A natto fermentation edge control mechanism-data hybrid regulation method and system

Abstract

This invention relates to the field of food fermentation and edge computing technology, specifically to a method and system for edge control mechanism-data hybrid regulation of natto fermentation. The key technical points include: constructing a multi-stage natto fermentation mechanism model library and lightweight reconstructing it into ONNX format, then deploying it to a field edge controller; collecting real-time fermentation data locally through the edge controller, completing preprocessing and dynamic correction of model parameters, and using a built-in lightweight optimization algorithm to locally solve for multi-objective optimal process parameters; finally, issuing control commands through industrial communication to form a fully closed-loop control at the edge, requiring no cloud intervention throughout the process. This invention achieves rapid response and precise control of the fermentation process, significantly shortening the latency of model correction and control command issuance, meeting the real-time adjustment requirements of parameters such as temperature, humidity, and oxygen supply during natto fermentation, adapting to the industrial environment of food fermentation, and capable of independent and stable operation, suitable for large-scale industrial-grade natto fermentation production.

Description

Technical Field

[0001] This invention relates to the field of food fermentation and edge computing technology, specifically to a natto fermentation edge control mechanism - data mixing and regulation method and system. Background Technology

[0002] Natto is rich in nattokinase, γ-polyglutamic acid and probiotics, and has extremely high nutritional and medicinal value. As the natto industry transforms towards large-scale and standardized production, the real-time control precision and batch consistency during the fermentation process have become core requirements.

[0003] Currently, the main technical challenges in regulating natto fermentation are as follows: 1. Existing technologies mostly rely on human experience or simple automated equipment, and only adjust single parameters such as temperature and humidity. They lack systematic modeling of multi-stage coupled processes such as microbial growth and product synthesis. Since the environmental requirements of different fermentation stages are different, traditional methods cannot respond to changes in state in a timely manner, resulting in a coefficient of variation of key indicators of multiple batches of production as high as 8%-15%, and poor product consistency.

[0004] 2. Existing intelligent control systems are mostly based on cloud architecture. Although cloud computing power is strong, it is highly dependent on communication networks, which poses a risk of control failure due to network interruption. At the same time, the additional latency introduced by data uploading, calculation and command feedback cannot meet the control needs of rapid response to parameters such as oxygen content and temperature at the fermentation site.

[0005] 3. Existing fermentation mechanism models are usually high-complexity, full-precision nonlinear models with large computational loads, making them difficult to deploy directly on resource-constrained field edge controllers. Currently, there is a lack of technical solutions in industrial settings that deeply integrate lightweight mechanism models with edge-end online calibration, optimization, and closed-loop execution.

[0006] 4. The fermentation environment of natto is characterized by high temperature, high humidity and strong electromagnetic interference, which requires the control system to have extremely high local operation stability and data security without relying on external networks.

[0007] In summary, how to construct an edge-end intelligent control system that is adapted to the industrial environment, can achieve low latency, high security and multi-batch stability is a key problem that urgently needs to be solved in the current industrial production of natto fermentation. To solve the above problems, this application proposes a natto fermentation edge control mechanism-data hybrid control method and system. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a natto fermentation edge control mechanism-data mixing regulation method and system, solving the problems mentioned in the background technology.

