, and ensure smooth control action;

[0110] S4: Set up actuator control and feedback correction mechanisms. The ventilation system uses a mass flow controller to regulate the flow of the air and oxygen mixture. The stirring system uses a variable frequency motor to drive a three-layer pitched-blade turbine stirrer. The speed is dynamically adjusted according to the fermentation stage. After each control action is completed, the feedback correction mechanism is activated to correct the prediction model parameters.

[0111] S5: Establish a full-process optimization mechanism to ensure the stability of the fermentation process by setting up a dual early warning mechanism. After the fermentation batch is completed, analyze the correlation between the fermentation indicators and the model control parameters, and update the model prediction control parameters based on the Bayesian optimization algorithm.

[0112] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A fermentation optimization system for Monascus to enhance immune efficacy based on data analysis, characterized by: include: Multi-source data edge pre-processing module: deploys multiple sensors to collect a variety of real-time data, and uses edge computing servers to pre-process the collected real-time data; Bacteria oxygen consumption rate prediction model training module: Based on historical fermentation data, a bacteria oxygen consumption rate prediction model is established through machine learning algorithms; Dissolved oxygen dynamic control strategy module: Establishes a dissolved oxygen dynamic control strategy and uses model predictive control to dynamically adjust the ventilation volume Q and stirring rate N. It also introduces fuzzy control rules to handle nonlinear disturbances, triggers fuzzy controller adjustments, and ensures smooth control actions. Actuator and Feedback Correction Module: This module sets up actuator control and feedback correction mechanisms. The ventilation system uses a mass flow controller to regulate the flow of the air and oxygen mixture. The stirring system uses a variable frequency motor to drive a three-layer pitched-blade turbine impeller, dynamically adjusting the speed according to the fermentation stage. After each control action is completed, the feedback correction mechanism is activated to correct the prediction model parameters. Full-process closed-loop optimization module: Establish a full-process optimization mechanism and ensure the stability of the fermentation process by setting up a dual early warning mechanism. After the fermentation batch is completed, analyze the correlation between fermentation indicators and model control parameters, and update the model prediction control parameters based on the Bayesian optimization algorithm.

2. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 1, characterized in that: The specific process of deploying multiple sensors is as follows: Various sensors are deployed in a circular array at different heights in the fermentation tank, 5 cm, 15 cm, and 25 cm from the bottom of the tank. The sensors include: fluorescence quenching fiber optic dissolved oxygen sensor, thermocouple temperature sensor, glass electrode pH sensor, and optical density sensor.

3. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 1, characterized in that: The multiple real-time data include: real-time dissolved oxygen concentration DO data collected by a fluorescence quenching fiber optic dissolved oxygen sensor, real-time temperature T data collected by a thermocouple temperature sensor, real-time pH value data collected by a glass electrode pH sensor, and real-time bacterial concentration X data collected by an optical density sensor.

4. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 1, characterized in that: The specific process of establishing the bacterial oxygen consumption rate prediction model through the machine learning algorithm is as follows: Collect data from at least 50 batches of Monascus fermentation process, with a fermentation period of ≥48 hours per batch and a sampling frequency of 10 minutes per batch. Ensure that each batch contains ≥1000 data points, including dissolved oxygen concentration (DO), temperature (T), pH value, bacterial concentration (X), and oxygen consumption rate (OUR). Construct feature engineering to extract DO, T, pH, X, and historical oxygen consumption rate OUR data at the current time and in the previous 30 minutes as input features. The output structure of the bacterial oxygen consumption rate prediction model is output: the predicted value of oxygen consumption rate OUR in the next 30 minutes, and construct a time series dataset; The prediction model is constructed by combining the long short-term memory network (LSTM) with the attention mechanism (Attention). The root mean square error (RMSE) is used as the loss function, and the model is trained using the Adam optimizer until the RMSE is less than 5%.

5. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 1, characterized in that: The specific process of dynamically adjusting the ventilation volume Q and the stirring rate N is as follows: Set the dissolved oxygen target value, Set the dissolved oxygen target value 20%~30% air saturation is used as the control benchmark; The model predictive control (MPC) optimization problem is solved by using the predicted oxygen consumption rate (OUR) for the next 30 minutes as input and the model predictive control (MPC) algorithm to solve the optimization problem, that is, to determine the regulation strategy of the ventilation volume (Q) and the stirring rate (N). Based on the predicted value of oxygen consumption rate OUR output by the bacterial oxygen consumption rate prediction model, the optimal control quantity is solved through rolling optimization to balance the dissolved oxygen deviation and the change range of the control quantity.

6. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 5, characterized in that: The specific process of dynamically adjusting the ventilation volume Q and the stirring rate N is as follows: The specific process of introducing fuzzy control rules to handle nonlinear disturbances, triggering fuzzy controller adjustments, and ensuring the smoothness of control actions is as follows: When the deviation between the measured DO and the target value exceeds ±10% of the target value, the fuzzy controller is triggered to dynamically adjust the ventilation volume Q and stirring rate N according to the change trend of the collected real-time dissolved oxygen concentration DO data; The control cycle is set to 30 seconds, that is, the adjustment is performed every 30 seconds to ensure smooth control action and avoid drastic fluctuations in fermentation parameters.

7. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 1, characterized in that: The specific process of the ventilation system using a mass flow controller to regulate the flow of air and oxygen mixed gas is as follows: A mass flow controller (MFC) with an accuracy of ±1% FS was used to adjust the flow of the mixed gas of air and oxygen. A microporous distributor was deployed at the gas inlet. The aperture diameter of the microporous distributor was 50 , ensure that the average diameter of bubbles is less than 2mm, and improve the dissolved oxygen transfer efficiency.

8. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 1, characterized in that: The stirring system uses a variable frequency motor to drive a three-layer pitched blade turbine stirring paddle. The specific process of dynamically adjusting the speed according to the fermentation stage is as follows: A variable frequency motor is used to drive a three-layer inclined blade turbine agitator. The speed is dynamically adjusted according to the fermentation stage. If the fermentation stage is in the exponential growth phase, the speed is set to 300-500 rpm; if the fermentation stage is in the stable phase, the speed is set to 200-300 rpm.

9. The Monascus fermentation optimization system for enhancing immune efficacy based on data analysis according to claim 1, characterized in that: The specific process of correcting the prediction model parameters is as follows: By adjusting the weight layer parameters of the long short-term memory network (LSTM) and the attention mechanism (Attention) network, the deviation between the subsequent predicted values ​​and the measured values ​​is reduced, thereby achieving the purpose of correcting the parameters of the bacterial oxygen consumption rate prediction model.

10. A fermentation optimization method for enhancing the immune effect of Monascus based on data analysis, characterized by: include: S1: Deploy multiple sensors to collect multiple real-time data, and use edge computing servers to pre-process the collected real-time data; S2: Based on historical fermentation data, a prediction model for bacterial oxygen consumption rate is established using a machine learning algorithm. S3: Establish a dynamic control strategy for dissolved oxygen, use model predictive control to achieve dynamic adjustment of ventilation volume Q and stirring rate N, and introduce fuzzy control rules to handle nonlinear disturbances, trigger fuzzy controller adjustment, and ensure smooth control action; S4: Set up actuator control and feedback correction mechanisms. The ventilation system uses a mass flow controller to regulate the flow of the air and oxygen mixture. The stirring system uses a variable frequency motor to drive a three-layer pitched-blade turbine stirrer. The speed is dynamically adjusted according to the fermentation stage. After each control action is completed, the feedback correction mechanism is activated to correct the prediction model parameters. S5: Establish a full-process optimization mechanism to ensure the stability of the fermentation process by setting up a dual early warning mechanism. After the fermentation batch is completed, analyze the correlation between the fermentation indicators and the model control parameters, and update the model prediction control parameters based on the Bayesian optimization algorithm.

Images