[0009] The above-mentioned technical objective of the present invention is achieved through the following technical solution: On the one hand, this invention proposes a natto fermentation edge control mechanism-data hybrid regulation method, including the following steps: S1. Construct a multi-stage mechanism model library for natto fermentation. The model library includes a Logistic model for describing the growth kinetics of Bacillus natto, a Michaelis-Menten equation model for describing the synthesis kinetics of nattokinase, and a non-competitive inhibition model for describing the synthesis kinetics of γ-polyglutamic acid. The model library is then refactored in a lightweight manner to adapt to the computing resources of the edge controller. S2. Compile the lightweight mechanism model library into the ONNX standard format and deploy it to the edge controller at the natto fermentation site to complete model initialization and load the initial industrial production parameters of each model. S3. The edge controller connects to the field sensing module through its peripheral interface to collect real-time data on temperature, humidity, oxygen content, viable Bacillus natto count, nattokinase activity, and γ-polyglutamic acid production throughout the fermentation cycle. The edge controller performs local preprocessing to remove outliers and standardize the data. S4. Based on the real-time data collected locally by the edge controller, the key parameters of each model in the lightweight mechanism model library deployed therein are dynamically corrected locally, and the growth rate constant of the Logistic model, the enzyme activity synthesis rate of the Michaelis equation model, and the product synthesis rate of the non-competitive inhibition model are updated online, so that the fitting error between the model prediction value and the measured value is controlled within the preset threshold range. S5. A multivariate constraint optimization algorithm is built into the edge controller. The number of viable bacteria, natto kinase activity and γ-polyglutamic acid yield are set as multi-objective optimization constraints. The algorithm is solved locally based on the corrected lightweight mechanism model and the optimal process parameters for each stage of natto fermentation are output. S6. The edge controller sends control commands to the field execution equipment through the industrial communication protocol, automatically adjusts the temperature control, humidity control and ventilation system of the fermenter, and receives status feedback and key indicator data after control from the execution equipment, forming an edge-end intelligent control process of field acquisition - edge modeling - local optimization - real-time control - closed-loop feedback.

[0010] Preferably, the edge controller is an industrial-grade edge computing controller equipped with an embedded processor, which has local data acquisition, model inference, algorithm operation and industrial equipment communication functions, and supports the deployment of ONNX and / or TensorFlow Lite lightweight model formats, adapting to the high temperature and high humidity industrial environment of food fermentation workshops.

[0011] Preferably, the lightweight model reconstruction in step S1 includes: simplifying higher-order nonlinear terms of the model, pruning redundant computation nodes of the model, and quantifying the accuracy of model parameters, so as to reduce the amount of computation per inference of the model by more than 60%, thereby meeting the local real-time computing requirements of the edge controller.

[0012] Preferably, the model local dynamic correction method in step S4 is as follows: a lightweight least squares fitting algorithm is used in the edge controller to locally fit the measured values ​​of key indicators collected in real time with the model prediction values, calculate the parameter correction values ​​and complete the online update, and the correction process is completed independently in the edge controller with a computation delay of no more than 500ms.

[0013] Preferably, the industrial communication protocol in step S6 is one or more of Modbus RTU, Modbus TCP, Profinet, or EtherNet / IP, and the edge controller communicates directly with the PLC controller or actuator of the fermenter temperature control equipment, humidity control equipment, and ventilation equipment through the industrial communication protocol, with the delay in issuing control commands not exceeding 300ms.

[0014] Preferably, the natto fermentation is divided into three stages: logarithmic growth phase, stationary phase, and decline phase, and the optimal process parameters output by the edge controller for each stage are adjustable within the following ranges: Logarithmic growth phase: heating rate 1-2℃ / h, oxygen content not less than 18%, temperature maintained at 40-42℃; Stable period: humidity 90%–95%, ventilation every 2–4 hours for 10–15 minutes, oxygen content 18%–20%; Decline phase: Cooling rate of 2-3℃ / h, reducing the temperature to 20-25℃.

[0015] On the other hand, this invention proposes a natto fermentation edge control mechanism-data hybrid control system, including an edge control master unit, a field sensor acquisition unit, and an industrial execution unit. The edge control master unit, the field sensor acquisition unit, and the industrial execution unit are all deployed at the natto fermentation production site and are connected to each other via industrial communication. The edge control main unit is an industrial-grade edge controller, which has a built-in lightweight mechanism model library, a local data preprocessing module, a model dynamic correction module, a multi-objective lightweight optimization algorithm module and an industrial communication module. It supports ONNX model deployment and local independent operation, and has data storage, status display and parameter configuration functions. The field sensing and acquisition unit includes a temperature sensor, a humidity sensor, a dissolved oxygen sensor, a microbial viable count detection sensor, and an enzyme activity detection sensor. The field sensing and acquisition unit is directly connected to the peripheral interface of the edge control main unit and is used to collect fermentation process data. The industrial execution unit includes fermenter temperature control equipment, humidity control equipment, ventilation equipment, and supporting PLC controller and / or actuator. The industrial execution unit communicates with the edge control master unit through Modbus RTU, Modbus TCP, Profinet or EtherNet / IP protocol to receive control commands and execute actions, while feeding back the equipment operating status and process parameter data after control.

[0016] Preferably, the edge control main unit also has a cloud-based lightweight interaction module. The cloud-based lightweight interaction module can upload the fermentation process data, model running status, control records and production data at the edge to the cloud server via 4G, 5G or industrial Ethernet for remote monitoring and historical traceability. The cloud server is only used for data storage and display and does not participate in model reasoning and control decisions.

[0017] Preferably, the field sensing acquisition unit and the edge control main unit are connected by shielded twisted pair cable or industrial Ethernet with an anti-interference capability of not less than 2000V, and the industrial execution unit and the edge control main unit are connected by industrial bus and / or wireless industrial communication with a communication distance of not less than 50m.

[0018] In summary, the present invention has the following main beneficial effects: 1. This invention integrates the construction of the mechanism model, lightweight deployment, on-site data acquisition, dynamic parameter correction, optimization solution, and execution control of the natto fermentation process into a unified process completed locally on the edge controller. This forms an intelligent control process of on-site data acquisition, edge modeling, local optimization, real-time regulation, and closed-loop feedback. This avoids the high dependence on network communication in traditional cloud control modes, reduces the control risks caused by network interruptions, transmission delays, or remote computing failures, and improves the independence and stability of system operation. At the same time, by lightweight reconstruction of the Logistic model, the Michaelis-Menten equation model, and the non-competitive inhibition model and conversion to the ONNX format, the mechanism model can be adapted to the limited computing resources of the industrial-grade edge controller. Combined with local dynamic correction and multi-objective constraint optimization, it realizes rapid response and precise control of the fermentation process, significantly shortens the latency of model correction and control command issuance, and meets the real-time adjustment requirements of parameters such as temperature, humidity, and oxygen supply during natto fermentation.

[0019] 2. This invention, through continuous online correction and phased optimization of key parameters during fermentation, can effectively reduce process fluctuations between different batches, improve the stability and consistency of Bacillus natto viable count, nattokinase activity, and γ-polyglutamic acid yield, and facilitate large-scale, standardized production. Since core data is collected, processed, and controlled locally at the edge, remote transmission links are reduced, data security is improved, and the risk of information leakage and communication packet loss is reduced. In addition, the industrial-grade edge controller and industrial communication method used are suitable for the high temperature, high humidity, and strong electromagnetic interference environment of food fermentation workshops, and have strong engineering applicability, scalability, and promotion value. It can also reduce manual intervention, improve production efficiency, and reduce labor costs. Attached Figure Description

[0020] Figure 1 This is a diagram of the overall architecture of the control system of the present invention; Figure 2 This is a schematic diagram of the closed-loop control process at the edge of natto fermentation according to the present invention; Figure 3 This is a schematic diagram of the construction and deployment process of the lightweight mechanism model library of the present invention. Detailed Implementation

[0021] 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 some embodiments of the present invention, but not all embodiments. Based on the described 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.

[0022] The following embodiments are used to illustrate the present invention, but should not be used to limit the scope of protection of the present invention. The conditions in the embodiments can be further adjusted according to specific conditions, and simple improvements to the method of the present invention under the premise of the concept of the present invention are all within the scope of protection claimed by the present invention.

[0023] Example 1 This embodiment takes an industrial-grade natto fermentation production line as an example. It uses an edge controller to complete mechanism modeling, real-time data acquisition, dynamic correction, optimization solution and closed-loop control locally, without relying on the cloud to participate in control decision-making.

[0024] 1. System Hardware Configuration This embodiment uses an industrial-grade edge controller as the main edge control unit, preferably the Advantech UNO-2484G industrial-grade edge controller, which is equipped with an Intel Core i5 embedded processor, supports local model deployment, industrial communication and edge computing functions, and its protection level meets the requirements of the food fermentation workshop application environment.

[0025] The fermentation equipment uses a 500L industrial-grade natto fermentation tank, equipped with a temperature control system, humidity control system, ventilation system, and PLC controller. The on-site sensor acquisition unit includes: Temperature sensor with a measurement accuracy of ±0.1℃; Humidity sensor with a measurement accuracy of ±1%RH; Dissolved oxygen sensor with a measurement accuracy of ±0.5%; Microbial viable cell count detection module; Enzyme activity detection module.

[0026] The edge control master unit is directly connected to the field sensing and acquisition unit through a peripheral interface. The edge control master unit is connected to the industrial execution unit using the Modbus TCP industrial communication protocol. The external network is disconnected throughout the system operation, and only the local closed-loop control link is retained to ensure that the core inference and control do not rely on the cloud.

[0027] 2. Construction of Mechanism Model Library This embodiment first constructs a multi-stage mechanism model library for natto fermentation, which includes at least the following models: Bacillus natto growth kinetics model The Logistic model is used to describe the bacterial growth process and characterize the dynamic changes in bacterial count over time.

[0028] Nattokinase Synthesis Kinetic Model The Michaelis-Menten equation model was used to describe the synthesis of nattokinase under the action of substrates.

[0029] γ-polyglutamic acid synthesis kinetic model A non-competitive inhibition model was used to describe the synthesis process of γ-polyglutamic acid and its variation under the influence of environmental factors.

[0030] To adapt to the limited computing resources of the edge controller, the above model is refactored in a lightweight manner, specifically including: The high-order nonlinear coupling terms are simplified; redundant computing nodes are pruned; and model parameters are fixed-point or low-precision quantized to reduce the computational load of a single inference, enabling the model to run in real time within the edge controller.

[0031] Subsequently, the lightweight model library is compiled into the ONNX standard format and deployed to the local storage and computing area of ​​the edge controller. At the same time, the corresponding initial parameters for industrial production are loaded, including the initial cell concentration, initial substrate concentration, initial temperature and humidity conditions, and ventilation settings.

[0032] 3. On-site data acquisition and preprocessing Throughout the fermentation process, the edge controller collects on-site process parameters and key quality data at a frequency of 20 minutes per cycle, including: Fermentation ambient temperature; Fermentation environment humidity; Oxygen content or dissolved oxygen level in the fermentation environment; The number of viable Bacillus natto bacteria; Nattokinase activity; γ-polyglutamic acid production.

[0033] The collected data is first preprocessed locally on the edge controller, including outlier removal, noise smoothing, and data standardization. Outlier removal can be accomplished by using threshold judgment combined with a sliding window method. Data standardization unifies data of different dimensions to the same scale to facilitate subsequent model correction and optimization.

[0034] 4. Local dynamic calibration of the model The edge controller performs online dynamic correction of key parameters in the model library based on local real-time data. Specifically, using the measured data and model prediction data within the current sampling period as input, it calculates the parameter correction values ​​using a lightweight least squares fitting algorithm and updates the following parameters online: The growth rate constant in the Logistic model; Enzyme activity synthesis rate in the Michaelis-Menten equation model; Product synthesis rate in the non-competitive inhibition model.

[0035] In this embodiment, the calibration process is completed independently and locally on the edge controller without interacting with the cloud. The latency of a single calibration operation is controlled within 500ms. In this way, the fitting error between the model prediction value and the measured value is controlled within a preset threshold range, ensuring the accuracy of subsequent optimization solutions.

[0036] 5. Multi-objective constrained optimization solution A multivariate constrained optimization algorithm is built into the edge controller to solve the corrected lightweight mechanism model locally. The optimization objective includes at least the following: The number of viable Bacillus natto bacteria is not less than 10. 9 CFU / g; Nattokinase activity is not less than 1800 FU / g; The yield of γ-polyglutamic acid is not less than 2.5g / 100g.

[0037] Optimization variables include, but are not limited to: Temperature setpoint; Heating rate; Humidity setting; Ventilation cycle; Ventilation duration; Oxygen content control value.

[0038] The optimization algorithm is calculated locally on the edge controller and outputs the optimal process parameters for each stage. This solution process does not rely on cloud computing power and can quickly provide control strategies based on real-time data acquisition, meeting the requirements of rapid response in the fermentation process.

[0039] 6. Industrial Execution and Closed-Loop Feedback The edge controller sends the optimized process parameters to the PLC controller and the execution equipment via the Modbus TCP industrial communication protocol. The execution equipment automatically adjusts the temperature, humidity and ventilation systems of the fermenter. After execution, the equipment status, environmental parameters and key process parameters are fed back to the edge controller. The edge controller continues to perform the next round of correction and optimization based on the feedback data, forming a complete control link of "on-site acquisition - edge modeling - local optimization - real-time control - closed-loop feedback".

[0040] In this embodiment, the delay in issuing control commands is controlled within 300ms to meet the real-time adjustment requirements of temperature, humidity and oxygen supply parameters during natto fermentation.

[0041] Example 2 Based on the system in Example 1, this embodiment controls the natto fermentation process in three stages: the logarithmic growth phase, the stationary phase, and the decline phase.

[0042] 1. Control of the logarithmic growth phase (0-10h) During the logarithmic growth phase, the edge controller loads the Logistic growth model, with the initial growth rate constant set to 0.34h⁻¹, and the initial control parameters set as follows: Temperature: 40℃; Oxygen content: 18%; Humidity: 90%~92%.

[0043] When fermentation reached 10 hours, the edge controller collected the following locally measured data: Temperature: 41.0℃; Humidity: 92.0%; Oxygen content: 19.0%; viable bacteria count: 9.998 × 10⁻⁶ 8 CFU / g.

[0044] The edge controller uses the least squares method to correct the model parameters, updating the growth rate constant to 0.36h⁻¹, with a fitting error of 2.1% and a correction time of 320ms.

[0045] After local optimization, the optimal parameters are output as follows: Heating rate: 1.5℃ / h; Temperature control range: 41~42℃; Oxygen content: 18%–20%; Ventilate for 5 minutes every 1 hour.

[0046] Subsequently, the edge controller sends the above parameters to the execution unit and completes real-time adjustments.

[0047] 2. Stabilization period control (10-20 hours) During the stabilization period, the edge controller simultaneously loads the Mie equation model and the non-competitive suppression model, and performs joint correction using real-time data. The initial control strategy is set as follows: Humidity: 90%–95%; Oxygen content: 18%–20%; Ventilate for 10-15 minutes every 2-4 hours.

[0048] The measured results collected after 20 hours of fermentation are as follows: Nattokinase activity: 1825 FU / g; γ-polyglutamic acid yield: 2.51g / 100g; viable bacteria count: 1.02 × 10⁻⁶ 9 CFU / g.

[0049] The edge controller completed the online update of model parameters. The correction of enzyme activity synthesis rate took 380ms and the correction of product synthesis rate took 410ms, with corresponding fitting errors of 3.5% and 4.2%, respectively.

[0050] After local optimization, the optimal parameters are output as follows: Humidity: 93%~95%; Oxygen content: 19%–20%; Ventilate for 12 minutes every 3 hours.

[0051] 3. Decline period control (20-24h) During the decay phase, the edge controller primarily slows down cell death and stabilizes the final quality through cooling control. The control parameters are set as follows: Cooling rate: 2-3℃ / h; Final temperature: 20-25℃.

[0052] The edge controller automatically performs cooling and ventilation adjustments based on real-time feedback data during the later stages of fermentation, ensuring a smooth end to the fermentation process and yielding a natto product with stable final quality.

[0053] Example 3 This embodiment is used to verify the continuous stability of the method and system of the present invention. Using the edge controller, sensing and acquisition unit and industrial execution unit described in Embodiment 1, 10 batches of 500L industrial-grade natto fermentation tank were produced continuously. The fermentation cycle of each batch was 24 hours. The entire production process was disconnected from the external network, and all modeling, correction, optimization and control were completed locally at the edge.

[0054] 1. Verification Method After each batch of fermentation, the viable count of Bacillus natto, the activity of nattokinase, and the yield of γ-polyglutamic acid were tested, and the overall quality score and the coefficient of variation between batches were calculated.

[0055] 2. Measured Results Note: All testing methods comply with national standards for the food fermentation industry. The coefficient of variation is calculated using the formula CV = (standard deviation / average) × 100%. All batches meet the preset constraints of this invention (viable count ≥ 9.8 × 10⁻⁶). 8 CFU / g, nattokinase activity ≥1800FU / g, γ-polyglutamic acid production ≥2.5g / 100g).

[0056] 3. Results Analysis From the above data, we can see that: The number of viable Bacillus natto, nattokinase activity, and γ-polyglutamic acid yield of 10 consecutive batches of products consistently met the preset requirements, and the coefficient of variation between batches was low, indicating that the present invention can achieve high consistency and high stability in the natto fermentation process. No control failures due to network interruption occurred during local control at the edge. The average delay of issuing control commands was about 210ms, and the average time for model correction was about 350ms, which met the real-time control requirements of industrial sites. Compared with traditional experience-based control, production efficiency is increased by 22%, labor costs are reduced by 30%, and the accuracy of process parameter control reaches ±0.1℃ for temperature, ±1%RH for humidity, and ±0.5% for oxygen content. The parameter deviation of the same batch is ≤±0.5℃ and ±2%RH, which verifies the accuracy of control and industrial economy.

[0057] Example 4 This embodiment further illustrates the scalability of the system architecture. The edge control main unit can optionally be configured with a cloud-based lightweight interaction module, which uploads fermentation process data, model operating status, and control records to the cloud server via 4G, 5G, or industrial Ethernet. The cloud server is only used for data storage, remote monitoring, and historical tracing, and does not participate in model inference or control decisions.

[0058] In actual operation, the core closed-loop control of the system is still completed locally by the edge controller. Even if the cloud communication is interrupted, the system can still maintain normal operation, thereby ensuring the continuity, stability and safety of the natto fermentation production process.

[0059] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that, unless otherwise defined, the technical or scientific terms used in this invention should be understood in the ordinary sense by those skilled in the art to which this invention pertains, and the terms "comprising" or "including" or similar terms used in this invention mean that the element or object preceding the word covers the element or object listed after the word and its equivalents.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for edge control mechanism and data-based regulation in natto fermentation, characterized in that, Includes the following steps: S1. Construct a multi-stage mechanism model library for natto fermentation. The model library includes a Logistic model for describing the growth kinetics of Bacillus natto, a Michaelis-Menten equation model for describing the synthesis kinetics of nattokinase, and a non-competitive inhibition model for describing the synthesis kinetics of γ-polyglutamic acid. The model library is then refactored in a lightweight manner to adapt to the computing resources of the edge controller. S2. Compile the lightweight mechanism model library into the ONNX standard format and deploy it to the edge controller at the natto fermentation site to complete model initialization and load the initial industrial production parameters of each model. S3. The edge controller connects to the field sensing module through its peripheral interface to collect real-time data on temperature, humidity, oxygen content, viable Bacillus natto count, nattokinase activity, and γ-polyglutamic acid production throughout the fermentation cycle. The edge controller performs local preprocessing to remove outliers and standardize the data. S4. Based on the real-time data collected locally by the edge controller, the key parameters of each model in the lightweight mechanism model library deployed therein are dynamically corrected locally, and the growth rate constant of the Logistic model, the enzyme activity synthesis rate of the Michaelis equation model, and the product synthesis rate of the non-competitive inhibition model are updated online, so that the fitting error between the model prediction value and the measured value is controlled within the preset threshold range. S5. A multivariate constraint optimization algorithm is built into the edge controller. The number of viable bacteria, natto kinase activity and γ-polyglutamic acid yield are set as multi-objective optimization constraints. The algorithm is solved locally based on the corrected lightweight mechanism model and the optimal process parameters for each stage of natto fermentation are output. S6. The edge controller sends control commands to the field execution equipment through the industrial communication protocol, automatically adjusts the temperature control, humidity control and ventilation system of the fermenter, and receives status feedback and key indicator data after control from the execution equipment, forming an edge-end intelligent control process of field acquisition - edge modeling - local optimization - real-time control - closed-loop feedback.

2. The method according to claim 1, characterized in that, The edge controller is an industrial-grade edge computing controller equipped with an embedded processor. It has local data acquisition, model inference, algorithm calculation and industrial equipment communication functions, and supports ONNX and / or TensorFlow Lite lightweight model format deployment, making it suitable for the high temperature and high humidity industrial environment of food fermentation workshops.

3. The method according to claim 1, characterized in that, The lightweight model reconstruction in step S1 includes: simplifying higher-order nonlinear terms of the model, pruning redundant computation nodes of the model, and quantifying the accuracy of model parameters, so as to reduce the amount of computation per inference of the model by more than 60%, thereby meeting the local real-time computing requirements of the edge controller.

4. The method according to claim 1, characterized in that, The model local dynamic correction method in step S4 is as follows: the least squares lightweight fitting algorithm is used in the edge controller to locally fit the measured values ​​of key indicators collected in real time with the model prediction values, calculate the parameter correction values ​​and complete the online update, and the correction process is completed independently in the edge controller with a computation delay of no more than 500ms.

5. The method according to claim 1, characterized in that, The industrial communication protocol in step S6 is one or more of Modbus RTU, Modbus TCP, Profinet, or EtherNet / IP, and the edge controller communicates directly with the PLC controller or actuator of the fermenter temperature control equipment, humidity control equipment, and ventilation equipment through the industrial communication protocol, with the delay in issuing control commands not exceeding 300ms.

6. The method according to claim 1, characterized in that, The natto fermentation process is divided into three stages: logarithmic growth phase, stationary phase, and decline phase. The optimal process parameters output by the edge controller for each stage are adjustable within the following ranges: Logarithmic growth phase: heating rate 1-2℃ / h, oxygen content not less than 18%, temperature maintained at 40-42℃; Stable period: humidity 90%–95%, ventilation every 2–4 hours for 10–15 minutes, oxygen content 18%–20%; Decline phase: Cooling rate of 2-3℃ / h, reducing the temperature to 20-25℃.

7. A natto fermentation edge control mechanism-data mixing and regulation system for implementing the method of any one of claims 1 to 6, characterized in that, The system includes an edge control main unit, a field sensor acquisition unit, and an industrial execution unit. All three units are deployed at the natto fermentation production site and are connected to each other via industrial communication. The edge control main unit is an industrial-grade edge controller, which has a built-in lightweight mechanism model library, a local data preprocessing module, a model dynamic correction module, a multi-objective lightweight optimization algorithm module and an industrial communication module. It supports ONNX model deployment and local independent operation, and has data storage, status display and parameter configuration functions. The field sensing and acquisition unit includes a temperature sensor, a humidity sensor, a dissolved oxygen sensor, a microbial viable count detection sensor, and an enzyme activity detection sensor. The field sensing and acquisition unit is directly connected to the peripheral interface of the edge control main unit and is used to collect fermentation process data. The industrial execution unit includes fermenter temperature control equipment, humidity control equipment, ventilation equipment, and supporting PLC controller and / or actuator. The industrial execution unit communicates with the edge control master unit through Modbus RTU, Modbus TCP, Profinet or EtherNet / IP protocol to receive control commands and execute actions, while feeding back the equipment operating status and process parameter data after control.

8. The system according to claim 7, characterized in that, The edge control main unit also has a cloud-based lightweight interaction module. The cloud-based lightweight interaction module can upload the fermentation process data, model running status, control records and production data at the edge to the cloud server via 4G, 5G or industrial Ethernet for remote monitoring and historical traceability. The cloud server is only used for data storage and display and does not participate in model reasoning and control decisions.

9. The system according to claim 7, characterized in that, The field sensing acquisition unit and the edge control main unit are connected by shielded twisted pair cable or industrial Ethernet with an anti-interference capability of not less than 2000V. The industrial execution unit and the edge control main unit are connected by industrial bus and / or wireless industrial communication, with a communication distance of not less than 50m.

Images